Amber14

User Manual: Amber14

Open the PDF directly: View PDF .
Page Count: 824

Download
Open PDF In Browser	View PDF

Amber 14
Reference Manual
Principal contributors to the current codes:
David A. Case (Rutgers)
Tom Darden (OpenEye)
Thomas E. Cheatham III (Utah)
Carlos Simmerling (Stony Brook)
Adrian Roitberg (Florida)
Junmei Wang (UT Southwestern Medical Center)
Robert E. Duke (NIEHS and UNC-Chapel Hill)
Ray Luo (UC Irvine)
Daniel R. Roe (Utah)
Ross C. Walker (SDSC, UCSD)
Scott LeGrand (Amazon)
Jason Swails (Rutgers)
David Cerutti (Rutgers)
Joe Kaus (UCSD)
Robin Betz (UCSD)
Romain M. Wolf (Novartis)
Kenneth M. Merz (Michigan State)
Gustavo Seabra (Recife, Brazil)
Pawel Janowski (Rutgers)

Andreas W. Götz (SDSC, UCSD)
István Kolossváry (Budapest and D.E. Shaw)
Francesco Paesani (UCSD)Jian Liu (Berkeley)
Xiongwu Wu (NIH)
Thomas Steinbrecher (Karlsruhe)
Holger Gohlke (Düsseldorf)
Nadine Homeyer (Düsseldorf)
Qin Cai (UC Irvine)
Wes Smith (UC Irvine)
Dave Mathews (Rochester)
Romelia Salomon-Ferrer (SDSC, UCSD)
Celeste Sagui (North Carolina State)
Volodymyr Babin (North Carolina State)
Tyler Luchko (Rutgers)
Sergey Gusarov (NINT)
Andriy Kovalenko (NINT)
Josh Berryman (U. of Luxembourg)
Peter A. Kollman (UC San Francisco)

For more information, please visit http://ambermd.org/contributors

Acknowledgments
Research support from DARPA, NIH, ONR, DOE and NSF is gratefully acknowledged, along with support
from NVIDIA, Amazon and Exxact. Many people helped add features to various codes; these contributions are
described in the documentation for the individual programs; see also http://ambermd.org/contributors.html.
Recommended Citation:
• When citing Amber 14 in the literature, the following citation should be used:
D.A. Case, V. Babin, J.T. Berryman, R.M. Betz, Q. Cai, D.S. Cerutti, T.E. Cheatham, III, T.A. Darden, R.E.
Duke, H. Gohlke, A.W. Goetz, S. Gusarov, N. Homeyer, P. Janowski, J. Kaus, I. Kolossváry, A. Kovalenko,
T.S. Lee, S. LeGrand, T. Luchko, R. Luo, B. Madej, K.M. Merz, F. Paesani, D.R. Roe, A. Roitberg, C. Sagui,
R. Salomon-Ferrer, G. Seabra, C.L. Simmerling, W. Smith, J. Swails, R.C. Walker, J. Wang, R.M. Wolf, X.
Wu and P.A. Kollman (2014), AMBER 14, University of California, San Francisco.
Peter Kollman died unexpectedly in May, 2001. We dedicate Amber to his memory.
Notes
• We thank Chris Bayly and Merck-Frosst, Canada for permission to include charge increments for the AM1BCC charge scheme.
• Some of the force field routines were adapted from similar routines in the MOIL program package: R. Elber,
A. Roitberg, C. Simmerling, R. Goldstein, H. Li, G. Verkhivker, C. Keasar, J. Zhang and A. Ulitsky, "MOIL:
A program for simulations of macromolecules" Comp. Phys. Commun. 91, 159-189 (1995).
• The cifparse routines to deal with mmCIF formatted files were written by John Westbrook, and are distributed with permission. See cifparse/README for details.
Cover illustration: The cover shows all atom bilayer self assembly of 128 DOPC phospholipids simulated using
the GPU version of pmemd. The lipid molecules were represented with the Lipid14 force field parameters, along
with TIP3P waters and 0.15 M KCl. The lipids are presented as stick models, with the head group phosphorus
atoms highlighted as orange spheres. Water and ions have been removed for clarity. Taken from Skjevik, Å. A.;
Madej, B. D.; Dickson, C. J.; Teigen, K.; Walker, R.C.; Gould, I.R., J. Am. Chem. Soc. 2014, in review.

Contents
Contents

Introduction and Installation

1. Introduction

1.1. Information flow in Amber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2. List of programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2. Installation

2.1. Applying Updates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2. Contacting the developers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

II.

Amber force fields

3.1. Specifying which force field you want in LEaP . . . .
3.2. The ff14SB force field . . . . . . . . . . . . . . . . .
3.3. The ff14ipq protein force field . . . . . . . . . . . . .
3.4. The Duan et al. (2003) force field . . . . . . . . . . .
3.5. The Yang et al. (2003) united-atom force field . . . . .
3.6. Force fields related to semi-empirical QM . . . . . . .
3.7. The GLYCAM force fields for carbohydrates and lipids
3.8. Lipid Force Fields . . . . . . . . . . . . . . . . . . . .
3.9. Ions . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10. Solvent models . . . . . . . . . . . . . . . . . . . . .
3.11. CHAMBER . . . . . . . . . . . . . . . . . . . . . . .
3.12. Obsolete force field files . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

4. The Generalized Born/Surface Area Model

5. PBSA

59
61
65

Introduction . . . . . . . . . . . . . . . . . . . . . .
Usage and keywords . . . . . . . . . . . . . . . . .
Example inputs and demonstrations of functionalities
Visualization functions in pbsa . . . . . . . . . . . .
pbsa in sander and NAB . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.

6. Reference Interaction Site Model

Introduction . . . . . . .
Practical Considerations
Work Flow . . . . . . .
rism1d . . . . . . . . . .

29
30
32
33
34
34
34
42
44
46
47
52
57

4.1. GB/SA input parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2. ALPB (Analytical Linearized Poisson-Boltzmann) . . . . . . . . . . . . . . . . . . . . . . . . .

6.1.
6.2.
6.3.
6.4.

23
25

3. Molecular mechanics force fields

5.1.
5.2.
5.3.
5.4.
5.5.

15
18

.
.
.
.

65
68
76
79
86
89

.
.
.
.

89
94
95
95

CONTENTS
6.5. 3D-RISM in NAB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.6. rism3d.snglpnt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.7. 3D-RISM in sander . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7. Empirical Valence Bond

7.1.
7.2.
7.3.
7.4.
7.5.
7.6.

111

Introduction . . . . . . . . . . . . . . . . .
General usage description . . . . . . . . . .
Biased sampling . . . . . . . . . . . . . . .
Quantization of nuclear degrees of freedom
Distributed Gaussian EVB . . . . . . . . .
EVB input variables and interdependencies

.
.
.
.
.
.

8. sqm: Semi-empirical quantum chemistry

111
112
115
117
117
119
125

8.1. Available Hamiltonians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.2. Dispersion and hydrogen bond correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.3. Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9. QM/MM calculations

9.1.
9.2.
9.3.
9.4.
9.5.

133

Built-in semiempirical NDDO methods and SCC-DFTB . . . . . .
Interface for ab initio and DFT methods . . . . . . . . . . . . . . .
Adaptive solvent QM/MM simulations . . . . . . . . . . . . . . . .
Adaptive buffered force-mixing QM/MM . . . . . . . . . . . . . .
SEBOMD: SemiEmpirical Born-Oppenheimer Molecular Dynamics

.
.
.
.
.

10. paramfit

10.1.
10.2.
10.3.
10.4.
10.5.

Usage . . . . . . . .
The Job Control File
Multiple molecule fits
Fitting Forces . . . .
Examples . . . . . .

171

.
.
.
.
.

III. System preparation
Cleaning up Protein PDB Files for AMBER . .
Residue naming conventions . . . . . . . . . .
Chains, Residue Numbering, Missing Residues
pdb4amber . . . . . . . . . . . . . . . . . . .
reduce . . . . . . . . . . . . . . . . . . . . . .

185

.
.
.
.
.

Introduction . . . . . . . . . . . . . . . . . . . . . .
Concepts . . . . . . . . . . . . . . . . . . . . . . .
Running LEaP . . . . . . . . . . . . . . . . . . . . .
Basic instructions for using LEaP to build molecules
Commands . . . . . . . . . . . . . . . . . . . . . .
Building oligosaccharides, lipids and glycoproteins .

.
.
.
.
.
.

12. LEaP

12.1.
12.2.
12.3.
12.4.
12.5.
12.6.

172
173
179
179
180

183

11. Preparing PDB Files

11.1.
11.2.
11.3.
11.4.
11.5.

133
142
153
158
165

185
186
187
187
190
191

13. Reading and modifying Amber parameter files

191
191
195
200
201
217
225

13.1. Understanding Amber parameter files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
13.2. ParmEd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

CONTENTS
14. Antechamber and GAFF

14.1.
14.2.
14.3.
14.4.
14.5.
14.6.
14.7.

255

Principal programs . . . . . . . . . . . . . . .
A simple example for antechamber . . . . . . .
Using the components.cif file from the PDB . .
Programs called by antechamber . . . . . . . .
Miscellaneous programs . . . . . . . . . . . .
New Development of Antechamber And GAFF
Metal Center Parameter Builder (MCPB) . . .

.
.
.
.
.
.
.

15. Setting up crystal simulations

15.1.
15.2.
15.3.
15.4.

UnitCell . .
PropPDB .
AddToBox .
ChBox . . .

.
.
.
.

255
259
262
262
266
268
269
271

.
.
.
.

16. Using the AMOEBA Force Field with AMBER

271
271
271
273
275

16.1. Installing TINKER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
16.2. Preparing the system with TINKER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276

IV. Running simulations

279

17. sander

281

17.1. Introduction . . . . . . . . . . . . . . . . . . .
17.2. File usage . . . . . . . . . . . . . . . . . . . .
17.3. Example input files . . . . . . . . . . . . . . .
17.4. Namelist Input Syntax . . . . . . . . . . . . .
17.5. Overview of the information in the input file . .
17.6. General minimization and dynamics parameters
17.7. Potential function parameters . . . . . . . . . .
17.8. Varying conditions . . . . . . . . . . . . . . .
17.9. File redirection commands . . . . . . . . . . .
17.10.Getting debugging information . . . . . . . . .
17.11.multisander (and multipmemd) . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.

18. pmemd and pmemd.amoeba

18.1.
18.2.
18.3.
18.4.
18.5.
18.6.
18.7.
18.8.

Introduction . . . . . . . . . . . . . . . .
Functionality . . . . . . . . . . . . . . .
PMEMD-specific namelist variables . . .
Slightly changed functionality . . . . . .
Parallel performance tuning and hints . .
GPU Accelerated PMEMD . . . . . . . .
Intel® Many Integrated Core Architecture
pmemd.amoeba . . . . . . . . . . . . . .

19. Atom and Residue Selections

281
282
283
284
285
285
294
300
304
304
307
309

.
.
.
.
.
.
.
.

309
309
311
312
313
313
319
321
325

19.1. Amber Masks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
19.2. "Atom Expressions" in NAB Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
19.3. GROUP Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328

CONTENTS
20. Sampling configuration space

20.1.
20.2.
20.3.
20.4.
20.5.
20.6.

Self-Guided Langevin dynamics .
Accelerated Molecular Dynamics .
Targeted MD . . . . . . . . . . .
Multiply-Targeted MD (MTMD) .
Nudged elastic band calculations .
Low-MODe (LMOD) methods . .

333

.
.
.
.
.
.

21. Free energies

21.1.
21.2.
21.3.
21.4.
21.5.
21.6.
21.7.

349

Thermodynamic integration . . . . . . . . . . . . . . . . . . . . . . . .
Absolute Free Energies using EMIL . . . . . . . . . . . . . . . . . . .
Linear Interaction Energies . . . . . . . . . . . . . . . . . . . . . . . .
Umbrella sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Replica Exchange Molecular Dynamics (REMD) . . . . . . . . . . . .
Adaptively biased MD, steered MD, and umbrella sampling with REMD
Steered Molecular Dynamics (SMD) and the Jarzynski Relationship . .

.
.
.
.
.
.
.

22. Constant pH calculations

22.1.
22.2.
22.3.
22.4.
22.5.
22.6.
22.7.

Background . . . . . . . . . . . . . . . . . . . . . .
Preparing a system for constant pH . . . . . . . . . .
Running at constant pH . . . . . . . . . . . . . . . .
Analyzing constant pH simulations . . . . . . . . . .
Extending constant pH to additional titratable groups
Constant pH MD Replica Exchange . . . . . . . . .
cphstats . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

23.1. Distance, angle and torsional restraints . . . . . . . . . . .
23.2. NOESY volume restraints . . . . . . . . . . . . . . . . .
23.3. Chemical shift restraints . . . . . . . . . . . . . . . . . .
23.4. Pseudocontact shift restraints . . . . . . . . . . . . . . . .
23.5. Direct dipolar coupling restraints . . . . . . . . . . . . . .
23.6. Residual CSA or pseudo-CSA restraints . . . . . . . . . .
23.7. Preparing restraint files for Sander . . . . . . . . . . . . .
23.8. Getting summaries of NMR violations . . . . . . . . . . .
23.9. Time-averaged restraints . . . . . . . . . . . . . . . . . .
23.10.Multiple copies refinement using LES . . . . . . . . . . .
23.11.Some sample input files . . . . . . . . . . . . . . . . . . .
23.12.X-ray Crystallography Refinement using SANDER . . . .
23.13.EMAP restraints for rigid and flexible fitting into EM maps

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.

23. NMR, X-ray, and cryo-EM/ET refinement

393
393
395
398
400
401
401
409

24. LES

349
358
362
363
364
381
388
393

.
.
.
.
.
.
.

24.1.
24.2.
24.3.
24.4.
24.5.
24.6.

333
336
338
339
341
344

410
415
416
417
419
420
421
428
428
429
429
433
434
437

Preparing to use LES with Amber . . . . . . . . . . . . .
Using the ADDLES program . . . . . . . . . . . . . . . .
More information on the ADDLES commands and options
Using the new topology/coordinate files with SANDER . .
Using LES with the Generalized Born solvation model . .
Case studies: Examples of application of LES . . . . . . .

437
438
440
441
442
442

CONTENTS
25. Quantum dynamics

25.1.
25.2.
25.3.
25.4.
25.5.
25.6.

447

Path-Integral Molecular Dynamics . . . . . . . . .
Centroid Molecular Dynamics (CMD) . . . . . . .
Ring Polymer Molecular Dynamics (RPMD) . . .
Linearized semiclassical initial value representation
Reactive Dynamics . . . . . . . . . . . . . . . . .
Isotope effects . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.
.
.

26. mdgx

26.1.
26.2.
26.3.
26.4.
26.5.
26.6.
26.7.
26.8.

447
451
454
454
459
462
467

Input and Output . . . . . . . . . . . . . . . . .
Installation . . . . . . . . . . . . . . . . . . . .
Special Algorithmic Features of mdgx . . . . . .
Customizable Virtual Site Support in mdgx . . .
Restrained Electrostatic Potential Fitting in mdgx
Bonded Term Fitting in mdgx . . . . . . . . . . .
Thermodynamic Integration . . . . . . . . . . .
Future Directions and Goals of the mdgx Project

.
.
.
.
.
.
.
.

Analysis of simulations

467
468
468
469
472
475
477
477

479

27. mdout_analyzer.py and ambpdb

481

27.1. ambpdb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
28. cpptraj

483

28.1. Running cpptraj . . . . . . . . . . . . . . . . .
28.2. General Concepts . . . . . . . . . . . . . . . .
28.3. Data Sets and Data Files . . . . . . . . . . . .
28.4. Data File Options . . . . . . . . . . . . . . . .
28.5. Coordinates as a Data Set (COORDS Data Sets)
28.6. General Commands . . . . . . . . . . . . . . .
28.7. Topology File Commands . . . . . . . . . . . .
28.8. Trajectory File Commands . . . . . . . . . . .
28.9. Actions that Modify Topology/Coordinates . .
28.10.Action Commands . . . . . . . . . . . . . . .
28.11.Matrix and Vector Actions . . . . . . . . . . .
28.12.Data Set Analysis Commands . . . . . . . . .
28.13.Coordinate Analysis Commands . . . . . . . .
28.14.Matrix and Vector Analysis . . . . . . . . . . .
28.15.Matrix/Vector Analysis Examples . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.

29. MMPBSA.py

29.1.
29.2.
29.3.
29.4.

569

Introduction . . . . . . . . . . . . . . . . . .
Preparing for an MM/PB(GB)SA calculation
Running MMPBSA.py . . . . . . . . . . . .
Python API . . . . . . . . . . . . . . . . . .

.
.
.
.

30. MM_PBSA

30.1.
30.2.
30.3.
30.4.

General instructions . . . . . . . . . . . .
Input explanations . . . . . . . . . . . . .
Auxiliary programs used by MM_PBSA .
APBS as an alternate PB solver in Sander

483
484
488
490
492
494
498
502
506
513
548
550
556
562
566

569
570
572
584
591

.
.
.
.

591
592
598
598

CONTENTS
31. FEW

31.1.
31.2.
31.3.
31.4.
31.5.
31.6.

601

Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview of workflow steps and minimal input . . . . . . . . . . . . . . .
Common setup of molecular dynamics simulations . . . . . . . . . . . . .
Workflow for automated MM-PBSA & MM-GBSA calculations (WAMM)
Linear interaction energy workflow (LIEW) . . . . . . . . . . . . . . . . .
Thermodynamic integration workflow (TIW) . . . . . . . . . . . . . . . .

.
.
.
.
.
.

32. XtalAnalyze

601
603
605
610
615
619
627

32.1. XtalAnalyze.sh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627
32.2. XtalPlot.sh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
32.3. md2map.sh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630

VI. NAB and AmberLite

633

33. NAB: Introduction

33.1. Background . . . . . . . . . . . . .
33.2. Methods for structure creation . . .
33.3. Compiling nab Programs . . . . . .
33.4. Parallel Execution . . . . . . . . . .
33.5. First Examples . . . . . . . . . . .
33.6. Molecules, Residues and Atoms . .
33.7. Creating Molecules . . . . . . . . .
33.8. Residues and Residue Libraries . . .
33.9. Atom Names and Atom Expressions
33.10.Looping over atoms in molecules . .
33.11.Points, Transformations and Frames
33.12.Creating Watson Crick duplexes . .

635

.
.
.
.
.
.
.
.
.
.
.
.

34.1. Language Elements . . . . . . . . . . . . . . . . .
34.2. Higher-level constructs . . . . . . . . . . . . . . .
34.3. Statements . . . . . . . . . . . . . . . . . . . . . .
34.4. Structures . . . . . . . . . . . . . . . . . . . . . .
34.5. Functions . . . . . . . . . . . . . . . . . . . . . .
34.6. Points and Vectors . . . . . . . . . . . . . . . . . .
34.7. String Functions . . . . . . . . . . . . . . . . . . .
34.8. Math Functions . . . . . . . . . . . . . . . . . . .
34.9. System Functions . . . . . . . . . . . . . . . . . .
34.10.I/O Functions . . . . . . . . . . . . . . . . . . . .
34.11.Molecule Creation Functions . . . . . . . . . . . .
34.12.Creating Biopoloymers . . . . . . . . . . . . . . .
34.13.Fiber Diffraction Duplexes in NAB . . . . . . . . .
34.14.Reduced Representation DNA Modeling Functions
34.15.Molecule I/O Functions . . . . . . . . . . . . . . .
34.16.Other Molecular Functions . . . . . . . . . . . . .
34.17.Debugging Functions . . . . . . . . . . . . . . . .
34.18.Time and date routines . . . . . . . . . . . . . . .
34.19.Computational resource consumption functions . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

34. NAB: Language Reference

636
637
639
639
640
643
643
644
646
647
648
649
659

659
661
667
669
670
671
672
672
673
673
676
677
677
678
679
679
681
682
682

CONTENTS
35. NAB: Rigid-Body Transformations

35.1.
35.2.
35.3.
35.4.
35.5.

683

Transformation Matrix Functions . . . . . . . .
Frame Functions . . . . . . . . . . . . . . . .
Functions for working with Atomic Coordinates
Symmetry Functions . . . . . . . . . . . . . .
Symmetry server programs . . . . . . . . . . .

.
.
.
.
.

Metric Matrix Distance Geometry . . . . . . . . . . . .
Creating and manipulating bounds, embedding structures
Distance geometry templates . . . . . . . . . . . . . . .
Bounds databases . . . . . . . . . . . . . . . . . . . . .

.
.
.
.

Basic molecular mechanics routines . . . . . . . . . . . .
NetCDF read/write routines . . . . . . . . . . . . . . . . .
Typical calling sequences . . . . . . . . . . . . . . . . . .
Second derivatives and normal modes . . . . . . . . . . .
Low-MODe (LMOD) optimization methods . . . . . . . .
Using the Hierarchical Charge Partitioning (HCP) method

.
.
.
.
.
.

.
.
.
.
.

36. NAB: Distance Geometry

36.1.
36.2.
36.3.
36.4.

691

37. NAB: Molecular mechanics and dynamics

37.1.
37.2.
37.3.
37.4.
37.5.
37.6.

Duplex Creation Functions . .
nab and Distance Geometry . .
Building Larger Structures . .
Wrapping DNA Around a Path
Other examples . . . . . . . .

691
692
696
698
701

38. NAB: Sample programs

38.1.
38.2.
38.3.
38.4.
38.5.

683
683
684
684
686

701
710
713
714
715
727
729

.
.
.
.
.

39. amberlite: Some AmberTools-Based Utilities

39.1. Introduction . . . . . . . . . . . . . . . . . .
39.2. Coordinates and Parameter-Topology Files . .
39.3. pytleap: Creating Coordinates and ParameterTopology Files . . . . . . . . . . . . . . . .
39.4. Energy Checking Tool: ffgbsa . . . . . . . .
39.5. Energy Minimizer: minab . . . . . . . . . . .
39.6. Molecular Dynamics "Lite": mdnab . . . . .
39.7. MM(GB)(PB)/SA Analysis Tool: pymdpbsa .
39.8. Examples and Test Cases . . . . . . . . . . .

729
730
738
744
750
751

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 751
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 752
.
.
.
.
.
.

.
.
.
.
.
.

754
757
757
758
759
766

Bibliography

775

Index

814

Part I.

Introduction and Installation

1. Introduction
Amber is the collective name for a suite of programs that allow users to carry out molecular dynamics simulations, particularly on biomolecules. None of the individual programs carries this name, but the various parts work
reasonably well together, and provide a powerful framework for many common calculations.[1, 2] The term amber
is sometimes used to refer to the empirical force fields that are implemented here.[3, 4] It should be recognized
however, that the code and force field are separate: several other computer packages have implemented the Amber
force fields, and other force fields can be implemented with the Amber programs. Further, the force fields are in
the public domain, whereas the codes are distributed under a license agreement.
The Amber software suite is divided into two parts: AmberTools14, a collection of freely available programs
mostly under the GPL license, and Amber14, which is centered around the pmemd simulation program, and which
continues to be licensed as before, under a more restrictive license. Amber14 (2014) represents a significant change
from the most recent previous version, Amber12, which was released in April, 2012. (We have moved to numbering
releases by the last two digits of the calendar year, so there is no version 13.) Please see http://ambermd.org for
an overview of the most important changes.
AmberTools is a set of programs for biomolecular simulation and analysis. They are designed to work well
with each other, and with the “regular” Amber suite of programs. You can perform many simulation tasks with
AmberTools, and you can do more extensive simulations with the combination of AmberTools and Amber itself.
Most components of AmberTools are released under the GNU General Public License (GPL). A few components
are in the public domain or have other open-source licenses. See the README file for more information.
Everyone should read (or at least skim) this chapter. Even if you are an experienced Amber user, there may be
things you have missed, or new features, that will help. There are also tips and examples on the Amber Web pages
at http://ambermd.org. Although Amber may appear dauntingly complex at first, it has become easier to use over
the past few years, and overall is reasonably straightforward once you understand the basic architecture and option
choices. In particular, we have worked hard on the tutorials to make them accessible to new users. Thousands of
people have learned to use Amber; don’t be easily discouraged.
If you want to learn more about basic biochemical simulation techniques, there are a variety of good books to
consult, ranging from introductory descriptions,[5–7] to standard works on liquid state simulation methods,[8–10]
to multi-author compilations that cover many important aspects of biomolecular modelling.[11–15] Looking for
"paradigm" papers that report simulations similar to ones you may want to undertake is also generally a good idea.
If you are new to this field, Chapter 13 provides a basic introduction to force fields, along with details of how the
parameters are encoded in Amber files.

1.1. Information flow in Amber
Understanding where to begin in AmberTools is primarily a problem of managing the flow of information in
this package — see Fig. 1.1. You first need to understand what information is needed by the simulation programs
(sander, pmemd, mdgx or nab). You need to know where it comes from, and how it gets into the form that the
energy programs require. This section is meant to orient the new user and is not a substitute for the individual
program documentation.
Information that all the simulation programs need (see the circles in Fig. 1.1):
1. Cartesian coordinates for each atom in the system. These usually come from X-ray crystallography, NMR
spectroscopy, or model-building. They should generally be in Protein Data Bank (PDB) format. The program
LEaP provides a platform for carrying out many of these modeling tasks, but users may wish to consider
other programs as well. Generally, editing of these files is needed, and the pdb4amber and reduce scripts
can do some of this.

1. Introduction

pdb4amber,
reduce

pdb

antechamber,
MCPB,
LEaP

force
field
info

prmtop
prmcrd

ParmEd

NMR or
XRAY info

MMPBSA.py,
amberlite,
FEW

sander,
nab, mdgx,
pmemd

mdin
info

mdout_analyzer,
cpptraj

Figure 1.1.: Basic information flow in Amber
2. Topology: Connectivity, atom names, atom types, residue names, and charges. This information comes from
the database, which is found in the $AMBERHOME/dat/leap/lib directory, and is described in Chapter 3. It
contains topology for the standard amino acids as well as N- and C-terminal charged amino acids, DNA,
RNA, and common sugars and lipids. Topology information for other molecules (not found in the standard
database) is kept in user-generated “residue files”, which are generally created using antechamber.
3. Force field: Parameters for all of the bonds, angles, dihedrals, and atom types in the system. The standard
parameters for several force fields are found in the $AMBERHOME/dat/leap/parm directory; see Chapter 3
for more information. These files may be used “as is” for proteins and nucleic acids, or users may prepare
their own files that contain modifications to the standard force fields.
4. Commands: The user specifies the procedural options and state parameters desired. These are specified in
input files (named mdin by default) or in “driver” programs written in the NAB language.

1.1.1. Preparatory programs
LEaP is the primary program to create a new system in Amber, or to modify existing systems. It is available as

the command-line program tleap or the GUI xleap. It combines the functionality of prep, link, edit and parm
from much earlier versions of Amber.
ParmEd provides a simple way to extract information about the parameters defined in a parameter-topology file.

It can also be used to check that the parameter-topology file is valid for complex systems, and it can also
make simple modifications to this file very quickly.

1.1. Information flow in Amber
antechamber is the main program to develop force fields for drug-like molecules or modified amino acids using

the general Amber force field (GAFF). These can be used directly in LEaP, or can serve as a starting point
for further parameter development.
MCPB provides a means to build, prototype, and validate MM models of metalloproteins. It uses the bonded plus

electrostatics model to expand existing pairwise additive force fields.
paramfit allows the generation of bonded force field parameters for any molecule by fitting to quantum data.

1.1.2. Simulation programs
sander (now part of AmberTools) is the basic energy minimizer and molecular dynamics program. This program

relaxes the structure by iteratively moving the atoms down the energy gradient until a sufficiently low average
gradient is obtained. The molecular dynamics portion generates configurations of the system by integrating
Newtonian equations of motion. MD will sample more configurational space than minimization, and will
allow the structure to cross over small potential energy barriers. Configurations may be saved at regular
intervals during the simulation for later analysis, and basic free energy calculations using thermodynamic
integration may be performed. More elaborate conformational searching and modeling MD studies can also
be carried out using the sander module. This allows a variety of constraints to be added to the basic force
field, and has been designed especially for the types of calculations involved in NMR, Xray or cryo-EM
structure refinement.
pmemd (part of Amber) is a version of sander that is optimized for speed and for parallel scaling; the pmemd.cuda

variant runs on GPUs. The name stands for “Particle Mesh Ewald Molecular Dynamics,” but this code can
now also carry out generalized Born simulations. The input and output have only a few changes from sander.
mdgx is a molecular dynamics engine with functionality that mimics some of the features in sander and pmemd,

but featuring simple C code and an atom sorting routine that simplifies the flow of information during force
calculations. The principal purpose of mdgx is to provide a tool for redesign of the basic molecular dynamics
algorithms and models, and for supporting new models for parameter development.
NAB (Nucleic Acid Builder) is a language that can be used to write programs to perform non-periodic simulations,

most often using an implicit solvent force field.

1.1.3. Analysis programs
mdout_analyzer.py is a simple-to-run Python script that will provide summaries of information that is in the

output files from sander or pmemd.
cpptraj is the main trajectory analysis utility (written in C++) for carrying out superpositions, extractions of

coordinates, calculation of bond/angle/dihedral values, atomic positional fluctuations, correlation functions,
analysis of hydrogen bonds, etc. It has many new features in version 14: see Chap. 28 for more information.
pbsa is an analysis program for solvent-mediated energetics of biomolecules. It can be used to perform both

electrostatic and non-electrostatic continuum solvation calculations with input coordinate files from molecular dynamics simulations and other sources (in the pqr format). It also supports visualization of solventmediated electrostatic potentials in various visualization programs.
MMPBSA.py is a python script that automates energy analysis of snapshots from a molecular dynamics simulation

using ideas generated from continuum solvent models. (There is also an older perl script, called mm_pbsa.pl,
that has similar functionality.)
FEW (Free energy workflow) automates free energy calculations of protein-ligand binding using TI, MM/PBSA-

type, or LIE calculations.

1. Introduction
amberlite is small set of NAB programs and python scripts that implement a limited set of MD simulations and

mm-pbsa (or mm-gbsa) analysis, aimed primarily at the analysis of protein-ligand interactions. These tools
can be useful in their own right, or as a good introduction to Amber and a starting point for more complex
calculations.
XtalAnalyze a set of utilities for analyzing crystal simulation trajectories. See Chapter 32 for more information.

1.2. List of programs
Amber is comprised of a large number of programs designed to aid you in your computational studies of chemical
systems, and the number of released tools grows regularly. This section provides a list of the main programs
included with AmberTools. Each program included in the suite is listed here with a very brief description of its
main function along with which chapter in the manual a more thorough description can be found. Note: there are
some additional programs that are part of the MTK++ suite that are not listed here; see that documentation for
more information.
AddToBox A program for adding solvent molecules to a crystal cell. See Subsection 15.3.
ChBox A program for changing the box dimensions of an Amber restart file. See Subsection 15.4.
CheckMD A program for automated checking of an MD simulation. Run the program without options for usage

statement.
MCPB A semi-automated tool for metalloprotein parametrization. See Section 14.7.
MMPBSA.py A program to post-process trajectories to calculate binding free energies according to the MM/PBSA

approximation. See Chapter 29.
MTKppConstants Lists the constants used in MTK++. Run the program without arguments to get the full list.
PropPDB A program for propagating a PDB structure. See Subsection 15.2
UnitCell A program for recreating a crystallographic unit cell from a PDB structure. See Subsection 15.1
acdoctor A tool to diagnose what may be causing antechamber to fail. See Subsection 14.5.1
am1bcc A program called by antechamber to calculate AM1-BCC charges during ligand parametrization. It

can be used as a standalone program, with the options printed when you enter the program name with no
arguments. See Section 14.4
ambpdb A program to convert an Amber system (prmtop and inpcrd/restart) into a PDB, MOL2, or PQR file. See

Section 27.1
ante-MMPBSA.py A program to create the necessary, self-consistent prmtop files for MMPBSA with a single

starting topology file. See Subsection 29.2.2
antechamber A program for parametrizing ligands and other small molecules. See Chapter 14
atomtype A program called by antechamber to judge the atom types in an input structure. It can be used as a

standalone program. If you provide no arguments, it prints out the usage statement. See Section 14.4
bondtype A program called by antechamber to judge what types of bonds exist in a given input structure. It can

be used as a standalone program. If you provide no arguments, it prints out the usage statement. See Section
14.4
chamber A program to convert a CHARMM psf file to an Amber topology (prmtop) file. See Section 3.11
charmmlipid2amber.x A script that converts a PDB created with the CHARMM-GUI lipid builder into one recog-

nized by Amber and AmberTools programs. If you provide no arguments, it prints out the usage statement.

1.2. List of programs
cpinutil.py A program to create a constant pH input (CPin) file from a PDB file. If you provide no arguments,

you get the usage statement.
cpptraj A versatile program for trajectory post-processing and data analysis. See Chapter 28
cphstats A program that computes protonation state statistics from constant pH simulations. See Section 22.7
elsize A program that estimates the effective electrostatic size of a given input structure. See Section 4.2.1
espgen A program called by antechamber to generate ESP files during ligand or small molecule parametrization.

If you provide no arguments, it prints out the usage statement.
hcp_getpdb A program that adds necessary sections to a topology (prmtop) file so it can be used for the HCP GB

approximation. See Section 37.6
matextract Part of the symmetry definition programs, used to print matrices dumped to stdin to stdout. See

Subsection 35.5.5
matgen Generate symmetry-transformation matrices. Part of the symmetry definition programs. See Subsection

35.5.1
matmerge Merges symmetry-transformation matrices into one matrix transformation matrix. Part of the symme-

try definition programs. See Subsection 35.5.3
matmul Multiplies matrices. Part of the symmetry definition programs. See Subsection 35.5.4
mdgx An explicit solvent, PME molecular dynamics engine. See Chapter 26
mdnab Implicit solvent MD program written in NAB as part of Amberlite (Chapter 39)
mdout_analyzer.py A script that allows you to rapidly analyze and graph data from sander/pmemd output files.

See Section 27
minab Implicit solvent minimization program written in NAB as part of Amberlite (Chapter 39)
mmpbsa_py_energy A NAB program written to calculate energies for MMPBSA using either GB or PB solvent

models. It can be used as a standalone program that mimics the imin=5 functionality of sander, but it is
called automatically inside MMPBSA. See MMPBSA mdin files as example input files for this program.
Providing the –help or -h flags prints the usage message.
mmpbsa_py_nabnmode A NAB program written to calculate normal mode entropic contributions for MMPBSA.

This can really only be used by MMPBSA.
molsurf A program that calculates a molecular surface area based on input PQR files and a probe radius. Providing

no arguments prints the usage message.
nab Stands for Nucleic Acid Builder. NAB is really a compiler that provides a convenient molecular programming

language loosely based on C. See Chapter 33 and other related chapters.
new_crd_to_dyn Sets up a Tinker-style .DYN file from a coordinate file. This is part of the Tinker-to-Amber

conversion program set.
new_to_old_crd Converts new Tinker-style coordinate files to old Amber-style coordinates. Part of the Tinker-

to-Amber conversion program set.
paramfit Improves force field parameters by fitting to quantum data. See Chapter 10
parmcal Calculates parameters for given angles and bonds interactively. See Subsection 14.5.2
parmchk2 A program that analyzes an input force field library file (mol2 or amber prep), and extracts relevant

parameters into an frcmod file. See Subsection 14.1.2

1. Introduction
parmed.py A program for querying and manipulating prmtop files. See Section 13.2
pbsa A finite difference Poisson-Boltzmann solver. See Chapter 5
pdb4amber A program to prepares PDB files for use in leap. See Section 11.4
pmemd A performance- and parallel-optimized dynamics engine implementing a subset of sander’s functionality
pmemd.cuda A GPU-accelerated version of pmemd
prepgen A program used as part of antechamber that generates an Amber prep file. Use no arguments to print the

usage message. See Section 14.4
pymdpbsa A full analysis tool for MD(GB,PB)/SA computations. See Section 39.7
pytleap A user-friendly wrapper for simple leap and antechamber runs to prepare topology and coordinate files

for Amber. See Section 39.3
reduce A program for adding or removing hydrogen atoms to a PDB. See Section 11.5
residuegen A program to automate the generation of an Amber residue template (i.e. Amber prep file). See

Subsection 14.5.3
resp A program typically called by antechamber and R.E.D tools to perform a Restrained ElectroStatic Potential

calculation for calculating partial atomic charges. Use no arguments to get the usage message
respgen A program called by antechamber to generate RESP input files. See Section 14.4
rism1d A 1D-RISM solver. See Section 6.4
rism3d.snglpnt A 3D-RISM solver for single point calculations. See Section 6.6
sander The main engine used for running molecular simulations with Amber. Originally an acronym standing for

Simulated Annealing with Nmr-Derived Energy Restraints.
softcore_setup.py A program to aid in softcore TI setup for sander. Use no arguments to get a usage message.
sqm Semiempirical (or Stand-alone) Quantum Mechanics solver. See Chapter 8
tinker_to_amber Converts a Tinker analout and parameter file into an Amber-compatible topology file.
tleap A script that calls teLeap with specific setup command-line arguments. See Chapter 12
transform Applies matrix transformations to a structure. Part of the symmetry definition programs. See Subsec-

tion 35.5.6
tss_init A program to do some matrix stuff. See Section 35.5
tss_main A program to do some matrix stuff. See Section 35.5
tss_next A program to do some matrix stuff. See Section 35.5
ucpp A program to do some source code preprocessing. You should never actually use this program—it is used

by nab.
xaLeap A graphical program for creating Amber topology files. This program is called through the xleap script,

so you should never actually invoke this program directly.
xleap A script that calls xaLeap with specific setup command-line arguments. See Chapter 12
xparmed.py A graphical front-end to ParmEd functionality (i.e., parameter file editing and querying). See Section

13.2

2. Installation
This chapter gives an overview of how to install and test your distribution. The Amber web page (http://ambermd.org)
has additional instructions and hints for various common operating systems. Look for the “Running Amber on ....”
links. Once you have downloaded the distribution files, do the following:
1. First, extract the files in some location (we use /home/myname as an example here):
cd /home/myname
tar xvfj AmberTools14.tar.bz2
tar xvfj Amber14.tar.bz2

# (Note: extracts in an
#
“amber14” directory)
# (only if you have licensed Amber 14!)

2. Next, set your AMBERHOME environment variable:
export AMBERHOME=/home/myname/amber14
setenv AMBERHOME /home/myname/amber14

# (for bash, zsh, ksh, etc.)
# (for csh, tcsh)

Be sure to change the “/home/myname” above to whatever directory is appropriate for your machine, and be
sure that you have write permissions in the directory tree you choose.
3. Next, you may need to install some compilers and other libraries. Details depend on what OS you have, and
what is already installed. Package managers can greatly simplify this task. For example for Debian-based
Linux systems (such as Ubuntu), the following command should get you what you need:
sudo apt-get install csh flex gfortran g++ xorg-dev \
zlib1g-dev libbz2-dev patch python-tk python-matplotlib

See http://ambermd.org/ubuntu.html for more information, and for requirements for other variants of Linux,
and for Macintosh OSX.
4. Now, in the AMBERHOME directory, run the configure script:
cd $AMBERHOME
./configure --help

will show you the options. Choose the compiler and flags you want; for most systems, the following should
work:
./configure gnu

Don’t choose any parallel options at this point. (You may need to edit the resulting config.h file to change any
variables that don’t match your compilers and OS. The comments in the config.h file should help.) This step
will also check to see if there are any updates and bug fixes that have not been applied to your installation,
and will apply them (unless you ask it not to). If the configure step finds missing libraries, go back to Step 3.
5. The configure step will create two resource files in the AMBERHOME directory: amber.sh and amber.csh.
These sourceable scripts will set up your shell environment correctly for Amber:
source /home/myname/amber14/amber.sh # for bash, zsh, ksh, etc.
source /home/myname/amber14/amber.csh # for csh, tcsh

Of course, /home/myname/amber14 should be adjusted for your AMBERHOME. Adding these commands
to your login resource file (e.g., ~/.bashrc, ~/.cshrc, ~/.zshrc, etc.) will set up your environment every time
you start a new shell.

2. Installation
6. Then,
make install

will compile the codes. If this step fails, try to read the error messages carefully to identify the problem.
7. This can be followed by
make test

which will run tests and will report successes or failures.
Where "possible FAILURE" messages are found, go to the indicated directory under $AMBERHOME/AmberTools/test or $AMBERHOME/test, and look at the "*.dif" files. Differences should involve round-off in
the final digit printed, or occasional messages that differ from machine to machine (see below for details).
As with compilation, if you have trouble with individual tests, you may wish to comment out certain lines in
the Makefiles (i.e., $AMBERHOME/AmberTools/test/Makefile or $AMBERHOME/test/Makefile), and/or
go directly to the test subdirectories to examine the inputs and outputs in detail. For convenience, all of the
failure messages and differences are collected in the $AMBERHOME/logs directory; you can quickly see
from these if there is anything more than round-off errors.
The nature of molecular dynamics, is such that the course of the calculation is very dependent on the order
of arithmetical operations and the machine arithmetic implementation, i.e., the method used for round-off.
Because each step of the calculation depends on the results of the previous step, the slightest difference
will eventually lead to a divergence in trajectories. As an initially identical dynamics run progresses on
two different machines, the trajectories will eventually become completely uncorrelated. Neither of them
are "wrong;" they are just exploring different regions of phase space. Hence, states at the end of long
simulations are not very useful for verifying correctness. Averages are meaningful, provided that normal
statistical fluctuations are taken into account. "Different machines" in this context means any difference in
floating point hardware, word size, or rounding modes, as well as any differences in compilers or libraries.
Differences in the order of arithmetic operations will affect round-off behavior; (a + b) + c is not necessarily
the same as a + (b + c). Different optimization levels will affect operation order, and may therefore affect
the course of the calculations.
All initial values reported as integers should be identical. The energies and temperatures on the first cycle
should be identical. The RMS and MAX gradients reported in sander are often more precision sensitive
than the energies, and may vary by 1 in the last figure on some machines. In minimization and dynamics
calculations, it is not unusual to see small divergences in behavior after as little as 100-200 cycles.
Note: If you have untarred the Amber14.tar.bz2 file, then steps 1-6 will install and test both AmberTools
and Amber; otherwise it will just install and test AmberTools. If you license Amber later, just come back and
repeat steps 1-6 again.
8. If you are new to Amber, you should look at the tutorials and this manual and become familiar with how
things work. If and when you wish to compile parallel (MPI) versions of Amber, do this:
cd $AMBERHOME
./configure -mpi <....other options....>
make install
# Note the value below may depend on your MPI implementation
export DO_PARALLEL=”mpirun -np 2”
make test
# Note, some tests, like the replica exchange tests, require more
# than 2 threads, so we suggest that you test with either 4 or 8
# threads as well
export DO_PARALLEL=”mpirun -np 8”
make test

2.1. Applying Updates
This assumes that you have installed MPI and that mpicc and mpif90 are in your PATH. Some MPI
installations are tuned to particular hardware (such as infiniband), and you should use those versions if you
have such hardware. Most people can use standard versions of either mpich or openmpi. To install one of
these, use one of the simple scripts that we have prepared:
cd $AMBERHOME/AmberTools/src
./configure_mpich
OR
./configure_openmpi

Follow the instructions of these scripts, then return to beginning of step 7.
Amber 14 brings with it an additional flag (-intelmpi) to enable use of the Intel® MPI Library. Use this
instead of -mpi in step 8 and ensure that you have installed the Intel® MPI Library, and that mpiicc and
mpiifort are in your PATH.
Some notes about the parallel programs in AmberTools:
1. The MPI version of nab is called mpinab, by analogy with mpicc or mpif90: mpinab is a compiler that
will produce an MPI-enabled executable from source code written in the NAB language. Before compiling
mpinab, be sure that you are familiar with the serial version of nab and that you really need a parallel version.
If you have shared-memory nodes, the OpenMP version might be a better alternative. See Section 33.4 for
more information. (Note that mpinab is primarily designed to write driver routines that call MPI versions of
the energy functions; it is not set up to write your own, novel, parallel codes.)
a) The MPI version of MMPBSA.py is called MMPBSA.py.MPI, and requires the package mpi4py to run.
If it is not present in your Python standard library already, it will be built along with MMPBSA.py.MPI
and placed in the $AMBERHOME/bin directory. If you have problems with MMPBSA.py.MPI, see if
you get the same problems with the serial version, MMPBSA.py, to see if it is an issue with the parallel
version or MMPBSA.py in general. Because we do not make or maintain the mpi4py source code,
MMPBSA.py.MPI will not be available on platforms on which mpi4py cannot be built.
2. NAB and Cpptraj can also be compiled using OpenMP:
./configure -openmp <....other options....>
make openmp

Note that the OpenMP version of NAB has the same name as the single-threaded version. See section 33.4
for information on running the OpenMP version of NAB and section 28.2.6 for information on running the
OpenMP version of Cpptraj.
3. See Section 18.6.5 for information about installing the GPU-accelerated versions of pmemd.

2.1. Applying Updates
For most users, simply running the configure script and responding ‘yes’ to the update request will automatically
download and apply all patches. This section describes the main updating script responsible for managing updates.
We suggest that you at least skim the first section on the basic usage—particularly the note about the --version
flag for if/when you ask for help on the mailing list.

2.1.1. Basic Usage
Updates to AmberTools and Amber are downloaded, applied, and managed automatically using the Python script
update_amber (it was patch_amber.py for AmberTools 12). This script requires Python 2. The configure script in
$AMBERHOME automatically uses update_amber to search for available updates to AmberTools 13 (and Amber 12
when present) unless explicitly disabled with the --no-updates flag (it must be the first option to configure). If
any are available, you will be asked if you want them downloaded and applied. This script resides in $AMBERHOME
and can be executed from anywhere (it will verify that AMBERHOME is set properly), but if moved from AMBERHOME,
it will not work. There are 3 main operating modes, or actions, that you can perform with them:

2. Installation
• $AMBERHOME/update_amber --check-updates : This option will query the Amber website for any updates that have been posted that have not been applied to your installation. If you think you have found a
bug, this is helpful to try first before emailing with problems since your bug may have already been fixed.
• $AMBERHOME/update_amber --version : This option will return which patches have been applied to the
current tree so far. When emailing the Amber list with problems, it is important to have the output of this
command, since that lets us know exactly which updates have been applied.
• $AMBERHOME/update_amber --update : This option will go to the Amber website, download all updates
that have not been applied to your installation, and apply them to the source code. Note that you will
have to recompile any affected code for the changes to take effect!

2.1.2. Advanced options
update_amber has additional functionality as well that allows more intimate control over the patching process.
For a full list of options, use the --full-help command-line option. These are considered advanced options.
• $AMBERHOME/update_amber --download-patches : Only download patches, do not apply them
• $AMBERHOME/update_amber --apply-patch= : This will apply a third-party patch
• $AMBERHOME/update_amber --reverse-patch= : Reverses a third-party patch file that was applied via the --apply-patch option (see above).
• $AMBERHOME/update_amber --show-applied-patches : Shows details about each patch that has been
applied (including third-party patches)
• $AMBERHOME/update_amber --show-unapplied-patches : Shows details about each patch that has been
downloaded but not yet applied.
• $AMBERHOME/update_amber --remove-unapplied : Deletes all patches that have been downloaded but
not applied. This will force update_amber to download a fresh copy of that patch.
• $AMBERHOME/update_amber --update-to AmberTools/#,Amber/# : This command will apply all patches
necessary to bring AmberTools up to a specific version and Amber up to a specific version. Note, no updates will ever be reversed using this command. You may specify only an AmberTools version or an Amber
version (or both, comma-delimited). No patches are applied to an omitted branch.
• $AMBERHOME/update_amber --revert-to AmberTools/#,Amber/# : This command does the same as
--update-to described above, except it will only reverse patches, never apply them.
update_amber will also provide varying amounts of information about each patch based on the verbosity setting.
The verbose level can be set with the --verbose flag and can be any integer between 0 and 4, inclusive. The
default verbosity level changes based on how many updates must be described. If only a small number of updates
need be described, all details are printed out. The more updates that must be described, the less information is
printed. If you manually set a value on the command-line, it will override the default. These values are described
below (each level prints all information from the levels before plus additional information):
• 0: Print out only the name of the update file (no other information)
• 1: Also prints out the name of the program(s) that are affected
• 2: Also prints out the description of the update written by the author of that update.
• 3: Also prints the name of the person that authored the patch and the date it was created.
• 4: Also prints out the name of every file that is modified by the patch.

2.2. Contacting the developers

2.1.3. Internet Connection Settings
If update_amber ever needs to connect to the internet, it will check to see if http://ambermd.org can be contacted
within 10 seconds. If not, it will report an error and quit. If your connection speed is particularly slow, you can
lengthen this timeout via the --timeout command-line flag (where the time is given in seconds).
Proxies By default, update_amber will attempt to contact the internet through the same mechanism as
programs like wget and curl. For users that connect to the internet through a proxy server, you can either set the
http_proxy environment variable yourself (in which case you can ignore the rest of the advice about proxies
here), or you can configure update_amber to connect to the internet through a proxy. To set up update_amber to
connect to the internet through a proxy, use the following command:
$AMBERHOME/update_amber --proxy=

You can often find your proxy address from your IT department or the preferences in your favorite (configured)
web browser that you use to surf the web. If your proxy is authenticated, you will also need to set up a user:
$AMBERHOME/update_amber --proxy-user=

If you have set up a user name to connect to your proxy, then you will be asked for your proxy password the first
time update_amber attempts to utilize an online resource. (For security, your password is never stored, and will
need to be retyped every time update_amber runs).
You can clear all proxy information using the --delete-proxy command-line flag—this is really only necessary
if you no longer need to connect through any proxy, since each time you configure a particular proxy user or server
it overwrites whatever was set before.
If you would like to download Amber patches from another website or even a folder on a local filesystem,
you can use the --amber-updates and --ambertools-updates command-line flags to specify a particular web
address (must start with http://) or a local folder (use an absolute path). You can use the --reset-remotes
command-line flag to erase these settings and return to the default Amber locations on http://ambermd.org.
If you set up online mirrors and never plan on connecting directly to http://ambermd.org, you can change
the web address that update_amber attempts to connect to when it verifies an internet connection using the
--internet-check command-line option.
Mirrors

2.2. Contacting the developers
Please send suggestions and questions to amber@ambermd.org. You need to be subscribed to post there; to
subscribe, go to http://lists.ambermd.org/mailman/listinfo/amber. You can unsubscribe from this mailing list on
the same site.

Part II.

Amber force fields

3. Molecular mechanics force fields
Amber is designed to work with several simple types of force fields, although it is most commonly used with
parametrizations developed by Peter Kollman and his co-workers and “descendents”. The traditional parametrization uses fixed partial charges, centered on atoms. The current recommended force field for proteins and nucleic
acids is ff14SB, although ff03.r1 or ff14ipq (both for proteins) are also commonly used; descriptions are given below. Less commonly used modifications add polarizable dipoles to atoms, so that the charge description depends
upon the environment; such potentials are called “polarizable” or “non-additive”. Examples are ff02 and ff02EP:
the former has atom-based charges (as in the traditional parametrization), and the latter adds in off-center charges
(or “extra points”), primarily to help describe better the angular dependence of hydrogen bonds. Major updates to
these are under development, but were not ready for the Amber14 release in April, 2014.
An alternative is to use force fields originally developed for the CHARMM or tinker (AMOEBA) codes; these
require a different setup procedure, which is described in Sections 3.11 (for CHARMM) and Chapter 16 and
Section 18.8 (for AMOEBA). Force fields for carbohydrates and lipids are also discussed below. Chapter 13
provides a basic introduction to force fields, along with details of how the parameters are encoded in Amber files.

3.1. Specifying which force field you want in LEaP
Various combinations of the above files make sense, and we have moved to an “ff” (force field) nomenclature
to identify these; examples would then be ff94 (which was the default in Amber 5 and 6), ff99, etc. The most
straightforward way to specify which force field you want is to use one of the leaprc files in
$AMBERHOME/dat/leap/cmd. The syntax is
xleap -s -f

Here, the -s flag tells LEaP to ignore any leaprc file it might find, and the -f flag tells it to start with commands
for some other file. Here are the combinations we support and recommend:
File name
leaprc.ff14SB
leaprc.ff14ipq
leaprc.ff03.r1
leaprc.ff03ua
leaprc.ff02
leaprc.gaff
leaprc.GLYCAM_06j-1
leaprc.GLYCAM_06EPb
leaprc.lipid11
leaprc.lipid14

Original Charge Scheme
Cornell et al., 1994
Cerutti et al., 2013
Duan et al. 2003
Yang et al. 2003
reduced charges
none
Woods et al.
"
Skjevik et al., 2012
"

Parameters
see Sec. 3.2
see Sec. 3.3
parm99.dat+frcmod.ff03
parm99.dat+frcmod.ff03+frcmod.ff03ua
parm99.dat+frcmod.ff02pol.r1
gaff.dat
GLYCAM_06j.dat
GLYCAM_06EPb.dat
lipid11.dat [16]
lipid14.dat [17]

Notes: The second column indicates the origin of the charge scheme used in the force field.
1. There is no default leaprc file. If you make a link from one of the files above to a file named leaprc, then that
will become the default. For example:
cd $AMBERHOME/dat/leap/cmd
ln -s leaprc.ff14SB leaprc

will provide a good default for many users; after this you could just invoke tleap or xleap without any
arguments, and it would automatically load the ff14SB force field. A file named leaprc in the working
directory overrides any other such files that might be present in the search path.

3. Molecular mechanics force fields
2. Most of the choices in the above table are for additive (non-polarizable) simulations; you should use saveAmberParm
to save the prmtop file.
3. The ff02 entries in the above table are for non-additive (polarizable) force fields. Use saveAmberParmPol to
save the prmtop file. Note that POL3 is a polarizable water model, so you need to use saveAmberParmPol
for it as well.
4. There is also a leaprc.gaff file, which sets you up for the GAFF (“general” Amber) force field. This is
primarily for use with Antechamber (see Chapter 14), and does not load any topology files.
5. There are some leaprc files for older force fields in the $AMBERHOME/dat/leap/cmd/oldff directory. We no
longer recommend these combinations, but we recognize that there may be reasons to use them, especially
for comparisons to older simulations. See Section 3.12.
6. Nucleic acid residues in ff14SB use the new (version 3) PDB nomenclature: “DC” is used for deoxy-cytosine,
and “C” for cytosine in RNA, etc. Earlier force fields (which are not recommended!) use “RC” for the RNA
version. If you want a single, nucleoside, use “CN”, etc. For a single nucleotide, use the following command
in LEaP:
cnuc = sequence { OHE C3 }

and analogs for other bases. Note that this will construct a protonated 5’ phosphate group, which may not
be what you want.
7. The General Amber Force Field (gaff) is discussed in Chap. 14.

3.2. The ff14SB force field
leaprc.ff14SB
parm10.dat
frcmod.ff14SB
amino12.lib
amino12nt.lib
amino12ct.lib
nucleic12.lib
leaprc.phosaa10
leaprc.modrna08
leaprc.ff14SBonlysc
frcmod.ff99SB14

This will load the files listed below
ff10 force field parameters
ff14SB modifications to parm10.dat
topologies and charges for amino acids
same, for N-terminal amino acids
same, for C-terminal amino acids
topologies and charges for nucleic acids
This will load parameters for phosphorylated amino acids
This will load parameters for modified RNA nucleotides
This is the same as leaprc.ff14SB, but will additionally load:
ff99SB backbone parameters with ff14SB atom types

3.2.1. Proteins
ff14SB is a continuing evolution of the ff99SB force field, primarily developed in the Simmerling group at Stony
Brook University.[18] Several groups had noticed that the older ff94 and ff99 parameter sets did not provide a good
energy balance between helical and extended regions of peptide and protein backbones. Another problem is that
many of the ff94 variants had incorrect treatment of glycine backbone parameters. ff99SB improved this behavior,
presenting a careful reparametrization of the backbone torsion terms in ff99 and achieves much better balance of
four basic secondary structure elements (PP II, β , αL , and αR ). A detailed explanation of the parametrization as
well as an extensive comparison with many other variants of fixed-charge Amber force fields is given in the reference above. Briefly, dihedral term parameters were obtained through fitting the energies of multiple conformations
of glycine and alanine tetrapeptides to high-level ab initio QM calculations. We have shown that this force field
provides much improved proportions of helical versus extended structures. In addition, it corrected the glycine
sampling and should also perform well for β -turn structures, two things which were especially problematic with
most previous Amber force field variants.

3.2. The ff14SB force field
Since 2006, a number of limitations of the ff99SB parameter sets became evident, and a new round of parameter
optimization was undertaken. The changes mainly involve torsional parameters for the backbone and side chains.
For backbones, experimental scalar coupling data for small solvated peptides became available [19] against which
ff99SB was compared.[20] As ff99SB backbone dihedrals were fit based on gas-phase quantum data, we felt that
slight empirical adjustments were worth pursuing. This was done to improve agreement with scalar coupling data,
and we observed that this also improved stabilities of helical peptides.
The side chain dihedral parameters of ff99SB were the same as those of ff99. Residues such as isoleucine,
leucine, aspartate, and asparagine (cf. ff99SB-ILDN) sample conformations different from those indicated by experiments. We therefore calibrated the dihedral corrections of the amino acid side chains against ab initio quantum
mechanical energy surfaces. As ff14SB is an additive model, a key objective was to minimize dependence of
ff14SB side chain parameters on particular backbone conformations. Therefore, side chain corrections were raised
against potential energy surfaces including multiple backbone conformations. Moreover, the method of generating fitting data was adjusted to minimize backbone-dependence, including restraint of all backbone dihedrals and
re-optimization of bonds and angles with each model. Whereas ff12SB also included restraints on all side chain dihedrals, restraint of one dihedral per side chain bond in ff14SB was found to further reduce backbone-dependence.
Together with new corrections for the backbone and the four amino acids addressed in ff99SB-ILDN, this work
offers updated side chain dihedral corrections for lysine, arginine, glutamate, glutamine, methionine, serine, threonine, valine, tryptophan, cysteine, phenylalanine, tyrosine, and histidine. ff14SB enhances reproduction of experimentally indicated geometries over ff99SB.
ff14SBonlysc, where sc stands for side chains, includes ff99SB backbone parameters with updated side chain
parameters that were derived from ab initio quantum mechanics calculations (as were the ff99SB backbone corrections). This model is slightly different from ff14SB, which includes the ff14SBonlysc parameters as well as a
small empirical correction to backbone parameters that was designed to improve agreement between NMR data
and simulations in TIP3P water for short peptides. We are currently exploring whether this empirical correction
also improves simulations in other water models, such as the GBneck2 (igb=8) model. [21] Currently, it appears
that igb=8 may work best with the fully quantum mechanics-based dihedral parameters included in ff14SBonlysc.
Simulations performed in explicit water most likely benefit from the empirical corrections included in ff14SB.

3.2.2. Nucleic acids
As with proteins, many features of the current force fields, including partial atomic charges, Lennard-Jones
parameters, and most bond and angle terms, date back to force fields developed in the 1990’s, and overviews of
this work are available.[22, 23] The next breakthrough’s in the Amber nucleic acid force field development came
from observations from relatively longer simulations on the 50-100 ns time scale in the early 2000’s.[24, 25] These
simulations found systematic over-population of γ = trans backbone geometries in nucleic acids. High level QM
calculations were performed on models of sugars and phosphates, specifically a sugar-phosphate model[26] and
a sugar-phosphate-sugar model,[27] which ultimately led to the ff99-bsc0 parameterization.[26] For simulation
of canonical DNA and RNA structures, the ff99-bsc0 parameterization has proven rather successful. For noncanonical structures, particular those with loops or bulges, or χ flips, some anomalies have been noted. With
RNA, incorrect loop geometries, backbone sub-state populations and sugar pucker populations were observed in
longer simulations. In addition to not being able to always maintain south puckers where found in RNA structures,
multiple groups noticed a tendency for the RNA backbone to shift, putting χ into the high-anti region which leads
to an opening of the duplex structure into a ladder-like configuration. Again, QM methods at various levels were
employed to improve the χ distribution using relevant model systems. The most tested χ modifications are the
“OL” modifications used in ff14SB.[28, 29] An alternative available with Amber is the Yildirim χ modifications
(and also related modifications called TOR which alter ε/ζ as well)[30–32], and a systematic assessment and
validation of these newer χ modifications is underway on a large series of RNA tetraloop structures. Note that
small changes to a particular dihedral may lead to alteration in properties of related dihedrals, and may have
unintended consequences. For example, the ff99-bsc0 modifications tend to lock RNA sugar puckers mainly in the
north, even with nucleotides in particular sequence contexts that prefer southern conformations. Moreover, the χ
modifications tend to further destabilize γ = trans. This suggests that to reliably improve the nucleic acid dihedrals,
a more systematic approach across many dihedrals with simultaneous fitting may be more appropriate. Moreover,
we no longer fully support the idea that parameters are transferable between DNA and RNA, or between purines

3. Molecular mechanics force fields
Name
ff94
ff98
ff99
bsc0
χOL3
ε/ζ OL1
χOL4

Modification
Original force field file
Modified charge set
Updated charge set
Barcelona α/γ backbone modification
χmodification tuned for RNA
ε/ζ modification for DNA
χ modification tuned for DNA

Notes
Obsolete
Obsolete
Current
[26]
[28, 29]
improvement for DNA, no effects for RNA [34]
[33]

Table 3.1.: Force field name and modifications for simulating nucleic acids.

and pyrimidines. For example, the ff99-OL modifications (with or without ff99-bsc0) improve the modeling of
RNA, but lead to issues with DNA, most notably with quadruplex structures. Therefore recent work has focused
on separate χ modifications for DNA.[33]
A new set of parameters for the ε/ζ dihedral torsion for DNA have been developed using QM methods that
include the solvation effects implicitly.[34] This set of parameters called ε/ζ OL1 (not to be confused with the
χ modification) have been tested with several double-stranded DNA systems including the Dickerson-Drew dodecamer, A-tracts, CG-rich duplexes, Z-DNA and G-quadruplexes. The modification increases the population of
BII substate by stabilizing the ε/ζ = g-/t state. Additionally, higher values for the helical twist are present in the
tested systems. In combination with the χ modification for DNA (χ OL4, [33]), the force field generates structures that suggest a better agreement with NMR data. The reader should pay careful attention to the use of the χ
modifications, since the naming convention of the authors is the same for RNA and DNA. Details of the different
modifications available for DNA are presented in Table 3.1.
Considering the multiple small modifications available, it is very important to do extensive testing and benchmarking of any simulations done with these variations. This is ongoing work from different research groups
involved in the simulation of nucleic acids. For DNA, we recommend the combination of ff99 + bsc0 + ε/ζ OL1
+ χ OL4. (The appropriate leaprc file will be made available soon as an update to AmberTools.) For RNA, we
recommend using ff14SB, which is identical to ff12 and is a combination of ff99 + bsc0 + χOL3.

3.2.3. Modified amino acids and nucleotides
Parameters for phosphorylated amino acids [35] can be obtained by typing “source leaprc.phophaa10” in
your leap.in file. Many post-translational modifications are also available at http://selene.princeton.edu/FFPTM/.
Parameters for common modifications for RNA nucleotides [36]can by loaded with “source leaprc.modrna08”.
Pointers to other sets of Amber-compatible force fields may be found at the Amber web site, http://ambermd.org/.

3.3. The ff14ipq protein force field
leaprc.ff1ripq
parm14ipq.dat
amino14ipq.lib
aminont14ipq.lib
aminoct14ipq.lib

This will load the files listed below
force field parameters
topologies and charges for amino acids
same, for N-terminal amino acids
same, for C-terminal amino acids

The ff14ipq force field features a complete rederivation of torsion potentials and nonbonded parameters within
the family of Amber fixed-charge force fields. Charges were derived by an extended IPolQ method, [37] which
we deemed necessary in order to simultaneously fit torsion parameters. In the extended methodology, two sets of
charges are fitted: one for the systems in vacuum, the other for systems in the condensed phase. In this manner, the
extended IPolQ methodology derives the condensed phase charges by fitting to the average electrostatic potential
of polarized and unpolarized molecules as it originally did, but also derives a set of charges specifically to fit
the data for unpolarized molecules. The two sets of charges are derived in the same linear least squares fitting

3.4. The Duan et al. (2003) force field
problem, with restraint equations weakly coupling the corresponding charges together. This creates charge sets
for each phase related by a minimal perturbation, which can be assumed to be the effective, average polarization
of the molecules when they enter solution. The charge set appropriate to the vacuum phase is then used when
fitting torsion potentials to vacuum phase quantum mechanical energies, and the torsion potentials are transferred
directly for use with the condensed-phase charge set in actual simulations, following the earlier assumption that the
effective polarization of the molecules, and thereby any energetic consequences of entering the condensed phase,
are captured in the charge perturbation.
As a consequence of the fitting protocol, it is most appropriate to use the TIP4P-Ew water model with ff14ipq.
However, because nearly all fixed-charge water models will produce similar reaction field potentials within solvated
biomolecules, it is likely that other common water models that are compatible with long-ranged electrostatics can
also be paired with the protein model.
In the original IPolQ derivation, some changes to the polar atom Lennard-Jones σ radii in ff94 were introduced
to improve agreement with experimental hydration free energies of amino acid side chain analogs; these are carried
over to ff14ipq. However, these larger radii frustrated fitting of internal energies, and were therefore applied only
between the polar atoms and surrounding water.
Like the charge set, the torsion potentials were also derived by an iterative scheme in which the model learns
from its performance in previous generations, building to a well converged parameter set. Preliminary testing on
α-helical, β -sheet, and small globular proteins indicates that the force field yields stable simulations of major
protein structures and even captures the instability of some small peptide systems observed experimentally. The
torsion potentials were derived by fitting against over 60,000 gas-phase energies of dipeptides, tripeptides, and
tetrapeptides calculated at the MP2/cc-pVTZ level. The original fitting set consisted of 28,000 structures, but
artifacts in subsequent simulations led us to examine the effect of reintroducing structures optimized by the fitted
force field back into its own fitting data. Most of the original 28,000 structures, which were obtained usng restraints
to sample particular torsion angles, were energy minimized without restraints by the first generation of the ff14ipq
force field. Energies of the resulting structures were calculated at the MP2/cc-pVTZ level and compared to the
ff14ipq predictions; this revealed many minor inconsistencies and some gaping discrepancies created by holes
in the parameter sampling. When ff14ipq predictions were in error, it was nearly always because the ff14ipqoptimized structures were more stable than MP2/cc-pVTZ would predict. In extreme cases the over-prediction
exceeded 15 kcal/mol for amino acid dipeptides. By incorporating the new energies and structures into the fitting
data, nearly all of these errors were fixed in the second generation of the torsion parameters, and by the third (and
final) generation, ff14ipq was able to maintain agreement with its MP2 benchmark while optimizing small peptide
structures.
Beyond this, there has so far been no "tinkering" with ff14ipq parameters: all of the numbers come straight from
quantum mechanical data. We would like to think that the promising results we have seen on small oligopeptides
and globular proteins is a result of how carefully we crafted the data set, but ff14ipq must gain acceptance over a
much longer process as other investigators use the force field to model proteins in solution. We anticipate that there
will now be a lineage of models based on ff14ipq, as there has been with the Cornell charge and Lennard_Jones
parameters, [38], whose current incarnation is ff14SB, described above. We have plans to incorporate the results
of future simulations into parameter refinement while explicitly maintaining a basis in quantum mechanical energies. If successful, ff14ipq may be the last major force field of its kind–future models will add virtual sites and
polarization–but the techniques we developed to fit this ab-initio force field should prove useful in making future
models.

3.4. The Duan et al. (2003) force field
frcmod.ff03
all_amino03.in
all_aminont03.in
all_aminoct03.in

For proteins: changes to parm99.dat, primarily in the
phi and psi torsions.
Charges and atom types for proteins
For N-terminal amino acids
For C-terminal amino acids

The ff03 force field [39, 40] is a modified version of ff99 (described below). The main changes are that charges
are now derived from quantum calculations that use a continuum dielectric to mimic solvent polarization, and that

3. Molecular mechanics force fields
the φ and ψ backbone torsions for proteins are modified, with the effect of decreasing the preference for helical
configurations. The changes are just for proteins; nucleic acid parameters are the same as in ff99.
The original model used the old (ff94) charge scheme for N- and C-terminal amino acids. This was what was
distributed with Amber 9, and can still be activated by using oldff/leaprc.ff03. More recently, new libraries for the
terminal amino acids have been constructed, using the same charge scheme as for the rest of the force field. This
newer version (which is recommended for all new simulations) is accessed by using leaprc.ff03.r1.

3.5. The Yang et al. (2003) united-atom force field
frcmod.ff03ua

uni_amino03.in
uni_aminont03.in
uni_aminoct03.in

For proteins: changes to parm99.dat, primarily in the
introduction of new united-atom carbon types and new
side chain torsions.
Amino acid input for building database
NH3+ amino acid input for building database.
COO- amino acid input for building database.

The ff03ua force field [41] is the united-atom counterpart of ff03. This force field uses the same charging scheme as
ff03. In this force field, the aliphatic hydrogen atoms on all amino acid side-chains are united to their corresponding
carbon atoms. The aliphatic hydrogen atoms on all alpha carbon atoms are still represented explicitly to minimize
the impact of the united-atom approximation on protein backbone conformations. In addition, aromatic hydrogens
are also explicitly represented. Van der Waals parameters of the united carbon atoms are refitted based on solvation
free energy calculations. Due to the use of an all-atom protein backbone, the φ and ψ backbone torsions from ff03
are left unchanged. The sidechain torsions involving united carbon atoms are all refitted. In this parameter set,
nucleic acid parameters are still in all atom and kept the same as in ff99.

3.6. Force fields related to semi-empirical QM
ParmAM1 and parmPM3 are classical force field parameter sets that reproduce the geometry of proteins
minimized at the semi-empirical AM1 or PM3 level, respectively.[42] These new force fields provide an
inexpensive, yet reliable, method to arrive at geometries that are more consistent with a semi-empirical treatment
of protein structure. These force fields are meant only to reproduce AM1 and PM3 geometries (warts and all) and
were not tested for use in other instances (e.g., in classical MD simulations, etc.) Since the minimization of a
protein structure at the semi-empirical level can become cost-prohibitive, a “preminimization” with an
appropriately parametrized classical treatment will facilitate future analysis using AM1 or PM3 Hamiltonians.

3.7. The GLYCAM force fields for carbohydrates and lipids
GLYCAM06 is a consistent and transferable parameter set for modeling carbohydrates,[43] lipids,[44] and
glycoconjugates.[45, 46] The core philosophy of the force field development process is that parameters should
be: (1) be transferable to all carbohydrate ring formations and sizes, (2) be self-contained and therefore readily
transferable to many quadratic force fields, (3) not require specific atom types for α- and β -anomers, (4) be readily
extendible to carbohydrate derivatives and other biomolecules, (5) be applicable to monosaccharides and complex
oligosaccharides, and (6) be rigorously assessed in terms of the relative accuracy of its component terms.
When combining GLYCAM06 with AMBER parameters for other biomolecules, parameter orthogonality is
ensured by assigning unique atom types for GLYCAM. In order to facilitate combining GLYCAM06 with other
AMBER parameter sets for other biomolecules, a variation on the GLYCAM atom types has been introduced in
which the new name consists of an uppercase letter followed by second character, either a number or lowercase
letter. For example the GLYCAM “CG” atom type has been changed to “Cg”; “HO” is now represented as “Ho”,
and so forth.

3.7. The GLYCAM force fields for carbohydrates and lipids
As soon as new parameters are generated, or alterations are made to existing parameters, a new version of
GLYCAM is released. Updated versions that introduce new functionality are denoted using a letter suffix (i.e.
GLYCAM06a, 06b, etc.). Each release is accompanied with an associated text file that summarizes the new
functionality or alteration. For example, a particularly important update, released in GLYCAM06e, altered the
endo-anomeric torsion term (Cg-Os-Cg-Os) in order to more accurately reproduce the populations arising from
ring flips (4 C1 to 1 C4 etc.). This particular case suggested the need to be able to independently characterize the
exo- and endo-anomeric effect, which was achieved by assigning different atom types (Oa and Oe) to represent the
endo-anomeric and exo-anomeric oxygen atoms, respectively.
In another important update (GLYCAM06g), a small van der Waals term was applied to all hydroxyl hydrogen
atoms (Ho) to address a rare, but catastrophic, situation that can arise during MD simulations. In certain carbohydrate (and potentially other) configurations, a hydroxyl proton may be structurally constrained to being very close
to a carboxylate moiety. During an MD simulation of such a system, an oscillatory motion can begin between the
hydroxyl proton and the negative charge site, leading ultimately to failure of the simulation as the proton collapses
onto the negatively charged moiety. The small van der Waals term (Ho, R* = 0.2000 Å, ε = 0.0300 kcal/mol)
is just large enough to add sufficient repulsion to prevent this behavior, while not being large enough to perturb
properties such as hydrogen bond lengths.
The GLYCAM force field family, especially, GLYCAM06, has been extensively employed in simulations of
biomolecules by the larger scientific community.[47–50] The updated GLYCAM parameters and documentation
are available for download at the GLYCAM-Web site (www.glycam.org). Also available on the website are tools
for simplifying the generation of structure and topology files for performing simulations of oligosaccharides,
glycoconjugates and glycoproteins. GLYCAM-Web has been integrated into several glycomics databases, such as
the Consortium for Functional Glycomics (www.functionalglycomics.org).

GLYCAM06 force field
Always check glycam.org/params for more recent versions and new functionalities.
GLYCAM_06j.dat
GLYCAM_06j-1.prep
GLYCAM_lipids_06h.prep
leaprc.GLYCAM_06j-1

Parameters for oligosaccharides
Structures and charges for glycosyl residues
Structures and charges for some lipid residues
LEaP configuration file for use of GLYCAM06
with carbohydrates alone or in combination
with the ff12SB or ff14SB force field.
GLYCAM_amino_06j_12SB.lib
Glycoprotein libraries compatible
GLYCAM_aminoct_06j_12SB.lib with ff12SB and ff14SB.
GLYCAM_aminont_06j_12SB.lib

GLYCAM06EP force field using lone pairs (extra points)
GLYCAM_06EPb.dat
GLYCAM_06EPb.prep
leaprc.GLYCAM_06EPb

Parameters for oligosaccharides
Structures and charges for glycosyl residues
LEaP configuration file for GLYCAM-06EP

GLYCAM Force Field Parameters Download Page
http://www.glycam.org/params

GLYCAM_06j-1.prep contains prep entries for all carbohydrate residues and GLYCAM_lipids_06h.prep contains
prep entries for lipid residues. GLYCAM_06EPb.prep contains prep entries for all carbohydrate residues available
for modeling with extra points.
For linking glycans to proteins, libraries containing modified amino acid residues (Ser, Thr, Hyp, and Asn) must
be loaded. To build a glycoprotein using ff12SB or ff14SB, GLYCAM_amino_06j_12SB.lib GLYCAM_aminont_06j_12SB.lib
and GLYCAM_aminoct_06j_12SB.lib must be loaded and the desired protein force field must also be loaded.
Amino acid libraries designed for linking carbohydrates modeled with extra points are not currently available.

3. Molecular mechanics force fields
Version

Release Date

Contributors

15 Feb., 2014

BLF

27 Aug., 2013

AKN

20 Oct., 2010

MBT, BLF

20 Oct., 2010

MBT

3 Feb., 2009

MBT

28 May, 2008

MBT

d
c

12 May, 2008
21 Feb., 2008

SPK, MBT, ABY
MBT, ABY

10 Jan., 2008

MBT, ABY

24 Apr., 2005

ABY

Change Summary
Modified all parameters to be compatible with ff10, ff13,
ff12SB and ff14SB. These files may not be compatible
with older protein and nucleic acid force fields.
Added two new monosaccharides to the prep file.
*Changed atom type naming to be orthogonal to other
force fields. Added HO van der Waals parameters. Set
protein-related parameter values to their parm99
counterparts. Updated N-sulfation parameters.
* 1,4-scaling terms added to parameter file. Angle and
torsion updates for pyranose rings, N-sulfate, phosphate
and sialic acid.
* Corrected a typo in O-Acetyl term
* Updated glycosidic linkage terms to optimize ring
puckering in pyranoses
Terms for thiol glycosidic linkages
* Additional (published) terms for lipid simulations[44]
Alkanes, alkenes, amide and amino groups for lipid
simulations[44]
Sulfates & phosphates for carbohydrates

Table 3.2.: Version change summary for the GLYCAM-06 force field. *Previously released parameters were
changed. See full release notes at glycam.org/params. SPK: Sameer P. Kawatkar. MBT: Matthew
B. Tessier. ABY: Austin B. Yongye. BLF: B. Lachele Foley. AKN: Anita K. Nivedha

3.7.1. File versioning
Beginning on 15 September, 2011, a new versioning system was implemented for Glycam parameters. Files
produced before that date will not necessarily conform to the new system. In the new system, all files containing
parameters are versioned. Users should check their contents and replace them with recent versions as appropriate.
The new versioning system employs letters and numbers. If a parameter set contains new functionality (e.g.,
the addition of new parameters) or fundamental changes (e.g., atom type name reassignments), a letter will be
appended to its name. If the new version contains corrections (e.g., for typographical errors), its name will be
appended with a number. See glycam.org/params for more documentation and examples.
Researchers are also encouraged to read the version change documentation available on the GLYCAM Parameters download page under “Documents.” In this document, the changes specific to each version release are detailed.
The changes are also summarized here in Table 3.2.

3.7.2. Atom type name changes
Beginning with versions g, Glycam atom type names will adopt a standard designed to keep them from overlapping with other force fields. In most cases, Glycam’s type names will consist of two characters, one upper-case
followed by one lower-case. Because of this, leaprc files, lib files and prep files from versions prior to g will be
incompatible with current versions.
Note that some type names will not reflect the new Glycam type standard, despite being present in the Glycam
force field files, for example in the files for linking glycans to amino acid residues. In these cases, Glycam will use
the type name appropriate to the external force field. Parameters will be introduced only to the extent necessary
to provide a link between the force fields. Since the associated parameters will also include Glycam types, they
should only affect the intersections between the two force fields.
Beginning with versions j, atom type names for linking to amino acids are compatible with ff10, ff13, ff12SB
and ff14SB. Older versions of protein and nucleic acid force fields might not be compatible.

3.7. The GLYCAM force fields for carbohydrates and lipids

3.7.3. General information regarding parameter development
In GLYCAM-06,[43] the torsion terms have now been entirely developed by fitting to quantum mechanical
data (B3LYP/6-31++G(2d,2p)//HF/6-31G(d)) for small-molecules. This has converted GLYCAM-06 into an additive force field that is extensible to diverse molecular classes including, for example, lipids and glycolipids. The
parameters are self-contained, such that it is not necessary to load any AMBER parameter files when modeling
carbohydrates or lipids. To maintain orthogonality with AMBER parameters for proteins, notably those involving
the CT atom type, tetrahedral carbon atoms in GLYCAM are called Cg (C-GLYCAM, CG in previous releases).
Thus, GLYCAM and AMBER may be combined for modeling carbohydrate-protein complexes and glycoproteins.
More information on atom type names is available in 3.7.2 . Because the GLYCAM-06 torsion terms were derived by fitting to data for small, often highly symmetric molecules, asymmetric phase shifts were not required
in the parameters. This has the significant advantage that it allows one set of torsion terms to be used for both
α- and β -carbohydrate anomers regardless of monosaccharide ring size or conformation. A molecular development suite of more than 75 molecules was employed, with a test suite that included carbohydrates and numerous
smaller molecular fragments. The GLYCAM-06 force field has been validated against quantum mechanical and
experimental properties, including: gas-phase conformational energies, hydrogen bond energies, and vibrational
frequencies; solution-phase rotamer populations (from NMR data); and solid-phase vibrational frequencies and
crystallographic unit cell dimensions.

3.7.4. Scaling of electrostatic and nonbonded interactions
As in previous versions of GLYCAM,[2] the parameters were derived for use without scaling 1-4 non-bonded
and electrostatic interactions. Thus, in sander, pmemd, and so on, the simulation parameters scnb and scee should
typically be set to unity. We have shown that this is essential in order to properly treat internal hydrogen bonds,
particularly those associated with the hydroxymethyl group, and to correctly reproduce the rotamer populations
for the C5-C6 bond.[51] Beginning with Amber 11, it is now possible to employ mixed scaling of the scnb and
scee parameters. Anyone wishing to simulate systems containing both carbohydrates and proteins should use the
new mixed scaling capability. To do this, any scaling factors that differ from the default must be included in the
parameter file. Beginning with the GLYCAM_06g parameter file shipped with Amber 11, these factors are already
included. Anyone wishing to employ earlier parameter sets must modify the files.

3.7.5. Development of partial atomic charges
As in previous versions of GLYCAM, the atomic partial charges were determined using the RESP formalism,
with a weighting factor of 0.01,[43, 52] from a wavefunction computed at the HF/6-31G(d) level. To reduce
artifactual fluctuations in the charges on aliphatic hydrogen atoms, and on the adjacent saturated carbon atoms,
charges on aliphatic hydrogens (types HC, H1, H2, and H3) were set to zero while the partial charges were fit
to the remaining atoms.[53] It should be noted that aliphatic hydrogen atoms typically carry partial charges that
fluctuate around zero when they are included in the RESP fitting, particularly when averaged over conformational
ensembles.[43, 54] In order to account for the effects of charge variation associated with exocyclic bond rotation,
particularly associated with hydroxyl and hydroxylmethyl groups, partial atomic charges for each sugar were
determined by averaging RESP charges obtained from 100 conformations selected evenly from 10-50 ns solvated
MD simulations of the methyl glycoside of each monosaccharide, thus yielding an ensemble averaged charge
set.[43, 54]

3.7.6. Carbohydrate parameters for use with the TIP5P water model
In order to extend GLYCAM to simulations employing the TIP-5P water model, an additional set of carbohydrate
parameters, GLYCAM-06EP, has been derived in which lone pairs (or extra points, EPs) have been incorporated
on the oxygen atoms.[55] The optimal O-EP distance was located by obtaining the best fit to the HF/6-31g(d)
electrostatic potential. In general, the best fit to the quantum potential coincided with a negligible charge on the
oxygen nuclear position. The optimal O-EP distance for an sp3 oxygen atom was found to be 0.70 Å; for an sp2
oxygen atom a shorter length of 0.3 Åwas optimal. When applied to water, this approach to locating the lone pair

3. Molecular mechanics force fields

Carbohydrate
Arabinose
Lyxose
Ribose
Xylose
Allose
Altrose
Galactose
Glucose
Gulose
Idose
Mannose
Talose
Fructose
Psicose
Sorbose
Tagatose
Fucose
Quinovose
Rhamnose
Galacturonic Acid
Glucuronic Acid
Iduronic Acid
N-Acetylgalactosamine
N-Acetylglucosamine
N-Acetylmannosamine
Neu5Ac
KDN
KDO

Pyranose
α/β , D / L
yes
yes
yes
yes
yes
yes
yes
yes
yes
a
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes, b
a,b
a,b

Furanose
α/β , D / L
yes
yes
yes
yes

a
a

yes
yes
yes
yes

yes,b
a,b
a,b

Table 3.3.: Current Status of Monosaccharide Availability in GLYCAM. (a) Currently under development. (b) Only
one enantiomer and ring form known.

3.7. The GLYCAM force fields for carbohydrates and lipids
positions and assigning the partial charges yielded a model that was essentially indistinguishable from TIP-5P.
Therefore, we believe this model is well suited for use with TIP-5P.[55] The new files are named 06EP (originally
04EP), as they have been corrected for numerous typographical errors and updated to match current naming and
residue structure conventions.

3.7.7. Carbohydrate Naming Convention in GLYCAM
In order to incorporate carbohydrates in a standardized way into modeling programs, as well as to provide a standard for X-ray and NMR protein database files (pdb), we have developed a three-letter code nomenclature. The
restriction to three letters is based on standards imposed on protein data bank (PDB) files by the RCSB PDB Advisory Committee (www.rcsb.org/pdb/pdbac.html), and for the practical reason that all modeling and experimental
software has been developed to read three-letter codes, primarily for use with protein and nucleic acids.
As a basis for a three-letter PDB code for monosaccharides, we have introduced a one-letter code for monosaccharides (Table 3.4).[56] Where possible, the letter is taken from the first letter of the monosaccharide name.
Given the endless variety in monosaccharide derivatives, the limitation of 26 letters ensures that no one-letter
(or three-letter) code can be all encompassing. We have therefore allocated single letters firstly to all 5- and 6carbon, non-derivatized monosaccharides. Subsequently, letters have been assigned on the order of frequency of
occurrence or biological significance.
Using three letters (Tables 3.5 to 3.7), the present GLYCAM residue names encode the following content:
carbohydrate residue name (Glc, Gal, etc.), ring form (pyranosyl or furanosyl), anomeric configuration (α or β ,
enantiomeric form (D or L) and occupied linkage positions (2-, 2,3-, 2,4,6-, etc.). Incorporation of linkage position
is a particularly useful addition, since, unlike amino acids, the linkage cannot otherwise be inferred from the
monosaccharide name. Further, the three-letter codes were chosen to be orthogonal to those currently employed
for amino acids.

3. Molecular mechanics force fields

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

Carbohydratea
D -Arabinose
D -Lyxose
D -Ribose
D -Xylose
D -Allose
D -Altrose
D -Galactose
D -Glucose
D -Gulose
D -Idose
D -Mannose
D -Talose
D -Fructose
D -Psicose
D -Sorbose
D -Tagatose
D -Fucose (6-deoxy D -galactose)
D -Quinovose (6-deoxy D -glucose)
D -Rhamnose (6-deoxy D -mannose)
D -Galacturonic Acid
D -Glucuronic Acid
D -Iduronic Acid
D -N-Acetylgalactosamine
D -N-Acetylglucosamine
D -N-Acetylmannosamine
N-Acetyl-neuraminic Acid
KDN
KDO
N-Glycolyl-neuraminic Acid

One letter codeb
A
D
R
X
N
E
L
G
K
I
M
T
C
P
Bd
J
F
Q
H
Od
Zd
Ud
Vd
Yd
Wd
Sd
KNc,d
KOc,d
SGc,d

Common Abbreviation
Ara
Lyx
Rib
Xyl
All
Alt
Gal
Glc
Gul
Ido
Man
Tal
Fru
Psi
Sor
Tag
Fuc
Qui
Rha
GalA
GlcA
IdoA
GalNac
GlcNAc
ManNAc
NeuNAc, Neu5Ac
KDN
KDO
NeuNGc, Neu5Gc

Table 3.4.: The one-letter codes that form the core of the GLYCAM residue names for monosaccharides a Users
requiring prep files for residues not currently available may contact the Woods group (www.glycam.org)
to request generation of structures and ensemble averaged charges. b Lowercase letters indicate Lsugars, thus L-Fucose would be “f”, see Table 3.7 . c Less common residues that cannot be assigned
a single letter code are accommodated at the expense of some information content. d Nomenclature
involving these residues will likely change in future releases.[56] Please visit www.glycam.org for the
most updated information.

3.7. The GLYCAM force fields for carbohydrates and lipids

Linkage Position
Terminalb
1-c
23462,32,42,63,43,64,62,3,42,3,62,4,63,4,62,3,4,6-

α−D-Glcp
Residue Name
0GAb
1GAc
2GA
3GA
4GA
6GA
ZGAd
YGA
XGA
WGA
VGA
UGA
TGA
SGA
RGA
QGA
PGA

β −D-Galp
Residue Name
0LB
1LB
2LB
3LB
4LB
6LB
ZLB
YLB
XLB
WLB
VLB
ULB
TLB
SLB
RLB
QLB
PLB

α−D-Arap
Residue Name
0AA
1AA
2AA
3AA
4AA

β −D-Xylp
Residue Name
0XB
1XB
2XB
3XB
4XB

ZAA
YAA

ZXB
YXB

WAA

WXB

TAA

TXB

Table 3.5.: Specification of linkage position and anomeric configuration in D-hexo- and D-pentopyranoses in threeletter codes based on the GLYCAM one-letter code a In pyranoses A signifies α-configuration; B = β .
b Previously called GA, the zero prefix indicates that there are no oxygen atoms available for bond
formation, i.e., that the residue is for chain termination. c Introduced to facilitate the formation of a
1–1´ linkage as in α-D-Glc-1-1´-α-D-Glc {1GA 0GA}. d For linkages involving more than one position,
it is necessary to avoid employing prefix letters that would lead to a three-letter code that was already
employed for amino acids, such as ALA.

Linkage position
Terminal
123···
etc.

α-D-Glc f
Residue name
0GD
1GD
2GD
3GD
···
etc.

β -D-Man f
Residue name
0MU
1MU
2MU
3MU
···
etc.

α-D-Ara f
Residue name
0AD
1AD
2AD
3AD
···
etc.

β -D-Xyl f
Residue name
0XU
1XU
2XU
3XU
···
etc.

Table 3.6.: Specification of linkage position and anomeric configuration in D-hexo- and Dpentofuranoses in threeletter codes based on the GLYCAM one-letter code. In furanoses D (down) signifies α; U (up) = β .

Linkage position
Terminal
123···
etc.

α-L-Glcp
Residue name
0gA
1gA
2gA
3gA
···
etc.

β -L-Manp
Residue name
0mB
1mB
2mB
3mB
···
etc.

α-L-Arap
Residue name
0aA
1aA
2aA
3aA
···
etc.

β -L-Xylp
Residue name
0xB
1xB
2xB
3xB
···
etc.

Table 3.7.: Specification of linkage position and anomeric configuration in L-hexo- and Lpentofuranoses in threeletter codes.

3. Molecular mechanics force fields

3.8. Lipid Force Fields
Biological processes in the human body are dependent on highly specific molecular interactions. The vast majority of the interactions take place in compartments within the cell, and an understanding of the behavior of the
membranes that compartmentalize and enclose the cell is therefore critical for rationalizing these processes. Biological membranes are complex structures formed mostly by lipids and proteins. For this reason lipid bilayers
have received a lot of attention both computationally and experimentally for many years.[57, 58] The vital role of
cell membranes is underlined by the estimation that over half of all proteins interact with membranes, either transiently or permanently.[59] Further, G protein-coupled receptors embedded in the membrane account for 50−60%
of present day drug targets, and membrane proteins as a whole make up around 70%.[60] Even so, only around
1300 unique resolved structures of membrane bound proteins, out of a total of 80,000 searchable entries, exist in
the Protein Data Bank reflecting the difficulties in studying membrane-associated proteins experimentally, making
them prime targets for simulation.
Prior to 2012, the only force field parameters for lipids distributed with AmberTools were part of the Glycam
force field.[44] Traditionally, lipid simulations with Amber have either employed the Charmm parameters, via
support for the Charmm force fields through the Chamber package[61]. Additionally there have been attempts to
adapt the General Amber Force Field (GAFF) with limited success.
In 2012, Amber greatly expanded support for simulation of lipids. This includes the development of a modular
framework for lipid simulations and initial parameterization within the LIPID11 force field[16] as well as a careful
refinement of the non-bonded parameters and associated torsion terms within the GAFF force field for specific
application to lipids.[62] The latter, GAFFLipid, is the first lipid parameter set based on the Amber force field
equation to support simulation of lipid bilayers in the tensionless NPT ensemble while the former, LIPID11,
provides the first modular framework for constructing lipid simulations that is analogous to the Amber amino
and nucleic acid force fields. Together these developments have made simulation of phospholipids with AMBER
substantially easier.
In 2014, LIPID14 was released as the latest Amber lipid force field. LIPID14 represents a major advancement
over the previous Amber compatible lipid force fields for lipid bilayer simulations in the NPT ensemble without
the need for an artificial constant surface tension term. LIPID14[17] is a new lipid force field that combines the
modular framework of LIPID11 as well as a number of refinements inspired by GAFFlipid. The modular nature of
the force field allows for many combinations of lipid head groups and tail groups as well as rapid parameterization
of further lipid types. In summary, several van der Waals and dihedral angle parameters have been refined to fit
experimental data and quantum energies as well as a new partial charge derivation for the head groups and tail
groups. The full parameterization details can be found in Dickson, et al[17]. The force field was validated on six
principle lipid bilayer types for a total of 0.5 microsecond each without applying a surface tension or constant area
term. The lipid bilayer structural features compare favorably with experimental measures such as area per lipid,
bilayer thickness, NMR order parameters, scattering data, and lipid lateral diffusion.

3.8.1. LIPID11: A modular lipid force field
Note: LIPID14 was released with Amber 14 as the latest Amber lipid force field.
Relevant files
leaprc.lipid11
lipid11.lib
lipid11.dat

loads the files below
atoms, charges, and topologies for LIPID11 residues
LIPID11 force field parameters

Description
LIPID11 is a modular force field for the simulation of phospholipids and cholesterol designed to be compatible
with the other pairwise additive Amber force fields.[16] Phospholipids are divided into interchangeable head group
and tail group “residues.”
Currently, there are seven tail group residues and eight head group residues supported, as well as cholesterol.
LEaP supports any combination of lipid residues. The supported LIPID11 residues and their residue names are
listed in Table 3.8.
Usage

3.8. Lipid Force Fields

Acyl chain

Head group

Other

Description
Palmitoyl (16:0)
Stearoyl (18:0)
Oleoyl (18:1 n-9)
Linoleoyl (18:2 n-6)
Linolenoyl (18:3 n-3)
Arachidonoyl (20:4 n-6)
Docosahexanoyl (22:6 n-3))
Phosphatidylcholine
Phosphatidylethanolamine
Phosphatidylserine
Phosphatidic acid (PHO4-)
Phosphatidic acid (PO42-)
R-phosphatidylglycerol
S-phosphatidylglycerol
Phosphatidylinositol
Cholesterol

LIPID11 Residue Name
PA
ST
OL
LEO
LEN
AR
DHA
PC
PE
PS
PHP2PGR
PGS
PI
CHL

Table 3.8.: LIPID11 residue names
Lipid 1

Lipid 2

...

sn-1 tail residue
head group residue
sn-2 tail residue
TER card
sn-1 tail residue
head group residue
sn-2 tail residue
TER card
...

Table 3.9.: LIPID11 PDB format for LEaP
source leaprc.lipid11

LIPID11 can be used alone or in conjunction with other Amber force fields. The order with which the various
AMBER force fields (FF12 for example) are loaded along with LIPID14 should not matter. For example, to load
ff14SB and LIPID11 in LEaP use:
source leaprc.ff14SB
source leaprc.lipid11

LIPID11 PDB format
LIPID11 atom names and types are defined in Skjevik, et al[16].
A properly formatted lipid PDB can be loaded into LEaP. Each phospholipid molecule in LIPID11 is made up
of three residues. Atoms from each residue must be in contiguous blocks and ordered as described below in each
molecule. A TER card must be appended after all the atoms for each molecule. Table 3.9 specifies the residue
format for the PDB file loaded by LEaP in order to correctly define linker atoms.
The connectivity (CONECT records) section of the PDB is redundant and should be removed prior to loading
into LEaP. The head group and tail residues are linked together by the LEaP program after loading the lipid PDB
file.
charmmlipid2amber.x: How to convert PDB files with lipid molecules with CHARMM residue/atom
names to LIPID11 residue/atom names
A simple script called charmmlipid2amber.x is available to convert a CHARMM-GUI membrane builder PDB
file to a LIPID11 PDB file ready to be loaded in LEaP for Amber simulations:

3. Molecular mechanics force fields

Acyl chain

Head group

Description
Lauroyl (12:0)
Myristoyl (14:0)
Palmitoyl (16:0)
Oleoyl (18:1 n-9)
Phosphatidylcholine
Phosphatidylethanolamine

LIPID14 Residue Name
LA
MY
PA
OL
PC
PE

Table 3.10.: LIPID14 residue names.

charmmlipid2amber.x input.pdb output.pdb

3.8.2. LIPID14: The Amber lipid force field
Relevant files
leaprc.lipid14
lipid14.lib
lipid14.dat

defines atom types and loads the files below
atoms, charges, and topologies for LIPID14 residues
LIPID14 force field parameters

Description
This force field represents the logical next step in development of an Amber lipid force field that includes the
modular nature of LIPID11 with the further parameter refinement inspired by GAFFLipid strategies to allow for
tensionless lipid bilayer simulations in Amber[17].
Currently there are two head groups and four tail groups available for use and LEaP supports any combination of
these residues. LIPID14 has been designed to be fully compatible with the other pairwise-additive protein, nucleic
acid, carbohydrate, and small molecule Amber force fields.
Usage
As in LIPID11, the new parameter set LIPID14 includes parameters for multiple head groups and tail groups.
Currently supported LIPID14 parameters are listed in Table 3.10. LIPID14 can be loaded into LEaP in a similar
way to the other Amber force fields. In LEaP, simply use the following command:
source leaprc.lipid14

LIPID14 PDB format
LIPID14 formatted PDB files follow the same format as the LIPID11 force field. The atom names and types are
defined in Dickson, et al [17]. See the LIPID11 section 3.8.1 for the specification of the PDB format to load in
LEaP.

3.9. Ions
frcmod.ionsjc_tip3p
Joung/Cheatham ion parameters for TIP3P water
frcmod.ionsjc_spce
same, but for SPC/E water
frcmod.ionsjc_tip4pew
same, but for TIP4P/EW water
frcmod.ionslrcm_hfe_tip3p
Li, Roberts, Chakravorty and Merz ion
parameters for TIP3P water (HFE set)
frcmod.ionslrcm_hfe_spce
same, but for SPC/E water
frcmod.ionslrcm_hfe_tip4pew same, but for TIP4P/EW water
frcmod.ionslrcm_cm_tip3p
Li, Roberts, Chakravorty and Merz ion
parameters for TIP3P water (CM set)
frcmod.ionslrcm_cm_spce
same, but for for SPC/E water
frcmod.ionslrcm_cm_tip4pew
same, but for TIP4P/EW water
frcmod.ionslrcm_iod
Li, Roberts, Chakravorty and Merz ion

3.9. Ions
parameters for TIP3P, SPC/E and TIP4P/EW
waters (IOD set)
frcmod.ionslm_1264_tip3p
Li/Merz ion parameters for TIP3P
water (12-6-4 set)
frcmod.ionslm_1264_spce
same, but for SPC/E water
frcmod.ionslm_1264_tip4pew
same, but for TIP4P/EW water
frcmod.ionsff99_tip3p
Older monovalent ion parameters from ff94/ff99
atomic_ions.lib
topologies for monoatomic ions (new naming scheme)
ions94.lib
topologies for ions with the old naming scheme

In the past, for alkali ions with TIP3P waters, Amber has provided the values of Aqvist,[63] adjusted for Amber’s
nonbonded atom pair combining rules to give the same ion-OW potentials as in the original (which were designed
for SPC water); these values reproduce the first peak of the radial distribution for ion-OW and the relative free
energies of solvation in water of the various ions. Note that these values would have to be changed if a water model
other than TIP3P were to be used. Rather arbitrarily, Amber also included chloride parameters from Dang.[64]
These are now known not to work all that well with the Aqvist cation parameters, particularly for the K/Cl pair.
Specifically, at concentrations above 200 mM, KCl will spontaneously crystallize; this is also seen with NaCl at
concentrations above 1 M.[65] These “older” parameters are now collected in frcmod.ionsff99_tip3p, but are not
recommended except to reproduce older simulations.
Subsequently, Joung and Cheatham have created a more consistent set of parameters, fitting solvation free energies, radial distribution functions, ion-water interaction energies and crystal lattice energies and lattice constants
for non-polarizable spherical ions.[66, 67] These have been separately parametrized for each of three popular water
models, as indicated above.
Li, Roberts, Chakravorty and Merz have developed the 12-6 Lennard-Jones parameters for 24 divalent metal
ions within TIP3P, SPC/E and TIP4P/EW waters for PME simulations.[68] The experimental values they tried
to reproduce are the experimental Hydration Free Energy (HFE) values, Ion-Oxygen Distance (IOD) values and
Coordination Number (CN) values of the first solvation shell. Similar to Joung and Cheatam’s work, three water
models were treated separately for the parameter design, as indicated in the name of frcmod files. Moreover, it
was found that it is hard to reproduce the three experimental values simultaneously due to the underestimation of
the nonbonded (Electrostatic plus Lennard-Jones) model. Therefore, three different parameter sets are designed
for each water model to meet different requirements, which are HFE set (to reproduce experimental HFE), IOD
set (to reproduce experimental IOD) and a compromise (CM) set of the former two (to reproduce the experimental relative HFE and CN values). Since the ion with certain parameter could reproduce similar IOD values in
the three water models, so the IOD set parameters of three water models were designed identical and shown in
frcmod.ionslrcm_iod file.
A new Lennard-Jones (LJ) type model based on a 12-6-4 potential has been developed and parameterized by
Li and Merz for 16 divalent metal ions with three widely used water models (TIP3P, SPC/E and TIP4P/EW).[69]
It arises from the consideration that the underestimation of the 12-6 LJ nonbonded model mainly comes from its
tendency to underestimate the charge-induced dipole interactions. Since the charge-induced dipole interaction is
4
proportional to r−4 , a new term with format Cr4 was added to the 12-6 LJ potential yeilding a 12-6-4 LJ-type
potential. These parameters reproduce the experimental HFE, IOD and CN values at the same time without significant compromise. The VDW parameters which specified designed for the divalent metal ions with 12-6-4 LJ-type
nonbonded model are shown as the 12-6-4 set in the above. These frcmod files can be used to generate an original prmtop file. After obtaining the original prmtop file, you can use the add12_6_4 command in parmed.py to
generate a prmtop with the additional C4 terms with the flag “LENNARD_JONES_CCOEF”. Please see Subsection13.2.2 in the manual for detailed information. After obtaining the prmtop with the additional C4 term, you can
use SANDER or PMEMD to run the simulation. There is an additional variable “lj1264” in the namelist of the
input file. When setting lj264 = 1, the “LENNARD_JONES_CCOEF” information in the prmtop file will be read
as C4 terms and 12-6-4 LJ-type potential will be employed in the simulation.
Please note: Many leaprc files load the atomic_ions.lib file, but you will still need to explicitly load a frcmod file
that matches the water model you are using.

3. Molecular mechanics force fields

3.10. Solvent models
solvents.lib
frcmod.tip4p
frcmod.tip4pew
frcmod.tip5p
frcmod.spce
frcmod.pol3
frcmod.meoh
frcmod.chcl3
frcmod.nma
frcmod.urea

library for water, methanol, chloroform, NMA, urea
Parameter changes for TIP4P.
Parameter changes for TIP4PEW.
Parameter changes for TIP5P.
Parameter changes for SPC/E.
Parameter changes for POL3.
Parameters for methanol.
Parameters for chloroform.
Parameters for N-methyacetamide.
Parameters for urea (or urea-water mixtures).

Amber now provides direct support for several water models. The default water model is TIP3P.[70] This model
will be used for residues with names HOH or WAT. If you want to use other water models, execute the following
leap commands after loading your leaprc file:
WAT = PL3 (residues named WAT in pdb file will be POL3)
loadAmberParams frcmod.pol3 (sets the HW,OW parameters to POL3)

(The above is obviously for the POL3 model.) The solvents.lib file contains TIP3P,[70] TIP3P/F,[71]
TIP4P,[70, 72] TIP4P/Ew,[73, 74] TIP5P,[75] POL3[76] and SPC/E[77] models for water; these are called TP3,
TPF, TP4, T4E, TP5, PL3 and SPC, respectively. By default, the residue name in the prmtop file will be WAT,
regardless of which water model is used. If you want to change this (for example, to keep track of which water
model you are using), you can change the residue name to whatever you like. For example,
WAT = TP4
set WAT.1 name "TP4"

would make a special label in PDB and prmtop files for TIP4P water. Note that Brookhaven format files allow at
most three characters for the residue label, which is why the residue names above have to be abbreviated.
Amber has two flexible water models, one for classical dynamics, SPC/Fw[78] (called “SPF”) and one for
path-integral MD, qSPC/Fw[79] (called “SPG”). You would use these in the following manner:
WAT = SPG
loadAmberParams frcmod.qspcfw
set default FlexibleWater on

Then, when you load a PDB file with residues called WAT, they will get the parameters for qSPC/Fw. (Obviously,
you need to run some version of quantum dynamics if you are using qSPC/Fw water.)
The solvents.lib file, which is automatically loaded with many leaprc files, also contains pre-equilibrated boxes
for many of these water models. These are called POL3BOX, QSPCFWBOX, SPCBOX, SPCFWBOX, TIP3PBOX,
TIP3PFBOX, TIP4PBOX, TIP4PEWBOX, and TIP5PBOX. These can be used as arguments to the solvateBox or
solvateOct commands in LEaP.
In addition, non-polarizable models for the organic solvents methanol, chloroform and N-methylacetamide are
provided,[80] along with a box for an 8M urea-water mixture. The input files for a single molecule are in
$AMBERHOME/dat/leap/prep, and the corresponding frcmod files are in $AMBERHOME/dat/leap/parm.
Pre-equilibrated boxes are in $AMBERHOME/dat/leap/lib. For example, to solvate a simple peptide in methanol,
you could do the following:
source leaprc.ff14SB (get a standard force field)
loadAmberParams frcmod.meoh (get methanol parameters)
peptide = sequence { ACE VAL NME } (construct a simple peptide)
solvateBox peptide MEOHBOX 12.0 0.8 (solvate the peptide with meoh)
saveAmberParm peptide prmtop prmcrd
quit

Similar commands will work for other solvent models.

3.11. CHAMBER

3.10.1. OPC WATER MODEL
OPC is a new non-polarizable, 4-point, 3-charge rigid water model[81]. Geometrically, it resembles TIP4P-like
models, although the values of OPC point charges and charge-charge distances are quite different. The model has
a single VDW center on the oxygen nucelus. The model is constructed based on the concept of optimal point
charge approximation[82]; the central idea of OPC is to distribute the point charges to best reproduce the 3 lowest
order multipole moments of water molecule in liquid phase. The optimal values for the dipole μ and the square
quadtupole moment QT [83] are determined as best fit values that reproduce key experimental properties of water
in liquid phase. The low dimensionality of the parameter space μ-QT permits a vitually exhaustive search. The
linear quadrupole and the octupole moments[84] are fixed to values obtained from high quality QM calcuations
[83].
Table below summarizes key properties of OPC water (liquid phase) at 298.16K compared to experiment (TMD
= Temperature of Density Maximum). The computations were performed with AMBER-12 on GPUs, using timestep of 2 fs, under periodic BC, PME, 8A cut-off. SHAKE was used to constrain hydrogens.
μ [D]

ρ [g cm−3 ]

Self diffusion [10−9 m2 s−1 ]

ε(0)

∆Hvap [kcalmol −1 ]

TMD [K]

OPC

2.48

0.997±0.0002

2.3±0.02

78.4±2

10.569±0.002

273-277

2.80

exp

2.5-2.9

0.997[85]

2.299±0.04[86]

78.4[87]

10.52[88]

277[88]

2.80[89]

rOO RDF 1st peak position [A]

No specific ion parameters have yet been developed for OPC water; however, based on our very limited study
of ion-ion interactions (Na+,Cl-) in OPC, it appears that the Joung/Cheatham [67] ion parameters for TIP4P-EW
might be acceptable. You can access the OPC model through the offbox.off file (to set a solvent box), and the
frcmod.opc file (to load the parameters).

3.11. CHAMBER
CHAMBER (CHarmm↔AMBER) is a tool which enables the use of the CHARMM force field within AMBER’s molecular dynamics engines (MDEs). If you make use of this tool, please cite Ref. [61]. There are two
components to CHAMBER:
1. The tool ($AMBERHOME/bin/chamber) which converts a CHARMM psf, associated coordinated file, parameter and topology to a CHARMM force field enabled version of AMBER’s prmtop and inpcrd.
2. The additional code within sander and pmemd to evaluate the extra CHARMM energies and forces.
AMBER[38] and CHARMM[90, 91] are two approaches to the parametrization of classical force fields that find
extensive use in the modeling of biological systems. The high similarity in the functional form of the two potential
energy functions used by these force fields, Eq.(3.1 and 3.2), gives rise to the possible use of one force field within
the other MDE.
VAMBER

∑

k (r − req )2 +

bonds

angles

"
+∑

i< j

VCHARMM

∑

Ai j Bi j
qi q j
−
+
∑
R12
R6i j
i< j εRi j
ij

kb (b − b0 )2 +

∑

(3.1)

kθ (θ − θ0 )2 +

angles

∑

nonbonded

∑

impropers

Rmini j
ri j

∑

kφ [1 + cos (nφ − δ )]

dihedrals

ku (u − u0 )2 +

∑

Urey−Bradley

Vn
[1 + cos(nφ − γ)] 6
dihedrals 2

∑

bonds

k (θ − θeq )2 +

∑

−

k (ω − ω0 )2 + ∑ VCMAP

Rmini j
ri j

φ ,ψ

6 #
+

qi q j
εri j

(3.2)

3. Molecular mechanics force fields
For the implementation of the CHARMM force field within Amber, parameters that are of the same energy
term can be directly translated. However, there are differences in the functional forms of the two potentials, with
CHARMM having three additional bonded terms. With respect to the 1-4 non-bonded interactions, CHARMM
scales these in a different manner: the electrostatic scaling factor (scee) is 1.0 in CHARMM but 1.2 in Amber,
while the van der Waals scaling factor (scnb) is 1.0 within CHARMM but 2.0 in Amber. Additionally, CHARMM
uses a different set of parameters in the Lennard-Jones equation for the van der Waals interaction if the two atoms
are bonded 1-4 to each other.
The first additional bonded term is CHARMM’s two-body Urey-Bradley term, which extends over all 1-3 bonds.
The second is a four-body quadratic improper term. The final additional term is a cross term, named CMAP,
[92, 93], which is a function of two sequential protein backbone dihedrals. This term originates from differences
observed between classically calculated two-dimensional φ /ψ peptide free energy surfaces using the CHARMM22
force field and those of experiment. CMAP is a numerical energy correction which essentially transforms the 2D
φ /ψ classical energy map to match that of a QM calculated map.
Support for these extra terms has required the development of extra sections to Amber’s extensible prmtop
format to accommodate this new information as well as modifications of the precision of existing sections. For example, the CHARMM parameter file stores the equilibrium angle (θ0 , Eq.3.2) parameter in degrees in its parameter
file, while Amber stores it in radians in the prmtop. However, during the conversion with chamber, this becomes
inexact when converted to radians. Within CHARMM this is done internally at runtime and the inexactness is
determined by the variable type that will hold the result of this conversion. However, for Amber, this conversion
is done at the chamber execution stage, and as a result is limited by the precision to which that specific parameter
is written to the prmtop file. Hence the precision of the ANGLE_EQUIL_VALUE has been increased; similar
changes were carried out for the CHARGE and VDW sections for the same reasons. Specifically, the modified
sections of the prmtop format and the additions to it are as follows:
%FLAG CTITLE

The keyword CTITLE is used in place of TITLE to specify that this is a CHAMBER prmtop.
%FLAG FORCE_FIELD_TYPE
%FORMAT(i2,a78)
1 CHARMM 31 *>>>>>>>>CHARMM22 All-Hydrogen Topology File for Proteins <<

This section described the force field in use. The initial integer specifies the number of lines to be read. The
keyword CHARMM here indicates that this is the CHARMM force field.
%FLAG CHARGE
%COMMENT Atomic charge multiplied by sqrt(332.0716D0) (CCELEC)
%FORMAT(3e24.16)

The default format for charge has been changed from 5e16.8 to 3e24.16
%FLAG CHARMM_UREY_BRADLEY_COUNT
%COMMENT V(ub) = K_ub(r_ik - R_ub)**2
%COMMENT Number of Urey Bradley terms and types
%FORMAT(2i8)

This additional section describes the number of CHARMM Urey-Bradley terms present and the total number of
Urey-Bradley types in use.
%FLAG CHARMM_UREY_BRADLEY
%COMMENT List of the two atoms and its parameter index
%COMMENT in each UB term: i,k,index
%FORMAT(10i8)

This additional section lists the atom indexes and parameter lookup index for each of the Urey-Bradley terms.
%FLAG CHARMM_UREY_BRADLEY_FORCE_CONSTANT
%COMMENT K_ub: kcal/mole/A**2

3.11. CHAMBER
%FORMAT(5e16.8)

This additional section lists the force constant for each of the Urey-Bradley types.
%FLAG CHARMM_UREY_BRADLEY_EQUIL_VALUE
%COMMENT r_ub: A
%FORMAT(5e16.8)

This additional section lists the equilibrium value for each of the Urey-Bradley types.
%FLAG SCEE_SCALE_FACTOR
%FORMAT(5e16.8)

This additional section lists a unique value of scee for each dihedral. This overides the default or &cntrl values
set for scee and in the case of the CHARMM force field will always be 1.0 for all dihedrals.
%FLAG SCNB_SCALE_FACTOR
%FORMAT(5e16.8)

This is the analogous additional term for scnb
%FLAG CHARMM_NUM_IMPROPERS
%COMMENT Number of terms contributing to the
%COMMENT quadratic four atom improper energy term:
%COMMENT V(improper) = K_psi(psi - psi_0)**2
%FORMAT(10i8)

This additional section lists the number of CHARMM improper terms present.
%FLAG CHARMM_IMPROPERS
%COMMENT List of the four atoms in each improper term
%COMMENT i,j,k,l,index i,j,k,l,index
%COMMENT where index is into the following two lists:
%COMMENT CHARMM_IMPROPER_{FORCE_CONSTANT,IMPROPER_PHASE}
%FORMAT(10i8)

This additional section lists the atom indices and index into the parameter arrays for each of the CHARMM
improper terms.
%FLAG CHARMM_NUM_IMPR_TYPES
%COMMENT Number of unique parameters contributing to the
%COMMENT quadratic four atom improper energy term
%FORMAT(i8)

This additional section lists the number of types present for the CHARMM impropers.
%FLAG CHARMM_IMPROPER_FORCE_CONSTANT
%COMMENT K_psi: kcal/mole/rad**2
%FORMAT(5e16.8)

This additional section lists the force constant for each CHARMM improper types.
%FLAG CHARMM_IMPROPER_PHASE
%COMMENT psi: degrees
%FORMAT(5e16.8)

This additional section lists the equilibrium phase angle for each of the CHARMM improper types.
%FLAG LENNARD_JONES_ACOEF
%FORMAT(3e24.16)

The default format for the Lennard Jones A and B coefficients has been changed from 5e16.8 to 3e24.16.

3. Molecular mechanics force fields
%FLAG LENNARD_JONES_14_ACOEF
%FORMAT(3e24.16)

This additional section and the corresponding BCOEF section provide the alternative parameters for 1-4 VDW
interactions in the CHARMM force field.
In concert with these prmtop additions, the appropriate modifications have to be made within sander and pmemd
to enable the calculation of the energy and derivatives corresponding to these new terms. The intention behind the
approach of creating a CHARMM enabled prmtop file is that the use of this prmtop file should be transparent
to the user. Once a CHARMM prmtop file is produced by chamber, the sander and pmemd dynamics engines
automatically detect the presence of CHARMM parameters in the prmtop file and automatically select the correct
parameters and code paths.
WARNING: The use of an unpatched Amber molecular dynamics engine with a chamber-generated prmtop file
will give undefined behavior, leading to incorrect results. If you see the following error at runtime:
ERROR: Flag "TITLE" not found in PARM file

it most likely means that you are using an old pmemd or sander executable.
A difficulty that has been encountered with the chamber generated prmtop files is visualisation with VMD.
The format of the chamber generated prmtop is valid with respect to AMBER’s prmtop %FLAG, %FORMAT
paradigm, however, VMD does not take into account a flag’s corresponding format specification since it has, a
priori, set each flag to a specific format. Hence, when the format of an existing flag is modified in a prmtop, VMD
fails to recognise this and incorrectly uses its hardcoded value instead.
Chamber has the ability to write an additional version of the prmtop (vmd_prmtop) file, that is compatible with
VMD. The general strategy here, is to use this additional vmd_prmtop file only for viewing purposes with VMD,
and use the correct prmtop for calculations with SANDER and PMEMD. The compatible vmd_prmtop file is
correct with respect to topology, but an incorrect with respect to certain parameters; for example %CHARGE has
been truncated to the old format and %COMMENT has been removed.
If one specifics the -vmd flag, an additional prmtop file, named vmd_prmtop, is generated. This can then be
used with VMD in the following ways:
vmd -parm7 vmd_prmtop -rst7 file.inpcrd
vmd -parm7 vmd_prmtop -mdcrd trajectory.mdcrd
vmd -parm7 vmd_prmtop -netcdf trajectory.nc

3.11.1. Usage
Here is the set of options returned from running the chamber binary:
Usage: chamber [args]
args for input are
-top
-param
-psf
-crd
Note: -crd can specify a pdb, a CHARMM crd or CHARMM rst file.
The filetype is auto detected.
args for output are

-p

-inpcrd

args for options are:
-cmap / -nocmap (Required option. Specifies
whether CMAP terms should be

3.11. CHAMBER
included or excluded.)
-str file1 file2 ... (for loading additional
CHARMM topology, parameter, and
stream files, for data not found in
-top and -param files)
-tip3_flex (allow angle in water)
-box a b c
Set the Orthorhombic lattice parameters a b c
for the generated inpcrd file.
-verbose
(lots of progress messages)
-vmd
Write a VMD compatible form of the prmtop file
-radius_set (GB radius set) options are:
0 Bondi radii (bondi)
1 Amber 6 modified Bondi radii (amber6)
<2> modified Bondi radii (mbondi)
6 H(N)-modified Bondi radii (mbondi2)
arg for help (this message) -h

Typical usage would be as follows:
$AMBERHOME/bin/chamber -cmap -top top_all22_prot.inp \
-param par_all22_prot.inp -psf foo.psf -crd foo.coor \
-p foo.prmtop -inpcrd foo.inpcrd -box 48.37 40.15 35.21

Often the CHARMM topology and parameter files being loaded with -top and -param
don’t contain all the necessary data, as the CHARMM .psf file may have been built with multiple files. For
example, in the c36 topology and corresponding param files (i.e top_all36_prot.rtf, par_all36_prot.prm) water
and ions are not included as they are in older force field files, thus, if the latter are desired, an additional file
containing them must be ’streamed’ in (i.e. toppar_water_ions.str) when generating the .psf. For cases like this,
chamber allows the reading of additional topology, parameter, and stream files that are needed using the -str flag.
An alternative, of course, would be to manually add any missing parameters to the primary topology and param
files by copying and pasting data from any supplementary files (i.e. TIP3 water params for c36), though this is
more error prone and time consuming. With the -str flag multiple files can be read in and the order is not
necessary. Below is a sample chamber input for system containing a protein, a sugar, and water and ions requiring
5 CHARMM topology and param files total:
Stream Files (-str flag)

$AMBERHOME/bin/chamber -top top_all36_prot.rtf -param par_all36_prot.prm \
-str toppar_water_ions.str top_all36_carb.rtf par_all36_carb.prm \
-psf prot.psf -crd prot.coor -p prot.prmtop -inpcrd prot.inpcrd

3.11.2. Validation
Starting with version c36a2 of CHARMM, a command (frcdump) has been implemented which provides a
validation route for alternate implementations of the CHARMM force fields. For a given system, this command
writes the various force field potential energy contributions, as well as the energy gradient experienced by each
atom, to a file using a specific format and to a high precision. The same formatted output can also be generated by
the AMBER MDEs to facilitate comparison and to validate that the CHARMM force field is being implemented
correctly in Amber’s MDEs.
An example section of a charmm script that will write this output to a file called charmm_gold_c36a2 is as
follows:
open unit 20 form write name charmm_gold_c36a2
frcdump unit 20
close unit 20

3. Molecular mechanics force fields
The analogous mdin section for Amber is as follows:
&debugf
do_charmm_dump_gold = 1,
/

Given this directive, the Amber MDE will stop after evaluating the potential energy of a system and write the
energy and forces pertaining to this to a (hardcoded) file called charmm_gold in the same directory as the mdin
file. The reader is invited to examine the various example test calculations within the $AMBERHOME/test/chamber/dev_tests/ directory for in depth examples of the above. For such testing, it is recommended that both the
CHARMM binary and the Amber MDE binaries be compiled with the same compiler. Given that CHARMM
support within Amber and the chamber software is still somewhat experimental, the user is advised to carry out
such a comparison before running a long production run.

3.11.3. Known limitations / Issues
This is a non-exhaustive list of the current known bugs and/or limitations with chamber:
• CHARMM polarization models are not supported. (IPOL /= 0)
• The ability to read CHARMM restart files is not currently supported.
• The mdout file will contain extra potential energy fields pertaining to the CHARMM terms. This may break
or confuse third party scripts that parse such outputs.
• Third party scripts and/or tools which do not correctly parse the extensible prmtop format may have issues
with a chamber-generated prmtop file.
• The potential energy decomposition components (self, reciprocal, direct, adjusted) of the Particle Mesh
Ewald energy generated in the charmm_gold file when the do_charmm_dump_gold = 1 mdin option in
Amber do not match with the breakdown used in CHARMM, however, the summation and resulting forces
do match.
If other issues are found, the chamber authors would be very grateful if these could be reported to them, either via
the Amber mailing list and/or directly to the authors. Please ensure that prior to reporting an issue, the chamber
binary passes the test cases provided with AmberTools. Please provide a standalone example of the problem with
all input files present and a script reproducing the sequence of commands that triggers the problem. The posting
of large files (> 2 MB) to the Amber mailing list is not recommended; instead one should make the files available
on a website somewhere and provide a link to it with the posting to the list.

3.12. Obsolete force field files
The following files are included for historical interest. We do not recommend that these be used any more for
molecular simulations. The leaprc files that load these files have been moved to $AMBERHOME/dat/leap/parm/oldff.

3.12.1. The Weiner et al. (1984,1986) force fields
all.in
allct.in
allnt.in
uni.in
unict.in
unint.in
parm91X.dat

All atom database input.
All atom database input, COO- Amino acids.
All atom database input, NH3+ Amino acids.
United atom database input.
United atom database input, COO- Amino acids.
United atom database input, NH3+ Amino acids.
Parameters for 1984, 1986 force fields.

3.12. Obsolete force field files
The ff86 parameters are described in early papers from the Kollman and Case groups.[94, 95] [The “parm91”
designation is somewhat unfortunate: this file is really only a corrected version of the parameters described in
the 1984 and 1986 papers listed above.] These parameters are not generally recommended any more, but may
still be useful for vacuum simulations of nucleic acids and proteins using a distance-dependent dielectric, or for
comparisons to earlier work. The material in parm91X.dat is the parameter set distributed with Amber 4.0. The
STUB nonbonded set has been copied from parmuni.dat; these sets of parameters are appropriate for united atom
calculations using the “larger” carbon radii referred to in the “note added in proof” of the 1984 JACS paper. If
these values are used for a united atom calculation, the parameter scnb must be defined in the prmtop file and
should be set to 8.0; for all-atom calculations it should be 2.0. The scee parameter should be defined in the prmtop
file and set to 2.0 for both united atom and all-atom variants. Note that the default value for scee is now 1.2 (the
value for 1994 and later force fields); this must be explicitly defined in the prmtop file when using the earlier force
fields.
parm91X.dat is not recommended. However, for historical completeness a number of terms in the non-bonded
list of parm91X.dat should be noted. The non-bonded terms for I (iodine), CU (copper) and MG (magnesium) have
not been carefully calibrated, but are given as approximate values. In the STUB set of non-bonded parameters, we
have included parameters for a large hydrated monovalent cation (IP) that represent work by Singh et al.[96] on
large hydrated counterions for DNA. Similar values are included for a hydrated anion (IM).
The non-bonded potentials for hydrogen-bond pairs in ff86 use a Lennard-Jones 10-12 potential. If you want to
run sander with ff86 then you will need to recompile, adding -DHAS_10_12 to the Fortran preprocessor flags.

3.12.2. The Cornell et al. (1994) force field
all_nuc94.in
all_amino94.in
all_aminoct94.in
all_aminont94.in
nacl.in
parm94.dat
parm96.dat
parm98.dat

Nucleic acid input for building database.
Amino acid input for building database.
COO- amino acid input for database.
NH3+ amino acid input for database.
Ion file.
1994 force field file.
Modified version of 1994 force field, for proteins.
Modified version of 1994 force field, for nucleic acids.

Contained in ff94 are parameters from the so-called “second generation” force field developed in the Kollman
group in the early 1990s.[38] These parameters are especially derived for solvated systems, and when used with an
appropriate 1-4 electrostatic scale factor, have been shown to perform well at modeling many organic molecules.
The parameters in parm94.dat omit the hydrogen bonding terms of earlier force fields. This is an all-atom force
field; no united-atom counterpart is provided. 1-4 electrostatic interactions are scaled by 1.2 instead of the value
of 2.0 that had been used in earlier force fields.
Charges were derived using Hartree-Fock theory with the 6-31G* basis set, because this exaggerates the dipole
moment of most residues by 10-20%. It thus “builds in” the amount of polarization which would be expected in
aqueous solution. This is necessary for carrying out condensed phase simulations with an effective two-body force
field which does not include explicit polarization. The charge-fitting procedure is described in Ref [38].
The ff96 force field [97] differs from parm94.dat in that the torsions for φ and ψ have been modified in response to ab initio calculations [98] which showed that the energy difference between conformations were quite
different than calculated by Cornell et al. (using parm94.dat). To create parm96.dat, common V1 and V2 parameters were used for φ and ψ, which were empirically adjusted to reproduce the energy difference between
extended and constrained alpha helical energies for the alanine tetrapeptide. This led to a significant improvement
between molecular mechanical and quantum mechanical relative energies for the remaining members of the set of
tetrapeptides studied by Beachy et al. Users should be aware that parm96.dat has not been as extensively used
as parm94.dat, and that it almost certainly has its own biases and idiosyncrasies, including strong bias favoring
extended β conformations.[18, 99, 100]
The ff98 force field [101] differs from parm94.dat in torsion angle parameters involving the glycosidic torsion
in nucleic acids. These serve to improve the predicted helical repeat and sugar pucker profiles.

3. Molecular mechanics force fields

3.12.3. The Wang et al. (1999) force field
parm99.dat
all_amino94.in
all_amino94nt.in
all_amino94ct.in
all_nuc94.in
gaff.dat
all_modrna08.lib
all_modrna08.frcmod

Basic force field parameters
topologies and charges for amino acids
same, for N-terminal amino acids
same, for C-terminal amino acids
topologies and charges for nucleic acids
Force field for general organic molecules
topologies for modified nucleosides
parameters for modified nucleosides

The ff99 force field [102] points toward a common force field for proteins for “general” organic and bio-organic
systems. The atom types are mostly those of Cornell et al. (see below), but changes have been made in many
torsional parameters. The topology and coordinate files for the small molecule test cases used in the development
of this force field are in the parm99_lib subdirectory. The ff99 force field uses these parameters, along with the
topologies and charges from the Cornell et al. force field, to create an all-atom nonpolarizable force field for
proteins and nucleic acids.
There are more than 99 naturally occurring modifications in RNA. Amber force field parameters for all these
modifications have been developed to be consistent with ff94 and ff99.[36] The modular nature of RNA was taken
into consideration in computing the atom-centered partial charges for these modified nucleosides, based on the
charging model for the “normal” nucleotides.[103] All the ab initio calculations were done at the Hartree-Fock
level of theory with 6-31G(d) basis sets, using the GAUSSIAN suite of programs. The computed electrostatic
potential (ESP) was fit using RESP charge fitting in antechamber. Three-letter codes for all of the fitted nucleosides
were developed to standardize the naming of the modified nucleosides in PDB files. For a detailed description of
charge fitting for these nucleosides and an outline for the three letter codes, please refer to Ref. [36].
The AMBER force field parameters for 99 modified nucleosides are distributed in the form of library files. The
all_modrna08.lib file contains coordinates, connectivity, and charges, and all_modrna08.frcmod contains information about bond lengths, angles, dihedrals and others. The AMBER force field parameters for the 99 modified
nucleosides in RNA are also maintained at the modified RNA database at http://ozone3.chem.wayne.edu.

3.12.4. The 2002 polarizable force fields
frcmod.ff02pol.r1
parm99.dat

parm99EP.dat
frcmod.ff02pol.r1
all_nuc02.in
all_amino02.in
all_aminoct02.in
all_aminont02.in
all_nuc02EP.in
all_amino02EP.in
all_aminoct02EP.in
all_aminont02EP.in

Recommended initialization file
Force field, for amino acids and some organic molecules;
can be used with either additive or
non-additive treatment of electrostatics.
Like parm99.dat, but with "extra-points": off-center
atomic charges, somewhat like lone-pairs.
Updated torsion parameters for ff02.
Nucleic acid input for building database, for a nonadditive (polarizable) force field without extra points.
Amino acid input ...
COO- amino acid input ...
NH3+ amino acid input ....
Nucleic acid input for building database, for a nonadditive (polarizable) force field with extra points.
Amino acid input ...
COO- amino acid input ...
NH3+ amino acid input ....

The ff02 force field is a polarizable variant of ff99. (See Ref. [104] for a recent overview of polarizable force fields.)
Here, the charges were determined at the B3LYP/cc-pVTZ//HF/6-31G* level, and hence are more like “gas-phase”
charges. During charge fitting the correction for intramolecular self polarization has been included.[80] Bond
polarization arising from interactions with a condensed phase environment are achieved through polarizable dipoles

3.12. Obsolete force field files
attached to the atoms. These are determined from isotropic atomic polarizabilities assigned to each atom, taken
from experimental work of Applequist. The dipoles can either be determined at each step through an iterative
scheme, or can be treated as additional dynamical variables, and propagated through dynamics along with the
atomic positions, in a manner analogous to Car-Parinello dynamics. Derivation of the polarizable force field
required only minor changes in dihedral terms and a few modification of the van der Waals parameters.
Subsequently, a set up updated torsion parameters has been developed for the ff02 polarizable force field.[105]
These are available in the frcmod.ff02pol.r1 file.
The user also has a choice to use the polarizable force field with extra points on which additional point charges
are located; this is called ff02EP. The additional points are located on electron donating atoms (e.g. O,N,S), which
mimic the presence of electron lone pairs.[106] For nucleic acids we chose to use extra interacting points only on
nucleic acid bases and not on sugars or phosphate groups.
There is not (yet) a full published description of this, but a good deal of preliminary work on small molecules
is available.[80, 107] Beyond small molecules, our initial tests have focused on small proteins and double helical
oligonucleotides, in additive TIP3P water solution. Such a simulation model, (using a polarizable solute in a nonpolarizable solvent) gains some of the advantages of polarization at only a small extra cost, compared to a standard
force field model. In particular, the polarizable force field appears better suited to reproduce intermolecular interactions and directionality of H-bonding in biological systems than the additive force field. Initial tests show ff02EP
behaves slightly better than ff02, but it is not yet clear how significant or widespread these differences will be.

4. The Generalized Born/Surface Area Model
The generalized Born solvation model can be used instead of explicit water for non-polarizable force fields; it
has been most widely tested on ff99SB, but in principle could be used with other non-polarizable force fields, such
as ff03. To estimate the total solvation free energy of a molecule, ∆Gsolv , one typically assumes that it can be
decomposed into the "electrostatic" and "non-electrostatic" parts:
∆Gsolv = ∆Gel + ∆Gnonel

(4.1)

where ∆Gnonel is the free energy of solvating a molecule from which all charges have been removed (i.e. partial
charges of every atom are set to zero), and ∆Gel is the free energy of first removing all charges in the vacuum,
and then adding them back in the presence of a continuum solvent environment. Generally speaking, ∆Gnonel
comes from the combined effect of two types of interaction: the favorable van der Waals attraction between the
solute and solvent molecules, and the unfavorable cost of breaking the structure of the solvent (water) around the
solute. In the current Amber codes, this is taken to be proportional to the total solvent accessible surface area (SA)
of the molecule, with a proportionality constant derived from experimental solvation energies of small non-polar
molecules, and uses a fast LCPO algorithm [108] to compute an analytical approximation to the solvent accessible
area of the molecule.
The Poisson-Boltzmann approach described in the next section has traditionally been used in calculating ∆Gel .
However, in molecular dynamics applications, the associated computational costs are often very high, as the
Poisson-Boltzmann equation needs to be solved every time the conformation of the molecule changes. Amber
developers have pursued an alternative approach, the analytic generalized Born (GB) method, to obtain a reasonable, computationally efficient estimate to be used in molecular dynamics simulations. The methodology has
become popular,[109–116] especially in molecular dynamics applications,[117–120] due to its relative simplicity
and computational efficiency, compared to the more standard numerical solution of the Poisson-Boltzmann equation. Within Amber GB models, each atom in a molecule is represented as a sphere of radius Ri with a charge qi
at its center; the interior of the atom is assumed to be filled uniformly with a material of dielectric constant 1. The
molecule is surrounded by a solvent of a high dielectric ε (80 for water at 300 K). The GB model approximates
∆Gel by an analytical formula,[109, 121]

qi q j
1
exp[−κ fGB ]
∆Gel = − ∑
1−
(4.2)
2 i j fGB (ri j , Ri , R j )
ε
where ri j is the distance between atoms i and j , the Ri are the so-called effective Born radii, and fGB () is a certain
smooth function of its arguments. The electrostatic screening effects of (monovalent) salt are incorporated [121]
via the Debye-Huckel screening parameter κ.
A common choice [109] of fGB is

1/2
fGB = ri2j + Ri R j exp(−ri2j /4Ri R j )

(4.3)

although other expressions have been tried.[112, 122] The effective Born radius of an atom reflects the degree of its
burial inside the molecule: for an isolated ion, it is equal to its van der Waals (VDW) radius ρi . Then one obtains
the particularly simple form:

q2
1
∆Gel = − i 1 −
(4.4)
2ρi
ε
where we assumed κ = 0 (pure water). This is the famous expression due to Born for the solvation energy of
a single ion. The function fGB () is designed to interpolate, in a clever manner, between the limit ri j → 0, when
atomic spheres merge into one, and the opposite extreme ri j → ∞, when the ions can be treated as point charges

4. The Generalized Born/Surface Area Model
obeying the Coulomb’s law.[115] For deeply buried atoms, the effective radii are large, Ri ρi , and for such atoms
one can use a rough estimate Ri ≈ Li , where Li is the distance from the atom to the molecular surface. Closer to
the surface, the effective radii become smaller, and for a completely solvent exposed side-chain one can expect Ri
to approach ρi .
The effective radii depend on the molecule’s conformation, and so have to be re-computed every time the conformation changes. This makes the computational efficiency a critical issue, and various approximations are normally
made that facilitate an effective estimate of Ri . In particular, the so-called Coulomb field approximation, or CFA,
is often used, which replaces the true electric displacement around the atom by the Coulomb field. Within this
assumption, the following expression can be derived:[115]
1
θ (|r| − ρi )r−4 dr
(4.5)
4π
where the integral is over the solute volume surrounding atom i. For a realistic molecule, the solute boundary
(molecular surface) is anything but trivial, and so further approximations are made to obtain a closed-form analytical expression for the above equation, e.g. the so-called pairwise de-screening approach of Hawkins, Cramer
and Truhlar,[123] which leads to a GB model implemented in Amber with igb=1. The 3D integral used in the
estimation of the effective radii is performed over the van der Waals (VDW) spheres of solute atoms, which implies a definition of the solute volume in terms of a set of spheres, rather than the complex molecular surface,[124]
commonly used in the PB calculations. For macromolecules, this approach tends to underestimate the effective
radii for buried atoms,[115] arguably because the standard integration procedure treats the small vacuum–filled
crevices between the van der Waals (VDW) spheres of protein atoms as being filled with water, even for structures with large interior.[122] This error is expected to be greatest for deeply buried atoms characterized by large
effective radii, while for the surface atoms it is largely canceled by the opposing error arising from the Coulomb
approximation, which tends [110, 114, 125] to overestimate Ri .
The deficiency of the model described above can, to some extent, be corrected by noticing that even the optimal packing of hard spheres, which is a reasonable assumption for biomolecules, still occupies only about three
quarters of the space, and so "scaling-up" of the integral by a factor of four thirds should effectively increase the
underestimated radii by about the right amount, without any loss of computational efficiency. This idea was developed and applied in the context of pH titration,[115] where it was shown to improve the performance of the GB
approximation in calculating pKa values of protein sidechains. However, the one-parameter correction introduced
in Ref. [115] was not optimal in keeping the model’s established performance on small molecules. It was therefore
proposed [120] to re-scale the effective radii with the re-scaling parameters being proportional to the degree of the
atom’s burial, as quantified by the value Ii of the 3D integral. The latter is large for the deeply buried atoms and
small for exposed ones. Consequently, one seeks a well-behaved re-scaling function, such that Ri ≈ (ρi−1 − Ii )−1
for small Ii , and Ri > (ρi−1 − Ii )−1 when Ii becomes large. The following simple, infinitely differentiable re-scaling
function was chosen to replace the model’s original expression for the effective radii:
−1
R−1
i = ρi −

−1
−1
2
3
R−1
i = ρ̃i − ρi tanh(αΨ − β Ψ + γΨ )

(4.6)

where Ψ = Ii ρ̃i , and α, β , γ are treated as adjustable dimensionless parameters which were optimized using the
guidelines mentioned earlier (primarily agreement with the PB). Currently, Amber supports two GB models (
termed OBC ) based on this idea. These differ by the values of α, β , γ, and are invoked by setting igb to either
igb=2 or igb=5. The details of the optimization procedure and the performance of the OBC model relative to the
PB treatment and in MD simulations on proteins is described in Ref. [120]; an independent comparison to the PB
in calculating the electrostatic part of solvation free energy on a large data set of proteins can be found in Ref.
[126].
Our experience with generalized Born simulations is mainly with ff12 or ff03; the current GB models are not
compatible with polarizable force fields. Replacing explicit water with a GB model is equivalent to specifying a
different force field, and users should be aware that none of the GB options (in Amber or elsewhere) is as mature
as simulations with explicit solvent; user discretion is advised! For example, it was shown that salt bridges are too
strong in some of these models [127, 128] and some of them provide secondary structure distributions that differ
significantly from those obtained using the same protein parameters in explicit solvent, with GB having too much
α-helix present.[129, 130] The combination of the ff14SB force field with igb=8 gives the best results for proteins;

4.1. GB/SA input parameters
1
mbondi

2
mbondi2

5
mbondi2

7
bondi

8
mbondi3

Table 4.1.: Recommended radii sets for various GB models. For values of igb given in the top row, the string in the
second row should be entered in LEaP as “set default PBRadii xxx”.

ff14SB and igb=1 is recommended for nucleic acids (see[131]for an evaluation of GB models for DNA).
The generalized Born models used here are based on the "pairwise" model introduced by Hawkins, Cramer and
Truhlar,[123, 132] which in turn is based on earlier ideas by Still and others.[109, 114, 125, 133] The so-called
overlap parameters for most models are taken from the TINKER molecular modeling package (http://tinker.wustl.edu).
The effects of added monovalent salt are included at a level that approximates the solutions of the linearized
Poisson-Boltzmann equation.[121] The original implementation was by David Case, who thanks Charlie Brooks
for inspiration. Details of our implementation of generalized Born models can be found in Refs. [134, 135].

4.1. GB/SA input parameters
As outlined above, there are several "flavors" of GB available, depending upon the value of igb. The version
that has been most extensively tested corresponds to igb=1; the "OBC" models (igb=2 and 5) are newer, but appear to give significant improvements and are recommended for most projects (certainly for peptides or proteins).
The newest, most advanced, and least extensively tested model, GBn (igb=7), yields results in considerably better
agreement with molecular surface Poisson-Boltzmann and explicit solvent results than the "OBC" models under
many circumstances.[130] The GBn model was parameterized for peptide and protein systems and is not recommended for use with nucleic acids. A modification on the GBn model (igb=8) further improves agreement
between Poisson-Boltzmann and explicit solvent data compared to the original formulation (igb=7).[21] Users
should understand that all (current) GB models have limitations and should proceed with caution. Generalized
Born simulations can only be run for non-periodic systems, i.e. where ntb=0. The nonbonded cutoff for GB calculations should be greater than that for PME calculations, perhaps cut=16. The slowly-varying forces generally do
not have to be evaluated at every step for GB, either nrespa=2 or 4.
igb
= 0 No generalized Born term is used. (Default)
= 1 The Hawkins, Cramer, Truhlar[123, 132] pairwise generalized Born model is used, with param-

eters described by Tsui and Case.[134] This model uses the default radii set up by LEaP. It is
slightly different from the GB model that was included in Amber6. If you want to compare to
Amber 6, or need to continue an ongoing simulation, you should use the command "set default
PBradii amber6" in LEaP, and set igb=1 in sander. For reference, the Amber6 values are those
used by an earlier Tsui and Case paper.[118] Note that most nucleic acid simulations have used
this model, so you take care when using other values. Also note that Tsui and Case used an offset
(see below) of 0.13 Å, which is different from its default value.
= 2 Use a modified GB model developed by A. Onufriev, D. Bashford and D.A. Case; the main idea

was published earlier,[115] but the actual implementation here[120] is an elaboration of this
initial idea. Within this model, the effective Born radii are re-scaled to account for the interstitial
spaces between atom spheres missed by the GBHCT approximation. In that sense, GBOBC is
intended to be a closer approximation to true molecular volume, albeit in an average sense. With
igb=2, the inverse of the effective Born radius is given
by:

−1
2
3
R−1
i = ρ i − tanh αΨ − β Ψ + γΨ /ρi
where ρ i = ρi − o f f set, and Ψ = Iρi , with I given in our earlier paper. The parameters α,
β , and γ were determined by empirical fits, and have the values 0.8, 0.0, and 2.909125. This

4. The Generalized Born/Surface Area Model
corresponds to model I in Ref [120]. With this option, you should use the LEaP command "set
default PBradii mbondi2" to prepare the prmtop file.
= 3 or 4 These values are unused; they were used in Amber 7 for parameter sets that are no longer

supported.
= 5 Same as igb=2, except that now α, β , γ are 1.0, 0.8, and 4.85. This corresponds to model II in

Ref [120]. With this option, you should use the command "set default PBradii mbondi2" in
setting up the prmtop file, although "set default PBradii bondi" is also OK. When tested in MD
simulations of several proteins,[120] both of the above parameterizations of the "OBC" model
showed equal performance, although further tests [126] on an extensive set of protein structures
revealed that the igb=5 variant agrees better with the Poisson-Boltzmann treatment in calculating
the electrostatic part of the solvation free energy.
= 6 With this option, there is no continuum solvent model used at all; this corresponds to a non-

periodic, "vacuum", model where the non-bonded interactions are just Lennard-Jones and Coulomb
interactions. This option is logically equivalent to setting igb=0 and eedmeth=4, although the
implementation (and computational efficiency) is not the same.
= 7 The GBn model described by Mongan, Simmerling, McCammon, Case and Onufriev[136] is

employed. This model uses a pairwise correction term to GBHCT to approximate a molecular
surface dielectric boundary; that is to eliminate interstitial regions of high dielectric smaller than
a solvent molecule. This correction affects all atoms and is geometry-specific, going beyond
the geometry-free, "average" re-scaling approach of GBOBC , which mostly affects buried atoms.
With this method, you should use the bondi radii set. The overlap or screening parameters in
the prmtop file are ignored, and the model-specific GBn optimized values are substituted. The
model carries little additional computational overhead relative to the other GB models described
above.[136] This method is not recommended for systems involving nucleic acids.
= 8 Same GB functional form as the GBn model (igb=7), but with different parameters.[21] The

offset, overlap screening parameters, and gbneckscale are changed. In addition, individual α,
β , and γ parameters are introduced for each of the elements H, C, N, O, S. Parameters for other
elements have not been optimized, and the values used are those from igb=5. An option is given
to specify individual parameters for P, though these are not included by default.
The following are the default parameters sander uses with igb=8:
Sh=1.425952, Sc=1.058554, Sn=0.733599,
So=1.061039, Ss=-0.703469, Sp=0.5,
offset=0.195141, gbneckscale=0.826836,
gbalphaH=0.788440, gbbetaH=0.798699, gbgammaH=0.437334,
gbalphaC=0.733756, gbbetaC=0.506378, gbgammaC=0.205844,
gbalphaN=0.503364, gbbetaN=0.316828, gbgammaN=0.192915,
gbalphaOS=0.867814, gbbetaOS=0.876635, gbgammaOS=0.387882,
gbalphaP=1.0, gbbetaP=0.8, gbgammaP=4.85

where Sh, Sc, Sn, So, Ss and Sp are scaling parameters, gbalphaX, gbbetaX, gbgammaX are the
α, β , γ set for element X. gbalphaOS, gbbetaOS, gbgammaOS is the α, β , γ set for O and S.
The phosphorus parameters are not optimized and are simply taken as the parameters used in the
OBC-2 model (igb=5). mbondi3 radii are recommended with igb=8 and can be employed with
the LEaP command "set default PBradii mbondi3".
=10 Calculate the reaction field and nonbonded interactions using a numerical Poisson-Boltzmann

solver. This option is described in the Chapter 5. Note that this is not a generalized Born
simulation, in spite of its use of igb; it is rather an alternative continuum solvent model.
intdiel

Sets the interior dielectric constant of the molecule of interest. Default is 1.0. Other values have not
been extensively tested.

extdiel

Sets the exterior or solvent dielectric constant. Default is 78.5.

4.2. ALPB (Analytical Linearized Poisson-Boltzmann)
saltcon

Sets the concentration (M) of 1-1 mobile counterions in solution, using a modified generalized Born
theory based on the Debye-Hückel limiting law for ion screening of interactions.[121] Default is 0.0
M (i.e. no Debye-Hückel screening.) Setting saltcon to a non-zero value does result in some increase
in computation time.

rgbmax

This parameter controls the maximum distance between atom pairs that will be considered in carrying
out the pairwise summation involved in calculating the effective Born radii. Atoms whose associated
spheres are farther way than rgbmax from given atom will not contribute to that atom’s effective Born
radius. This is implemented in a "smooth" fashion (thanks mainly to W.A. Svrcek-Seiler), so that when
part of an atom’s atomic sphere lies inside rgbmax cutoff, that part contributes to the low-dielectric
region that determines the effective Born radius. The default is 25 Å, which is usually plenty for
single-domain proteins of a few hundred residues. Even smaller values (of 10-15 Å) are reasonable,
changing the functional form of the generalized Born theory a little bit, in exchange for a considerable
speed-up in efficiency, and without introducing the usual cut-off artifacts such as drifts in the total
energy.
The rgbmax parameter affects only the effective Born radii (and the derivatives of these values with
respect to atomic coordinates). The cut parameter, on the other hand, determines the maximum distance for the electrostatic, van der Waals and "off-diagonal" terms of the generalized Born interaction.
The value of rgbmax might be either greater or smaller than that of cut: these two parameters are
independent of each other. However, values of cut that are too small are more likely to lead to artifacts
than are small values of rgbmax; therefore one typically sets rgbmax <= cut.

rbornstat

If rbornstat = 1, the statistics of the effective Born radii for each atom of the molecule throughout the
molecular dynamics simulation are reported in the output file. Default is 0.

offset

The dielectric radii for generalized Born calculations are decreased by a uniform value "offset" to give
the "intrinsic radii" used to obtain effective Born radii. Default is 0.09 Å.

gbsa

Option to carry out GB/SA (generalized Born/surface area) simulations. For the default value of 0,
surface area will not be computed and will not be included in the solvation term. If gbsa = 1, surface
area will be computed using the LCPO model.[108] If gbsa = 2, surface area will be computed by
recursively approximating a sphere around an atom, starting from an icosahedra. Note that no forces
are generated in this case, hence, gbsa = 2 only works for a single point energy calculation and is
mainly intended for energy decomposition in the realm of MM_GBSA.

surften

Surface tension used to calculate the nonpolar contribution to the free energy of solvation (when gbsa
= 1), as Enp = surften*SA. The default is 0.005 kcal/mol/A2 .[137]

rdt

This parameter is only used for GB simulations with LES (Locally Enhanced Sampling). In GB+LES
simulations, non-LES atoms require multiple effective Born radii due to alternate descreening effects
of different LES copies. When the multiple radii for a non-LES atom differ by less than RDT, only a
single radius will be used for that atom. See Chapter 24 for more details. Default is 0.0 Å.

4.2. ALPB (Analytical Linearized Poisson-Boltzmann)
Like the GB model, the ALPB approximation [138, 139] can be used to replace the need for explicit solvent,
with similar benefits (such as enhanced conformational sampling) and caveats. The basic ALPB equation that
approximates the electrostatic part of the solvation free energy is

1
1
1
αβ
1 1
−
q
q
+
∆Gel ≈ ∆Gal pb = −
i j
2 εin εex 1 + αβ ∑
fGB
A
ij
where β = εin /εex is the ratio of the internal and external dielectrics, α=0.571412, and A is the so-called effective
electrostatic size of the molecule, see the definition of Arad below. Here fGB is the same smooth function as in
the GB model. The GB approximation is then just the special case of ALPB when the solvent dielectric is infinite;

4. The Generalized Born/Surface Area Model
however, for finite values of solvent dielectric the ALPB tends to be more accurate. For aqueous solvation, the
accuracy advantage offered by the ALPB is still noticeable, and becomes more pronounced for less polar solvents.
Statistically significant tests on macromolecular structures [139] have shown that ALPB is more likely to be a
better approximation to PB than GB. At the same time, ALPB has virtually no additional computational overhead
relative to GB. However, users should realize that at this point the new model has not yet been tested nearly as
extensively as the GB model, and is therefore in its experimental stage. The model can potentially replace GB in
the energy analysis of snapshots via the MM-GB/SA scheme. The electrostatic screening effects of monovalent
salt are currently introduced into the ALPB in the same manner as in the GB, and are determined by the parameter
saltcon .
alpb

Flag for using ALPB to handle electrostatic interactions within the implicit solvent model.
= 0 No ALPB (default).
= 1 ALPB is turned on. Requires that one of the GB models is also used to compute the effective

Born radii, that is one must set igb=1,2,5, or 7. The ALPB uses the same sets of radii as required
by the particular GB model.
arad

Effective electrostatic size (radius) of the molecule. Characterizes its over-all dimensions and global
shape, and is not to be confused with the effective Born radius of an atom. An appropriate value of
Arad must be set if alpb=1: this can be conveniently estimated for your input structure with the utility
elsize that comes with the main distribution. The default is 15 Å. While Arad may change during the
course of a simulation, these changes are usually not very large; the accuracy of the ALPB is found
to be rather insensitive to these variations. In the current version of Amber Arad is treated as constant
throughout the simulation, the validity of this assumption is discussed in Ref. [139]. Currently, the
effective electrostatic size is only defined for "single-connected" molecules. However, the ALPB
model can still be used to treat the important case of complex formation. In the docked state, the
compound is considered as one, with its electrostatic size well defined. When the ligand and receptor
become infinitely separated, each can be assigned its own value of Arad.

4.2.1. elsize
NAME
elsize - Given the structure, estimates its effective electrostatic size
(parameter Arad ) need by the ALPB model.

SYNOPSIS
Usage: elsize input-pqr-file [-options]
-det an estimate based on structural invariants. DEFAULT.
-ell an estimate via elliptic integral (numerical).
-elf same as above, but via elementary functions.
-abc prints semi-axes of the effective ellipsoid.
-tab prints all of the above into a table without header.
-hea prints same table as -tab but with a header.
-deb prints same as -tab with some debugging information.
-xyz uses a file containing only XYZ coordinates.

DESCRIPTION

elsize is a program originally written by G. Sigalov to estimate the effective electrostatic size of a structure via a
quick, analytical method. The algorithm is presented in detail in Ref. .[139] You will need your structure in a pqr
format as input, which can be easily obtained from the prmtop and inpcrd files using ambpdb utility described
above:
ambpdb -p prmtop -pqr < inpcrd > input-file-pqr

4.2. ALPB (Analytical Linearized Poisson-Boltzmann)
After that you can simply do: elsize input-file-pqr , the value of electrostatic size in Angstroms will be output on
stdout. The source code is in the src/etc/ directory, its comments contain more extensive description of the options
and give an outline of the algorithm. A somewhat less accurate estimate uses just the XYZ coordinates of the
molecule and assumes the default radius size of for all atoms:
elsize input-file-xyz

This option is not recommended for very small compounds. The code should not be used on structures made up
of two or more completely disjoint" compounds – while the code will still produce a finite value of Arad , it is not
very meaningful. Instead, one should obtain estimates for each compound separately.

5. PBSA
Several efficient finite-difference numerical solvers, both linear [140, 141] and nonlinear,[142] are implemented
in pbsa for various applications of the Poisson-Boltzmann method. In the following, a brief introduction is given
on the method, the numerical solvers, and numerical energy and force calculations. This is followed by a detailed
description of the usage and keywords. Finally example input files are explained for typical pbsa applications. For
more information on the background and how to use the method, please consult cited references and online Amber
tutorial pages.

5.1. Introduction
Solvation interactions, especially solvent-mediated dielectric screening and Debye-Hückel screening, are essential determinants of the structure and function of proteins and nucleic acids.[143] Ideally, one would like to
provide a detailed description of solvation through explicit simulation of a large number of solvent molecules and
ions. This approach is frequently used in molecular dynamics simulations of solution systems. In many applications, however, the solute is the focus of interest, and the detailed properties of the solvent are not of central
importance. In such cases, a simplified representation of solvation, based on an approximation of the mean-force
potential for the solvation interactions, can be employed to accelerate the computation.
The mean-force potential averages out the degrees of freedom of the solvent molecules, so that they are often
called implicit or continuum solvents. The formalism with which implicit solvents can be applied in molecular
mechanics simulations is based on a rigorous foundation in statistical mechanics, at least for additive molecular
mechanics force fields. Within the formalism, it is straightforward to understand how to decompose the total meanfield solvation interaction into electrostatic and non-electrostatic components that scale quite differently and must
be modeled separately (see for example [144]).
The Poisson-Boltzmann (PB) solvents are a class of widely used implicit solvents to model solvent-mediated
electrostatic interactions.[143] They have been demonstrated to be reliable in reproducing the energetics and conformations as compared with explicit solvent simulations and experimental measurements for a wide range of
systems.[143] In these models, a solute is represented by an atomic-detail model as in a molecular mechanics force
field, while the solvent molecules and any dissolved electrolyte are treated as a structure-less continuum. The
continuum treatment represents the solute as a dielectric body whose shape is defined by atomic coordinates and
atomic cavity radii.[145] The solute contains a set of point charges at atomic centers that produce an electrostatic
field in the solute region and the solvent region. The electrostatic field in such a system, including the solvent
reaction field and the Coulombic field, may be computed by solving the PB equation:[146, 147]
∇ [ε(r)∇φ (r)] = −4πρ(r) − 4πλ (r) ∑ zi ci exp(−zi φ (r)/kB T )

(5.1)

where ε(r) is the dielectric constant, φ (r) is the electrostatic potential, ρ(r) is the solute charge, λ (r) is the Stern
layer masking function, zi is the charge of ion type i, ci is the bulk number density of ion type i far from the solute,
kB is the Boltzmann constant, and T is the temperature; the summation is over all different ion types. The salt
term in the PB equation can be linearized when the Boltzmann factor is close to zero. However, the approximation
apparently does not hold in highly charged systems. Thus, it is recommended that the full nonlinear PB equation
solvers be used in such systems.
The non-electrostatic or non-polar (NP) solvation interactions are typically modeled with a term proportional to
the solvent accessible surface area (SASA).[137] An alternative and more accurate method to model the non-polar
solvation interactions is also implemented in pbsa.[148] The new method separates the non-polar solvation interactions into two terms: the attractive (dispersion) and repulsive (cavity) interactions. Doing so significantly improves
the correlation between the cavity free energies and solvent accessible surface areas or molecular volumes enclosed

5. PBSA
by SASA for branched and cyclic organic molecules.[149] This is in contrast to the commonly used strategy that
correlates total non-polar solvation energies with solvent accessible surface areas, which only correlates well for
linear aliphatic molecules.[137] In the alternative method, the attractive free energy is computed by a numerical integration over the solvent accessible surface area that accounts for solvation attractive interactions with an infinite
cutoff.[150]

5.1.1. Numerical solutions of the PB equation
In pbsa both the linear form and the full nonlinear form of the PB equation are supported. Many strategies
may be used to discretize the PB equation, but only the finite-difference (FD) method, or more rigorously, the
finite-volume method [151–153] is fully supported in pbsa for both the linear and nonlinear PB equations. A FD
method involves the following steps: mapping atomic charges to the FD grid points (termed grid charges below);
assigning non-periodic/periodic boundary conditions, i.e., electrostatic potentials on the boundary surfaces of the
FD grid; and applying a dielectric model to define the high-dielectric (i.e., water) and low-dielectric (i.e., solute
interior) regions and mapping it to the FD grid edges.
These steps allow the partial differential equation to be converted into a linear or nonlinear system with the
electrostatic potential on grid points as unknowns, the charge distribution on the grid points as the source, and
the dielectric constant on the grid edges (and the salt-related term for the linear case) wrapped into the coefficient
matrix, which is a seven-banded symmetric matrix. In pbsa, four common linear FD solvers are implemented:
modified ICCG, geometric multigrid, conjugate gradient, and successive over-relaxation (SOR).[141] In addition,
we have also implemented six nonlinear FD solvers: Inexact Newton(NT)/modified ICCG, NT/geometric multigrid, conjugate gradient, and SOR and its improved versions - adaptive SOR and damped SOR.[142]
In addition to the FD method, a new discretization strategy is also introduced to solve the linear PB equation.[154]
The Immersed Interface method (IIM) is a second-order accurate numerical method developed for systems with
interface, i.e. solute/solvent boundary in this case. In the IIM discretization scheme, the linear equations on regular
grid points, i.e. grid points away from the interface, are the same as the standard finite-difference method, but the
linear equations on irregular grid points, i.e. grid points nearby the interface, are constructed by minimizing the
magnitude of the local truncation error in the discretization of the PB equation.[155] It can be proven that the
errors of calculated potentials are at the order of O(h2 ) on the regular grid points and O(h) on the irregular grid
points.[155]

5.1.2. Numerical interpretation of energy and forces
PB solvents approximate the solvent-induced electrostatic mean-force potential by computing the reversible
work in the process of charging the atomic charges in a solute molecule or complex. The charging free energy is a
function of the electrostatic potential φ , which can be computed by solving the linear or nonlinear system.
It has been shown (see for example [144]) that the total electrostatic energy of a solute molecule can be approximated through the FD approach by subtracting the self FD Coulombic energy (GFD
coul,shel f ) and the short-range FD
FD
FD
Coulombic energy (Gcoul,short ) from the total FD electrostatic energy (Gcoul,total ), and adding back the analytical
short-range Coulombic energy (Gana
coul,short ). The self FD Coulombic energy is due to interactions of grid charges
within one single atom. The self energy exists even when the atomic charge is exactly positioned on one grid point.
It also exists in the absence of solvent and any other charges. It apparently is a pure artifact of the FD approach
and must be removed. The short-range FD Coulombic energy is due to interactions between grid charges in two
different atoms that are separated by a short distance, usually less than 14 grid units. The short-range Coulombic
energy is inaccurate because the atomic charges are mapped onto their eight nearest FD grids, thus causing deviaFD
tion from the analytical Coulomb energy. The correction of GFD
coul,shel f and Gcoul,short is made possible by the work
of Luty and McCammon’s analytical approach to compute FD Coulombic interactions.[156]
Therefore, the PB electrostatic interactions include both Coulombic interactions and reaction field interactions
for all atoms of the solute. The total electrostatic energy is given in the energy component EEL in the output file.
The term that is reserved for the reaction field energy, EPB, is zero if this method is used. If you want to know how
much of EEL is the reaction field energy, you can set the BCOPT keyword (to be explained below) to compute the
reaction field energy only by using a Coulombic field (or singularity) free formulation.[157]

5.1. Introduction
When the full nonlinear Poisson-Boltzmann equation is used, an additional energy term, the ionic energy, should
also be included. This energy term disappears in the symmetrical linear system because the effects due to opposite
ions cancel out. It is currently approximated by calculation up to the space boundary of the FD grid. It should
be noted that the NBUFFER keyword may need increasing to obtain good precision in the ionic energy for small
molecules with a large FILLRATIO.
An alternative method of computing the electrostatic interactions is also implemented in pbsa. In this method,
the reaction field energy is computed directly after the induced surface charges are first computed at the dielectric
boundary (i.e., the surface that separates solute and solvent). These surface charges are then used to compute the
reaction field energy,[143] and is given as the EPB term. It has been shown that doing so improves the convergence
of reaction field energy with respect to the FD grid spacing. However, a limitation of this method is that the
Coulombic energy has to be recomputed analytically with a pairwise summation procedure. When this method
is used, the EEL term only gives the Coulombic energy with a cutoff distance provided in the input file. The
two ways of computing electrostatic interactions are controlled by the keywords ENEOPT and FRCOPT to be
described below.
The non-polar solvation free energy is returned by the ECAVITY term, which is either the total non-polar
solvation free energy or the cavity solvation free energy in the two different models described above. The EDISPER
term returns the dispersion solvation free energy. Of course it is zero if the total non-polar solvation free energy
has been returned by ECAVITY. The word INP can be used to choose one of the two treatments of non-polar
solvation interactions.[148] Specifically, you can use SASA to correlate total non-polar solvation free energy, i.e.,
Gnp = NP_T ENSION × SASA + NP_OFFSET as in PARSE.[137] You can also use SASA to correlate the cavity
term only and use a surface-integration approach to compute the dispersion term.[148] i.e., Gnp = Gdisp + Gcavity ,
with Gcavity = CAV ITY _T ENSION × SASA +CAV ITY _OFFSET . See the discussion of keywords in 8.2.8. These
options are described in detail in Ref. [148].
Finally, in this release, the PB forces are now correctly interpreted for the widely used SES molecular surface
definition, i.e., the partition of dielectric boundary pressure/force can now reproduce the virtual work principle.
This is achieved by proper decomposition of the dielectric boundary force on the reentrant portion of the molecular
surface. Specifically, the molecular surface is computed more accurately by considering the cases when the solvent
probe touches three atoms simultaneously. Next the reentrant force is also distributed onto the three atoms forming
the reentrant surface following the virtual work principle.[158]

5.1.3. Numerical accuracy and related issues
Note that the accuracy of any numerical PB procedure is determined by the discretization resolution specified
in the input, i.e., the grid spacing. The convergence criterion for the iteration procedures also plays some role for
the numerical PB solvers. Finally the accuracy is highly dependent upon the methods used for computing total
electrostatic interactions. In Lu and Luo,[144] the accuracy of the first method for total electrostatic interactions is
discussed in detail. In Ref.[158] the accuracy of the second method is discussed.
It is recommended that the second method for total electrostatic interactions be used for most calculations.
Apparently the cutoff distance for charge-charge interactions strongly influences the accuracy of electrostatic interactions. The default setting is infinity, i.e., no cutoff is used. In this method, the convergence of the reaction
field energy with respect to the grid spacing is much better than that of the first method. Our experience shows
that the reaction field energies converge to within ~2% for tested proteins at the grid spacing of 0.5 Å when the
weighted harmonic average of dielectric constants is used at the solute/solvent interface (when SMOOTHOPT =
1, see below).[159]
The reaction field energies computed with the second method (when SMOOTHOPT = 2) are also in excellent
agreement (differences in the order of 0.1%) with those computed with the Delphi program which uses the same
method for energy calculation. For example, see the computational set up documented in test case pbsa_delphi in
this release.[160]
The accuracy of non-polar solvation energy depends on the quality of SASA which is computed numerically by
representing each atomic surface by spherically distributed dots. Thus a higher dot density gives more accurate
atomic surface and molecular surface. However, it is found that the default setting for the dot density is quite
sufficient for typical applications.[148] Should you encounter any memory allocation error for surface calculation,
you are advised to use a coarser surface dot resolution if the physical memory of your computer is limited.

5. PBSA
Numerical solvation calculations are memory intensive for macromolecules due to the fine grid resolution required for sufficient accuracy. Thus, the efficiency of pbsa depends on how much memory is allocated for it and
the performance of the memory subsystem. The option that is directly related to its memory allocation is the FD
grid spacing for the PB equation and the surface dot resolution for molecular surface. Apparently the geometric
dimension and the number of atoms are also important for predicting the memory usage. In general for a typical
computer configuration with 8GB memory, the geometric dimension can be as large as 180 × 180 × 180 Å3 at the
default grid spacing of 0.5 Å before the computer responds too slowly.

5.2. Usage and keywords
5.2.1. File usage
pbsa has a very similar user interface as the Amber/sander program, though much simpler.
pbsa [-O] -i mdin -o mdout [-p prmtop -c inpcrd]/[-pqr pqr]

Starting from the 2014 release, pbsa supports the free format pqr file. Once the pqr reading is enabled, the default
Amber file reading and processing would be bypassed. Here is a brief description of the files mentioned above.
mdin input control data for the run.
mdout output user readable state info and diagnostics “-o stdout” will send output to stdout (to the terminal)

instead of to a file.
prmtop input molecular topology, force field, atom and residue names, and (optionally) periodic box type.
inpcrd input initial coordinates and (optionally) velocities and periodic box size.
pqr input initial coordinates, atomic charges and radii in the free format pqr.

A few comments on the “free-formatted” pqr file used by pbsa. First all fields are delimited by spaces only. Second
there is no strict format requirement as in a standard pdb file. This more liberal style is to accomodate pqr files of
different origins. pbsa reads data on a per-line basis using the following format:
Tag AtomNumber AtomName ResidueName ChainID ResidueNumber XYZ Charge Radius

Tag A string specifying either ATOM or HETATM. Lines with other strings are ignored.
AtomNumber The sequence no of the atom, which is reset to start from 1.
AtomName The atom name.
ResidueName The residue name.
ChainID The chain ID of the atom, optional, which is ignored.
ResidueNumber The sequence no. of the residue, which is ignored.
XYZ The floating numbers representing the atomic coordiantes (in Angstrom).
Charge A float number providing the atomic charge (in electron).
Radius A float number providing the atomic radius (in Angstrom).

Finally it is worth to point out that it is apparently very hard to know whether the charge and radius fields are
swapped as in the Delphi generated pqr file. Here we have assumed that the data are in the plain P.Q.R. order.
Please make sure you are following the same convention in generating the pqr files.

5.2. Usage and keywords

5.2.2. Basic input options
The layout of the input file is in the same way as that of Amber/sander for backward compatibility with previous
releases in Amber. The keywords are put in the the namelist of &cntrl for basic controls and &pb for more detailed
manipulation of the numerical procedures. This subsection discusses the basic keywords, either retained from
sander or newly created to invoke different energetic analyses. To reduce confusion most keywords from sander
have been removed from the namelist so they can no longer be read since the current implementation in pbsa only
performs single-structure calculations with the coordinates from inpcrd and exits. However, the current release is
compatible with the mdin file generated with the mmpbsa script in previous releases in Amber. Users interested in
energy minimization and molecular dynamics with the PB implementation are referred to sander in the release of
Amber. Nevertheless, for purposes of validation and development, the atomic forces can be dumped out in a file
when requested as described below.
The numerical electrostatic procedures can be turned on by setting IPB to either 1, 2 or 4. The flag IGB = 10 is
phased out in this release. The numerical non-polar procedures can be turned on by setting INP to either 1 or 2.
The backward compatible flag NPOPT is also phased out in this release.
imin

Flag to run minimization. Both options give the same output energies though the output formats are
slightly different. This option is retained from previous releases in the Amber package for backward
compatibility. The current release of pbsa only supports single point energy calculation.
= 0 No minimization. Dynamics is available with sander and NAB.
= 1 Single point energy calculation. Default. Multiple-step PB minimization is also available with

sander and NAB.
ntx

Option to read the coordinates from the “inpcrd” file. Only options 1 and 2 are supported in this
releases. Other options will cause pbsa to issue a warning though it does not affect the energy calculation.
= 1 X is read formatted with no initial velocity information. Default.
= 2 X is read unformatted with no initial velocity information.

ipb

Option to set up a dielectric model for all numerical PB procedures. IPB = 1 corresponds to a classical
geometric method, while a level-set based algebraic method is used when IPB > 2. The default IPB
is 2.
= 0 No electrostatic solvation free energy is computed.
= 1 The dielectric interface between solvent and solute is built with a geometric approach.
= 2 The dielectric interface is implemented with the level set function. Use of a level set function

simplifies the calculation of the intersection points of the molecular surface and grid edges and
leads to more stable numerical calculations. Default.
= 4 The dielectric interface is also implemented with the level set function. However, the linear

equations on the irregular points are constructed using the IIM. In this option, The dielectric
constant do not need to be smoothed, that is, SMOOTHOPT is useless. Only the linear PB
equation is supported, that is, NPBOPT = 0. And the different solvers are used to solve the
generated linear equation set, that is, the meaning of SOLVOPT is changed as shown below.
inp

Option to select different methods to compute non-polar solvation free energy.
= 0 No non-polar solvation free energy is computed.
= 1 The total non-polar solvation free energy is modeled as a single term linearly proportional to the

solvent accessible surface area, as in the PARSE parameter set, that is, if INP = 1, USE_SAV
must be equal to 0. See Introduction.
= 2 The total non-polar solvation free energy is modeled as two terms: the cavity term and the dis-

persion term. The dispersion term is computed with a surface-based integration method [148]

5. PBSA
closely related to the PCM solvent for quantum chemical programs.[150] Under this framework, the cavity term is still computed as a term linearly proportional to the molecular solventaccessible-surface area (SASA) or the molecular volume enclosed by SASA. Default.
Once the above basic options are specified, pbsa can proceed with the default options to compute the solvation free
energies with the input coordinates. Of course, this means that you only want to use default options for default
applications.
More PB options described below can be defined in the &pb namelist, which is read immediately after the
&cntrl namelist. We have tried hard to make the defaults for these parameters appropriate for calculations of
solvated molecular systems. Please use caution when changing any default options.

5.2.3. Options to define the physical constants
epsin

Sets the dielectric constant of the solute region, default to 1.0. The solute region is defined to be the
solvent excluded volume.

epsout

Sets the implicit solvent dielectric constant, default to 80. The solvent region is defined to be the space
not occupied the solute region. i.e., only two dielectric regions are allowed in the current release.

epsmemb

Sets the membrane dielectric constant. Only used if membraneopt > 0, does nothing otherwise. Value
used should be between epsin and epsout or there may be errors. Defaults to 1.0.

smoothopt Instructs PB how to set up dielectric values for finite-difference grid edges that are located across the
solute/solvent dielectric boundary.
= 0 The dielectric constants of the boundary grid edges are always set to the equal-weight harmonic

average of EPSIN and EPSOUT.
= 1 A weighted harmonic average of EPSIN and EPSOUT is used for boundary grid edges. The

weights for EPSIN and EPSOUT are fractions of the boundary grid edges that are inside or
outside the solute surface.[161] Default.
= 2 The dielectric constants of the boundary grid edges are set to either EPSIN or EPSOUT depending

on whether the midpoints of the grid edges are inside or outside the solute surface.
istrng

Sets the ionic strength (in mM) for the PB equation. Default is 0 mM. Note the unit is different from
that (in M) in the generalized Born methods implemented in Amber. Note also that we are only dealing
with symmetrical solution, so the ionic strength should be equal to the square of the valence of the
symmetrical ions times the ion concentration (in mM).

pbtemp

Temperature (in K) used for the PB equation, needed to compute the Boltzmann factor for salt effects;
default is 300 K.

radiopt

Option to set up atomic radii.
= 0 Use radii from the prmtop file for both the PB calculation and for the NP calculation (see INP).
= 1 Use atom-type/charge-based radii by Tan and Luo [162] for the PB calculation. Note that the

radii are optimized for Amber atom types as in standard residues from the Amber database. If
a residue is built by antechamber, i.e., if GAFF atom types are used, radii from the prmtop file
will be used. Please see [162] on how these radii are optimized. The procedure in [162] can also
be used to optimize radii for nonstandard residues. These optimized radii can be read in if they
are incorporated into the radii section of the prmtop file (of course via RADIOPT = 0). Default.
dprob

Solvent probe radius for molecular surface used to define the dielectric boundary between solute and
solvent. DPROB = 1.4 by default.

iprob

Mobile ion probe radius for ion accessible surface used to define the Stern layer. Default to 2.0 Å.

5.2. Usage and keywords
sasopt

Option to determine which kind of molecular surfaces to be used in the Poisson-Boltzmann implicit
solvent model. Default is 0.
= 0 Use the solvent excluded surface as implemented by[160]
= 1 Use the solvent accessible surface. Apparently, this reduces to the van der Waals surface when

the dprobe is set to zero.
= 2 Use the smooth surface defined by a revised density function.[163] This must be combined with

IPB > 2.
saopt

Option to compute the surface area of a molecule. Default is 0. Once the computation is enabled, the
surface area will be reported in the output file with the subtitle “Total molecular surface”. Note that
only the surface areas for the solvent excluded surface and the solvent accessible surface are supported
in this release.
= 0 Do not compute any surface area.
= 1 Use the field-view method to compute the surface area.[164]

triopt

Option to add trimer arc dots for a more accurate and lower memory mapping method of the analytical
solvent excluded surface. See[160]
= 0 Trimer arc dots are not used.
= 1 Trimer arc dots are used. Default.

arcres

pbsa uses a numerical method to compute solvent accessible arcs. See [160]. The ARCRES keyword
gives the resolution (in the unit of Å) of dots used to represent these arcs, default to 0.25 Å. These
dots are first checked against nearby atoms to see whether any of the dots are buried. The exposed
dots represent the solvent accessible portion of the arcs and are used to define the dielectric constants
on the grid edges. It should be pointed out that ARCRES should be reduced to (0.125 Å) when
the TRIOPT option is turned off to achieve a similar accuracy in the reaction field energies. More
generally, ARCRES should be set to max(0.125 Å, 0.5h) when the TRIOPT option is turned on, or
max(0.0625 Å, 0.25h) when the TRIOPT option is turned off (h is the grid spacing).

5.2.4. Options for Implicit Membranes
membraneopt Option to turn implicit membrane on and off. Membrane is implemented as a slab like region with
same dielectric constant as solute. Other membrane setup schemes will be made available in the future.
= 0 No implicit membrane used (default).
= 1 Use a slab-like implicit membrane.

mthick

Membrane thickness in Å, default to 20.0.

mctrdz

Distance in Å to offset membrane along the z direction. Default is 0 - membrane centered at the center
of the finite difference grid.

poretype

Option to control use of exclusion region for channel proteins. Only cylindrical region is supported
currently.
= 0 Do not use a cylindrical exclusion region (Default).
= 1 Use cylindrical exclusion region.

poreradius Controls the radius, in Å, of the cylindrical exclusion region.

5. PBSA

5.2.5. Options to select numerical procedures
npbopt

Option to select the linear or the full nonlinear PB equation.
= 0 Linear PB equation is solved. Default.
= 1 Nonlinear PB equation is solved.

solvopt

Option to select iterative solvers.
= 1 Modified ICCG or Periodic (PICCG) if bcopt = 10 is. Default. If IPB = 4, an algebraic multigrid

solver is used.
= 2 Geometric multigrid. A four-level v-cycle implementation is applied. Each dimension of the

finite-difference grid is 24 ×n-1. If IPB = 4, preconditioned GMRES.
= 3 Conjugate gradient (Periodic version available under bcopt = 10). This option requires a large

MAXITN to converge. If IPB = 4, preconditioned BiCG.
= 4 SOR. This option requires a large MAXITN to converge.
= 5 Adaptive SOR. This is only compatible with NPBOPT = 1. This option requires a large MAXITN

converge.
= 6 Damped SOR. This is only compatible with NPBOPT = 1. This option requires a large MAXITN

to converge.
accept

Sets the iteration convergence criterion (relative to the initial residue). Default to 0.001.

maxitn

Sets the maximum number of iterations for the finite difference solvers, default to 100. Note that
MAXITN has to be set to a much larger value, like 10,000, for the less efficient solvers, such as
conjugate gradient and SOR, to converge.

fillratio

The ratio between the longest dimension of the rectangular finite-difference grid and that of the solute.
Default is 2.0. It is suggested that a larger FILLRATIO, for example 4.0, be used for a small solute,
such as a ligand molecule. Otherwise, part of the small solute may lie outside of the finite-difference
grid, causing the finite-difference solvers to fail.

space

Sets the grid spacing for the finite difference solver; default is 0.5 Å.

nbuffer

Sets how far away (in grid units) the boundary of the finite difference grid is away from the solute
surface; default is 0 grids, i.e., automatically set to be at least a solvent probe or ion probe (diameter)
away from the solute surface.

nfocus

Set how many successive FD calculations will be used to perform an electrostatic focussing calculation on a molecule. Default to 2, the maximum. When NFOCUS = 1, no focusing is used. It is
recommended that NFOCUS = 1 when the multigrid solver is used.

fscale

Set the ratio between the coarse and fine grid spacings in an electrostatic focussing calculation. Default
to 8.

npbgrid

Sets how often the finite-difference grid is regenerated; default is 1 step. For molecular dynamics
simulations, it is recommended to be set to at least 100. Note that the PB solver effectively takes
advantage of the fact that the electrostatic potential distribution varies very slowly during dynamics
simulations. This requires that the finite-difference grid be fixed in space for the code to be efficient.
However, molecules do move freely in simulations. Thus, it is necessary to set up the finite-difference
grid once in a while to make sure a molecule is well within the grid.

5.2. Usage and keywords

5.2.6. Options to compute energy and forces
ENEOPT is the option to set a method to compute electrostatic energy and forces, and DBFOPT is phased out
in this release.
bcopt

Boundary condition options.
= 1 Boundary grid potentials are set as zero. Total electrostatic potentials and energy are computed.
= 5 Computation of boundary grid potentials using all grid charges. Total electrostatic potentials and

energy are computed. Default.
= 6 Computation of boundary grid potentials using all grid charges. Reaction field potentials and

energy are computed with the charge singularity free formulism.[157]
= 10 Periodic boundary condition is used. Total electrostatic potentials and energy are computed. Can

be used to switch ICCG and CG to PICCG and PCG. Should only be used with charge neutral
systems.
eneopt

Option to compute total electrostatic energy and forces.
= 1 Compute total electrostatic energy and forces with the particle-particle particle-mesh (P3M) pro-

cedure outlined in Lu and Luo.[144] In doing so, energy term EPB in the output file is set to
zero, while EEL includes both the reaction field energy and the Coulombic energy. The van
der Waals energy is computed along with the particle-particle portion of the Coulombic energy.
The electrostatic forces and dielectric boundary forces can also be computed.[144] This option
requires a non-zero CUTNB and BCOPT = 5.
= 2 Use dielectric boundary surface charges to compute the reaction field energy. Default. Both

the Coulombic energy and the van der Waals energy are computed via summation of pairwise
atomic interactions. Energy term EPB in the output file is the reaction field energy. EEL is the
Coulombic energy.
= 3 Similar to the first option above, a P3M procedure is applied for both solvation and Coulombic

energy and forces for larger systems.
frcopt

Option to compute and output electrostatic forces to a file named force.dat in the working directory.
= 0 Do not compute or output atomic and total electrostatic forces. This is default.
= 1 Reaction field forces are computed by trilinear interpolation. Dielectric boundary forces are com-

puted using the electric field on dielectric boundary. The forces are output in the unit of kcal/mol·Å.
= 2 Use dielectric boundary surface polarized charges to compute the reaction field forces and dielec-

tric boundary forces [158] The forces are output in the unit of kcal/mol·Å.
= 3 Reaction field forces are computed using dielectric boundary polarized charge. Dielectric bound-

rary forces are computed using the electric field on dielectric boundary. [165] The forces are
output in the unit of kcal/mol·Å.
scalec

Option to compute reaction field energy and forces.
= 0 Do not scale dielectric boundary surface charges before computing reaction field energy and

forces. Default.
= 1 Scale dielectric boundary surface charges using Gauss’s law before computing reaction field en-

ergy and forces.
cutfd

Atom-based cutoff distance to remove short-range finite-difference interactions, and to add pairwise
charge-based interactions, default is 5 Å. This is used for both energy and force calculations. See Eqn
(20) in Lu and Luo.[144]

5. PBSA
cutnb

Atom-based cutoff distance for van der Waals interactions, and pairwise Coulombic interactions when
ENEOPT = 2. Default to 0. When CUTNB is set to the default value of 0, no cutoff will be used
for van der Waals and Coulombic interactions, i.e., all pairwise interactions will be included. When
ENEOPT = 1, this is the cutoff distance used for van der Waals interactions only. The particle-particle
portion of the Coulombic interactions is computed with the cutoff of CUTFD.

nsnba

Sets how often atom-based pairlist is generated; default is 1 step. For molecular dynamics simulations,
a value of 5 is recommended.

5.2.7. Options for visualization and output
phiout

pbsa can be used to output spatial distribution of electrostatic potential for visualization.
= 0 No potential file is printed out. Default.
= 1 Electrostatic potential is printed out in a file named pbsa.phi in the working directory. Please refer

to examples in the next section on how to display electrostatic potential on molecular surface.
phiform

Controls the format of the electrostatic potential file.
= 0 The electrostatic potential (kT/mol·e) is printed in the Delphi binary format. Default.
= 1 The electrostatic potential (kcal/mol·e) is printed in the Amber ASCII format.
= 2 The electrostatic potential (kcal/mol·e) is printed in the DX volumetric data format for use with

VMD.
outlvlset

pbsa can be set to write the total level set, used in locating interfaces between regions of differing
dielectric constant, to a DX format volumetric data file. This option will control printing of the total
level set (i.e. both solute-solvent and membrane level sets combined if membrane present)
= false No level set file printed out. Default.
= true Level set printed out in a file named pbsa_lvlset.dx

outmlvlset pbsa can be set to write the membrane level set, used in locating interfaces between regions of differing
dielectric constant, to a DX format volumetric data file. This option controls printing a separate file
for the membrane level set. Does nothing if membraneopt is not turned on.
= false No level set file printed out. Default.
= true Level set printed out in a file named pbsa_lvlset.dx

npbverb

When set to 1, turns on verbose mode in pbsa; default is 0.

5.2.8. Options to select a non-polar solvation treatment
decompopt Option to select different decomposition schemes when INP = 2. See [148] for a detailed discussion
of the different schemes. The default is 2, the σ decomposition scheme, which is the best of the three
schemes studied.[148] As discussed in Ref. [148], DECOMPOPT = 1 is not a very accurate approach
even if it is more straightforward to understand the decomposition.
= 1 The 6/12 decomposition scheme.
= 2 The σ decomposition scheme. Default
= 3 The WCA decomposition scheme.

use_rmin

The option to set up van der Waals radii. The default is to use rmin to improve the agreement with
TIP3P [148].
= 0 Use atomic van der Waals σ values.
= 1 Use atomic van der Waals rmin values. Default.

5.2. Usage and keywords
sprob

Solvent probe radius for solvent accessible surface area (SASA) used to compute the dispersion term,
default to 0.557 Å in the σ decomposition scheme as optimized in Ref. [148] with respect to the
TIP3P solvent and the PME treatment. Recommended values for other decomposition schemes can be
found in Table 4 of [148]. If USE_SAV = 0 (see below), SPROB can be used to compute SASA for the
cavity term as well. Unfortunately, the recommended value is different from that used in the dispersion
term calculation as documented in Ref. [148] Thus two separate pbsa calculations are needed when
USE_SAV = 0, one for the dispersion term and one for the cavity term. Therefore, please carefully
read Ref. [148] before proceeding with the option of USE_SAV = 0. Note that SPROB was used for
ALL three terms of solvation free energies, i.e., electrostatic, attractive, and repulsive terms in previous
releases in Amber. However, it was found in the more recent study [148] that it was impossible to use
the same probe radii for all three terms after each term was calibrated and validated with respect to the
TIP3P solvent. [148, 162]

vprob

Solvent probe radius for molecular volume (the volume enclosed by SASA) used to compute nonpolar cavity solvation free energy, default to 1.300 Å, the value optimized in Ref. [148] with respect
to the TIP3P solvent. Recommended values for other decomposition schemes can be found in Tables
1-3 of Ref. [148].

rhow_effect Effective water density used in the non-polar dispersion term calculation, default to 1.129 for DECOMPOPT = 2, the σ scheme. This was optimized in Ref. [148] with respect to the TIP3P solvent in
PME. Optimized values for other decomposition schemes can be found in Table 4 of Ref. [148].
use_sav

The option to use molecular volume (the volume enclosed by SASA) or to use molecular surface
(SASA) for cavity term calculation. The default is to use the molecular volume enclosed by SASA.
Recent study shows that the molecular volume approach transfers better from small training molecules
to biomacromolecules.
= 0 Use SASA to estimate cavity free energy.
= 1 Use the molecular volume enclosed by SASA. Default.

cavity_surften The regression coefficient for the linear relation between the total non-polar solvation free energy
(INP = 1) or the cavity free energy (INP = 2) and SASA/volume enclosed by SASA. The default value
is for INP = 2 and set to the best of three tested schemes as reported in Ref. [148], i.e. DECOMPOPT
= 2, USE_RMIN = 1, and USE_SAV = 1. See recommended values in Tables 1-3 for other schemes.
cavity_offset The regression offset for the linear relation between the total non-polar solvation free energy (INP =
1) or the cavity free energy (INP = 2) and SASA/volume enclosed by SASA. The default value is for
INP = 2 and set to the best of three tested schemes as reported in Ref. [148], i.e. DECOMPOPT = 2,
USE_RMIN = 1, and USE_SAV = 1. See recommended values in Tables 1-3 for other schemes.
maxsph

pbsa uses a numerical method to compute solvent accessible surface area.[148] MAXSPH variable
gives the approximate number of dots to represent the maximum atomic solvent accessible surface,
default to 400. These dots are first checked against covalently bonded atoms to see whether any of the
dots are buried. The exposed dots from the first step are then checked against a non-bonded pair list
with a cutoff distance of 9 to see whether any of the exposed dots from the first step are buried. The
exposed dots of each atom after the second step then represent the solvent accessible portion of the
atom and are used to compute the SASA of the atom. The molecular SASA is simply a summation
of the atomic SASA’s. A molecular SASA is used for both PB dielectric map assignment and for NP
calculations.

5.2.9. Options to enable active site focusing
Active site focusing is an extension to the electrostatic focusing method. Electrostatic focusing can be regarded
as a multi-level FDPB calculation (two levels currently implemented) in which a coarse-grid solution is conducted
to set up the boundary condition for the requested fine-grid solution. In the original implementation of electrostatic

5. PBSA
focusing, the fine grid always covers all the solute atoms. However in the enhanced implementation, the fine grid
is allowed to cover only a local region of interest, such as an enzyme active site or ligand docking site. In such
applications, most or all of the protein atoms are held frozen during a calculation while only the active site side
chain and the substrate ligand are allowed to move. In principle, energies computed with the local electrostatic focusing method should correlate with those computed with the original electrostatic focusing method if the movable
substrate/ligand atoms are well within the local region of interest. The “active site” or the local region is specified
as a rectangular box by the following six variables:
xmax

The upper boundary of the box in x direction.

xmin

The lower boundary of the box in x direction, XMAX has to be greater than XMIN.

ymax

The upper boundary of the box in y direction.

ymin

The lower boundary of the box in y direction, YMAX has to be greater than YMIN.

zmax

The upper boundary of the box in z direction.

zmin

The lower boundary of the box in z direction, ZMAX has to be greater than ZMIN.

Of course, these keywords are zero by default, i.e. the original electrostatic focusing would be invoked if these
keywords remain to be the default value of zero.

5.2.10. Options to enable multiblock focusing
In order to handle large molecular systems with typical computer hardwares available to our end users, the
basic principle of the electrostatic focusing discussed in the previous subsection is extended for the multiblock
electrostatic focusing method. Briefly, the time-limiting step of FDPB, the fine-grid calculation, is divided into a
series of smaller jobs, with each solving only a small local region of a large molecular system. Once all the smaller
jobs are finished, the solutions are combined to obtain the final energy for the large molecular system. Note
that this is an approximated method, just like the original electrostatic focusing method. In this implementation,
overlapping/padding grid points are used to preserve accuracy. Most of the settings for this feature are hidden from
end users except the dimensions of the multi-blocks. [166]
Before your production runs, please activate NPBVERB = 1 and check in the mdout file to see if your multiblock settings are indeed reasonable. Here are some hints. First, the blocksize should be around 643 to 963 for
typical computers with 8GB memory. Secondly, the grid dimension, xm, should be divisible by (ngrdblkx − 1), or
slightly larger, for the x direction. The same applies for y and z directiosn as well. Keep in mind that the incentive
for choosing this method is to be able to work with large systems on typical computer hardwares.
ngrdblkx

The number of fine-grid points for a focusing block in x direction, (ngrdblkx − 1) should be divisible
by FSCALE.

ngrdblky

The number of fine-grid points for a focusing block in y direction, (ngrdblky − 1) should be divisible
by FSCALE.

ngrdblkz

The number of fine-grid points for a focusing block in z direction, (ngrdblkz − 1) should be divisible
by FSCALE.

pbsa can also be run in parallel environment with pbsa.MPI executable but for multiblock focusing only. Do make
sure that the number of nodes is less than the number of focusing blocks.

5.3. Example inputs and demonstrations of functionalities
5.3.1. Single-point calculation of solvation free energies
Normally the default pbsa options are capable of dealing with most situations. Users should be fully aware of
the meaning of an option before they change its default value. In all the following example inputs, only the

5.3. Example inputs and demonstrations of functionalities
options that are different from their default values will be shown, and the explanations on the changes will be
given in detail. Here is a sample input file that might be used to perform single structure calculations.
Sample single point PB calculation
&cntrl
/
&pb
npbverb=1, istrng=150, fillratio=1.5, saopt=1,
/

Note that NPBVERB = 1 above. This generates much detailed information in the output file for the PB and NP
calculations. A useful printout is atomic SASA data for both PB and NP calculations which may or may not
use the same atomic radius definition. Since the FD solver for PB is called twice to perform electrostatic focus
calculations, two PB printouts are shown for each single point calculation. For the PB calculation, a common error
message can be generated when FILLRATIO is set to the default value of 2.0 for small molecules. This may cause
a solute to lie outside of the focusing finite-difference grid.
In this example INP is not set and equal to the default value of 2, which calls for non-polar solvation calculation
with the new method that separates cavity and dispersion interactions. The EDISPER term gives the dispersion
solvation free energy, and the ECAVITY term gives the cavity solvation free energy. The default options for the
NP calculation are set to the recommended values for the σ decomposition scheme and to use molecular volume
to correlate with cavity free energy. You can find recommended values for other decomposition schemes and other
options in Tables 1-4 of Ref. [148]. If INP is set to 1, the ECAVITY term would give the total non-polar solvation
free energy.
Finally, a few words on the RADIOPT option, set to the default value of 1 instructing PB to use the optimized
values instead of reading the radii from the prmtop file. Starting this release, the RADIOPT option only controls
the radius definition for the PB calculation. The INP=2 calculation automatically uses the default values, such as
atomic radii and solvent probes as optimized in Ref. [148]. On the other hand, the INP=1 calculation is allowed to
use whatever radii that a user decides to use.
The ion strength option ISTRNG is set to 150 in unit mM, a typical value for a physiological environment. The
FILLRATIO option is set to 1.5 because the biomolecule is relatively large. We set saopt to 1 because we need the
information of the molecular surface area (the molecular surface is defined as the solvent excluded surface since
SASOPT is set to its default value 0).

5.3.2. Implicit membrane model
pbsa now supports inclusion of an implicit membrane region in implicit solvation calculations. This feature
is enabled by setting MEMBRANEOPT to 1 (default value is 0, for off). The membrane will extend the solute
dielectric region to include a slab-like planar region running parallel to the xy plane. The thickness is controlled
by the MTHICK option. The default is 20 Å. The membrane region will be centered on the center of the finitedifference grid by default, and can be offset along the z-axis using the MCTRDZ option (default is 0). Neither
option will have any effect unless MEMBRANEOPT is set to 1. The dielectric constant can be controlled using
epsmemb. We set the membrane interior dielectric constant to a value of 4.0 in this example. This is four times
that of the solute which defaults to 1 (same as vacuum). The value of epsmemb should always be set to a value
greater than or equal to epsin (solute dielectric constant) and less than epsout (solvent dielectric constant). These
default to 1.0 and 80.0 respectively.
When using the implicit membrane model, only SASOPT = 2, i.e. the smooth molecular surface based on the
revised density function, is currently supported. It is also suggested that periodic boundary conditions be used to
avoid unphysical edge effects. This is currently supported under the conjugate gradient solvers: Periodic
Conjugate Gradient (PCG) and Periodic Incomplete Cholesky Conjugate Gradient (PICCG), and can be
accomplished by setting IPB = 2 (default), BCOPT = 10, and SOLVOPT = 1 (PICCG, default) or SOLVOPT = 3
(PCG). In addition, ENEOPT needs to be set to 1 because the charge-view method (ENEOPT = 2) has not been
verified for this application.
Sample single point PB calculation with membrane region

5. PBSA
&cntrl
inp=0
/
&pb
radiopt=0, nfocus=1, maxitn=200,
bcopt=10, eneopt=1, solvopt=1,
sasopt=2, membraneopt=1, epsmemb = 4.0
outlvlset=true, outmlvlset=true
/

The MAXITN option is set to a bigger value, 200, than the default one, 100, because the conjugate gradient
method, when applied to periodic boundary conditions, seem to require slightly more iterations than non-periodic
conjugate gradient solvers.
To aid in visualization of the dielectric model, the level set function, which is used to locate the interfacial
surfaces between regions of differing dielectric constant, can be written to output files. Output of the total level
set function, including both the solute-solvent and membrane contributions, can be written to a DX formatted
volumetric data file by setting the OUTLVLSET option to “true”. The membrane contribution can be written to a
separate file by setting the OUTMLVLSET option to “true”. This may take a good deal of extra time, so be sure to
leave it off if you don’t want / need to visualize the levelset surface. Accordingly, NFOCUS is set to 1 because we
want the electrostatic potential and the level set function in both the solute and the solvent region.
Finally, if calculations need to be performed on a protein which includes a solvent filled channel region, this
region should be excluded from the membrane dielectric region. This can be accomplished by setting PORETYPE
= 1 to allow definition of a cylindrical exclusion region. This region will be centered upon the center of mass of
the solute and will extend the entire length of the membrane. Its radius may be controlled using PORERADIUS =
r, where r is the desired radius in angstroms. An initial visualization of the system is generally required to facilitate
selection of an appropriate radius (see section 8.4).

5.3.3. Single point calculation of forces
Since pbsa is released for single point calculations in AmberTools, no energy minimization or molecular
dynamics is supported. However, the PB procedure can be invoked to print out the numerical electrostatic forces
for developmental purposes. Here is a sample input:
Sample PB force computation
&cntrl
inp=0
/
&pb
npbverb=1, radiopt=0, frcopt=2
/

Note that INP is set to 0 to turn off non-polar solvation interactions. RADIOPT = 0 means the atomic radii from the
topology files will be used. FRCOPT is set to 2, i.e., induced surface charges are used to compute the electrostatic
energy and forces. Since CUTNB is equal to the default value of zero, an infinite cutoff distance is used for both
Coulombic and van der Waals interactions.

5.3.4. Comparing with Delphi results
Under identical condition, pbsa is highly consistent with Delphi in term of computed reaction field energies. In
this subsection, we briefly go over the details on how you can obtain comparable energies from both programs.
Apparently, you need coordinates, atomic charges, and atomic radii that have exactly the same numerical values
but in both the Amber format and the Delphi format, i.e., the pqr format.
For a Delphi computation with the following input parameters:

5.4. Visualization functions in pbsa
salt=0.150
ionrad=2.0
exdi=80.0
indi=1.0
scale=2.0
prbrad=1.5
perfil=50
bndcon=4
linit=1000

A comparable computation in pbsa can be obtained by using the following input file:
Sample PB for delphi comparison
&cntrl
ipb=1, inp=0
/
&pb
istrng=150, ivalence=1, iprob=2.0, dprob=1.5,
radiopt=0, bcopt=5, smoothopt=2, nfocus=1,
/

IPB is set to 1 to make sure pbsa is using the exactly same surface defination as Delphi. Note that the values of
exdi, indi, prbrad, and ionrad in Delphi should be consistent with the values of EPSOUT, EPSIN, DPROB, and
IPROB in pbsa, respectively. In Delphi salt=0.150 is set in the unit of M, while in pbsa ISTRNG = 150 is in
the unit of mM. In Delphi the grid spacing is set as the number of grids per Å, i.e., scale=2.0, while in pbsa the
grid spacing is set straight in Å as SPACE = 0.5. In Delphi the grid dimension is set as percentage of the solute
dimension over the grid dimension, i.e., perfil=50, which is equivalent to the ratio of solute dimension over grid
dimension set as FILLRATIO = 2 in pbsa. Finally, Delphi sets the boundary condition by bndcon=4 and pbsa sets
the boundary condition as BCOPT = 5; both programs mean to use the Debye-Huckel limitation behavior for each
atomic charged sphere. There are additional options in pbsa that do not have corresponding counterparts in Delphi.
For example, SMOOTHOPT is used to instruct the program to use a specific dielectric boundary smoothing option,
which is equivalent to that used in Delphi when set to 2. (see Section 5.2.3).

5.4. Visualization functions in pbsa
pbsa can produce volumetric data files to allow visualization of electrostatic potential and level set maps. There
are two points to note before continuing.
1. The data files generated can become quite large if small grid spacings are used since they will scale as the
cube of the inverse of grid spacing
2. Unless singularity removal methods are used, the potential at grid nodes corresponding to atom centers may
be quite large when compared to the potential at the molecular / atomic surface. This will often result in poor
contrast during visualization of the potential map, particularly when it is used as a color map for a molecular
surface.
These two points should be kept in mind when determining grid spacing. For visualization purposes, a grid spacing
of about one angstrom should provide good results. If finer spacing is needed, singularity removal (BCOPT = 6)
can be used to prevent poor contrast that could result from the presence of singularities. Lastly, when using grid
spacings of 0.5 Å or lower, the output files may become quite large (tens, or even hundreds of megabytes each)
and may take a significant amount of time (up to several seconds each) to generate.

5.4.1. Visualization of electrostatic potential using PyMol
pbsa can produce an electrostatic potential map for visualization in PyMol when setting PHIOUT = 1. By
default, pbsa outputs a file pbsa.phi in the Delphi binary format. The sample input file is listed below:

5. PBSA
Sample PB visualization input
&cntrl
inp=0
/
&pb
npbverb=1, space=1.,
phiout=1, phiform=0
/

To be consistent with the surface routine of PyMol, the option PHIOUT = 1 instructs pbsa to use the radii as defined
in PyMol. The finite-difference grid is also set to be cubic as in Delphi. The default DPROB value is equal to that
used in PyMol, 1.4 Å. A large grid spacing, e.g. 1 Å or higher, is recommended for visualization purposes, as
commented above.
Here is an example of loading the potential map in PyMol. First load the molecule in the form of prmtop and
inpcrd. In our case we need to rename our prmtop file to molecule.top and inpcrd file to molecule.rst and load the
molecule with commands
PyMol> load molecule.top
PyMol> load molecule.rst

The molecule will appear as an object “molecule”. Next display the surface of the molecule in the PyMol menu
by clicking “S” and then select surface. Now import the potential map generated by pbsa with the command in
PyMol
PyMol> load pbsa.phi

to create a value map object called “pbsa”. After this, create a value ramp called e_lvl from the potential map with
the command
PyMol> ramp_new e_lvl, pbsa, [-7, 0, 7]

You can assign surface_color to the e_lvl ramp with the command
PyMol> set surface_color, e_lvl, molecule

This will display the surface with the color scale according to the potential. You can adjust the value scale, such as
[-5, 0, 5], to change the color scale and use “rebuild” command to redraw the surface.

5.4.2. Writing electrostatic potential to DX format volumetric data file
To visualize the pbsa potential using VMD, you will need to set the output to DX format by changing
PHIFORM = 0 to PHIFORM = 2.
Sample PB visualization input
&cntrl
inp=0
/
&pb
npbverb=1, space=1., sasopt=2,
phiout=1, phiform=2
/

The program will now generate a file called pbsa_phi.dx. This format should be automatically recognized by
VMD. It can be either loaded directly into your molecule or as a separate file.

5.4. Visualization functions in pbsa

5.4.3. Loading DX format electrostatic potential data in VMD
1. go to the “File” menu in the VMD Main window.
2. Select “New Molecule...”.
• This will bring up the “Molecule File Browser” window
3. Click on the “Browse...” button in the “Molecule File Browser” window
4. Select the file “pbsa_phi.dx” that was generated by pbsa using the file selection dialogue that pops up.
• The “Determine file type:” drop down menu should now read “DX”.
5. Click the “Load” button.
VMD will, by default, display the data with an isosurface representation.

5.4.4. Changing the representation model
1. Select “Representations...” from the “Graphics” menu in the “VMD Main” window
• The “Graphical Representations” window should pop up
2. Select the object corresponding to the volumetric data you loaded from the “Selected Molecule” pull down
menu
3. Click on the representation you wish to change
• There should be one present for the isosurface being displayed
4. Click on the “Draw style” tab if it is not already selected
5. Select “Volume” from the “Coloring Method” pull down menu if it is not already chosen
• Another pull down menu will appear next to it.
• If you have multiple data files loaded for the same object you can choose which is used to color your
chosen draw method representation here
6. The “Drawing Method” pull down menu will let you choose a different visual representation model.
• To directly visualize potential data, use either “Isosurface” or “Volume Slice”
• VMD can also be used to visualize the corresponding electric field by choosing “Field Lines”.
Displayed below are Volume Slice representations of electrostatic potential maps generated for an aquaporin system. Computations were run using the periodic conjugate gradient solver for a 1 Å grid spacing, and FILLRATIO
of 2.0. For the systems using implicit water, the charge singularity removal methodology was used.
From Left to right: Vacuum, Water only, Water and 20 Å slab-like membrane, Water and 20 Å slab-like membrane with 6 Å cylindrical channel region removed.

Often, the data ranges will not be consistent between potential distributions for different implicit solvent setups.
E.g. the range of the electrostatic values seen for vacuum will likely be larger than the range for implicit water.
The range of values displayed can be set manually to provide consistent color scaling for comparison.

5. PBSA

5.4.5. Adjusting the color scale of the color map
1. Select “Colors...” from the “Graphics” menu in the “VMD Main” window
• This should cause the “Color Controls” window to pop up
2. Select the “Color Scale” tab
• The color scheme can be selected from the “Method” pull down menu
• The “Offset” and “Midpoint” sliders can be used to adjust the scaling of the color map.
– If singularities are present, it may be difficult to get a good scaling for volume maps generated
with fine grid spacings. In this case, either re-run with singularity removal on, or set the color
scale range manually as shown in the next section.
When singularity removal is not employed, the presence of singularities will cause the range of the electrostatic
potential distribution near the atom centers to be much wider than near the molecular surface. This typically results
in very poor contrast particularly for implicit solvent since the high dielectric constant in the solvent region will
amplify the effect. This can be compensated for by manually setting the Color Scale Data Range.

5.4.6. Changing the color scale range
1. Select desired representation to modify
2. Select “Volume” Coloring Method and Select the desired volumetric map to rescale from the pull down
menu.
• Each time you change the volumetric map being displayed, you will need to repeat this, so it is a good
idea to make multiple representations for each potential data set rather than switching between them
on the same representation.
3. Select the “Trajectory” tab
4. You should see the automatically computed range in the “Color Scale Data Range:” boxes. The left hand
box controls the minimum value for the range, the right hand box controls the maximum value for the range.
5. Set the minimum and maximum values as needed to improve the contrast. Often the inner 10% to 30% of
the total (automatic) range will give good contrast for a one angstrom grid spacing.
6. Click on the “Set” button when you are finished
7. To return to the automatic scaling that was originally calculated by VMD, click the “Autoscale” button.
Electrostatic potential data can also be used as a color map for other drawing methods. You will need to first load
the data into the molecule you wish to display.

5.4.7. Loading electrostatic potential data into an existing molecule
The names of the files are used as labels, so it is useful to rename them from “pbsa_phi.dx” to something more
descriptive before loading.
1. Select the molecule you wish to display the potential color map on in the “VMD Main” window
2. Go to the “File” menu in the VMD Main window.
3. Select “Load Data Into Molecule...”.
• This will bring up the “Molecule File Browser” window
4. Click on the “Browse...” button in the “Molecule File Browser” window

5.4. Visualization functions in pbsa
5. Select the file “pbsa_phi.dx” that was generated by pbsa using the file selection dialogue that pops up.
• The “Determine file type:” drop down menu should now read “DX”.
6. Click the “Load” button.
The data should now be loaded into the molecule you selected.

5.4.8. Using the electrostatic potential data as a color map
Once you have loaded a volumetric data file into a molecule, it can be used to generate a color map for any
representations of that molecules model.
1. Open the “Graphical Representations” window if it is not already open
• Select “Representations...” from the “Graphics” menu in the “VMD Main” window
2. Select the molecule you loaded the data into from the “Selected Molecule” pull down menu
3. Select the representation you wish to map the potential color map onto
4. Select the “Draw Style” tab if it is not already selected
5. Select “Volume” from the “Coloring Method” pull down menu
• Another pull down menu should appear next to it
• Choose the selection that corresponds to the data you just loaded, it should be the last one on the list if
it is the last one that was loaded.
VMD will attempt to automatically scale the color mapping used for Volumetric data that you load. The color scale
may be manually adjusted if needed (see previous section)

5.4.9. Loading and displaying the level set map
The level set used by pbsa to model the solute - solvent interface can be written to an output file in DX format
by setting OUTLVLSET to “true” in the input file.
Sample PB visualization input
&cntrl
inp=0
/
&pb
npbverb=1, space=1., sasopt=2,
phiout=1, phiform=2,
outlvlset=true
/

The level set will be written to a DX format volumetric data file named “pbsa_lvlset.dx”. This file can be used
to visualize the corresponding molecular surface. The level set file is loaded into VMD in the same manner as an
electrostatic potential data file. Cross sections can be viewed using the “Volume Slice” representation.
Shown below are the level sets for the aquaporin systems shown previously (no level set is shown for vacuum as
there is no dielectric interface being modeled in that system)
From left to right: Water, Water + Slab-like membrane, Water + Membrane with pore region

5. PBSA

5.4.10. Visualizing the molecular surface as an isosurface of the level set
The level set is constructed such that the molecular surface is the locus of all points where the level set is zero.
This allows us to use the Isosurface representation in VMD to display the solvent excluded surface by setting the
“Isovalue” to 0. Alternatively, if we wish to view the potential just outside the surface, we can set the “Isovalue”
to a number slightly higher than 0. E.g. 0.1 or 0.01.
1. Load the level set data file into the molecule.
• This is done using the same procedure as loading an electrostatic potential data file, but the level set
data file will be chosen instead of the potential data file.
2. Create a new Isosurface representation in the “Graphical Representations” window.
3. Select the volume map for the level set from the pull down menu
4. Choose an “Isovalue” at or slightly above 0.
5. Using the “Coloring Method” pull down menu, you may also use a previously loaded electrostatic potential
data file as a color map by selecting “Volume” and then selecting the appropriate volume map from the pull
down menu that appears.
• VMD will automatically assign color scale range every time.
• To compare multiple potential maps, it is often desirable to use the same color scale range for each.
The best way to do this is to make a new representation for each potential map and manually assign the
same color scale range to be identical for each (see previous section)
The examples below were generated for Aquaporin (1IH5 in the protein data bank) under various implicit solvent
options using a FILLRATIO of 2.0, grid spacing of 1Å. For each calculation, the periodic conjugate gradient
solver with singularity removal was used. The level set for the system modeling implicit water was used to build
the isosurfaces. The electrostatic potential data files were then overlayed as color maps with the color scale ranges
set to [-80000,80000].
From Left to right: Water only, Water + Slab Like Membrane, Water + Membrane with 6Å cylindrical pore.

5.4. Visualization functions in pbsa

5.4.11. Visualizing interior channels, voids, and solvent pockets
One of the common roles for membrane proteins is to act as a transmembrane channel, to allow specific substance to pass from one side of a membrane to another. Features such as solvent / ion channels or internal voids
will often be occluded from view by the exterior surface. One option that can allow these to be viewed is to use
the clipping plane tool in VMD.
1. Open the “Exensions” pull down menu in the “VMD Main” window and go to the “Visualization” submenu
and select “Clipping Plane Tool”.
2. The “Clip Tool” window should pop up.
3. The “Distance” slider allows clipping to be set
4. The “Normal” slider sets the normal of the clipping plane.
• The “flip” button on the right will let you clip from front to back, which will be useful to clip the
occluding exteriro surface from the view and reveal the interior.
The clipping tool was used to reveal the internal pore region for the aquaporin system setups used in the previous
section.
From Left to right: Water only, Water + Slab like Membrane, Water + Membrane with pore region excluded.

As an alternative, the level set map generated using PORTYPE=1 with the implicit membrane option will allow
a cylindrical region to be excluded from the membrane level set. The corresponding isosurface will show any
interior cavities or voids which fall within this region for isovalues at or slightly above 0 (since the level set at the
membrane-solute interface will be below 0). See the previous section for details on writing and loading the level
set file.
Shown below is the level set isosurface for the aquaporin system with implicit water plus a membrane with a
cylindrical region removed. The corresponding potential data was again overlayed as a color map. The surface of
the channel region, and the membrane-solvent interface planes are now clearly visible.

5.4.12. Importing / Modifying Atomic Radii to VMD from the prmtop file
Currently, VMD does not support loading radii for atoms directly from the prmtop file when it loads a molecule.
These values can be loaded relatively easily using the tkconsole, however. To do so:

5. PBSA
1. select “Tk Console” from the “Extensions” menu in the “VMD Main” window.
• The “VMD TkConsole” window will then open
2. Be sure that the atom you want to import radii for is the top molecule on the list in the VMD Main window.
If it is not, you will need to replace “top” with the appropriate ID
3. Type or copy and paste the following lines, but DO NOT hit enter yet.
set prot [atomselect “top” all]
$prot set radius {#RadiiList#}

4. You will now need to replace #RadiiList# with the one from the prmtop file.
a) Open the prmtop file for the molecule using a text editor
b) find the section that starts with “%FLAG RADII”
c) Highlight/Select the list of numbers that follows “%FORMAT(5E16.8)”
d) Copy the list (usually done by selecting “Copy” from the “Edit” menu in your text editor)
e) Go back to the “VMD TkConsole” window
f) Highlight/Select #RadiiList#
g) Select “Paste Ctrl-v” from the “Edit” menu in the “VMD TkConsole” window
5. Now hit return
• If this was successful, you should now have the correct radii for each atom in the molecule.
• you can have the console print the list of all radii by typing:
$prot get radius

• For a more human readable printout, use:
for {set ind 0} {$ind<[llength $rad]} {incr ind} \
{puts "Atom $ind radius is [lindex $rad $ind]"}

These radii are used by VMD to display the VDW surface (made by selecting “VDW” from the “Drawing
Method” pull down menu in the “Graphical Representations” window). One useful trick is to set them to be a
small amount larger (say .01 Å) than those used to generate the surface. This will ensure that the color map will
represent the external field just outside of the molecule. To modify the radii type or copy the following in the Tk
Console:
set rad [$prot get radius]
for {set ind 0} {$ind<[llength $rad]} {incr ind} \
{lset rad $ind [expr [lindex $rad $ind] +.01]}

The above code will increase all atomic radii by .01 angstroms. This can be changed if a different amount is
desired. (The code assumes you already followed steps 1 through 5 otherwise $prot will be undefined!)

5.5. pbsa in sander and NAB
5.5.1. Electrostatic forces/gradients in pbsa
Force calculation in the finite-difference Poisson-Boltzmann method is straightforward, though not a trivial
issue. It can be shown, by using the variation of the electrostatic free energy, that the electrostatic force density
consists of three components, viz., the reaction field force, the dielectric boundary force, and the ionic force. [167]
Since the ionic force is much smaller in absolute value than the other two components, we only include the reaction
field force and the dielectric boundary force in this release.

5.5. pbsa in sander and NAB
The reaction field force only exists where there are atomic charges, so that it is straightforward to be mapped
onto atoms. In contrast, the dielectric boundary force exists on the molecular surface where the dielectric constant
changes. The surface force, or pressure, cannot be easily mapped onto atoms. This is because a force-mapping
procedure from the molecular surface to atoms apparently needs the derivatives of molecular surface with respect
to atomic positions. However such derivatives do not exist for the widely used molecular surface definition, i.e.,
the solvent excluded surface (SES). We are actively developing an analytical molecular surface definition that is
consistent with the widely used SES definition for the numerical PB methods so that this difficulty will be overcome
in future releases.
Temporarily, a partial solution in the mapping of dielectric boundary force as described by Gilson et al[167]
is implemented for PB dynamics and minimization when the SES definition is used. The stability of the MD
simulation has been much improved with a more accurate mapping method of analytical SES.

5.5.2. Example for pbsa in sander
All pbsa functionalities are available in sander and all input options are exactly the same as in the standalone
pbsa. An apparent exception is IPB: you need to really set IPB to nonzero in order to invoke pbsa functionalities.
All other default values of PB options in sander are same as those in pbsa for single point calculations, whereas
there are some options that have different recommended or default values when PB minimization or dynamics is
enabled. These options are
space=0.25
arcres=0.125
fscale=4
eneopt=2
bcopt=6
frcopt=2

The SPACE, ARCRES and FSCALE are all set for higher resolution of the grid so that the force calculation can
be more accurate. The charge view method (ENEOPT = 2, FRCOPT = 2) is used here because it has been tested
to be able to run stable molecular dynamics simulations. Plus, BCOPT is set to 6 to remove charge singularity for
the same stability purpose. An example input for PBMD is given as follows
Sample PB visualization input
&cntrl
imin=0, ntx=1, irest=0,
ipb=2, ntb=0,
ntc=2, ntf=2,
tempi=100, temp0=100, ntt=3, gamma_ln=1,
nstlim=100000, dt=0.002,
ntpr=100, ntwr=100, ntwx=100,
/
&pb
npbgrid=500, nsnba=5,
/

IPB is explicitly set to 2 to enable PB dynamics. The NPBGRID option is set to 500, which means the finite
difference grid is regenerated every 500 dynamics steps. NSNBA = 5 means the atom-based pairlist is generated
every 5 steps. Please refer to Chapter 17 for the other &cntrl options. Note that the above input can be used with
sander only.

5.5.3. Example for pbsa in NAB
pbsa functionalities are available in NAB as a part of the standard build. However the available input options
are limited, please refer to the table in Section 37.1 for the list of available pbsa input options. The structures and

5. PBSA
parameters are supplied by NAB’s facility. Here is a sample of calls in a NAB program to the mm_options()
routine, in order to run pbsa:
mm_options("ntpr=1, cut=99.0"); // No solute-solute cutoff
mm_options("ipb=2"); // Use PBSA
mm_options("accept=0.000001"); // Convergence criterion
mm_options("dprob=1.6"); // Solvent probe radius for SASA
mm_options("radiopt=1"); // Useatom-type/charge-based radii
mm_options("fillratio=4"); // Coarse/Fine ratio of electrostatic focusing

6. Reference Interaction Site Model
In addition to explicit and continuum implicit solvation models, Amber also has a third type of solvation model
for molecular mechanics simulations, the reference interaction site model (RISM) of molecular solvation[168–
181]. In AmberTools, 1D-RISM is available as rism1d. 3D-RISM is available as an option in NAB, MMPBSA.py
and sander. rism3d.snglpnt is a simplified, standalone interface, ideal for calculating solvation thermodynamics on
individual structures and trajectories. Details specific to using sander and sander.MPI can be found in Chapter 17.

6.1. Introduction
RISM is an inherently microscopic approach, calculating the equilibrium distribution of the solvent, from which
all thermodynamic properties are then arrived at. Specifically, RISM is an approximate solution to the OrnsteinZernike (OZ) equation[169, 178, 179, 182, 183]
Z

h(r12 , Ω1 , Ω2 ) = c(r12 , Ω1 , Ω2 ) + ρ

dr3 dΩ3 c(r13 , Ω1 , Ω3 ) h(r32 , Ω3 , Ω2 ),

(6.1)

where r12 is the separation between particles 1 and 2 while Ω1 and Ω2 are their orientations relative to the vector
r12 . The two functions in this relation are h, the total correlation function, and c, the direct correlation function.
The total correlation function is defined as
hab (rab , Ωa , Ωb ) ≡ gab (rab , Ωa , Ωb ) − 1,
where gab is the pair-distribution function, which gives the conditional density distribution of species b about a. In
cases where only radial separation is considered, for example by orientational averaging over site α of species a
and site γ of species b, gives the familiar one dimensional site-site radial distribution function, gαγ (rαγ ).
For real mixtures, it is often convenient to speak in terms of a solvent, V, of high concentration and a solute, U,
of low concentration. A generic case of solvation is infinite dilution of the solute, i.e., ρ U → 0. We can rewrite
Equation (6.1), in the limit of infinite dilution, as a set of three equations:
hVV (r12 , Ω1 , Ω2 ) = cVV (r12 , Ω1 , Ω2 ) + ρ V

hUV (r12 , Ω1 , Ω2 ) = cUV (r12 , Ω1 , Ω2 ) + ρ V

hUU (r12 , Ω1 , Ω2 ) = cUU (r12 , Ω1 , Ω2 ) + ρ V

dr3 dΩ3 cVV (r13 , Ω1 , Ω3 ) hVV (r32 , Ω3 , Ω2 ),

(6.2)

dr3 dΩ3 cUV (r13 , Ω1 , Ω3 ) hVV (r32 , Ω3 , Ω2 ),

(6.3)

dr3 dΩ3 cUV (r13 , Ω1 , Ω3 ) hVU (r32 , Ω3 , Ω2 ).

(6.4)

Equation (6.3) is directly relevant for biomolecular simulations where we are often interested in the properties of
a single, arbitrarily complex solute in the solution phase. Solutions to Equation (6.3) can be obtained using 3DRISM. However, a solution to Equation (6.2) for pure solvent is a necessary prerequisite and is readily obtained
from 1D-RISM.
To obtain a solution to the OZ equations it is necessary to have a second equation that relates h and c or uniquely
defines one of these functions. The general closure relation is[182]
g(r12 , Ω1 , Ω2 ) = exp [−β u(r12 , Ω1 , Ω2 ) + h(r12 , Ω1 , Ω2 ) − c(r12 , Ω1 , Ω2 ) + b(r12 , Ω1 , Ω2 )]

(6.5)

u is the potential energy function for the two particles and b is known as the bridge function (a non-local functional,
representable as infinite diagrammatic series in terms of h [182]). It should be noted that u is the only point at
which the interaction potential enters the equations. Depending on the method used to solve the OZ equations, u

6. Reference Interaction Site Model
is generally an explicit potential. In principle, it should now be possible to solve our two equations. For example,
we may wish to use SPC/E as a water model. Inputting the relevant aspects of the SPC/E model into u, 1D-RISM
can be used to calculate the equilibrium properties of the SPC/E model. A different explicit water model will yield
different properties.
A fundamental problem for all OZ-like integral equation theories is the bridge function, which contains multiple
integrals that are readily solved only in special circumstances. In practice, an approximate closure relation must be
used. While many closures have been developed, at this time only three are implemented in 3D-RISM: hypernettedchain approximation (HNC), Kovalenko-Hirata (KH) and the partial series expansion of order-n (PSE-n).
For HNC, we set b = 0, giving[182]
gHNC (r12 , Ω1 , Ω2 ) = exp (−β u(r12 , Ω1 , Ω2 ) + h(r12 , Ω1 , Ω2 ) − c(r12 , Ω1 , Ω2 ))
= exp (t ∗ (r12 , Ω1 , Ω2 ))

(6.6)

where t ∗ is the renormalize-indirect correlation function. HNC works well in many situations, including charged
particles, but has difficulties when the size ratios of particles in the system are highly varied and may not always
converge on a solution when one should exist. Also, as the bridge term is generally repulsive, HNC allows particles
to approach too closely, overestimating non-Coulombic interactions[179].
KH is a combination of HNC and the mean spherical approximation (MSA), the former being applied to the
spatial regions of solvent density depletion (g < 1), including the repulsive core, and the latter to those of solvent
density enrichment (g > 1), such as association peaks[178, 179]
(

exp t ∗ (r12 , Ω1 , Ω2 )
for g(r12 , Ω1 , Ω2 ) ≤ 1
KH
.
(6.7)
g (r12 , Ω1 , Ω2 ) =
∗
1 + t (r12 , Ω1 , Ω2 )
for g(r12 , Ω1 , Ω2 ) > 1
Like HNC, KH handles Coulombic systems well but overestimates non-Coulombic interactions. Unlike HNC, it
does not have difficulties with highly asymmetric particle sizes and readily converges to stable solutions for almost
all systems of practical interest. The reliability of the KH closure makes it particularly suitable for molecular
mechanics calculations.
PSE-n offers the ability to interpolate between KH and HNC. Here, the exponential regions of solvent density
enrichment are treated as a Taylor expansion,

(
exp t ∗ (r12 , Ω1 , Ω2 )
for g(r12 , Ω1 , Ω2 ) ≤ 1
PSE-n
g
(r12 , Ω1 , Ω2 ) =
.
(6.8)
n (t ∗ (r ,Ω ,Ω ))i
12
1
2
/
i!
for g(r12 , Ω1 , Ω2 ) > 1
∑i=0
In the case of n = 1, the KH closure is obtained, while in the limit of n → ∞ HNC is recovered. This allows a
balance between the numerical stability of KH and the often better accuracy of HNC.

6.1.1. 1D-RISM
1D-RISM is used to calculate bulk properties of the solvent and is a prerequisite for 3D-RISM, for which the
primary result is the bulk solvent site-site susceptibility in reciprocal space, χ VV (k). As its name would suggest,
1D-RISM is a one-dimensional calculation. The six-dimensional OZ equations are reduced to one dimension (radial separation) via the fundamental RISM approximation[169–172, 182, 183], which produces the intramolecular
pair correlation matrix,
ωαγ (k) = sin(krαγ )/(krαγ )
(6.9)
where α and γ label the different atom types in the model. Note that atoms of the same type in RISM theory
have the same Lennard-Jones and Coulomb parameters. For example, most three site water models have two
RISM types, oxygen and hydrogen. Depending on the model, propane, C3 H8 , may have two carbon types and two

6.1. Introduction
hydrogen types. Equation (6.2) then becomes
Z

hαγ (r) = ∑

dr0 dr00 ωα µ ( r − r0 )cµν ( r0 − r00 ) ωνγ (r00 ) + ρν hνγ (r00 )

µν

1
(2π)3

h
i
eik·r dk ωc [1 − ρωc]−1 ω
αγ

∞

= ∑ ω(k)c(k)ω(k) [ρc(k)ω(k)]n .

(6.10)

Equation (6.10) must be complemented with one of the five closures currently supported by rism1d (see Subsection 6.4.1). In 1d, these are site-site closures and there is no orientational dependence. For example, the HNC
closure (Eq. (6.6)) becomes,

gHNC
αγ (r) = exp −β uαγ (r) + hαγ (r) − cαγ (r) .

(6.11)

Equation (6.10), with KH, HNC or PSE-n closures, is readily applicable to liquid mixtures, with site indices
of the site-site correlation functions enumerating interaction sites on all (different) species in the solution and the
intramolecular matrix (6.9) set equal to zero for sites α, γ belonging to different species.
A dielectrically consistent version of 1D-RISM theory (DRISM) enforces the proper dielectric asymptotics of
the site-site correlation functions, and so provides the self-consistent dielectric properties of electrolyte solution
with polar solvent and salt in a range of concentrations, including the given dielectric constant of the solution
[184].
The 1D-RISM integral equations are then solved for the site-site direct correlation function in an iterative manner, accelerated by the modified direct inversion of the iterative subspace (MDIIS) [179, 185]. All correlation
functions are represented as one-dimensional grids and the convolution integrals in Equation (6.10) are performed
in reciprocal space by making use of a fast Fourier transform applied to the short-range parts of all the correlations,
while the electrostatic asymptotics are separated out and Fourier transformed analytically [179–181].

6.1.2. 3D-RISM
With the results from 1D-RISM, a 3D-RISM calculation for a specific solute can be carried out. For 3D-RISM
calculations, only the solvent orientational degrees of freedom are averaged over and Equation (6.3) becomes[177,
178]
Z
VV 0
hUV
r − r0 χαγ
(r ),
(6.12)
dr0 cUV
γ (r) = ∑
α
α

where

VV (r)
χαγ

is the site-site susceptibility of the solvent, obtained from 1D-RISM and given by
VV
VV
χαγ
(r) = ωαγ
(r) + ρα hVV
αγ (r).

3D-RISM supports HNC, KH and PSE-n closures (see Sections 6.6.1, 37.1 and 29.3.1). As with the 1D-RISM
closures, these are constructed by analogy from Eqs. 6.6-6.8. For example, HNC becomes

UV
UV
gHNC,UV
(r) = exp −β uUV
(6.13)
γ
γ (r) + hγ (r) − cγ (r) .
As with 1D-RISM, correlation functions are represented on (3D) grids, convolution integrals are performed in
reciprocal space and a self-consistent solution is iteratively converged upon using the MDIIS accelerated solver.
There is one 3D grid for each solvent type for each correlation function. For example, for a solute in SPC/E water
UV
UV
there will be both gUV
H (r) and gO (r) grids. Each point on the gH (r) will give the fractional density of water
hydrogen a that location of real-space.
To properly treat electrostatic forces in electrolyte solution with polar molecular solvent and ionic species, the
electrostatic asymptotics of all the correlation functions (both the 3D and radial ones) are treated analytically [179,
180, 186]. The non-periodic electrostatic asymptotics are separated out in the direct and reciprocal space and the
remaining short-range terms of the correlation functions are discretized on a 3D grid in a non-periodic box large

6. Reference Interaction Site Model
enough to ensure decay of the short-range terms at the box boundaries [186]. The convolution of the short-range
terms in the integral equation (6.12) is calculated using 3D fast Fourier transform [187, 188]. Accordingly, the
electrostatic asymptotics terms in the thermodynamics integral (6.15) below are handled analytically and reduced
to one-dimensional integrals easy to compute [186].
With a converged 3D-RISM solution for hUV and cUV it is straightforward to calculate solvation thermodynamics.
From the perspective of molecular simulations, the most important thermodynamic values are the excess chemical
potential of solvation (solvation free energy), µ ex and the mean solvation force, fUV
i (Ri ), on each solute atom, i.
µ ex can be obtained through analytical thermodynamic integration for HNC,

Z
1 UV
1 UV 2
UV
(r)
−
hα (r) − cUV
h
(r)c
(r)
,
(6.14)
µ ex,HNC = kB T ∑ ραV dr
α
α
2
2 α
α
KH ,
µ ex,KH = kB T ∑ ραV
α

UV
1 UV 2
1 UV
UV
hα (r) Θ −hUV
(r)
−
c
(r)
−
h
(r)c
(r)
,
α
α
α
2
2 α

(6.15)

and PSE-n,
µ

ex,PSE-n

= kB T ∑
α

ραV

1 UV
1 UV 2
UV
hα (r) − cUV
α (r) − hα (r)cα (r)
2
2
#

(t ∗ (r))n+1
UV
−
Θ hα (r) , (6.16)
(n + 1)!

where Θ is the Heaviside function.
Analogous versions of Eqns. 6.6, 6.15 and 6.16 are used in 1D-RISM. While these are used for DRISM they
are have been derived for XRISM. Furthermore, these equations have been derived a number of different ways
with slightly different functional forms of the − 21 hc term [178, 189–192]. These different functional forms are
equivalent in XRISM but not in DRISM. The form introduced by Pettitt and Rossky [190] is the most popular
in the literature and the default selection in rism1d. It is possible to have rism1d evaluate and output all three
functional forms (see Output) but, for DRISM, none of these expressions are correct.
The force equation
Z
∂ µ ex
∂ uUV
α (r − Ri )
fUV
(R
)
=
−
=
−
ρ
drgUV
i
α
∑
i
α (r)
∂ Ri
∂ Ri
α
is valid for all closures with a path independent expression for the excess chemical potential, such as HNC, KH
and PSE-n closures implemented in 3D-RISM [168, 193–195].
In addition to closure specific expressions for the solvation free energy, other approximations also exist. The
Gaussian fluctuation (GF) approximation[196, 197] is given as

Z
1 UV
ex,GF
V
UV
UV
µ
= kB T ∑ ρα dr −cα (r) − hα (r) cα (r)
2
α
and has been shown to yield improved absolute solvation free energies for both polar and non-polar solutes[197,
198] but not necessarily for relative free energies[199]. It is not associated with a particular closure but is typically
used in place of the expression for a given closure.
Eqs. (6.14)-(6.16) give the total solvation free energy, ∆Gsol , but it is often useful to decompose this into electrostatic (solvent polarization), ∆Gpol , and non-electrostatic (dispersion and cavity formation), (∆Gdis + ∆Gcav ),
terms. Conceptually, we can divide the path of the thermodynamic integration into two steps: first the solute
without partial charges is inserted into the solvent (dispersion and cavity formation) and then partial charges are
introduced, which polarize the solvent,
µ ex = ∆Gsol = ∆Gpol + ∆Gdis + ∆Gcav .
∆Gsol is produced by a 3D-RISM calculation on the charged solute. ∆Gpol is then the difference of the two

6.1. Introduction
calculations. As a point of reference, generalized-Born and Poisson-Boltzmann methods calculate only ∆Gpol and,
typically, use a calculation involving solvent accessible surface area to predict ∆Gdis + ∆Gcav .

6.1.3. Analytic Temperature Derivatives
For the thermodynamic analysis of solvation, it is often useful to calculate the energetic and entropic contributions, ε solv and −T Ssolv respectively, to the solvation free energy. It has been shown that it is possible to analytically
decompose the solvation free energy into these two contributions when the solvation free energy has a closed analytical form, such as with HNC and KH closure [200]. In what follows, the analytical expression of energetic and
entropic contributions to the solvation free energy are derived in the framework of 1D-RISM theory with HNC
closure. The similar derivation can be applied to other closures as well as to the framework of 3D-RISM theory.
At this time, temperature derivatives are implemented for rism1d with HNC, KH and PSE-n closures.
The solvation free energy of species U in a solution consisting of N total species is expressed in the RISM-HNC
framework as
ex,U
U N
M
µHNC
= kB T ∑on
ργ dr
∑M=1 ∑on
α
γ

1
2

i
2
hαγ (r) − cαγ (r) − 21 hαγ (r)cαγ (r) .

The differentiation of the solvation free energy with respect to the temperature T leads to
ex,U
=
δT µHNC

"
ex,U
+ kB T
µHNC

U N
M
ργ dr hαγ (r) · δT hαγ (r) − δT cαγ (r) − 12 δT hαγ (r) · cαγ (r) − 12 hαγ (r) · δT cαγ (r) .
∑on
∑M=1 ∑on
α
γ

ex,U
ex,U
= −T Ssolv,U and therefore the above equation
= ε solv,U − T Ssolv,U , we have δT µHNC
where δT is T ∂∂T . Since µHNC
can be rearranged as

ε solv,U =

"
−kB T

U N
M
ργ
∑on
∑M=1 ∑on
α
γ

dr hαγ (r) · δT hαγ (r) − δT cαγ (r) −

#
1
1
2 δT hαγ (r) · cαγ (r) − 2 hαγ (r) · δT cαγ (r)

It is noted that the solvation energy ε solv,U can be viewed as consisting of two contributions: one arising from
creation of a polarized cavity (in pure solvent) and the other corresponding to the energy of embedding the solute
molecule into the cavity. The former is the solvent reorganization
energy and the latter is the average
R
solute-solvent interaction energy that is obtained as ∑α ∑γ ργ druαγ gαγ .
The temperature derivatives of correlation functionsδT h(r) and δT c(r) can be obtained by solving the temperature derivative of RISM-HNC equations
δT h(k) = w(k)δT c(k)w(k) + ρw(k)δT c(k)h(k) + ρw(k)c(k)δT h(k)
and
δT hαγ (r) =

uαγ (r)
kB T

i
+ δT hαγ (r) − δT cαγ (r) (hαγ (r) + 1).

Some practical examples can be found in [201] and [202].

6. Reference Interaction Site Model

6.2. Practical Considerations
6.2.1. Computational Requirements and Parallel Scaling
Calculating a 3D-RISM solution for a single solute conformation typically requires about 100 times more computer time than the same calculation with explicit solvent or PB. While there are other factors to consider, such as
sampling confined solvent or overall efficiency of sampling in the whole statistical ensemble at once, this can be
prohibitive for many applications. Memory is also an issue as the 3D correlation grids require anywhere from a
few megabytes for the smallest solutes to gigabytes for large complexes. A lower bound and very good estimate
for the total memory required is





Total memory ≥ 8 bytes× Nbox N V 2NMDIIS + 1 + Ndecomp Npropagate 
| {z } |{z}
| {z } | {z }
c,residual

polar decomp past solutions






(Nbox + 2Ny Nz )
4
+ |{z}
1
+ |{z}
2 NV 
|{z}


asymptotics

FFT scratch

g,h

where Nbox = Nx × Ny × Nz is the total number of grid points, N V is the number of solvent atom species and NMDIIS
is the number of MDIIS vectors used to accelerate convergence. uUV , cUV and the residual of cUV are stored in
real-space only and require a full grid for each solvent. cUV and its residual also require NMDIIS grids for the
MDIIS routine (see the mdiis_nvec keyword) and Npropagate grids to make use of solutions from previous solute
configurations to improve the initial guess (see the npropagate keyword). If a polar/non-polar decomposition is
requested (see the polardecomp keyword) an additional set of grids for past solutions with no solute charges is
kept (Ndecomp = 2); by default this is turned off (Ndecomp = 1). The full real space grid plus an additional 2Ny Nx
grid points are needed (due to the FFT) for g and h for each solvent species and for the four grids required
to compute the long range asymptotics. Memory, therefore, scales linearly with Nbox while computation time
scales as O(Nbox log(Nbox )) due to the requirements of calculating the 3D fast Fourier transform (3D-FFT). To
overcome these requirements, two options are available beyond optimizations already in place, multiple time steps
and parallelization. Multiple time step methods are available only in sander (Chapter 17) and are applicable to
molecular dynamics calculations only. Parallelization is available for all calculations but is limited by system size
and computational resources.
Both sander and NAB have MPI implementations of 3D-RISM (see Section 6.5.5 for NAB compiling instructions) that distribute both memory requirements and computational load. As memory is distributed, the aggregate
memory of many computers can be used to perform calculations on very large systems. Memory distribution is
handled by the FFTW 3.3 library so decomposition is done along the z-axis. If a variable solvation box size is
used, the only consideration is to avoid specifying a large, prime number of processes (≥ 7). For fixed box sizes,
the number of grids points in each dimension must be divisible by two (a general requirement) and the number
of grid points in the z-axis must be divisible by the number of processes. sander.MPI also has the additional
consideration that the number of processes cannot be larger than the number of solute residues; NAB does not suffer
from this limitation.

6.2.2. Output
gUV , hUV and cUV files can be output for 3D-RISM calculations and are useful for visualization and calculation of
thermodynamic quantities. These use the ASCII Data Explorer (DX) file format (See http://ambermd.org/formats.html)

so there is one file for each solvent atom type for each requested frame. Each file is 348 + Nbox × 16 31 bytes,
which can quickly fill disk space. Also, very few visualization programs are capable of displaying both molecular
and volumetric trajectories.

6.3. Work Flow

6.2.3. Numerical Accuracy
Numerical accuracy depends on the specified residual tolerance for the solution and the solvation box physical
size and grid spacing. Almost all applications should use a grid spacing of 0.5 Å. A larger grid spacing quickly leads
to severe errors in thermodynamic quantities. Smaller grid spacing may be necessary for some applications (e.g.,
mapping potentials of mean force) but this is rare and computationally expensive. A buffer distance between the
solute and the edges of the solvent box should typically be 14 Å for water or larger for ionic solutions. Molecular
dynamics[168], minimization and trajectory post-processing[199] have different requirements for the maximum
residual tolerance. Molecular dynamics does well with a tolerance of 10−5 and npropagate=5. Minimization
requires tolerances of 10−11 or lower and drms ≥ 10−4 . Trajectory post-processing for MM/RISM type calculations
typically have high statistical noise from the trajectory itself and it is possible to use a tolerance of 10−3 and
npropagate=1. However, this should be compared against a tolerance of 10−5 on a subset of the data before
committing to this level of accuracy.

6.2.4. Solution Convergence

6.3. Work Flow
Using 3D-RISM with SANDER or NAB for molecular dynamics, minimization or snapshot analysis is very
similar to using implicit solvent models like GBSA or PBSA. However, some additional preliminary setup is
required, the extent of which depends on the solvent to be used.
3D-RISM requires detailed information of the bulk solvent in the form of the site-site susceptibility, χ VV , and
properties such as the temperature and partial charges. This is read in as an .xvv file, which is produced by a
1D-RISM calculation. If another 3D-RISM calculation is to be preformed with any details of the bulk solvent
changed (e.g., temperature or pressure) a new .xvv file must be produced. Examples of precomputed .xvv files
for SPC/E and TIP3P water can be found in $AMBERHOME/AmberTools/test/rism1d.
Special care must be taken when producing .xvv files for use with 3D-RISM, particularly with respect to grid
parameters. It is important that the spatial extent of the grid be large enough to capture the essential long range
features of the solvent while the spacing must be fine enough to sample the short-range structure. A grid spacing
of 0.025 Å is sufficient for most applications. The number of grid points required, which will determine the
physical length of the grid in Å, generally depends on the properties of the solvent. Low concentration aqueous
salt solutions typically require much larger grids than pure bulk water. A good indicator that the grid is large
enough is convergence of delhv0 in the .xvv file. When converged, delhv0 should retain four to five digits of
precision when the number of grid points is doubled.
1D-RISM calculations require details of the some bulk properties of the solvent, such as temperature and dielectric constant, and an explicit model of the molecular components. These are read in from one or more .mdl
files, depending on the composition of the solvent. Several .mdl files are included in the Amber11 distribution
and can be found in $AMBERHOME/dat/rism1d/mdl. These include many of the explicit models for solvent and
ions used with the Amber force fields. Other solvents models may be used by creating appropriate MDL files. See
http://ambermd.org/formats.html for format details.

6.4. rism1d
1D-RISM calculations are carried out with rism1d, and require only one input file with an .inp suffix. The
input file is listed on the command line without this suffix.
rism1d inputfile

Parameters for the calculation are read in from parameters name list.

6.4.1. Parameters
Note that these keywords are not case sensitive.

6. Reference Interaction Site Model
Theory

theory

[DRISM] The 1D-RISM theory to use.
DRISM Dielectrically consistent RISM (recommended).
XRISM Extended RISM.

closure

[KH] The type of closure to use.
KH Kovalenko-Hirata (recommended).
PSEn Partial serial expansion of order n. E.g., “PSE3”.
HNC Hyper-netted chain equation.
PY Percus-Yevick.

temperature_deriv [1] Solve another set of integral equations to calculate the temperature derivative. This typically
adds less than 50% to the compute time and yields an energy/entropy decomposition of the excess
chemical potential for all species and sites.
0 Do not calculate the temperature derivative.
1 Calculate the temperature derivative.
Grid Size

[0.025] Grid spacing in real space in Å.

[16384] Number of grid points. Should be a product of small prime factors (2, 3 and 5).

Output

outlist

[] Indicates what output files to produce. Output file names use the root name of the input file with
an extension listed below. This is a list of any combination of the following characters in any order,
upper or lower case.
U U VV (r) Solvent site-site potential in real space, inputfile.uvv (see http://ambermd.org/formats.html).
X χ VV (k) Solvent site-site susceptibility in reciprocal space. Required input for 3D-RISM, inputfile.xvv

(see http://ambermd.org/formats.html).
G GVV (r) Solvent site-site pair distribution function in real-space, inputfile.gvv (see http://ambermd.org/formats.html).
B BVV (r) Solvent site-site bridge correction in real space, inputfile.bvv (see http://ambermd.org/formats.html).
T Thermodynamic properties of the solvent, inputfile.therm (see http://ambermd.org/formats.html).
E exN VV (r), exN VV Solvent site-site running, inputfile.exnvv, and total, inputfile.n00 (see

http://ambermd.org/formats.html), excess coordination numbers in real space.
N

N VV (r) Solvent site-site running coordination numbers in real space, inputfile.nvv (see http://ambermd.org/formats.htm

Q exQVV Solvent site-site excess total charge of site γ about α, inputfile.q00 (see http://ambermd.org/formats.html).
S SVV (k) Solvent site-site structure factor in reciprocal space, inputfile.svv (see http://ambermd.org/formats.html).

rout

[0] Largest real space separation in Å for output files. If 0 then all grid points will be output.

kout

[0] Largest reciprocal space separation in Å-1 for output files. If 0 then all grid points will be output.

ksave

[-1] Output an intermediate solution every ksave steps. If ksave <= 0 then no intermediate restart
files are written. If any restart files are present at run time (.sav suffix) they are automatically used.
However, such files are non-portable binary files.

6.4. rism1d
progress

[1] Write the current residue to standard output every progress iteration. If progress <= 0 then
residue is not reported.

selftest

[0] If ‘1’, perform a self-consistency check and output the results to inputfile.self.test. Only
tests applicable to the input parameters and system are performed. The results will depend on the input
parameters (e.g., ‘tolerance’) used.

Species keywords

For each molecular species in the solvent mixture, a species name list should be provided.
density

[] (Required.) Density of the species in M. See ’units’ below.

units

[‘M’] Units for density value. Options are ‘M’ (molar), ‘mM’ (millimolar), ‘1/A^3’ (number per Å3 ),
‘g/cm^3’ (g/cm3 ) or ‘kg/m^3’ (kg/m3 ).

model

[] (Required.) Relative or absolute path to and name of the .mdl file with the parameters for this solvent
molecule.

Solution Convergence
rism1d uses MDIIS to accelerate convergence. The default parameters for this method are usually near optimal
but some systems can be difficult to converge. In such cases it may be useful to use a small step size (mdiis_del=0.1
or 0.2). Occasionally, the target tolerance of 10−12 can not be achieved. A tolerance of 10−10 to 10−11 is often
sufficient but it is advisable to check how sensitive your calculations are to this.

mdiis_nvec [20] Number of MDIIS vectors to use.
mdiis_del

[0.3] MDIIS step size.

tolerance

[1e-12] Target residual tolerance for the self-consistent solution.

maxstep

[10000] Maximum number of iterations to converge to a solution.

extra_precision [1] Controls the use of extra precision routines at key points in the 1D-RISM solver. This can
be useful for achieving low tolerances or for very large box lengths but increases computational cost.
Strongly recommended for solutions with charged particles (e.g., salts).
0

No extra precision routines are used.

Sensitive matrix multiplication and addition routines are done in extra precision. A small
computational cost is incurred.

Solvent Description

temperature [298.15] Temperature in Kelvin.
dieps

[] (Required.) Dielectric constant of the solvent.

nsp

[] (Required.) Number of species (molecules) in the solutions. Also indicates the number of species
name lists to follow.

Other

smear

[1.0] Charge smear parameter in Å for long range asymptotics corrections.

adbcor

[0.5] Numeric parameter for DRISM.

6. Reference Interaction Site Model

6.4.2. Example
Mixed ionic solvent.
&PARAMETERS
THEORY=’DRISM’, CLOSURE=’KH’,
!Theory
NR=16384, DR=0.025,
!Grid size and spacing
OUTLIST=’x’, ROUT=384, KOUT=0,
!Output
MDIIS_NVEC=20, MDIIS_DEL=0.3, TOLERANCE=1.e-12,
!MDIIS
KSAVE=-1,
!Check pointing
PROGRESS=1,
!Output frequency
MAXSTEP=10000,
!Maximum iterations
SMEAR=1, ADBCOR=0.5,
!Electrostatics
TEMPERATURE=310, DIEPS=78.497, NSP=3 !bulk solvent properties
/
&SPECIES
!SPC/E water
DENSITY=55.296d0,
!very close to 0.0333 1/A3
MODEL="../../../dat/rism1d/model/SPC.mdl"
/
&SPECIES
!Sodium
units=’mM’
DENSITY=100,
MODEL="../../../dat/rism1d/model/Na+.mdl"
/
&SPECIES
!Chloride
units=’g/cm^3’
DENSITY=35.45e-4,
MODEL="../../../dat/rism1d/model/Cl-.mdl"
/

6.5. 3D-RISM in NAB
3D-RISM functionality is available in NAB and is built as part of the standard install procedure. MPI functionality for 3D-RISM in NAB requires some additional information at compile time, described in Section 6.5.5. At
this time, standard molecular dynamics and minimization with non-polarizable force fields are supported.

6.5.1. Solvation Box Size
The non-periodic solvation box super-cell can be defined as variable or fixed in size. When a variable box size
is used, the box size will be adjusted to maintain a minimum buffer distance between the atoms of the solute and
the box boundary. This has the advantage of maintaining the smallest possible box size while adapting to changes
of solute shape and orientation. Alternatively, the box size and grid spacing can be explicitly specified at run-time
and used for the duration of the calculation.
Regardless of how the solvation box is defined, the “center” of the solute is placed in the middle of the box. The
center of the solute and how it is placed in the solvent box is controlled with the centering keyword. Generally,
centering=1 (center=center-of-mass) is the default and should be used for MD and centering=2 (center=center-ofgeometry) should be used for minimization. Center-of-mass and center-of-geometry are conserved quantities in
each method respectively.
Other options for solute centering are available for special situations. To restrict the absolute position of gridpoints to be integer multiples of the grid-spacing (e.g., (2.5 Å,3.0 Å) for a grid spacing of 0.5 Å) use centering=3

6.5. 3D-RISM in NAB
for center-of-mass and centering=4 for center-of-geometry. To perform centering only on the first calculation (i.e.,
first step of MD or minimization or first frame of a trajectory analysis), use the negative integer corresponding to
the desired center definition. This allows the solute to drift in the solvent box. Finally, with some care, it is possible
to achieve custom centering using centering=0. Here, no solute centering is performed and the solvent grid has an
+ dx, y-length
+ dy, z-length
+ dz). If you use centering=0, it is advisable to
origin of (0,0,0) and a center of ( x-length
2
2
2
use a fixed-size solvent box.

6.5.2. I/O
All 3D-RISM options, including input and output files, are specified using mm_options() (see Section 37.1).
Generated output files can be quite large and numerous. For each type of correlation, a separate file is produced
for each solvent atom type. The frequency that files are produced is controlled by the ntwrism parameter. For
every time step that output is produced, a new set of files is written with the time step number in the file name. For
example, a molecular dynamics calculation using an SPC/E water model with ntwrism=2 and guvfile=guv will
produce two files on time step ten: guv.O.10.dx and guv.H1.10.dx.

6.5.3. Examples
Molecular Dynamics
.
.
.
mm_options("ntpr=100, ntpr_md=100");
mm_options("dt=0.002");
mm_options("rattle=1");
mm_options("cut=999.0");

//Large time step
//Use RATTLE
//No solute-solute
//cut off
mm_options("rism=1");
//Use 3D-RISM-KH
mm_options("xvvfile=../rism1d/spc/spc.xvv.save"); //1D-RISM input
.
.
.

Minimization
.
.
.
mm_options("ntpr=1, cut=999.0");

//No solute-solute
//cut off
mm_options("rism=1");
//Use 3D-RISM-KH
mm_options("xvvfile=../rism1d/spc/spc.xvv.save"); //1D-RISM input
mm_options("tolerance=1e-11");
//Low tolerance
mm_options("solvcut=999.0");
//No solute-solvent
//cut off
mm_options("centering=2");
//Center solute
//using center//of-geometry
.
.
.

6.5.4. Thermodynamic Output
When nptrism6= 0 thermodynamic data about the solvent is output. This is presented as a table
solute_epot:

Total
Angle
Coulomb-14

LJ
Dihedral
Restraints

Coulomb
H-Bond
3D-RISM

Bond
LJ-14

6. Reference Interaction Site Model
Solute internal energy [kcal/mol] and its components. This is written as a single line.
rism_exchem:

Total

ExChem_1

ExChem_2

...

Excess chemical potential (solvation free energy) [kcal/mol] for the closure used and the contribution from each
solvent atom type.
rism_exchGF:

Total ExChem_GF_1 ExChem_GF_2

...

Excess chemical potential (solvation free energy) [kcal/mol] using the Gaussian fluctuation approximation and
the contribution from each solvent atom type.
rism_exEnUV:

Total

Energy_1

Energy_2

...

Average solute-solvent interaction energy [kcal/mol],
Z

∆Usol = ∑ ρα

UV
drgUV
α (r)uα (r),

and the contribution from each solvent atom type. Note that this is only a component of the solvation energy as it
does not include changes in the solvent-solvent interaction energy[203].
rism_volume:

PMV

Partial molar volume of the solute [Å3 ].
rism_exNumb:

ExNum_1

ExNum_2

...

Excess number of each atom type of solvent accumulated by the solute.
rism_exChrg:

Total

ExChg_1

ExChg_2

...

Excess charge [e] of each atom type of solvent accumulated by the solute.
rism_polar_:

Total

polar_1

polar_2

...

Solvent polarization contribution to the total excess chemical potential [kcal/mol] and the contribution from each
solvent atom type. Only present when polardecomp=1.
rism_apolar:

Total

apolar_1

apolar_2

...

Cavity formation and dispersion contribution to the total excess chemical potential [kcal/mol] and the
contribution from each solvent atom type. Only present when polardecomp=1.
rism_polGF_:

Total

polarGF_1

polarGF_2

...

Solvent polarization contribution to the Gaussian fluctuation total excess chemical potential [kcal/mol] and the
contribution from each solvent atom type. Only present when polardecomp=1.
rism_apolGF:

Total

apolarGF_1

apolarGF_2

...

Cavity formation and dispersion contribution to the Gaussian fluctuation total excess chemical potential [kcal/mol]
and the contribution from each solvent atom type. Only present when polardecomp=1.

6.5.5. Compiling MPI 3D-RISM
Executables compiled with mpinab and 3D-RISM must link to both C and Fortran MPI libraries, which is not the
default behaviour of most MPI compilers. As there are a wide variety of MPI implementations and no standards
for naming Fortran libraries, 3D-RISM is not included by default when compiling mpinab. The additional steps
required to include 3D-RISM in mpinab are

100

6.6. rism3d.snglpnt
1. If
a) you are using OpenMPI or MPICH2, proceed to step 2.
b) you are not using OpenMPI or MPICH2, identify the Fortran 77 libraries corresponding to your MPI
implementation. These will be found in the lib directory for your MPI implementation and will likely
contain “f” or “f77” in the file name. Set the XTRA_FLIBS environment variable to contain the compiler directive to link the library.
For example, the OpenMPI and MPICH2 library files are libmpi_f77.a and libfmpich.a respectively (the
suffix may vary) and XTRA_FLIBS could be explicitly set as:
OpenMPI export XTRA_FLIBS=-lmpi_f77
MPICH2 export XTRA_FLIBS=-lfmpich
2. Run configure and specify both -mpi and -rismmpi. For example:
./configure -mpi -rismmpi gnu

3. For dynamically linked executables (the default), set your LD_LIBRARY_PATH environment variable to the
location of your MPI library:
where

export LD_LIBRARY_PATH=$MPIHOME/lib

$MPIHOME is the base directory for you MPI installation.

6.6. rism3d.snglpnt
3D-RISM functionality is also available in the command line tools rism3d.snglpnt and rism3d.snglpnt.MPI installed at compile time. These programs perform single point 3D-RISM calculations on trajectories and individual
solute snapshots. No other processing is done to the structures so unwanted solvent molecules should be removed
before hand. Except for minimization and molecular dynamics, all 3D-RISM features are available. Thermodynamic data is always output (see Section 6.5.4). Note that these executables are built by NAB so please see Section
6.5.5 on ensuring rism3d.snglpnt.MPI is built.

6.6.1. Usage
3D-RISM specific command line keywords generally correspond to keyword options available in NAB’s mm_options
(see Section 37.1). If run without input, rism3d.snglpnt prints default settings for all parameters.
--pdb PDB file (Required, input.) PDB file for the solute. Coordinates are only used if a restart or trajectory file

is not supplied.
--prmtop prmtop file (Required, input.) Parameter topology file for the solute.
--rst restart file (Optional, input.) Coordinates for the solute in restart format.
--nc NetCDF file (Optional, input.) Trajectory for the solute in NetCDF format.
--xvv XVV file (Required, input.) Bulk solvent susceptibility file from 1D-RISM (see http://ambermd.org/formats.html).
--guv GUV root (Optional, output.) Root name for 3D solvent pair distribution files.
--cuv CUV root (Optional, output.) Root name for 3D solvent direct correlation files.
--huv HUV root (Optional, output.) Root name for 3D solvent total correlation files.
--uuv UUV root (Optional, output.) Root name for 3D solvent potential [kT ] files.
--asymp asymptotics root (Optional, output.) Root name for 3D real-space long range asymptotics for total and

direct correlation files. This will produce one file for each of C and H for each frame requested and
does not include the solvent site charge. Multiply the distribution by the solvent site charge to obtain
the long-range asymptotics for that site.

101

6. Reference Interaction Site Model
--quv QUV root (Optional, output.) Root name for 3D solvent charge density distribution files. This is the charge

density [e/ Å] at each grid point with contributions from all solvent types.
--chgdist charge distribution root (Optional, output.) Root name for 3D solvent charge distribution files. This

gives a point charge [e] at each grid point with contributions from all solvent types.
--volfmt

(Optional.) Format of volumetric data files. May be dx for DX files or xyzv for XYZV format (see
http://ambermd.org/formats.html).

--closure closure name (Optional.) A list of one or more of KH, HNC or PSEn where “n” is a positive integer.

If more than one closure is provided, the 3D-RISM solver will use the closures in order to obtain a
solution for the last closure in the list when no previous solutions are available. The solution for the
last closure in the list is used for all output. This can be useful for difficult to converge calculations
(see §6.2.4).
--closureorder closure order (Deprecated.) Specifies the order of the PSE-n closure if the closure name is

given as “PSE” or “PSEN” (no integers).
--noasympcorr (Optional.) Turn off long range asymptotic corrections for thermodynamic output only. Long-

range asymptotics are still used to calculate the solution.
--buffer distance (Optional.) Minimum distance between the solute and the edge of the solvent box. Use this
with --grdspc. Incompatible with --ng and --solvbox.
--solvcut distance (Optional.) Set solute-solvent interaction cut off distance. If no value is specified then the

buffer distance is used. If a buffer distance is not provided, the cut off must be explicitly set. Note that
Coulomb interactions are interpolated and not truncated beyond the cut off. See [168] for details.
--grdspc 3D grid spacing (Optional.) Comma separated linear grid spacings for x, y and z dimensions. Use this
with --buffer. Incompatible with --ng and --solvbox.
--ng 3D grid points (Optional.) Comma separated number of grid points for x, y and z dimensions. Use this with
--solvbox. Incompatible with --buffer and --grdspc.
--solvbox 3D box length (Optional.) Comma separated solvation box side length for x, y and z dimensions. Use
this with --ng. Incompatible with --buffer and --grdspc.
--tolerance residual target (Optional.) A list of maximum residual values for solution convergence. When

used in combination with a list of closures it is possible to define different tolerances for each of
the closures. This can be useful for difficult to converge calculations (see §6.2.4). For the sake of
efficiency, it is best to use as high a tolerance as possible for all but the last closure. Three formats of
list are possible.
one tolerance All closures but the last use a tolerance of 1. The last tolerance in the list is used by the
last closure. In practice this, is the most efficient.
two tolerances All closures but the last use the first tolerance in the list. The last tolerance in the list
is used by the last closure.
n tolerances Tolerances from the list are assigned to the closure list in order.
--mdiis_del step size (Optional.) MDIIS step size.
--mdiis_nvec # of vectors (Optional.) Number of previous iterations MDIIS uses to predict a new solution.
--maxstep step number (Optional.) Maximum number of iterative steps per solution.
--npropagate # old solutions (Optional.) Number of previous solutions to use in predicting a new solution.
--polarDecomp (Optional.) Decomposes solvation free energy into polar and non-polar components. Note that

this typically requires 80% more computation time.

102

6.7. 3D-RISM in sander
--centering method (Optional.) Select how solute is centered in the solvent box.

-4 Center-of-geometry with grid-point rounding. Center on first step only.
-3 Center-of-mass with grid-point rounding. Center on first step only.
-2 Center-of-geometry. Center on first step only.
-1 Center-of-mass. Center on first step only.
0 No centering. Dangerous.
1 Center-of-mass. Center on every step. Recommended for molecular dynamics.
2 Center-of-geometry. Center on every step. Recommended for minimization.
3 Center-of-mass with grid-point rounding.
4 Center-of-geometry with grid-point rounding.
--verbose level (Optional.)

0 No output.
1 Print the number of iterations required to converge.
2 Print convergence details for each iteration.

6.7. 3D-RISM in sander
In addition to explicit and continuum implicit solvation models, sander has a third type of solvation model
for molecular mechanics simulations, the reference interaction site model (RISM) of molecular solvation[168–
181, 183]. General information is given above; some features specific to sander are discussed here.

6.7.1. Multiple Time Step Methods for 3D-RISM
At this time, the computational cost of 3D-RISM is still prohibitive for performing calculations at each step of
molecular dynamics calculations. One of the most effective ways to reduce this computational burden is to reduce
the number of solutions calculated by using multiple time step (MTS) methods. Two MTS methods, r-RESPA and
force-coordinate extrapolation (FCE), are implemented for 3D-RISM and can be combined such that solutions are
only calculated once every 10 or 20 fs. At this time, these methods are only available for sander and not NAB.
r-RESPA[204, 205] and I-Verlet[206] impulse MTS algorithms are widely used methods to reduce the computational load of long-range interactions while maintaining the desirable properties of energy conservation and time
reversibility. Impulse MTS can be invoked for 3D-RISM independent of the existing r-RESPA implementation
using the RISMnRESPA variable. For typical biomolecular simulations, impulse MTS is limited to a maximum
step size of 5 fs[207]. Since the computational load of calculating all internal interactions of the solute is small
compared to the 3D-RISM calculation, it is recommend to use dt=0.001, nrespa=1 and RISMnRESPA=5.
To overcome the stability limitation of impulse MTS, FCE uses an efficient extrapolation method to predict the
forces for some time steps rather than computing a full 3D-RISM solution[168]. In this method, forces, {F}, on
N U solute atoms for the current time step tk are approximated as a linear combination of forces from the n previous
time steps obtained from 3D-RISM calculations,
{F}(k) =

∑ akl {F}(l) ,

l ∈ 3D-RISM steps.

(6.17)

l=1

The weight coefficients akl are obtained by expressing the current set of coordinates, {R}(k) , as a linear combination
of coordinates from the n previous time steps for which 3D-RISM calculations were performed. That is, the current
set of coordinates is projected onto the basis of n previous solute arrangements by minimizing the norm of the

103

6. Reference Interaction Site Model

}
Force

Time

solvation
forces

extrapolated
solvation forces

RISMnRESPA

FCEstride X RISMnRESPA

FCEnbasis

Figure 6.1.: Multiple time step methods in 3D-RISM. RISMnRESPA(= 5) is the number of base time steps between
application of solvation forces (exact or extrapolated). FCEnbasis(= 4) is the number of previous
solutions used to extrapolate forces, in this case four previous solutions. Once FCEnbasis solutions
have be calculated, exact 3D-RISM forces are calculated every FCEstride(= 2)×RISMnRESPA time
steps; solvation forces are otherwise obtained through extrapolation.
difference between the current 3 × N U matrix of coordinates {R}(k) and the corresponding linear combination of
the previous ones {R}(l) ,
n

minimize {R}(k) − ∑ akl {R}(l) .
l=1

Coefficients akl are then used in Equation (6.17)to extrapolate forces at the current intermediate time step. Similarly, the known coordinates for the current time step can be approximated from previous time steps as
{R}(k) =

∑ akl {R}(l) .
l=1

FCE MTS does not conserve energy and is not time reversible. However, 3D-RISM calculations can be reduced
to a frequency of once every 10 to 20 fs and stable dynamics achieved by using a Langevin thermostat with
gamma_ln=10 to 20 ps−1 . Combined impulse FCE MTS calculations (see Figure 6.1) start the simulation using
impulse MTS until the requested size for the basis set, FCEnbasis, is achieved. After a large enough basis set is
collected, 3D-RISM calculations are only performed once every FCEstride × RISMnRESPA time steps.

6.7.2. 3D-RISM in sander
Full 3D-RISM functionality is available in sander as part of the standard install procedure. However, some
methods available in sander are not compatible with 3D-RISM, such as QM/MM simulations. At this time, only
standard molecular dynamics, minimization and trajectory post-processing with non-polarizable force fields are
supported. With the exception of multiple time step features, 3D-RISM keywords in sander are identical to those
in NAB, rism3d.snglpnt and MMPBSA.py.
3D-RISM specific command line options for sander are
sander [standard options] -xvv xvvfile -guv guvroot -huv huvroot
-cuv cuvroot -uuv uuvroot -asymp asympfile
-quv quvroot -chgdist chgdistroot

xvvfile input description of bulk solvent properties, required for 3D-RISM calculations. Produced by rism1d.
guvroot output rootname for solute-solvent 3D pair distribution function, GUV (R). This will produce one file for

each solvent atom type for each frame requested.

104

6.7. 3D-RISM in sander
huvroot output rootname for solute-solvent 3D total correlation function, H UV (R). This will produce one file for

each solvent atom type for each frame requested.
cuvroot output rootname for solute-solvent 3D total correlation function, CUV (R). This will produce one file for

each solvent atom type for each frame requested.
uuvroot output rootname for solute-solvent 3D potential energy function, U UV (R), in units of kT . This will

produce one file for each solvent atom type for each frame requested.
asympfile output rootname for solute-solvent 3D long-range real-space asymptotics for C and H. This will pro-

duce one file for each of C and H for each frame requested and does not include the solvent site charge.
Multiply the distribution by the solvent site charge to obtain the long-range asymptotics for that site.
quvroot output rootname for solute-solvent 3D charge density distribution [e/ Å]. This will produce one file that

combines contributions from all solvent atom types for each frame requested.
chgdist output rootname for solute-solvent 3D charge distribution [e]. This will produce one file that combines

contributions from all solvent atom types for each frame requested.
Generated output files can be large and numerous. For each type of correlation, a separate file is produced for each
solvent atom type. The frequency that files are produced is controlled by the ntwrism parameter. Every time step
that output is produced, a new set of files is written with the time step number in the file name. For example, a
molecular dynamics calculation using an SPC/E water model with ntwrism=2 and -guv guv on the command line
will produce two files on time step ten: guv.O.10.dx and guv.H1.10.dx.
6.7.2.1. Keywords

With the exception of irism, which is found in the &cntrl name list, all 3D-RISM options are specified in the
&rism name list.

irism

[0] Use 3D-RISM. Found in &cntrl name list.
= 0 Off.
= 1 On.

Closure Approximation

closure

[KH] Comma separate list of closure approximations. If more than one closure is provided, the 3DRISM solver will use the closures in order to obtain a solution for the last closure in the list when no
previous solutions are available. The solution for the last closure in the list is used for all output.
= KH Kovalenko-Hirata (KH).
= HNC Hyper-netted chain equation (HNC).
=PSEn Partial series expansion of order-n (PSE-n), where “n” is a positive integer.

Long-range asymptotics are used to analytically account for solvent distribution beyond the solvent box. Long-range asymptotics are always used to when calculating solution but can be omitted for
the subsequent thermodynamic calculations, though it is not recommended.
Long-range asymptotics

asympcorr [T] Use long-range asymptotic corrections for thermodynamic calculations.
= T Use the long-range corrections.
= F Do not use long-range corrections.

105

6. Reference Interaction Site Model
Solvation Box The non-periodic solvation box super-cell can be defined as variable or fixed in size. When a
variable box size is used, the box size will be adjusted to maintain a minimum buffer distance between the atoms
of the solute and the box boundary. This has the advantage of maintaining the smallest possible box size while
adapting to changes of solute shape and orientation. Alternatively, the box size can be specified at run-time. This
box size will be used for the duration of the sander calculation.

solvcut

[buffer] Cut-off distance for solvent-solute potential and force calculations. If buffer < 0 and
solvcut is not explicitly set, solvcut = |buffer|. For minimization it is recommended to not use a
cut-off (e.g. solvcut=9999).

Variable Box Size

buffer

[14] Minimum distance in Å between the solute and the edge of the solvent box.
< 0 Use fixed box size (ng3 and solvbox).
>= 0 Buffer distance.

grdspc

[0.5,0.5,0.5] Linear grid spacing in Å.

Fixed Box Size

ng3

[] Sets the number of grid points for a fixed size solvation box. This is only used if buffer< 0.
nx,ny,nz

solvbox

Points for x, y and z dimensions.

[] Sets the size in Å of the fixed size solvation box. This is only used if buffer< 0.
lx,ly,lz

Box length in x, y and z dimensions.

Solution Convergence

tolerance

[1e-5] A list of maximum residual values for solution convergence. When used in combination with a
list of closures it is possible to define different tolerances for each of the closures. This can be useful
for difficult to converge calculations (see Subsection 6.4.1 for details). For the sake of efficiency, it is
best to use as high a tolerance as possible for all but the last closure. For minimization a tolerance of
1e-11 or lower is recommended. Three formats of list are possible.
one tolerance All closures but the last use a tolerance of 1. The last tolerance in the list is used by the
last closure. In practice this, is the most efficient.
two tolerances All closures but the last use the first tolerance in the list. The last tolerance in the list
is used by the last closure.
n tolerances Tolerances from the list are assigned to the closure list in order.

mdiis_del

[0.7] “Step size” in MDIIS.

mdiis_nvec [5] Number of vectors used by the MDIIS method. Higher values for this parameter can greatly
increase memory requirements but may also accelerate convergence.
mdiis_method [2] Specify implementation of the MDIIS routine.
= 0 Original. For small systems (e.g. < 643 grid points) this implementation may be faster than the

BLAS optimized version.
= 1 BLAS optimized.
= 2 BLAS and memory optimized.

maxstep

106

[10000] Maximum number of iterations allowed to converge on a solution.

6.7. 3D-RISM in sander
npropagate [5] Number of previous solutions propagated forward to create an initial guess for this solute atom
configuration.
= 0 Do not use any previous solutions
= 1..5 Values greater than 0 but less than 4 or 5 will use less system memory but may introduce

artifacts to the solution (e.g., energy drift).
Minimization and Molecular Dynamics

centering

[1] Controls how the solute is centered/re-centered in the solvent box.
= -4 Center-of-geometry with grid-point rounding. Center on first step only.
= -3 Center-of-mass with grid-point rounding. Center on first step only.
= -2 Center-of-geometry. Center on first step only.
= -1 Center-of-mass. Center on first step only.
= 0 No centering. Dangerous.
= 1 Center-of-mass. Center on every step. Recommended for molecular dynamics.
= 2 Center-of-geometry. Center on every step. Recommended for minimization.
= 3 Center-of-mass with grid-point rounding.
= 4 Center-of-geometry with grid-point rounding.

zerofrc

[1] Redistribute solvent forces across the solute such that the net solvation force on the solute is zero.
= 0 Unmodified forces.
= 1 Zero net force.

Trajectory Post-Processing

apply_rism_force [1] Calculate and use solvation forces from 3D-RISM. Not calculating these forces can save
computation time and is useful for trajectory post-processing.
= 0 Do not calculate forces.
= 1 Calculate forces.
Multiple Time Steps

Multiple time step features are only available in sander.

rismnrespa [1] rismnrespa × dt =RISM RESPA multiple time step size. 5 fs is the maximum time step. “1”
corresponds to no multiple time stepping.
fcestride

[0] fcestride × rismnrespa × dt = FCE multiple time step size. I.e., full 3D-RISM solutions
are performed every fcestride × rismnrespa steps. In between full solutions extrapolated force
impulses are applied every rismnrespa steps. Maximum step size for stable dynamics depends on
damping coefficient for Langevin dynamics. “1” corresponds to no multiple time stepping.
= 0 No FCE multiple time stepping.
= 1 Invokes the FCE code but yields the same trajectories as 0.
>= 1 Invoke FCE with 3D-RISM solutions every fcestride × rismnrespa steps.

fcenbasis

[10] Number of previous full solutions used to extrapolate new forces. If FCE is not used this can be
set to 1 to reduce memory usage.

fcecrd

[0] The coordinates used for the FCE method.
= 0 The absolute x, y, z position of each neighbour atom (with translations due to centering).

107

6. Reference Interaction Site Model
= 1 For predicting the forces on atom i, use the distance of each neighbour atom as the “coordinate”.

This has one third the number of coordinates to use in the prediction. Also, directional information is lost.
= 2 For predicting the forces on atom i, use the x, y, z position of each neighbour atom with atom i as

the origin. Recommended.
Output

ntwrism

[0] Indicates that solvent density grid should be written to file every ntwrism iterations.
= 0 No files written.
>= 1 Output every ntwrism time steps.

volfmt

[‘DX’] Format of volumetric data files. May be ’dx’ for DX files or ’xyzv’ for XYZV format. See the
AmberTools manual for more information.

verbose

[0] Indicates level of diagnostic detail about the calculation written to the log file.
= 0 No output.
= 1 Print the number of iterations used to converge.
= 2 Print details for each iteration and information about what FCE is doing every progress itera-

tions.
write_thermo [1] Print solvation thermodynamics in addition to standard sander output. The format is the same
as that found in NAB and rism3d.snglpnt.
polarDecomp [0] Decomposes solvation free energy into polar and non-polar components. Note that this typically
requires 80% more computation time.
= 0 No polar/non-polar decomposition.
= 1 Polar/non-polar decomposition.

progress

[1] Display progress of the 3D-RISM solution every kshow iterations. 0 indicates this information will
not be displayed. Must be used with verbose > 1.

6.7.2.2. Example
Molecular Dynamics (imin=0)
molecular dynamics with 3D-RISM and impulse MTS
&cntrl
ntx=1, ntpr=100, ntwx=1000,ntwr=1000,
nstlim=10000,dt=0.001,
!No shake or r-RESPA
ntt=3, temp0=300, gamma_ln=20,
!Langevin dynamics
ntb=0,
!Non-periodic
cut=999.,
!Calculate all
!solute-solute
!interactions
irism=1,
/
&rism
rismnrespa=5,
!r-RESPA MTS
fcenbasis=10,fcestride=2,fcecrd=2
!FCE MTS
/

108

6.7. 3D-RISM in sander
Minimization (imin=1)
Default XMIN minimization with 3D-RISM
&cntrl
imin=1, maxcyc=200,
drms=1e-3,
!RMS force. Can be as low as 1e-4
ntmin=3,
!XMIN
ntpr=5,
ntb=0,
!Non-periodic
cut=999.,
!Calculate all
!solute-solute interactions
irism=1
/
&rism
tolerance=1e-11,
!Low tolerance
solvcut=9999,
!No cut-off for
!solute-solvent interactions
centering=2
!Solvation box centering
!using center-of-geometry
/

Trajectory Post-Processing (imin=5)
Trajectory post-processing with 3D-RISM
&cntrl
ntx=1, ntpr=1, ntwx=1,
imin=5,maxcyc=1,
!Single-point energy calculation
!on each frame
ntb=0,
!Non-periodic
cut=9999.,
!Calculate all
!solute-solute interactions
irism=1
/
&rism
tolerance=1e-4,
!Saves some time compared to 1e-5
apply_rism_force=0,
!Saves some time. Forces are not used.
npropagate=1
!Saves some time and 4*8*Nbox bytes
!of memory compared to npropagate=5.
/

109

7. Empirical Valence Bond
7.1. Introduction
Chemical reactivity can be formulated within the empirical valence bond (EVB) model[208, 209], whereby
the reactive surface is defined as the lowest adiabatic surface obtained by diagonalization of the potential energy
matrix in the representation of non-reactive diabatic states. These diabatic states can be described by a force field
approach, such as Amber, or by a prescription incorporating information from ab initio calculations. The coupling
elements in the matrix embody all the physics needed for describing transitions between the diabatic states.
As an example, the intramolecular proton transfer reaction in malonaldehyde (Figure 7.1) can be described by a
two-state EVB matrix

V11 V12
V=
(7.1)
V21 V22
where valence bond state 1 represents the reactant state (RS) with the proton H9 bonded to O8 and valence bond
state 2 represents the product state (PS) with the proton bonded to O7. The matrix elements V11 and V22 are
simply the energies of the reactant and product systems. The off-diagonal elements of this symmetric matrix, i.e.
V12 = V21 , couple these diabatic states.
Amber provides several options for computing the V12 resonance integrals. In its simplest form, V12 is set to
a constant value which provides an EVB surface that reproduces experimental or ab initio barrier heights. More
flexibility can be introduced into V12 by employing an exponential or Gaussian function of the coordinates. It has
recently been shown [210, 211] that a linear combination of distributed Gaussian functions is the most accurate
and flexible form for V12 . With a set of distributed Gaussians, V12 can be fit to high-level electronic structure data
using the following form,
NDim
2
V12
(q) = ∑

∑

Bi jK g (q, qK , i, j, αK )

(7.2)

K i≥ j≥0

2
V12
(q) = [V11 (q) −V (q)] [V22 (q) −V (q)]

(7.3)

1
1
g (q, qK , 0, 0, αK ) = 1 + αK |q − qK |2 exp − αK |q − qK |2
2
2

(7.4)

9
H
8

9
H

O 7

C
3

C
C
1

5
H 6

O 7

C
3

C
C
1

H
2

5
H 6

Figure 7.1.: Intramolecular proton transfer in malonaldehyde.

111

7. Empirical Valence Bond

1
2
g (q, qK , i, 0, αK ) = (q − qK )i exp − αK |q − qK |
2

1
g (q, qK , i, j, αK ) = (q − qK )i (q − qK ) j exp − αK |q − qK |2
2

(7.5)
(7.6)

where g(q, qK , i, j, αK ) are s-, p-, and d-type Gaussians at a number of points, qK , on the potential energy surface,
NDim is the total number of internal coordinates, V is the ab initio energy and B is a vector of coefficients. It
is important to note that a nonstandard s-type Gaussian is employed to precondition the resulting set of linear
equations that is passed to a GMRES[212] (aka DIIS[213, 214]) solver. For a more exhaustive discussion of the
DG EVB method please see reference [211]. Additionally, the EVB facility in Amber can perform MD or energy
optimization on the EVB ground-state surface and biased sampling along a predefined reaction coordinate (RC).
Nuclear quantization based on the Feynman path integral formalism [215–217] is also possible.

7.2. General usage description
The EVB facility is built on top of the multisander infrastructure in Amber. (Section 17.11) As such, the user
will need to build the parallel version of sander in order to utilize the EVB feature. Information for each EVB
diabatic state is obtained from separate (simultaneous) instances of sander. The energies and forces of all the states
are communicated via MPI to the master node, which is responsible for computing the EVB energy and forces and
broadcasting these to the other nodes for the next MD step.
The required input files are (1) an EVB multisander group file containing per line all the command line options
for each sander job, (2) the mdin, coordinate, and parmtop files specified in the group file, and (3) the EVB input
files. At the top level, an EVB calculation is invoked as follows:
mpirun -np <# procs> sander.MPI -ng <# groups> -groupfile

The contents of the EVB group file is similar to that for a conventional multisander execution, with the addition
of a command line flag -evbin for specifying the name of the EVB input file. Below is an example of an EVB
group file:
# Malonaldehyde RS: H9 bonded to O8
-O -i mdin -p mr.top -c mr.crd -o mr.out -r mr.rst -evbin input.mr
# Malonaldehyde PS: H9 bonded to O7
-O -i mdin -p mp.top -c mr.crd -o mp.out -r mp.rst -evbin input.mp

Each line corresponds to a diabatic state, and comments are preceded by a # symbol in the first column of a line.
Now, it is important to notice in the above example that the starting configurations for both sander jobs are the
same, although the topology files are different. This constraint guarantees that the system starts in a physically
meaningful part of configuration space. Furthermore, it is critical that the atom numbers (delineating the atom
locations in the coordinate and parmtop files) are identical among the EVB diabatic states. In Figure 7.1, for
example, the atom numbers of the RS and PS malonaldehydes are identical. The only additional flag in the &cntrl
namelist of the mdin file is ievb, which has the following values
ievb

Flag to run EVB
=0

No effect (default)

Enable EVB. The value of imin specifies if the sander calculation is a molecular dynamics (imin=0) or an energy minimization (imin=1). The variable evb_dyn in the &evb
namelist of the EVB input file refines this choice to specify if the calculation type is on
the EVB ground-state surface, on a mapping potential, or on a biased potential.

The argument of the command line flag -evbin provides the name of the EVB input file. Corresponding to the
above group file example, the inputs for EVB state 1 are provided in the file input.mr and those for EVB state 2
are provided in input.mp. For the case of constant coupling between the EVB states, the file input.mr may look
like the following:

112

7.2. General usage description
# Malonaldehyde RS: proton (H9) bound to O8
&evb nevb = 2, nbias = 1, nmorse = 1, nmodvdw = 1, ntw_evb = 50,
xch_type
= "constant",
evb_dyn
= "egap_umb",
dia_shift(1)%st = 1, dia_shift(1)%nrg_offset = 0.0,
dia_shift(2)%st = 2, dia_shift(2)%nrg_offset = 0.0,
xch_cnst(1)%ist = 1, xch_cnst(1)%jst = 2,
xch_cnst(1)%xcnst = 12.5,
egap_umb(1)%ist = 1, egap_umb(1)%jst = 2,
egap_umb(1)%k = 0.005, egap_umb(1)%ezero = 0.0,
morsify(1)%iatom = 8, morsify(1)%jatom = 9, morsify(1)%D = 356.570,
morsify(1)%a = 1.046, morsify(1)%r0 = 1.000,
modvdw(1)%iatom = 9, modvdw(1)%jatom = 7,
/

and the file input.mp may appear as follows:
# Malonaldehyde PS: proton (H9) bound to O7
&evb nevb = 2, nbias = 1, nmorse = 1, nmodvdw = 1, ntw_evb = 50,
xch_type
= "constant",
evb_dyn
= "egap_umb",
dia_shift(1)%st = 1, dia_shift(1)%nrg_offset = 0.0,
dia_shift(2)%st = 2, dia_shift(2)%nrg_offset = 0.0,
xch_cnst(1)%ist = 1, xch_cnst(1)%jst = 2,
xch_cnst(1)%xcnst = 12.5,
egap_umb(1)%ist = 1, egap_umb(1)%jst = 2,
egap_umb(1)%k = 0.005, egap_umb(1)%ezero = 0.0,
morsify(1)%iatom = 7, morsify(1)%jatom = 9, morsify(1)%D = 356.570,
morsify(1)%a = 1.046, morsify(1)%r0 = 1.000,
modvdw(1)%iatom = 9, modvdw(1)%jatom = 8,
/

The above EVB files specify that the system is described by a two-state model, the coupling between the two-states
is a constant, and the dynamics is umbrella sampling along an energy gap RC. Since the reactant and product states
are identical by symmetry, no adjustments of the relative energies of the diabatic states are performed. The constant
value coupling between the two states is parameterized such that the EVB barrier reproduces the ab initio barrier of
~ 3 kcal/mol (RMP2/cc-pVTZ level). Lastly, the standard Amber harmonic bond interactions involving the proton
with the donor and acceptor oxygens are replaced by Morse functions and certain van der Waals interactions are
excluded.
This parameterization of the EVB surface to provide observables that match either results from high-level quantum chemistry calculations or experimental measurements is the trickiest aspect of the EVB model. However, after
the EVB surface has been calibrated, the user has access to reactive chemical dynamics simulation timescales and
lengthscales which would be otherwise inaccessible using conventional ab initio MD approaches. The distributed
Gaussian EVB framework provides a systematic procedure for computing V12 from ab initio data.
Now, let us suppose that the constant coupling prescription does not provide the detailed features needed to
describe the reaction pathway. Furthermore, we find that the coupling as a function of the coordinates can be
described adequately (from comparison to ab initio data) using a Gaussian functional form. How should one
modify the above EVB input files to obtain a more accurate reactive surface? We need to change the xch_type
variable from “constant” to “gauss” as well as replace the variable xch_cnst by the variable xch_gauss(:), which
contains the parameters for the Gaussian functional form. Of course, these parameters need to be optimized to
provide the more accurate surface. The modifications to the EVB input files look something like the following,
.
.
.
xch_type

= "constant",

113

7. Empirical Valence Bond
xch_type
= "gauss",
.
.
.
xch_cnst(1)%ist = 1, xch_cnst(1)%jst = 2,
xch_cnst(1)%xcnst = 12.5,
xch_gauss(1)%ist = 1, xch_gauss(1)%jst = 2,
xch_gauss(1)%iatom = 8, xch_gauss(1)%jatom = 7,
xch_gauss(1)%a = 11.0, xch_gauss(1)%sigma = 0.0447,
xch_gauss(1)%r0 = 2.3,
.
.
.

where the cross-through lines have been replaced by those below them. Access to the exponential functional form
or the distributed Gaussian approximation to V12 entails similar changes to the input files. Please see $AMBERHOME/test/evb for examples.

114

7.3. Biased sampling

Potential of Mean Force [kcal/mol]

0
-75

-50

-25

Collective Reaction Coordinate [kcal/mol]

Figure 7.2.: Potential of mean force along an energy gap RC for the intramolecular proton transfer in malonaldehyde as obtained from a series of mapping potential simulations.

7.3. Biased sampling
When a reactive event is described by an intrinsic high free energy barrier, molecular dynamics on the EVB
ground-state surface will not adequately sample the important transition state region. Under these conditions,
chemical reactions are rare events and sampling on the EVB surface effectively reduces to sampling on a diabatic
surface. One framework for enhancing the sampling of rare events is through modification of the system Hamiltonian with the addition of biasing potentials. The EVB facility in Amber offers several options for biased sampling:
(1) Ariel Warshel’s mapping potential approach[208] (2) Dave Case’s umbrella sampling on an energy gap RC (3)
umbrella sampling on a distance RC and (4) umbrella sampling on a difference of distances RC.
In the mapping potential framework, the system Hamiltonian (and hence, the molecular dynamics) is described
by the modified potential
Vλ = (1 − λ )Vii + λV f f

(7.7)

where Vii is the EVB matrix element for the initial state and V f f is the EVB matrix element for the final state. As
the value of the mapping potential parameter λ changes from 0 to 1, the system evolves from the initial state to the
final state. As an example, for λ = 0.50, the system Hamiltonian is an equal linear combination of the initial and
final states and molecular dynamics sample the region in the vicinity of the transition state. Each mapping potential
Vλ samples only a portion of the reaction coordinate. In practice, a series of mapping potentials are used to bias
the sampling across the entire range of the RC. The average distribution of the RC for each mapping potential is
then unbiased and the set of unbiased distributions are combined to give the potential of mean force (PMF) on the
EVB ground-state surface. Figure 7.2 shows a PMF for the malonaldehyde intramolecular proton transfer reaction
as obtained from 9 mapping potential simulations with λ ranging from 0.10 to 0.90 at 0.10 intervals.
In the umbrella sampling framework, the system Hamiltonian is described by the modified potential
(n)

(n)

Vbiased (q) = Vel0 (q) +Vumb (q)
h
i
1
(n) 2
= Vel0 (q) + k(n) RC(q) − RC0
2

(7.8)
(n)

where q is the set of system coordinates, k is the harmonic force constant parameter, and Vumb is an umbrella
potential that is added to the original system potential Vel0 (obtained from diagonalization of the EVB matrix) to
(n)
bias the sampling towards a particular value of the reaction coordinate RC0 . The superscript (n) denotes that a

115

7. Empirical Valence Bond

Potential of Mean Force [kcal/mol]

0
-75

-50

-25

Collective Reaction Coordinate [kcal/mol]

Figure 7.3.: Potential of mean force for the intramolecular proton transfer in malonaldehyde as obtained from a
series of umbrella sampling simulations along an energy gap RC. The distributions of the RC from all
the windows are combined using the WHAM procedure.

evb_dyn
“evb_map”
“egap_umb”
“bond_umb”
“dbonds_umb”
“qi_bond_pmf”
“qi_bond_dyn”
“qi_dbonds_pmf”
“qi_dbonds_dyn”

⇒
⇒
⇒
⇒
⇒
⇒
⇒
⇒

dependency
emap(:)
egap_umb(:)
bond_umb(:)
dbonds_umb(:)
bond_umb(:)
bond_umb(:)
dbonds_umb(:)
dbonds_umb(:)

Table 7.2.: Derived variable types for EVB.
series of biased simulations, each enhancing the sampling of a particular window of the RC, is required to map out
the entire PMF. The number of umbrella sampling windows as well as the choice of values for the force constant
parameter and the RC equilibrium position will depend ultimately on the nature of the free energy landscape of the
system in question.
Results from the biased samplings then can be unbiased and combined using the weighted histogram analysis
method (WHAM)[218–220] to generate the PMF describing chemistry on the physically relevant EVB groundstate potential energy surface, Vel0 . Figure 7.3 depicts the PMF for the malonaldehyde intramolecular proton
(n)
transfer that is obtained from 13 umbrella sampling simulations with RC0 spanning the range -60 kcal/mol to +60
kcal/mol at 10 kcal/mol intervals. The supporting program to generate the PMF from a set of mapping potential or
from a set of umbrella sampling simulations can be obtained from the Amber website, http://ambermd.org.
Biased sampling is accessed through the nbias and evb_dyn variables in the EVB input file. The variable nbias
specifies the number of biasing potentials to include in the system Hamiltonian. Mapping potential dynamics is
invoked using the assignment evb_dyn=“evb_map”. Biased sampling via umbrella potentials is invoked with the
assignment evb_dyn=“egap_umb”, evb_dyn=“bond_umb” or evb_dyn=“dbonds_umb”. Associated with each
choice of biased sampling approach is a derived type variable that provides the required parameters, as shown in
Table 7.2. Please see Section 7.6 for more details about the variable dependencies.

116

7.4. Quantization of nuclear degrees of freedom

7.4. Quantization of nuclear degrees of freedom
The EVB framework provides a computationally practical approximation to the electronic surface for modeling
chemical reactions involving classical atoms. The full Schrödinger equation, nevertheless, describes not only the
electrons but also the nuclei as a wave function. This quantum mechanical description of nuclei is particularly
important for capturing the nuclear dispersion of light particles, such as hydrogen. We provide quantization of the
nuclear degrees of freedom via coupling of the EVB facility with the Feynman Path-Integral Molecular Dynamics
function in Amber [215–217]. The current implementation utilizes the PIMD engine that is built on top of the
locally enhanced sampling (LES) infrastructure. As such, the user will need to build the parallel version of LES
sander in order to utilize EVB/LES-PIMD.
PIMD is invoked using the ipimd variable (and associated dependencies) in the &cntrl namelist of the mdin
input file (please consult Section 25.1.2). The requirements for EVB within the EVB/LES-PIMD context are
similar to those described for classical EVB but with the coordinate and parmtop files modified for a LES-type
calculation, where the number of LES copies correspond to the number of path integral slices. For example, a
classical EVB umbrella sampling on a difference of distances RC will have EVB input files similar to the above
examples but with the following modifications
.
.
.
evb_dyn
= "dbonds_umb",
.
.
.
dbonds_umb(1)%iatom = 8, dbonds_umb(1)%jatom = 9, dbonds_umb(1)%katom = 7,
dbonds_umb(1)%k = 100.000, dbonds_umb(1)%ezero = -.20,
.
.
.

EVB/LES-PIMD utilizes these same EVB input files. The EVB group file evb.grpfile, however, has been modified
to point to the LES coordinate and parmtop files
# 32-bead Malonaldehyde RS:
-O -i mdin -p mr_les.top -c
-evbin input.mr
# 32-bead Malonaldehyde PS:
-O -i mdin -p mp_les.top -c
-evbin input.mp

H9 bonded to O8
mr_les.crd -o mr_les.out -r mr_les.rst \
H9 bonded to O7
mr_les.crd -o mp_les.out -r mp_les.rst \

Additionally, the -nslice <# PIMD slices> variable must be passed to the sander executable:
mpirun -np 2 sander.LES.MPI -ng 2 -nslice 32 -groupfile evb.grpfile

Here, the atoms of the malonaldehyde system have been replicated into 32 copies using the addles utility (see
Section 25.1.2) and each of the EVB diabatic states now use the corresponding LES coordinate and parmtop files.
Nuclear quantization lowers the free energy barrier due to quantum mechanical effects, such as zero point motion
and tunneling. Figure 7.4 compares the PMFs for the malonaldehyde proton transfer reaction along a difference
of distances RC from classical EVB and EVB/PIMD umbrella sampling simulations. Currently, only the distance
and difference of distances RCs are supported in EVB/PIMD. The energy gap RC is not supported because the
theoretical formulation of quantum transition state theory based on an energy gap RC has not yet been worked out.

7.5. Distributed Gaussian EVB
As briefly mentioned in the Introduction to EVB, V12 can be fit to high-level electronic structure data using a
set of s-, p-, and d-type Gaussians as the fitting basis functions. The current incarnation of DG EVB is limited to
two-state gas-phase systems. Current efforts to extend this approach to the condensed phase will provide a
practical systematic procedure for constructing a reactive surface from ab initio information. The curious student
is encouraged to read the original papers on this method for the theoretical formulation[210, 211]. Here, we only

117

7. Empirical Valence Bond

Potential of Mean Force [kcal/mol]

0
-0.5

0.5

Difference of Bond Lengths [Å]

Figure 7.4.: PMFs as a function of the difference of bond lengths involving the proton with the donor and acceptor
oxygens in malonaldehyde. TheHcurve is from classical EVB, while the curve is from EVB/PIMD.
provide an example of this approach for constructing an ab initio-inspired surface describing the proton transfer
reaction in malonaldehyde. All the previously described EVB functionalities are accessible to this method. For
example, the key elements of the RS input.mr file for biased sampling along a distance RC on the DG EVB
surface may look something like the following:
.
.
.
nUFF = 1, nbias = 1,
dia_type = "ab_initio",
xch_type = "dist_gauss",
evb_dyn = "bond_umb",
bond_umb(1)%iatom = 7, bond_umb(1)%jatom = 9,
bond_umb(1)%k = 400.000, bond_umb(1)%ezero = 1.20,
dist_gauss%stype = "no_dihedrals",
dist_gauss%lin_solve = "diis",
dist_gauss%xfile_type = "gaussian_fchk",
ts_xfile(1) = "malonaldehydeTS_35.fchk",
min_xfile(1) = "malonaldehydeR_35.fchk",
min_xfile(2) = "malonaldehydeP_35.fchk",
dgpt_alpha(1) = 0.72,
dgpt_alpha(2) = 0.72,
dgpt_alpha(3) = 0.72,
UFF(1)%iatom = 7, UFF(1)%jatom = 9
.
.
.

These variables are described in Section 7.6. DG EVB is invoked through the xch_type variable, with dependencies on dist_gauss, ts_xfile(:), min_xfile(:), dgpt_alpha(:), and UFF(:). The ab initio data for the RS minimum
are contained in the file malonaldehydeR_35.fchk, those for the PS minimum are contained in malonaldehydeP_35.fchk, and those for the transition state are contained in malonaldehydeTS_35.fchk. These files are in the
Gaussian [221] formatted checkpoint file format (gaussian_fchk). The α parameter [see Eqs. (7.4-7.6)] associated with each of these configuration space points is specified in the variable dgpt_alpha(:). If we wish to include
additional ab initio data points along the reaction path, we can specify the file names for those points in the variable
xdg_xfile(:). The α parameters associated with these points can be specified in dgpt_alpha(:). It is important to
keep in mind that the α parameters are ordered as follows: dgpt_alpha(ts_xfile(1), min_xfile(1), min_xfile(2),

118

7.6. EVB input variables and interdependencies

Figure 7.5.: PMF as a function of the distance between atoms H9 and O7 in malonaldehyde. The potential energy
surface was constructed from ab initio data using the DG EVB approach.
xdg_xfile(:)). Lastly, the UFF variable requests the inclusion of a Universal Force Field [222] repulsive term in V11
between the transferred proton (H9) and the acceptor (O7). The input.mp file for the PS V22 is identical to the above,
but with the UFF variable changed to reflect the identity of the acceptor atom from the perspective of the product
state topology: UFF(1)%iatom = 8, UFF(1)%jatom = 9. In practice, the inclusion of this term to Vii provides a
more optimal DG EVB surface for molecular dynamics sampling. Figure 7.5 shows the PMF for shortening the
rH9−O7 distance of the malonaldehyde RS from 1.8 Å to 1.0 Å using umbrella sampling of this RC. Note that the
PMF is not symmetric because this choice of RC breaks the intrinsic symmetry of the reaction. The difference
of distances RC involving atoms O8, H9 and O7 does provide a symmetric PMF and this is shown in Figure 25.2
within the context of kinetic isotope effect (Section 25.6.5).

7.6. EVB input variables and interdependencies
The variables in the &evb namelist of the EVB input file are described below. The style of the input file is
similar to the traditional mdin used in a sander run. Assignment to character type variables need to be encapsulated
within quotation marks (for example, evb_dyn=“groundstate”). Array variables are denoted below by a colon
enclosed within parentheses [for example, dia_shift(:)]. Derived type variables can be assigned element-wise, i.e.,
dia_shift(1)%st = 1, dia_shift(1)%nrg_offset = 0.0. In the specifications below, the data type of each variable is
enclosed in {· · · }, while the size of each array variable is enclosed in [· · · ].
ntw_evb

{integer}. MD step interval for writing to the EVB output file evbout.

nevb

{integer}. Number of EVB states. For example, nevb = 3 specifies that the system is described by a
3 × 3 EVB matrix in the representation of three diabatic states. The EVB group file will contain three
lines of sander command line options specifying the mdin, coordinate, parmtop, and EVB input files.

nmorse

{integer}. Number of Amber harmonic bond interactions that will be changed to a Morse type interaction. Requires additional inputs from the variable morsify(:).

119

7. Empirical Valence Bond
nbias

{integer}. Number of biasing potentials to include in the system Hamiltonian. The supported biased
sampling approaches include (1) mapping potential, (2) umbrella sampling along an energy gap RC,
(3) umbrella sampling along a distance RC, and (4) umbrella sampling along a difference of distances
RC. See evb_dyn for associated dependencies.

nmodvdw {integer}. Number of van der Waals terms to exclude in the calculation of Vii . Requires additional
inputs from the variable modvdw(:).

nuff

{integer}. Number of Universal Force Field [222] repulsive terms to include in the harmonic expansion of Vii about the ab initio minimum. Requires additional inputs from the variable uff(:).

xch_type

{character*512}. Coupling element type.
= “constant” Vi j is a constant. Requires additional inputs from the variable xch_cnst(:).
h

i
(0,i j)
= “exp”
Vi j (rkl ) = Ai j exp −ui j rkl − rkl
. Requires additional inputs from the variable xch_exp(:).

2
(0,i j)
. Requires additional inputs from the variable
= “gauss” Vi j (rkl ) = Ai j exp − σ12 rkl − rkl
ij

xch_gauss(:).
= “dist_gauss” Vi j is described by the Schlegel-Sonnenberg distributed Gaussian approach. Requires
additional inputs from the variables dist_gauss, ts_xfile(:), min_xfile(:), xdg_xfile(:),
dgpt_alpha(:), uff(:).

evb_dyn

{character*512}. EVB dynamics type.
= “groundstate” Dynamics on the EVB ground-state potential energy surface.
= “evb_map” Biased sampling based on Ariel Warshel’s mapping potential approach. Requires additional inputs from the variable emap(:).
= “egap_umb” Umbrella sampling along an energy gap reaction coordinate. Requires additional
inputs from the variable egap_umb(:).
= “bond_umb” Umbrella sampling along a distance reaction coordinate. Requires additional inputs
from the variable bond_umb(:).
= “dbonds_umb” Umbrella sampling along a difference of two distances reaction coordinate. Requires additional inputs from the variable dbonds_umb(:).
= “qi_bond_pmf” For generating the QI joint distribution function along the distance RCs of the
P and P/2 slices (see Section 25.5.2). Requires additional inputs from the variable
bond_umb(:).
= “qi_bond_dyn” For sampling of the QI fv , F and G factors with the P and P/2 slices constrained
to the dividing surfaces along the distance RCs (see Section 25.5.2). Requires additional
inputs from the variable bond_umb(:).
= “qi_dbonds_pmf” For generating the QI joint distribution function along the difference of distances
RCs of the P and P/2 slices (see Section 25.5.2). Requires additional inputs from the
variable dbonds_umb(:).
= “qi_dbonds_dyn” For sampling of the QI fv , F and G factors with the P and P/2 slices constrained
to the dividing surfaces along the difference of distances RCs (see Section 25.5.2). Requires additional inputs from the variable dbonds_umb(:).

120

7.6. EVB input variables and interdependencies

dia_shift(:) {derived type}, [nevb]. Diabatic state energy shift.
%st

{integer}. Diabatic state index.

%nrg_offset {real}. Energy offset for EVB state.

xch_cnst(:) {derived type}, [nxch]. Constant coupling. The size of this derived type array is nxch, which is
calculated internally as nevb(nevb − 1)/2.
%ist

{integer}. Diabatic state index involved in the coupling.

%jst

{integer}. Diabatic state index involved in the coupling.

%xcnst

{real}. Constant exchange parameter.

xch_exp(:) {derived
for the exponential functional form of the coupling term, Vi j (rkl ) =
h type},
[nxch]. Parameters
i
(0,i j)
Ai j exp −ui j rkl − rkl
. The size of this derived type array is nxch, which is calculated internally
as nevb(nevb − 1)/2.
%ist

{integer}. Diabatic state index involved in the coupling.

%jst

{integer}. Diabatic state index involved in the coupling.

%iatom

{integer}. Index of atom involved in rkl .

%jatom

{integer}. Index of atom involved in rkl .

{real}. Ai j .

{real}. ui j .

%r0

{real}. rkl

(0,i j)

xch_gauss(:) {derived
[nxch]. Parameters
for the Gaussian functional form of the coupling term, Vi j (rkl ) =
type},

(0,i j) 2
1
. The size of this derived type array is nxch, which is calculated interAi j exp − σ 2 rkl − rkl
ij

nally as nevb(nevb − 1)/2.
%ist

{integer}. Diabatic state index involved in the coupling.

%jst

{integer}. Diabatic state index involved in the coupling.

%iatom

{integer}. Index of atom involved in rkl .

%jatom

{integer}. Index of atom involved in rkl .

{real}. Ai j .

%sigma

{real}. σi j .

%r0

{real}. rkl

(0,i j)

morsify(:) {derived type}, [nmorse]. Parameters used for converting the Amber harmonic bond interactions to

2
−α ri j −ri0j
the Morse type, VMorse (ri j ) = De 1 − e
. The components in the derived type are
%iatom

{integer}. Index of atom involved in ri j .

121

7. Empirical Valence Bond

emap(:)

%jatom

{integer}. Index of atom involved in ri j .

{real}. De .

{real}. α.

%r0

{real}. ri0j .

{derived type}, [nbias]. Mapping potential parameters required for the function Vλ = (1 − λ )Vii +
λV f f .
%ist

{integer}. Diabatic state index for the initial state.

%jst

{integer}. Diabatic state index for the final state.

%lambda

{real}. λ .

egap_umb(:) {derived type}, [nbias]. Umbrella potential parameters required for the function Vumb (RC) =
2
1
2 k [RC − RC0 ] , where RC = Vii −V f f .
%ist

{integer}. Diabatic state index for the initial state.

%jst

{integer}. Diabatic state index for the final state.

{real}. k.

%ezero

{real}. RC0 .

modvdw(:) {derived type}, [nmodvdw]. Exclude the van der Waals interactions between the specified atom pairs.
%iatom

{integer}. Index of atom involved in the non-bonded interaction.

%jatom

{integer}. Index of atom involved in the non-bonded interaction.

bond_umb(:) {derived type}, [nbias]. Umbrella potential parameters for the function Vumb (RC) = 12 k [RC − RC0 ]2 ,
where RC = ri j .
%iatom

{integer}. Index of atom involved in a distance.

%jatom

{integer}. Index of atom involved in a distance.

{real}. k.

%ezero

{real}. RC0 .

dbonds_umb(:) {derived type}, [nbias]. Umbrella potential parameters for the difference of two distances RC
where one of the atoms is common to both distances. Vumb (RC) = 12 k [RC − RC0 ]2 , where RC =
ri j − rk j .

122

%iatom

{integer}. Index of atom involved in a distance.

%jatom

{integer}. Index of the atom common to both distances.

%katom

{integer}. Index of atom involved in a distance.

{real}. k.

%ezero

{real}. RC0 .

7.6. EVB input variables and interdependencies
out_RCdot {logical}. Output the velocity of a free particle along the RC direction to the file evbout.

dist_gauss {derived type}. Schlegel-Sonnenberg distributed Gaussian specifications.
%stype

{character*512}. Coordinate selection type. Supported coordinate selection types include
“all_coords”, “bonds_only”,“no_dihedrals”,“react-product”,“react-ts-product”.

%stol

{real}. Coordinate selection tolerance for stype=“react-product” or stype=“react-tsproduct”. For stype=“react-product”, a particular internal coordinate is used in the DG
EVB procedure if the difference between the reactant and product structures is > stol. For
the case of stype=“react-ts-product”, the intersection of the selected set of coordinates
from react-ts > stol and product-ts > stol will be used for the DG EVB procedure.

%xfile_type {character*512}. File type of external ab initio data. Supported file types are “gaussian_fchk” and “EVB”.
ts_xfile(:)

{character*512}, [*]. Name of the file containing the ab initio data corresponding to the transition
state.

min_xfile(:) {character*512}, [*]. Name of the file containing the ab initio data corresponding to the minimum,
i.e. V11 and V22 .

xdg_xfile(:) {character*512}, [*]. Name of the file containing the ab initio data corresponding to additional points
along the IRC.

dgpt_alpha(:) {real}, [*]. Optimized α parameters associated with the distributed Gaussian data points.

uff(:)

{derived type}, [nuff]. Include a UFF repulsive term between the specified atom pairs in the harmonic
expansion of Vii about the ab initio minimum.
%iatom

{integer}. Index of atom involved in the non-bonded interaction.

%jatom

{integer}. Index of atom involved in the non-bonded interaction.

123

8. sqm: Semi-empirical quantum chemistry
AmberTools now contains its own quantum chemistry program, called sqm. This is code extracted from the
QM/MM portions of sander, but is limited to “pure QM” calculations. A principal current use is as a replacement
for MOPAC for deriving AM1-bcc charges, but the code is much more general than that. Right now, it is limited to
carrying out single point calculations and energy minimizations (geometry optimizations) for closed-shell systems.
It supports a wide variety of semi-empirical Hamiltonians, including many recent ones. An external electric field
generated by a set of point charges can be included for single point calculations. Our plan is to add capabilities to
subsequent versions. The major contributors are as follows:

• The original semi-empirical support was written by Ross Walker, Mike Crowley, and Dave Case,[223] based
on public-domain MOPAC codes of J.J.P. Stewart.
• SCC-DFTB support was written by Gustavo Seabra, Ross Walker and Adrian Roitberg,[224] and is based
on earlier work of Marcus Elstner.[225, 226]
• Support for third-order SCC-DFTB was written by Gustavo Seabra and Josh Mcclellan.
• Various SCF convergence schemes were added by Tim Giese and Darrin York.
• The PM6 Hamiltonian was added by Andreas Goetz and dispersion and hydrogen bond corrections were
added by Andreas Goetz and Kyoyeon Park.
• The extension for MNDO type Hamiltonians to support d orbitals was written by Tai-Sung Lee, Darrin York
and Andreas Goetz.
• The charge-dependent exchange-dispersion corrections of vdW interactions[227] was contributed by TaiSung Lee, Tim Giese, and Darrin York.
• The ability of reading user-defined parameters was added by Tai-Sung Lee and Darrin York.

8.1. Available Hamiltonians
Available MNDO-type semi-empirical Hamiltonians are PM3,[228] AM1,[229] RM1,[230] MNDO,[231]
PDDG/PM3,[232] PDDG/MNDO,[232] PM3CARB1,[233], PM3-MAIS[234, 235], MNDO/d[236–238], AM1/d
(Mg from AM1/d[239] and H, O, and P from AM1/d-PhoT[240]) and PM6[241].
Support is also available for the Density Functional Theory-based tight-binding (DFTB) Hamiltonian,[224, 242,
243] as well as the Self-Consistent-Charge version, SCC-DFTB.[225] DFTB/SCC-DFTB also supports approximate inclusion of dispersion effects,[244] as well as reporting CM3 charges [245] for molecules containing only
the H, C, N, O, S and P atoms and third-order corrections[246].
The elements supported by each QM method are:
• MNDO: H, Li, Be, B, C, N, O, F, Al, Si, P, S, Cl, Zn, Ge, Br, Cd, Sn, I, Hg, Pb
• MNDO/d: H, Li, Be, B, C, N, O, F, Na, Mg, Al, Si, P, S, Cl, Zn, Ge, Br, Sn, I, Hg, Pb
• AM1: H, C, N, O, F, Al, Si, P, S, Cl, Zn, Ge, Br, I, Hg
• AM1/d: H, C, N, O, F, Mg, Al, Si, P, S, Cl, Zn, Ge, Br, I, Hg
• PM3: H, Be, C, N, O, F, Mg, Al, Si, P, S, Cl, Zn, Ga, Ge, As, Se, Br, Cd, In, Sn, Sb, Te, I, Hg, Tl, Pb, Bi

125

8. sqm: Semi-empirical quantum chemistry
• PDDG/PM3: H, C, N, O, F, Si, P, S, Cl, Br, I
• PDDG/MNDO: H, C, N, O, F, Cl, Br, I
• RM1: H, C, N, O, P, S, F, Cl, Br, I
• PM3CARB1: H, C, O
• PM3-MAIS: H, O, Cl
• PM6: H, He, Li, Be, B, C, N, O, F, Ne, Na, Mg, Al, Si, P, S, Cl, Ar, K, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu,
Zn, Ga, Ge, As, Se, Br, Kr, Rb, Sr, Y, Zr, Nb, Mo, Tc, Ru, Rh, Pd, Ag, Cd, In, Sn, Sb, Te, I, Xe, Cs, Ba, La,
Lu, Hf, Ta, W, Re, Os, Ir, Pt, Au, Hg, Tl, Pb, Bi
• DFTB/SCC-DFTB: (Any atom set available from the www.dftb.org website)
The PM6 implementation has not been extensively tested for all available elements. Please check your results
carefully, possibly by comparison to other codes that implement PM6, in particular if transition metal elements
are present. SCF convergence may be more difficult to achieve for transition metal elements with partially filled
valence shells.
If the PM6 Hamiltonian is used in a QM/MM simulation with sander using electrostatic embedding (see Section
9) or if an electric field of external point charges is used, then the electrostatic interactions between QM and MM
atoms are modeled using the MNDO type core repulsion function for interactions between QM and MM atoms.
Parameters for the exponents α of the QM atoms are taken from PM3 (a default value of five is used for the
exponents α of the MM atoms as is the case for MNDO, AM1 and PM3). Since PM3 does not have parameters for
all elements that are supported by PM6, the missing exponents were defined in an ad hoc manner (see the source
code in $AMBERHOME/AmberToosl/src/sqm/qm2_parameters.F90, variable alp_pm6). The magnitude of the
coefficients α is probably not critical for the accuracy of QM/MM calculations but this should be tested on a case
by case basis. This does not affect QM calculations with sqm.
The DFTB/SCC-DFTB code was originally based on the DFT/DYLAX code by Marcus Elstner et al., but
has since been extensively re-written and optimized. In order to use DFTB (qm_theory=DFTB) a set of integral
parameter files are required. These are not distributed with Amber and must be obtained from the www.dftb.org
website and placed in the $AMBERHOME/dat/slko directory. Dispersion parameters for H, C, N, O, P and S
are available in the $AMBERHOME/dat/slko/DISPERSION.INP_ONCHSP file, and CM3 parameters for the same
atoms are in the $AMBERHOME/dat/slko/CM3_PARAMETERS.DAT file. Parameters for two parametrizations of
the third-order SCC-DFTB terms, namely SCC-DFTB-PA and SCC-DFTB-PR are distributed with Amber in the
files DFTB_3RD_ORDER_PA.DAT and DFTB_3RD_ORDER_PR.DAT, located in the same directory.

8.2. Dispersion and hydrogen bond correction
An empirical dispersion and hydrogen bonding correction is implemented for the MNDO type Hamiltonians
AM1 and PM6[247]. The empirical dispersion correction follows the formalism for DFT-D[248] and consists of a
physically sound r−6 term that is damped at short distances to avoid the short-range repulsion which can be written
as
Edis = −s6 ∑ fdamp (ri j , R0i j )C6,i j ri−6
(8.1)
j ,
ij

where ri j is the distance between two atoms i and j, R0i j is the equilibrium van der Waals (vdW) separation
derived from the atomic vdW radii, C6,i j the dispersion coefficient, and s6 a general scaling factor. The damping
function is given as
"
!#−1
ri j
0
fdamp (ri j , Ri j ) = 1 + exp −α
−1
.
(8.2)
sR R0i j

126

8.3. Usage
Bondi vdW radii[249] are used and for a pair of unlike atoms we have
R0i j =

R0ii + R0j j
R0ii + R0j j

(8.3)

For the C6 coefficients the following equation is used,
C6,i j = 2

2 C2 N
1/3
(C6,ii
6, j j eff,i Neff, j )
2 )1/3 + (C
2
1/3
(C6,ii Neff,
6, j j Neff,i )
j

(8.4)

where the Slater-Kirkwood effective number of electrons Neff,i and the C6 coefficients can easily be found in the
literature[248].
An empirical hydrogen bonding correction[247] that is transferable among different semiempirical Hamiltonians
and has been parametrized for use with the dispersion correction described above is also available. This correction
does not make the assumption of a specific acceptor/hydrogen/donor binding situation. Instead it considers the
hydrogen bond as a charge-independent atom- atom term between two atoms capable of serving as an acceptor
or donor (for example, O, N) and weights this by a function that accounts for the steric arrangement of the two
atoms and the favorable positioning of a hydrogen atom inbetween. A damping function corrects for long- and
short-range behavior,
CAB
(8.5)
EH−bond = 2 fgeom fdamp ,
rAB
fgeom = cos(θA )2 cos(φA )2 cos(ψA )2 cos(φB )2 cos(φB )2 cos(ψB )2 fbond ,
1
,
1 + exp[−60(rXH /1.2 − 1)]

1
1
1−
,
fdamp =
1 + exp[−100(rAB /2.4 − 1)]
1 + exp[−10(rAB /7.0 − 1)]
fbond = 1 −

(8.6)
(8.7)
(8.8)

CA +CB
.
(8.9)
2
Here, CA and CB are the atomic hydrogen bonding correction parameters and the (torsion) angles in the function
fgeom are defined similarly to an earlier hydrogen bond correction[250].
The hydrogen bond correction can be used both for single point energy calculations or geometry optimizations
with SQM and for molecular dynamics simulations with SANDER. However, we do not recommend the use for
molecular dynamics at present since cutoffs needed to be implemented for the calculation of fgeom of equation
(8.6). This and some other conditional evaluations give rise to discontinuities in the potential energy surface and
thus make this method unattractive for MD simulations.
CAB =

8.3. Usage
The sqm program uses the following simple command line:
sqm [-O] -i -o

As in other Amber programs, the “-O” flag allows the program to over-write the output file.
An example input file for running a simple minimization is shown here:
Run semi-empirical minimization
&qmmm
qm_theory=’AM1’,
qmcharge=0,
/
6
CG
-1.9590
0.1020
6
CD1
-1.2490
0.6020

0.7950
-0.3030

127

8. sqm: Semi-empirical quantum chemistry
6
6
6
6
1
16
1
1
1
1
1

CD2
CE1
C6
CZ
HE2
S15
H19
H29
H30
H31
H32

-2.0710
-0.6460
-1.4720
-0.7590
-1.5580
-2.7820
-3.5410
-0.7870
0.3730
-0.0920
-2.3790

0.8650
1.8630
2.1290
2.6270
2.7190
0.3650
0.9790
-0.0430
2.0450
3.5780
-0.9160

1.9630
-0.2340
2.0310
0.9340
2.9310
3.0600
3.2740
-0.9380
-0.7840
0.7810
0.9010

The &qmmm namelist contains variables that allow you to control the options used. Following that is one line per
atom, giving the atomic number, atom name, and Cartesian coordinates (free format). The variables in the &qmmm
namelist are these:
qm_theory Level of theory to use for the QM region of the simulation (Hamiltonian). Default is to use the semiempirical Hamiltonian PM3. Options are AM1, RM1, MNDO, PM3-PDDG, MNDO-PDDG, PM3CARB1, MNDO/d (same as MNDOD), AM1/d (same as AM1D), PM6, and DFTB. The dispersion
correction can be switched on for AM1 and PM6 by choosing AM1-D* and PM6-D, respectively. The
dispersion and hydrogen bond correction will be applied for AM1-DH+ and PM6-DH+.
dftb_disper Flag turning on (1) or off (0) the use of a dispersion correction to the DFTB/SCC-DFTB energy.
Requires qm_theory=DFTB. It is assumed that you have the file DISPERSION.INP_ONCHSP in your
$AMBERHOME/dat/slko/ directory. This file must be downloaded from the website www.dftb.org,
as described in the beginning of this chapter. Not available for the Zn atom. (Default = 0)
dftb_3rd_order Third order correctio to SCC-DFTB. Default=” (no third order correction).
= ’PA’ Use the SCC-DFTB-PA parametrization, which was developed for proton affinities. The pa-

rameters will be read from the $AMBERHOME/dat/slko/DFTB_3RD_ORDER_PA.DAT file.
= ’PR’ Use the SCC-DFTB-PR parametrization,

which was developed for phosphate hydrolysis reactions.
The parameters will be read from the $AMBERHOME/dat/slko/DFTB_3RD_ORDER_PR.DAT file.

= ’READ’ Parameters will be read from the mdin file, in a separate “dftb_3rd_order” namelist, which

must have the same format as the files above.
= ’filename’ Parameters will be read from the file specified by filename, in the “dftb_3rd_order”

namelist, which must have the same format as the files above.
dftb_chg

Flag to choose the type of charges to report when doing a DFTB calculation.
= 0 (default) - Print Mulliken charges
= 2 Print CM3 charges. Only available for H, C, N, O, S and P.

dftb_telec

Electronic temperature, in K, used to accelerate SCC convergence in DFTB calculations. The electronic temperature affects the Fermi distribution promoting some HOMO/LUMO mixing, which can
accelerate the convergence in difficult cases. In most cases, a low telec (around 100K) is enough.
Should be used only when necessary, and the results checked carefully. Default: 0.0K

dftb_maxiter Maximum number of SCC iterations before resetting Broyden in DFTB calculations. (default: 70 )
qmcharge

Charge on the QM system in electron units (must be an integer). (Default = 0)

spin

Multiplicity of the QM system. Currently only singlet calculations are possible and so the default
value of 1 is the only available option. Note that this option is ignored by DFTB/SCC-DFTB, which
allows only ground state calculations. In this case, the spin state will be calculated from the number
of electrons and orbital occupancy.

128

8.3. Usage
qmqmdx

Flag for whether to calculate QM-QM derivatives analytically or pseudo numerically. The default (and
recommended) option is to use ANALYTICAL QM-QM derivatives.
= 1 (default) - Use analytical derivatives for QM-QM forces.
= 2 Use numerical derivatives for QM-QM forces. Note: the numerical derivative code has not been

optimised as aggressively as the analytical code and as such is significantly slower. Numerical
derivatives are intended mainly for testing purposes.
verbosity

Controls the verbosity of QM/MM related output. Warning: Values of 2 or higher will produce a lot
of output.
= 0 (default) - only minimal information is printed - Initial QM geometry and link atom positions as

well as the SCF energy at every ntpr steps.
= 1 Print SCF energy at every step to many more significant figures than usual. Also print the number

of SCF cycles needed on each step.
= 2 As 1 but also print info about memory reallocations, number of pairs per QM atom. Also prints

QM core - QM core energy, QM core - MM charge energy and total energy.
= 3 As 2 but also print SCF convergence information at every step.
= 4 As 3 but also print forces onof the file QM atoms due to the SCF calculation and the coordinates

of the link atoms at every step.
= 5 As 4 but also print all of the info in kJ/mol as well as kcal/mol.

tight_p_conv Controls the tightness of the convergence criteria on the density matrix in the SCF.
=0 (default) - loose convergence on the density matrix (or Mulliken charges, in case of a SCC-DFTB

calculation). SCF will converge if the energy is converged to within scfconv and the largest
change in the density matrix is within 0.05*sqrt(scfconv).
= 1 Tight convergence on density(or Mulliken charges, in case of a SCC-DFTB calculation). Use

same convergence (scfconv) for both energy and density (charges) in SCF. Note: in the SCCDFTB case, this option can lead to instabilities.
scfconv

Controls the convergence criteria for the SCF calculation, in kcal/mol. In order to conserve energy
in a dynamics simulation with no thermostat it is often necessary to use a convergence criterion of
1.0d-9 or tighter. Note, the tighter the convergence the longer the calculation will take. Values tighter
than 1.0d-11 are not recommended as these can lead to oscillations in the SCF, due to limitations in
machine precision, that can lead to convergence failures. Default is 1.0d-8 kcal/mol. Minimum usable
value is 1.0d-14.

pseudo_diag Controls the use of ’fast’ pseudo diagonalisations in the SCF routine. By default the code will attempt
to do pseudo diagonalisations whenever possible. However, if you experience convergence problems
then turning this option off may help. Not available for DFTB/SCC-DFTB.
= 0 Always do full diagonalisation.
= 1 Do pseudo diagonalisations when possible (default).

pseudo_diag_criteria Float controlling criteria used to determine if a pseudo diagonalisation can be done. If the
difference in the largest density matrix element between two SCF iterations is less than this criteria
then a pseudo diagonalisation can be done. This is really a tuning parameter designed for expert use
only. Most users should have no cause to adjust this parameter. (Not applicable to DFTB/SCC-DFTB
calculations.) Default = 0.05
diag_routine Controls which diagonalization routine should be used during the SCF procedure. This is an advanced option which has no effect on the results but can be used to fine-tune performance. The speed
of each diagonalizer is both a function of the number and type of QM atoms as well as the LAPACK

129

8. sqm: Semi-empirical quantum chemistry
library that Sander was linked to. As such there is not always an obvious choice to obtain the best
performance. The simplest option is to set diag_routine = 0 in which case Sander will test each diagonalizer in turn, including the pseudo diagonalizer, and select the one that gives optimum performance.
This should ideally be the default behavior but this option has not been tested on sufficient architectures to be certain that it will always work. Not available for DFTB/SCC-DFTB.
= 0 Automatically select the fastest routine (recommended).
= 1 Use internal diagonalization routine (default).
= 2 Use lapack dspev.
= 3 Use lapack dspevd.
= 4 Use lapack dspevx.
= 5 Use lapack dsyev.
= 6 Use lapack dsyevd.
= 7 Use lapack dsyevr.

printcharges
= 0 Don’t print any info about QM atom charges to the output file (default)
= 1 Print Mulliken QM atom charges to output file every ntpr steps.

print_eigenvalues Controls printing of MO eigenvalues.
= 0 Do not print MO eigenvalues
= 1 Print MO eigenvalues at the end of a single point calculation or geometry optimization (default)
= 2 Print MO eigenvalues at the end of every SCF cycle (only NDDO methods, not DFTB)
= 3 Print MO eigenvalues during each step of the SCF cycle (only NDDO methods, not DFTB)

qxd

Flag to turn on (=.true.) or off (=.false., default) the charge-dependent exchange-dispersion corrections
of vdW interactions[227].

parameter_file
= ’PARAM.FILE’ Read user-defined parameters from the file ’PARAM.FILE’. The first three space-

separated entries (case insensitive) of each line will be interpreted as a user-modified parameter
in the sequence of parameter name, element name, and value. For example, a line contains “USS
Cl -111.6139480D0 “ will cause the USS parameter of the Cl element changed to -111.6139480.
A line beginning with “END” will stop the reading. This function currently only works for
MNDO, AM1, PM3, MNDO/d, and AM1/d. Also, when new nuclear core-core parameters (FN,
in PM3, AM1, and AM1/d) are re-defined, the number of FNN parameter sets (NUM_FN) also
needs to be defined. For example, if FNn3 (n = 1, 2, or 3) is defined, then NUM_FN needs to be
set to 3 or 4.
peptide_corr
= 0 Don’t apply MM correction to peptide linkages. (default)
= 1 Apply a MM correction to peptide linkages. This correction is of the form Esc f = Esc f +

htype (itype ) sin2 φ , where φ is the dihedral angle of the H-N-C-O linkage and htype is a constant
dependent on the Hamiltonian used. (Recommended, except for DFTB/SCC-DFTB.)
itrmax

130

Integer specifying the maximum number of SCF iterations to perform before assuming that convergence has failed. Default is 1000. Typically higher values will not do much good since if the SCF
hasn’t converged after 1000 steps it is unlikely to. If the convergence criteria have not been met after
itrmax steps the SCF will stop and the minimisation will proceed with the gradient at itrmax. Hence
if you have a system which does not converge well you can set itrmax smaller so less time is wasted

8.3. Usage
before assuming the system won’t converge. In this way you may be able to get out of a bad geometry
quite quickly. Once in a better geometry SCF convergence should improve.
maxcyc

Maximum number of minimization cycles to allow, using the xmin minimizer (see Section 37.5) with
the TNCG method. Default is 9999. Single point calculations can be done with maxcyc = 0.

ntpr

Print the progress of the minimization every ntpr steps; default is 10.

grms_tol

Terminate minimization when the gradient falls below this value; default is 0.02

ndiis_attempts Controls the number of iterations that DIIS (direct inversion of the iterative subspace) extrapolations will be attempted. Not available for DFTB/SCC-DFTB. The SCF does not even begin to exhaust
its attempts at using DIIS extrapolations until the end of iteration 100. Therefore, for example, if
ndiis_attempts=50, then DIIS extrapolations would be performed at end of iterations 100 to 150. The
purpose of not performing DIIS extrapolations before iteration 100 is because the existing code base
performs quite well for most molecules; however, if convergence is not met after 100 iterations, then it
is presumed that further iterations will not yield SCF convergence without doing something different,
i.e., DIIS. Thus, the implementation of DIIS in SQM is a mechanism to try and force SCF convergence for molecules that are otherwise difficult to converge. Default 0. Maximum 1000. Minimum
0. Note that DIIS will automatically turn itself on for 100 attempts at the end of iteration 800 even if
you did not explicitly set ndiis_attempts to a nonzero value. This is done as a final effort to achieve
convergence.
ndiis_matrices Controls the number of matrices used in the DIIS extrapolation. Including only one matrix is the
same as not performing an extrapolation. Including an excessive number of matrices may require a
large amount of memory. Not available for DFTB/SCC-DFTB. Default 6. Minimum 1. Maximum 20.
vshift

Controls level shifting (only NDDO methods, not DFTB). Virtual orbitals can be shifted up by vshift
(in eV) to improve SCF convergence in cases with small HOMO/LUMO gap. Default 0.0 (no level
shift).

errconv

SCF tolerance on the maximum absolute value of the error matrix, i.e., the commutator of the Fock
matrix with the density matrix. The value has units of hartree. The default value of errconv is sufficiently large to effectively remove this tolerance from the SCF convergence criteria. Not available for
DFTB/SCC-DFTB. Default 1.d-1. Minimum 1.d-16. Maximum 1.d0.

qmmm_int When running QM calculations in the sqm program, an electric field of external point charges can
be added. In this way, the electrostatic effect outside of the QM region can be modeled, making the
calculation a simplified QM/MM calculation without QM/MM vdW’s contribution. Like QM/MM
calculations (see Section 9), the method to couple QM and MM electrostatic interactions for external
charges and semiempirical Hamiltonians can be specified via the qmmm_int namelist variable.
The current implementation limits use of external charges to only single point energy calculations. To
run such a calculation, an additional field, which begins with #EXCHARGES and ends with #END,
is required to specify the external point charges in the input. Each external point charge must include
atomic number, atom name, X, Y, Z coordinates and the charge in units of the electron charge. An
example input looks like:
single point energy calculation (adenine), with external charges (thymine)
&qmmm
qm_theory = ’PM3’,
qmcharge = 0,
maxcyc = 0,
qmmm_int = 1,
/
7 N
1.0716177 -0.0765366
1.9391390
1 H
0.0586915 -0.0423765
2.0039181

131

8. sqm: Semi-empirical quantum chemistry
1 H
1.6443796
6 C
1.6739638
7 N
0.9350155
6 C
1.5490760
1 H
0.8794435
7 N
2.8531510
6 C
3.5646109
6 C
3.0747955
7 N
4.0885824
6 C
5.1829921
1 H
6.1882591
7 N
4.9294871
1 H
5.6035368
#EXCHARGESwill be
6 C -4.7106131
1 H -4.4267056
1 H -4.4439282
1 H -5.7883971
6 C -3.9917387
6 C -4.6136833
1 H -5.6909220
7 N -3.9211729
1 H -4.4017172
6 C -2.5395897
8 O -1.9416783
7 N -1.9256484
1 H -0.8838255
6 C -2.5361367
8 O -1.8674730
#END

132

-0.0347395
-0.0357766
-0.0279801
0.0012569
0.0050260
0.0258031
0.0195446
-0.0094480
-0.0054429
0.0253971
0.0375542
0.0412404
0.0648755

2.7619159
0.7424316
-0.3788916
-1.5808009
-2.4315709
-1.8409596
-0.7059872
0.5994562
1.5289786
0.7872176
1.1738824
-0.5567274
-1.3036811

0.0413373
0.9186178
-0.8302573
0.0505530
0.0219348
0.0169051
0.0269347
-0.0009646
-0.0036078
-0.0149474
-0.0291878
-0.0110593
-0.0216168
0.0074651
0.0112093

2.1738637
2.7530256
2.7695655
2.0247280
0.8663338
-0.3336520
-0.4227183
-1.5163659
-2.4004924
-1.5962357
-2.6573783
-0.3638948
-0.3784269
0.8766724
1.9120833

-0.03140
0.06002
0.05964
0.03694
-0.25383
0.03789
0.16330
-0.47122
0.35466
0.80253
-0.63850
-0.58423
0.35404
0.71625
-0.60609

9. QM/MM calculations
Sander supports the option of describing part of the system quantum mechanically in an approach known as a
hybrid (or coupled potential) QM/MM simulation. Semi-empirical neglect of diatomic overlap (NDDO)-type and
density functional tight binding (DFTB) Hamiltonians are supported natively by sander and the basic documentation (e.g. what Hamiltonians are implemented, description of the input parameters) can be found in Chapter 8.
Here we limit our description to those features that are unique to the QM/MM interface implemented in sander.
More advanced Hamiltonians based on ab initio wave function theory (WFT) and density functional theory (DFT)
are supported via an interface to external QM software packages the use of which is described in section 9.2.
The built-in semi-empirical QM/MM support was written by Ross Walker and Mike Crowley, [223] based
originally on public-domain MOPAC codes of J.J.P. Stewart. The QM/MM generalized Born implementation uses
the model described by Pellegrini and Field[251] while regular QM/MM Ewald support is based on the work
of Nam et al.[252] with QM/MM PME support based on the work of Walker et al.[223]. SCC-DFTB support
was written by Gustavo Seabra, Ross Walker and Adrian Roitberg,[224] and is based on earlier work of Marcus
Elstner.[225, 226] Support for third-order SCC-DFTB was written by Gustavo Seabra and Josh McClellan.

9.1. Built-in semiempirical NDDO methods and SCC-DFTB
When running a QM/MM simulation in sander the system is partitioned into two regions, a QM region consisting
of the atoms defined by either the qmmask or iqmatoms keyword, and a MM region consisting of all the atoms that
are not part of the QM region. For a typical protein simulation in explicit solvent the number of MM atoms will
be much greater than the number of QM atoms. Either region can contain zero atoms, giving either a pure QM
simulation or a standard classical simulation. For periodic simulations, the quantum region must be compact, so
that the extent (or diameter) of the QM region (in any direction) plus twice the QM/MM cutoff must be less than
the box size. Hence, you can define an "active site" to be the QM region, but in most cases could not ask that
all cysteine residues (for example) be quantum objects. The restrictions are looser for non-periodic (gas-phase
or generalized Born) simulations, but the codes are written and tested for the case of a single, compact quantum
region.
The partitioned system is characterized by an effective Hamiltonian which operates on the system’s wavefunction Ψ, which is dependent on the position of the MM and QM nuclei, to yield the system energy Ee f f :
He f f Ψ(xe , xQM , xMM ) = E e f f (xQM , xMM )Ψ(xe , xQM , xMM )

(9.1)

The effective Hamiltonian consists of three components - one for the QM region, one for the MM region and a
term that describes the interaction of the QM and MM regions, implying that likewise the energy of the system can
be divided into three components. If the total energy of the system is re-written as the expectation value of He f f
then the MM term can be removed from the integral since it is independent of the position of the electrons:
Ee f f = Ψ|HQM + HQM/MM |Ψ + EMM

(9.2)

In the QM/MM implementation in sander, EMM is calculated classically from the MM atom positions using
the Amber or CHARMM force field equation and parameters, whereas HQM is evaluated using the chosen QM
method.
The interaction term HQM/MM is more complicated. By default, sander uses an electrostatic embedding scheme
(also referred to as additive scheme) in which the interaction of the MM point charges with the electrons of the
QM system as well as the interaction between the MM point charges and the QM nuclei (atomic cores for semiempirical methods) is explicitly taken into account. In other words, the MM region polarizes the QM electron
density. For the case where there are no covalent bonds between the atoms of the QM and MM regions the

133

9. QM/MM calculations
interaction Hamiltonian is thus the sum of an electrostatic term and a Lennard-Jones (VDW) term and can be
written as
"
!#
A
B
HQM/MM = ∑ ∑ Qm helectron (xe , xMM ) − Qm Zq hcore (xQM , xMM ) + 12 − 6
(9.3)
rqm rqm
q m
where the subscripts e, m and q refer to the electrons, the MM nuclei and the QM nuclei respectively. Here Qm is
the charge on MM atom m, Zq is the core charge (nucleus minus core electrons) on QM atom q, rqm is the distance
between atoms q and m, and A and B are Lennard-Jones interaction parameters. For systems that have covalent
bonds between the QM and MM regions, the situation is more complicated, as discussed later.
A more approximate form of the interaction term HQM/MM is referred to as mechanical embedding (or subtractive
QM/MM scheme). In this case the interactions between the QM and the MM region are obtained within the same
classical approximation that is used for the MM region, that is
!#
"
Qm Qq
A
B
+ 12 − 6
(9.4)
HQM/MM = ∑ ∑
rqm
rqm rqm
q m
where Qq is the classical MM point charge assigned to an atom in the QM region. Mechanical embedding is useful
to impose steric constraints on the embedded QM system, however, the electron density is not polarized by the
MM environment. An additional complication of this approach is that the point charges that are assigned to the
atoms in the QM region have to represent the electrostatic potential of the QM region during the whole course of
a QM/MM simulation.
If one evaluates the expectation values in Eq. 9.2 over a single determinant built from molecular orbitals
φi = ∑ ci j χ j

(9.5)

where the ci j are molecular orbital coefficients and the χ j are atomic basis functions, the total energy depends upon
the ci j and on the positions xMM and xQM of the atoms. The energy is obtained by setting ∂ Ee f f /∂ ci j to zero which
leads to a self-consistent (SCF) procedure to determine the ci j , (with a modified Fock matrix that contains the
electric field arising from the MM charges in the case of electrostatic embedding). Once the energy is known, the
forces on the atoms can be obtained by taking the derivative of the energy expression with respect to the positions
of the QM and MM atoms.
The main subtlety that arises in the case of electrostatic embedding is that, for a periodic system, there are
formally an infinite number of QM/MM interactions; even for a non-periodic system, the (finite) number of such
interactions may be prohibitively large. These problems are addressed in a manner analogous to that used for pure
MM systems: a PME approach is used for periodic systems, and a (large) cutoff may be invoked for non-periodic
systems. Some details are discussed below.

9.1.1. The QM/MM interface and link atoms
The sections above dealt with situations where there are no covalent bonds between the QM and MM regions. In
many protein simulations, however, it is necessary to have the QM/MM boundary cut covalent bonds, and a number
of additional approximations have to be made. There are a variety of approaches to this problem, including hybrid
orbitals, capping potentials, and explicit link atoms. The last option is the method available in sander.
There are a number of ways to implement a link atom approach that deal with the way the link atom is positioned,
the way the forces on the link atom are propagated, and the way non-bonding interactions around the link atom are
treated. Each time an energy or gradient calculation is to be done, the link atom coordinates are re-generated from
the current coordinates of the QM and MM atoms making up the QM-MM covalent pair. The link atom is placed
along the bond vector joining the QM and MM atom, at a distance dL−QM from the QM atom. By default dL−QM is
set to the equilibrium distance of a methyl C-H atom pair (1.09 Å) but this can be set in the input file. The default
link atom type is hydrogen, but this can also be specified as an input.
Since the link atom position is a function of the coordinates of the "real" atoms, it does not introduce any new
degrees of freedom into the system. The chain rule is used to re-write forces on the link atom itself in terms

134

9.1. Built-in semiempirical NDDO methods and SCC-DFTB
of forces on the two real atoms that define its position. This is analogous to the way in which "extra points" or
"lone-pairs" are handled in MM force fields.
The remaining details of how the QM-MM boundary is treated are as follows: for the interactions surrounding
the link atom, the MM bond term between the QM and MM atoms is calculated classically using the classical
force field parameters, as are any angle or dihedral terms that include at least one MM atom. The Lennard-Jones
interactions between QM-MM atom pairs are calculated in the same way as described in the section above with
exclusion of 1-2 and 1-3 interactions and scaling of 1-4 interactions. What remains is to specify the electrostatic
interactions between QM and MM atoms around the region of the link atom.
A number of different schemes have been proposed for handling link-atom electrostatics. Many of these have
been tested or calibrated on (small) gas-phase systems, but such testing can neglect some considerations that are
very important for more extended, condensed-phase simulations. In choosing our scheme, we wanted to ensure
that the total charge of the system is rigorously conserved (at the correct value) during an MD simulation. Further,
we strove to have the Mulliken charge on the link atom (and the polarity of its bond to the nearest QM atom)
adopt reasonable values and to exhibit only small fluctuations during MD simulations. Link atoms interact with
the MM field in exactly the same way as regular QM atoms. That is they interact with the electrostatic field due
to all the MM atoms that are within the cutoff, with the exception of the MM link pair atoms (MM atoms that are
bound directly to QM atoms). VDW interactions are not calculated for link atoms. These are calculated between
all real QM atoms and all MM atoms, including the MM link pair atoms. For Generalized Born simulations the
effective Born radii for the link atoms are calculated using the intrinsic radii for the MM link pair atoms that they
are replacing.
In the case of electrostatic embedding the atoms that make up the QM region (including the MM link pair
atom) have their charges from the prmtop file essentially replaced with Mulliken charges. Hence it is important to
consider the issue of charge conservation. The QM region (including the link atoms) by definition must have an
integer charge. This is defined by the &qmmm namelist variable qmcharge. If the MM atoms (including the MM
link pair atoms) that make up the QM region have prmtop charges that sum to the value of qmcharge then there is
no problem. If not, there are two options for dealing with this charge, defined by the namelist variable adjust_q. A
value of 1 will distribute the difference in charge equally between the nearest nlink MM atoms to the MM link pair
atoms. A value of 2 will distribute this charge equally over all of the MM atoms in the simulation (excluding MM
link pair atoms).

9.1.2. A reformulated QM/MM interface for PM3
In the current version of Amber, a reformulated QM-MM core-charge potential (denoted as PM3/MM*) has
been implemented. This reformulated potential scales the interaction between a QM core and a MM charge for the
purpose of better description of the geometry and energy at the QM-MM interface:[253]

a
|qm | − f a ·Ram
core
· −e 1
+ e− f2 ·Ram
(9.6)
EQM/MM
= Za qm (sa sa , sm sm ) 1 +
qm
where Za is the effective core charge of QM atom a, qm is the partial charge on MM atom m, sa is an s orbital
on the QM atom, sm is a notional s orbital on the MM atom, Ram is the QM-MM interatomic distance, and f1a
and f2a are exponential scale factors which depend on the QM atom only. Optimal values for f1a and f2a were
determined based on the PM3 Hamiltonian, and are available for H, C, N and O atoms (so the QM region is
limited to these four atoms; but the MM region is not restricted). Application of this reformulated potential shows
improved prediction of geometry and interaction energy at the QM-MM interface for hydrogen bonded small
molecule complexes typical of biomolecular interactions, without significantly impacting the modeling of other
interaction types, such as dispersion dominant complexes.[253] In a QM/MM calculation, giving qmmm_int=3
along with qm_theory=PM3 will invoke this potential.
Based on PM3/MM*, further developments to the semi-empirical QM/MM coupling method have been introduced – PM3/MMX2 (qmmm_int=4 and qm_theory=PM3) – which shares the same QM core-MM charge equation
with the PM3/MM* model. In addition, a QM parameter, ρmm , is introduced to each type of QM atoms in order
to “fine-tune” the QM electron-MM charge interaction (Eq. 9.7). Although ρmm is a parameter for QM atom, the
subscript mm emphasizes that it is a MM-related property (eqn 3.xx). Parameters are currently avaiable for H, C,
N, O and S QM atoms (manuscript in preparation).

135

9. QM/MM calculations

electron
EQM/MM
= −qm (µa νa , sm sm ) = ∑ ∑ Mlaa k Mlmm k

(9.7)

`a `m

where
a m
Mla k Mlm k =

2la

2la +lm ∑ ∑

h
2 i−1/2
a
ri2j + ρlaa + ρmm

(9.8)

i=1 j=1

9.1.3. Generalized Born implicit solvent
The implementation of Generalized Born (GB) for QM/MM calculations is based on the method described by
Pellegrini and Field.[251] Here, the total energy is taken to be Ee f f from Eq. 9.2 plus Egb from Eq. 4.2. In Egb ,
charges on the QM atoms are taken to be the Mulliken charges determined from the quantum calculation; hence
these charges depend upon the molecular orbital coefficients ci j as well as the positions of the atoms.
As with conventional QM/MM simulations, one then solves for the ci j by setting ∂ Ee f f /∂ ci j = 0. This leads
to a set of SCF equations with a Fock matrix modified not only by the presence of MM atoms (as in "ordinary"
QM/MM simulations), but also modified by the presence of the GB polarization terms. Once self-consistency is
achieved, the resulting Mulliken charges can be used in the ordinary way to compute the GB contribution to the
total energy and forces on the atoms.

9.1.4. Ewald and PME
The support for long range electrostatics in QM/MM calculations using electrostatic embedding is based on a
modification of the Nam, Gao and York Ewald method for QM/MM calculations.[252] This approach works in a
similar fashion to GB in that Mulliken charges are used to represent long range interactions. Within the cut off,
interactions between QM and MM atoms are calculated using a full multipole treatment. Outside of the cut off the
interaction is based on pairwise point charge interactions. For semiempirical NDDO-type methods this leads to a
slight discontinuity at the QM/MM cut off boundary and thus a small energy drift during QM/MM MD simulations
in the NVE ensemble. This energy drift can be avoided by using a switching function at the cutoff (see below).
The implementation in Ref [252] uses an Ewald sum for both QM/QM and QM/MM electrostatic interactions.
This can be expensive for large MM regions, and thus sander uses a modification of this method by Walker and
Crowley[223] that uses a PME model (rather than an Ewald sum) for QM/MM interactions. This is controlled by
the qm_pme variable discussed below.
When running QM/MM Ewald or PME simulations in sander, if QM multipoles are involved in QM-MM
interactions (NDDO methods), a discontinuity in the QM-MM electrostatic potential occurs at the cut-off distance
due to the sudden change in the potential function (the difference between Eqs. 9.9 and 9.10) , thus resulting in
energy conservation problems in the simulation.
rcuto f f
EQM/MM
=

qm (Za − ∑ cµ µ )
r

(9.9)

(9.10)

This problem can be avoided by applying a switching function to smoothly connect the two different potentials.
The QM/MM electrostatic potential using a switching function can thus be written as:
r>cuto f f
r1) (default).

printdipole Controls whether the dipole moment shall be printed every ntpr steps.
= 0 Do not print the dipole moment (default).
= 1 Print the dipole moment of the QM region.
= 2 Print the total dipole moment of the QM and MM region.

writepdb
= 0 Do not write a PDB file of the selected QM region. (default).
= 1 Write a PDB file of the QM region. This option is designed to act as an aid to the user to allow easy

checking of what atoms were included in the QM region. When this option is set a crude PDB file
of the atoms in the QM region will be written on the very first step to the file qmmm_region.pdb.
vsolv

Controls whether solvent molecules shall be included into the QM region (requires settings in the
&vsolv namelist; see also section 9.3 on adaptive solvent QM/MM simulations, in particular the
namelist information in section 9.3.2.2).
= 0 Do not include solvent molecules into the QM region (default).
= 1 Include solvent molecules via simple solvent switching (requires &vsolv namelist).
= 2 Adaptive solvent QM/MM with fixed number of solvent molecules in A and T regions (requires

&vsolv and &adqmmm namelists).
= 3 Adaptive solvent QM/MM with fixed size of A and T regions (requires &vsolv and &adqmmm

namelists).
In addition to the above parameters, the following variables may be set, as described in Section 8.3:
qm_theory, dftb_disper, dftb_3rd_order , dftb_chg , dftb_telec , dftb_maxiter , qmcharge, spin, qmqmdx, verbosity, tight_p_conv, scfconv, pseudo_diag, pseudo_diag_criteria, diag_routine, printcharges, qxd, parameter_file,
peptide_corr, and itrmax.

140

9.1. Built-in semiempirical NDDO methods and SCC-DFTB

9.1.7. Link Atom Specific QM/MM &qmmm Namelist Variables
The following options go in the &qmmm namelist and control the link atom behaviour.
lnk_dis

Distance in Å from the QM atom to its link atom. Currently all link atoms must be placed at the
same distance. A negative value of lnk_dis specifies that the link atom should be placed directly on
top of the MM link pair atom. In this case the distance of the link atom from the QM region changes
as a function of time and the actual value of lnk_dis is ignored. Additionally this means that not all
link atoms will be placed at the same distance. Negative values of lnk_dis will work with regular
link atoms, such as hydrogen, but are really intended for use with pseudo atom / capping approaches.
Default = 1.09Å.

lnk_method This defines how classical valence terms that cross the QM/MM boundary are dealt with.
=1 (Default) in this case any bond, angle or dihedral that involves at least one MM atom, including

the MM link pair atom is included. This means the following (where QM = QM atom, MM =
MM atom, MML = MM link pair atom.):
Bonds = MM-MM, MM-MML, MML-QM
Angles = MM-MM-MM, MM-MM-MML, MM-MML-QM, MML-QM-QM
Dihedrals = MM-MM-MM-MM, MM-MM-MM-MML, MM-MM-MM-MML-QM, MM-

MML-QM-QM, MML-QM-QM-QM
=2 Only include valence terms that include a full MM atom, that is, count the MM link pair atom as

effectively being a QM atom. This option is designed to be used in conjunction with a pseudo
atom / capping type approach where the link atom is parameterized specifically to behave like a
uni-valent version of the MM atom it replaces. This option gives the following interactions:
Bonds = MM-MM, MM-MML
Angles = MM-MM-MM, MM-MM-MML, MM-MML-QM
Dihedrals = MM-MM-MM-MM, MM-MM-MM-MML, MM-MM-MML-QM, MM-MML-

QM-QM
lnk_atomic_no The atomic number of the link atoms. This selects what element the link atoms are to be. Default
= 1 (Hydrogen). Note this must be an integer and an atomic number supported by the chosen QM
theory.
adjust_q

This controls how charge is conserved during a QMMM calculation involving link atoms. When the
QM region is defined the QM atoms and any MM atoms involved in link bonds have their RESP
charges zeroed. If the sum of these RESP charges does not exactly match the value of qmcharge then
the total charge of the system will not be correct.
= 0 No adjustment of the charge is done.
= 1 The charge correction is applied to the nearest nlink MM atoms to MM atoms that form link pairs.

Typically this will be any MM atom that is bonded to a MM link pair atom (a MM atom that
is part of a QM-MM bond). This results in the total charge of QM+QMlink+MM equaling the
original total system charge from the prmtop file. Requires natom-nquant-nlink >= nlink and
nlink>0.
= 2 (default) - This option is similar to option 1 but instead the correction is divided among all MM

atoms (except for those adjacent to link atoms). As with option 1 this ensures that the total charge
of the QM/MM system is the same as that in the prmtop file. Requires natom-nquant-nlink >=
nlink.

141

9. QM/MM calculations

9.1.8. Charge-dependent exchange-dispersion corrections of vdW interactions
The sqm program provides a new charge-dependent energy model consisting of van der Waals (vdW) and polarization interactions between the quantum mechanical (QM) and molecular mechanical (MM) regions in a combined
QM/MM calculation. vdW interactions are commonly treated using empirical Lennard-Jones (L-J) potentials,
whose parameters are often chosen based on the QM atom type (e.g., based on hybridization or specific covalent
bonding environment). This strategy for determination of QM/MM nonbonding interactions becomes tedious to
parametrize and lacks robust transferability. Problems occur in the study of chemical reactions where the “atom
type” is a complex function of the reaction coordinate. This is particularly problematic for reactions, where atoms
or localized functional groups undergo changes in charge state and hybridization.
In sqm, this charge-dependent energy model was implemented based on a scaled overlap model for repulsive
exchange and attractive dispersion interactions that is a function of atomic charge. The model is chemically
significant since it properly correlates atomic size, softness, polarizability, and dispersion terms with minimal
one-body parameters that are functions of the atomic charge[227].
This “Charge-dependent exchange-dispersion corrections of vdW interactions” can be invoked by the
“qxd=.true.” switch in the &qmmm namelist. Note that this model currently does not have any effect on pure
quantum calculations through sqm, the qxd correction is only added to QM/MM interactions in sander. The default values of qxd parameters are set to reproduce the regular L-J interactions of typical atom types (HC for H,
C* for C, N for N, OW for O, and parameters for F and Cl are optimized[227]) when the charge dependence
parameters are zero. There are eight qxd parameters (symbols used in the reference[227] are indicated in the
parentheses): qxd_s (s), qxd_z0 (ζ (0)), qxd_zq (ζq ), qxd_d0 (α1 ), qxd_dq (3 × B), qxd_q0 (α2 ), qxd_qq (3 × B),
and qxd_neff (Ne f f (0)). All parameters can be modified through external user-defined parameter files (see the
usage of ’parameter_file’ in Section 8.3).

9.2. Interface for ab initio and DFT methods
In addition to the built-in semi-empirical methods sander also supports QM/MM simulations with ab initio wave
function theory (WFT) and density functional theory (DFT) potentials via an interface to external QM software
packages[254]. The implementation makes use of the existing QM/MM infrastructure that has been developed
earlier for the semi-empirical methods. Thus, much of AMBER’s previous QM/MM functionality such as the userfriendly link atom approach are available and the implementation remains simple and transparent to use without
any significant additional steps in the simulation setup as compared to semi-empirical QM/MM simulations. At
present the interface supports several well-known and widely used QM software packages. Mechanical embedding
is available for
• ADF (Amsterdam Density Functional) [255, 256]
• GAMESS-US [257, 258]
• NWChem [259]
Mechanical and electrostatic embedding is available for
• Gaussian [221]
• Orca [260]
• TeraChem [261]
While ADF, Gaussian and TeraChem are commercial programs, GAMESS-US, NWChem and Orca are available
at no cost for academic research. The interface has been written in a modular fashion and is easily extensible to
support other QM software packages. It is our intention to keep adding support for other software packages. If you
are interested in interfacing a specific program, please do not hesitate to contact us.
The interface was developed by Andreas Goetz (SDSC, UCSD) with help of Matthew Clark (SDSC) and support
by Ross Walker (SDSC, UCSD). Thanks are due to Christine Isborn and Todd Martinez (Stanford University) for
modifications to the TeraChem code to support this interface and to Mark Williamson (University of Cambridge)

142

9.2. Interface for ab initio and DFT methods
for an initial version of the module that supports NWChem. If you make use of this interface, please cite the
following work:
• A. W. Götz, M. A. Clark, R. C. Walker, An extensible interface for QM/MM molecular dynamics simulations
with AMBER, J. Comput. Chem. 35, 95-108 (2014), DOI: 10.1002/jcc.23444
If you are using the interface with the TeraChem code, please cite in addition the following work:
• C. M. Isborn, A. W. Götz, M. A. Clark, R. C. Walker, T. J. Martínez, Electronic Absorption Spectra from
MM and ab initio QM/MM Molecular Dynamics: Environmental Effects on the Absorption Spectrum of
Photoactive Yellow Protein, J. Chem. Theory Comput. 8, 5092-5106 (2012), DOI: 10.1021/ct3006826
Access to QM methods not available within Amber is also possible via the Amber interface to the PUPIL simulation
framework. For details, see refs. 262, 263. In what follows we will describe the new interface that is native to
sander.

9.2.1. Theory
As described in section 9.1, the Hamiltonian of a system that is partitioned into a QM region that is treated with
WFT and a classical region that is treated with MM consists of three components and the energy associated with
this Hamiltonian is obtained as the corresponding expectation value
E = hΨ|HQM + HQM/MM |Ψi + EMM .

(9.11)

A QM/MM calculation therefore requires not only to choose the WFT used in the QM region and the MM model
used for the MM region, but in addition also the form of the QM/MM Hamiltonian which describes the interaction
between the quantum and the classical region. The most simple approach is to neglect any electronic coupling
between the QM and the MM system and include only the classical non-bonded van der Waals (vdW) and electrostatic interactions between the QM and the MM atoms. This is useful to impose steric constraints on the embedded
QM system and commonly referred to as mechanical embedding. In most cases, however, it is better to allow for
an explicit polarization of the QM system due to the presence of the point charges on the MM atoms. This is
referred to as electronic embedding and the resulting interaction energy becomes

electronic
EQM/MM
=

ZA QB
QA
dr +
∑
|r
−
R
|
A
A∈QM,B∈MM RAB
A∈MM
"
12
#
σAB
σAB 6
+
∑ εAB RAB − RAB .
A∈QM,B∈MM
Z

∑

ρ(r)

(9.12)

This QM/MM energy expression also holds for DFT and the terms represent, in order, the electrostatic interaction
between the QM electron density and the MM point charges, the electrostatic interation between the QM point
charge nuclei and the MM point charges, and the van der Waals repulsion between the QM and MM atoms.
The forces acting on an atom A in a QM/MM calculation are given in terms of derivatives of the total energy
expression (9.11) with respect to the Cartesian coordinates of the atom,
FA = −∇A EQM − ∇A EQM/MM − ∇A EMM ,

(9.13)

where ∇A = ∂ /∂ RA = (∂ /∂ RxA , ∂ /∂ RyA , ∂ /∂ RzA ). If a QM and an MM program are coupled for QM/MM calculations, the QM program will calculate the QM forces −∇A EQM acting on QM atoms and the MM program the MM
forces −∇A EMM acting on the MM atoms. All that remains, is to calculate the forces acting on QM and MM atoms
due to the QM/MM interaction energy, −∇A EQM/MM . For mechanical embedding this will be entirely handled by
the MM program. For electronic embedding the forces are given as

143

9. QM/MM calculations

electronic
∇A EQM/MM
= ZA

QB (RA − RB )
+ ∑
∑
R3AB
B∈MM
B∈MM

= −ZA EMM (RA ) −

∂ ρ(r) QB
LJ
dr + ∑ ∇AVAB
∂ RA |r − RB |
B∈MM

ρ(r)EMM (r) dr +

∑

(9.14)

LJ
∇AVAB

B∈MM

for the derivatives with respect to the positions of the QM atoms A where EMM is the electric field generated by
LJ is the Lennard-Jones potential from (9.12) and
the MM point charges and and VAB

electronic
∇B EQM/MM
= QB

ZA (RB − RA )
+
R3AB
A∈QM

∑

= −QB EQM (RB ) +

∑

ρ(r)

LJ
∇BVAB

QB (RB − r)
mechanic
dr + ∇B EQM/MM
|r − RB |3

(9.15)

A∈QM

for the derivatives with respect to the positions of the MM atoms B where EQM is the electric field due to the QM
charge distribution. The contributions to the gradient due to the point charge interactions and due to the interaction
between the MM point charges and the QM electrons is evaluated by the QM program. Some QM programs do not
calculate the forces acting on the MM atoms (point charges) due to the presence of the QM system but in general
are able to return the electric field EQM at arbitrary points in space which is then used to obtain these forces. The
van der Waals repulsion (Lennard-Jones interaction) between QM and MM atoms is treated by AMBER in the
same way as for semiempirical NDDO-type and DFTB methods.

9.2.2. General Remarks
When using the AMBER interface to external QM software packages for performing WFT or DFT based QM or
QM/MM MD simulations, it is absolutely critical to be aware of the capabilities and limitations of the QM method
to be employed. In particular, QM based MD can be more tricky than MM based MD in the sense that it is more
likely that the QM program can fail for example due to SCF convergence problems. This can be the case if the
geometry of the QM region is far from its ground state equilibrium, for example because a simulation is started
from a bad geometry or performed at high temperature.
We have gone to large efforts and analyzed a large set of test simulations to provide the best default parameters
for the supported QM programs such that forces are computed with sufficient accuracy to guarantee energy coservation for constant energy MD simulations. Of particular importance are SCF convergence and associated integral
negelct thresholds and the size of the grid used for the numerical quadrature of the exchange-correlation (XC)
potential and energy for DFT calculations. However, other than providing appropriate input parameters, AMBER
does not have any control over the external program and it is at the user’s discretion to employ sensible input
parameters for the QM program and to prepare the system such that the simulations are started at a reasonable
starting structure.
In any case we highly recommend to write restart files frequently so that a simulation can be restarted without
loss of much computational time in the case that a simulation should crash. The interface also stores the last inand output files of the external QM program during each MD step. Should there be any problems with the QM
program, it is therefore possible to analyze the reasons and take appropriate countermeasures.
The interface requires data to be exchanged between sander and the QM program. The interface is based on
file exchange and system calls and, during each step of a geometry optimization or an MD simulation, writes an
input file for the external program, starts a single point gradient calculation with the external program, and reads
the energy and forces from the external program’s output file (binary ADF checkpoint or formatted GAMESS,
Gaussian, ORCA and TeraChem output file). Data communication via MPI is also implemented and currently
supported by TeraChem.

144

9.2. Interface for ab initio and DFT methods

9.2.3. Limitations
In principle, all types of simulations that are possible with sander are supported. There are, however, some
restrictions for simulations that require sander to run in parallel, in particular path integral molecular dynamics (PIMD) and replica exchange molecular dynamics (REMD), see the discussion of Parallelization below. The
interface to external QM programs also lacks some features regarding solvent models in comparison to the semiempirical MNDO and DFTB QM/MM implementation that is available in AMBER, the most critical ones are listed
here.
Generalized Born (GB) implicit solvent models are not supported if external QM programs
are used for the QM region.

Generalized Born

Particle Mesh Ewald (PME) and Periodic Boundary Conditions The PME approach for treating long-range
electrostatic QM/MM and QM/QM interactions in periodic systems is currently not supported. It is possible to
use periodic boundary conditions but a cutoff is used for the point charges to be included in the QM Hamiltonian
(determined by &qmmm namelist variable qmcut) thus truncating the long-range QM/MM electrostatic interactions
in (9.12). This leads to discontinuities in the potential energy surface and poor energy conservation for MD runs
in the NVE ensemble. The user may consider running non-periodic simulations with a cutoff that is larger than the
system size thus effectively including all interactions.

9.2.4. Performance Considerations
The computational cost of DFT is comparable to Hartree–Fock (HF) theory which is the simplest WFT method
that serves as zeroth order approximation for more elaborate correlated WFT methods such as Møller–Plesset perturbation theory, configuration interaction theory and coupled cluster theory. The calculations can be accelerated
by using density fitting approaches, sometimes called resolution-of identity (RI) approximation, which in the case
of DFT with exchange-correlation (XC) functionals that do not require admixture of exact HF-exchange, leads to
speedups of roughly one order of magnitude without compromising the accuracy of the results. Nevertheless, the
computational cost of DFT is in general two to three orders of magnitude higher than that of semiempirical QM
models. We recommend to carefully test the performance of the QM program to choose an optimal number of
processor counts for parallelized QM calculations. Typical simulation performance for typical QM system sizes of
tens of atoms will be on the order of a few picoseconds per day, depending on the underlying QM model chosen.

9.2.5. Parallelization
The MPI parallel executable sander.MPI can be used to run QM/MM MD simulations with external QM software
in which the MM portion of the calculation is parallelized. However, the computational cost of the MM part is
usually small compared to the cost of the QM part. In order to execute the QM part of the calculation in parallel,
the external QM program has to be instructed to do so, as described in the sections below.
In the case of PIMD or REMD simulations that require a separate energy and force evaluation for each group
at each time step, the parallelized executable sander.MPI has to be used. Multiple processes can be launched per
group to parallelize the MM calculations. Care has to be taken to choose the right number of parallel threads in
the external QM program. For example, on a machine with 32 cores, a simulation with 16 beads or replicas can
run the external QM program with 2 threads in parallel to make maximum use of the available processing cores.
If the available processors are spread over multiple nodes, special care has to be taken to ensure that the different
instances of the external QM program are launched on the correct nodes.
It is currently not possible to execute sander.MPI in parallel via MPI if the external QM program is also parallelized using MPI.

9.2.6. Usage
All that is required to use the interface is a working installation of AMBER and one or more of the supported
QM programs. In order to use the external program from within sander, the &cntrl namelist variable ifqnt = 1

145

9. QM/MM calculations
must be set to enable QM calculations and the &qmmm namelist variable qm_theory = ’EXTERN’ must be set to
enable the external interface. The &qmmm namelist variable qmmask or iqmatoms is used for selecting the QM
region just as for QM/MM calculations with the semiempirical NDDO-type and DFTB approaches that are
natively available in AMBER. Charge and spin multiplicity for the QM region need to be defined via the variables
qmcharge and spin, respectively, in the &qmmm namelist. For a QM MD simulation, the sander input file
therefore needs to contain

! example input for QM simulation with external QM program
&cntrl
...
ifqnt = 1,
! switch on QM/MM
/
&qmmm
qmmask = ’@*’,
! select QM atoms (here: make all QM)
qmcharge = 0,
! charge on QM region (default = 0)
spin = 1,
! spin multiplicity of QM region (default = 1)
qm_theory = ’EXTERN’, ! use external QM program
/

For QM/MM simulations with electronic embedding (this is the default) we recommend to include all MM point
charges as external electric field in the QM Hamiltonian to avoid problems with energy conservation. For nonperiodic simulations this can be achieved by setting the &qmmm namelist variable qmcut to a value larger than the
system size.
In addition either the &adf, &gms, &nw, &gau, &orc or &tc namelist must be present to use either ADF,
GAMESS, NWChem, Gaussian, ORCA or TeraChem, respectively, and to assign parameters for the external QM
program. Please refer to the ADF, GAMESS, NWChem, Gaussian, ORCA or TeraChem user manual for details on
settings for the ab initio or DFT calculations. A list of namelist variables and their default setting is given below.
The defaults have been chosen such that energy conserving MD simulations in the NVE ensemble are possible.
An exception is ADF for which good energy conservation cannot be reached at present, in particular with the DZ
basis set. NWChem has not been extensively tested.
Properties that are calculated along the trajectory are printed to property files with names adf_job.ext,
gms_job.ext, gau_job.ext, orc_job.ext and tc_job.ext, where ext is either dip for dipole moment (x,
y, z component and absolute value) or chg for atomic charges, where supported. These property files are only
written if requested and will be deleted at the beginning of a run, so back them up in case a trajectory needs to be
restarted.
All calculations with a spin multiplicity larger than one will automatically be performed in the framework of
an unrestricted formalism (as opposend to restricted open shell), that is with unrestricted HF (UHF), unrestricted
DFT (UDFT) and MP2 with a UHF reference wave function (UMP2).
In addition to controlling the external programs via the sander input file, you may supply a template input file
for the external program in order to provide additional input which is not supported by the external interface.
The format, name, and input requirements for the template file vary with the external program as detailed in the
corresponding program’s documentation below. If you are using your own template, please make sure that the
parameters of the QM method (like SCF convergence threshold and XC quadrature grid size) yield sufficiently
accurate forces.
9.2.6.1. AMBER/ADF

To use ADF with the external interface, ADF must be properly installed on the working machine. In particular,
the executable adf must be in the search path. Please note that by default the ZORA/QZ4P fit basis will be used
for all basis sets to maximize the accuracy of the gradients calculated by ADF and thus obtain the best possible
energy conservation.
Limitations

146

At present only mechanical embedding is supported.

9.2. Interface for ab initio and DFT methods
&adf Namelist variables

basis

Basis set type to be used in the DFT calculation. Valid standard basis set types are: SZ, DZ, DZP,
TZP, TZ2P, TZ2P+ and ZORA/QZ4P. (Default: basis = ’DZP’)

fit_type

Fit basis set type to be used for density fitting. Valid values are identical to the available basis sets
(SZ, DZ etc) in which case the fit basis corresponding to the AO Basis will be used; “standard” will
choose the fit basis that by default corresponds to the basis set chosen with the basis namelist variable.
(Default: fit_type = ’ZORA/QZ4P’)

core

Type of frozen core to use. Allowed values are: None, Small, Medium, Large. (Default: core =
’None’)

Exchange-correlation functional to be used. Popular choices are ’LDA VWN’, ’GGA BLYP’, ’GGA
PBE’, ’HYBRID B3LYP’ and ’HYBRID PBE0’. Consult the ADF manual for all available options.
(Default: xc = ’GGA BLYP’)

scf_iter

Maximum number of SCF cycles allowed. (Default: scf_iter = 50)

scf_conv

Threshold upon which to stop the SCF procedure. The tested error is the commutator of the Fock
matrix and the density matrix. Convergence is considered to be achieved if the maximum element of
the commutator (which is zero for an optimized wave function) is smaller than scf_conv. (Default:
scf_conv = 1.0d-06)

integration Numerical integration accuracy. (Default: integration = 5.0)
num_threads Number of threads (and thus CPU cores) for ADF to use. Note that this is not requied if you are
running in a queuing system as ADF will automatically use the full number of reserved cores. (Default:
num_threads = 0 [this causes ADF to use all available cores on a machine])
use_dftb

Specifies whether DFTB shall be used with ADF’s DFTB program dftb. If use_dftb = 1 then DFTB
will be used and only variables charge and scf_conv will be considered. (Default: use_dftb = 0 [do
not use DFTB, regular DFT calculation]) - works only with older DFTB versions (prior to 2011).

exactdensity The exact (as opposed to fitted) electron density is used for the evaluation of the exchange-correlation
potential if exactdensity = 1. (Default: exactdensity = 0)
use_template Determine whether or not to use a user-provided template file for running external programs. (Default: use_template = 0)
ntpr

Controls frequency of printing for dipole moment to file adf_job.dip (Defaults to &cntrl namelist
variable ntpr)

dipole

Toggles writing of dipole moment to file adf_job.dip (Default: dipole = 0)

An input file for QM or (mechanical embedding) QM/MM MD with ADF using the PBE functional
and the TZP basis set therefore would have to contain
Example

&adf
xc = ’GGA PBE’,
basis = ’TZP’,
/

147

9. QM/MM calculations
Template input file

The template file for ADF should be named adf_job.tpl and must contain the following

keywords:
BASIS ... END
SAVE TAPE21

You should not include the following (block) keywords in the template file as these are taken care of by sander:
UNITS
FRAGMENTS ... END
RESTART
GRADIENT
ATOMS ... END

9.2.6.2. AMBER/GAMESS-US

To use GAMESS with the external interface, GAMESS must be compiled on the target system. Make note of the
version number you specify during the GAMESS compilation process (default is 00 which makes the GAMESS
execution script rungms look for the executable gamess.00.x). If you use a different version number you must
specify it with the gms_version namelist variable. $GMS_PATH should be set to the path where the script rungms
is located (for example /opt/gamess/). We assume that the rungms script copies the output .dat files to the
directory from which GAMESS is invoked. If this is not the case, please modify the script rungms accordingly.
Only mechanical embedding is supported with GAMESS. The available QM models are limited to
HF, DFT and MP2 since only for these analytical gradients are available in GAMESS.
Limitations

&gms Namelist variables

basis

Basis set type to be used in the calculation. Presently supported are the Pople type basis sets STO-3G,
6-31G, 6-31G*, 6-31G**, 6-31+G*, 6-31++G*, 6-311G, 6-311G* and 6-311G**. Also supported are
the Karlsruhe valence triple zeta basis sets KTZV, KTZVP and KTZVPP (with none, one and two
polarization functions, respectively) and the Dunning-type correlation consistent basis sets CCn (n
= D, T, Q, 5, 6; officially called cc-pVnZ) and ACCn (as CCn but augmented with a set of diffuse
function, officially called aug-cc-pVnZ). (Default: basis = ’6-31G*’)

method

QM method to be used. At present, we support ’HF’ for Hartree–Fock, ’MP2’ for second order MøllerPlesset perturbation theory and any of the supported DFT functionals. Popular choices for for DFT
functionals include ’BP86’, ’BLYP’, ’PBE’, ’B3LYP’ or ’PBE0’. (Default: method = ’BP86’)

nrad

Number of radial points in the Euler-MacLaurin quadrature of the XC potential and energy density.
(Default: nrad = 96)

nleb

Number of angular points in the Lebedev grids for the numerical quadrature of the XC potential and
energy density. (Default: nleb = 590 [The GAMESS default of 302 is not accurate enough to conserve
energy])

scf_conv

SCF convergence threshold. Convergence is reached when the absolute density change between two
consecutive SCF cycles is less than scf_conv}. (Default: scf_conv = 1.0D-06)

maxit

Maximum number of SCF iterations. (Default: maxit = 50)

gms_version This is the version number specified when building GAMESS. (Default: gms_version = 00)
num_threads Number of threads (and thus CPU cores) for GAMESS to use. Note that GAMESS may require a
special setup in the rungms script to be able to run using multiple threads. Unless num_threads is
explicitly specified, GAMESS will only use one thread (run on one core). (Default: num_threads = 1)

148

9.2. Interface for ab initio and DFT methods
mwords

The maximum replicated memory which your job can use, on every node. This is given in units of
1,000,000 words (as opposed to 1024*1024 words), where a word is defined as 64 bits. You may
need to increase this value if GAMESS crashes due to not having enough memory allocated. (Default:
mwords = 50)

use_template Determine whether or not to use a user-provided template file for running external programs. (Default: use_template = 0)
ntpr

Controls frequency of printing for dipole moment and atomic charges to files gms_prop.ext (Defaults
to &cntrl namelist variable ntpr)

chelpg

CHELPG charges are calculated if chelpg = 1. These charges are written to the file gms_prop.chg
(Default: chelpg = 0)

dipole

Toggles writing of dipole moment to file gms_prop.dip (Default: dipole = 0)

An input file for QM or (mechanical embedding) QM/MM MD with GAMESS using the PBE
functional and the 6-31G** basis set that should run GAMESS on 16 CPU cores therefore would have to contain
Example

&gms
method = ’DFT’,
dfttyp = ’PBE’,
basis = ’6-31G**’,
num_threads = 16,
/

Template input file The template file for GAMESS should be named gms_job.tpl and the $CONTRL card must
contain the following keywords:
RUNTYP=GRADIENT
UNIT=ANGS
COORD=UNIQUE

You should not include the $DATA card in the template file as it is taken care of by sander.
9.2.6.3. AMBER/Gaussian

To use Gaussian with the interface, Gaussian03 or Gaussian09 must be properly installed on the system and the
g03 or g09 executable must be in the path.
A cutoff is applied to QM/MM interactions in QM/MM simulations using electrostatic embedding
with and without PBCs. This leads to discontinuities in the potential energy surface and poor energy conservation.
In the case of QM/MM simulations without PBCs, this cutoff (qmcut variable in the &qmmm namelist) can be set
to a number that is larger than the simulated system, thus effectively not applying a cutoff. This is recommended.
Limitations

&gau Namelist variables

basis

Basis set type to be used in the calculation. Any basis set that is natively supported by Gaussian can
be used. Examples are the single zeta, split valence or triple zeta Pople type basis sets ’STO-3G’, ’321G’, ’6-31G’ and ’6-311G’. The split-valence or triple zeta basis sets can be augmented with diffuse
functions on heavy atoms or additionally hydrogen by adding one or two plus signs, respectively, as in
’6-31++G’. Polarization functions on heavy atoms or additionally hydrogens are used by adding one
or two stars, respectively, as in ’6-31G**’. (Default: basis = ’6-31G*’)

149

9. QM/MM calculations
method

Method to be used in the calculation. Can either be one of the WFT models for which Gaussian
supports gradients, for example ’RHF’ or ’MP2’, or some supported DFT functional. Popular choices
are ’BLYP’, ’PBE’ and ’B3LYP’. (Default: method = ’BLYP’)

scf_conv

num_threads Number of threads (and thus CPU cores) for Gaussian to use. Unless num_threads is explicitly
specified, Gaussian will only use one thread (run on one core). (Default: num_threads = 1)
use_template Determine whether or not to use a user-provided template file for running external programs. (Default: use_template = 0)
ntpr

Controls frequency of printing for dipole moment to file gau_job.dip (Defaults to &cntrl namelist
variable ntpr)

dipole

Toggles writing of dipole moment to file gau_job.dip (Default: dipole = 0)

An input file for QM or QM/MM MD with Gaussian using the BP86 functional and the 6-31G**
basis set and running in parallel on 8 threads therefore would have to contain
Example

&gau
method = ’BP86’,
basis = ’6-31G**’,
num_threads = 8,
/

Template input file The template file for Gaussian should be named gau_job.tpl and should only contain the
route section of a Gaussian input file. The route section defines the method to be used and SCF convergence
criteria. Charge and spin multiplicity are provided by via the &qmmm namelist. For example for a B3LYP
calculation with 6-31G* basis set, the route section would be:
#P B3LYP/6-31G* SCF=(Conver=8)

Do not include any information about coordinates or point charge treatment since this will all be handled by sander.
Also, do not include any Link 0 Commands (line starting with %) since these are handled by sander. If you want
to run Gaussian in parallel, specify the number of processors via the num_threads variable in the &gau namelist.
9.2.6.4. AMBER/Orca

To use Orca with the interface, Orca must be properly installed on the system, the Orca executables need to
reside in a directory that is in the search path. For convenience of use, namelist parameters in general correspond
to Orca keywords, see the Orca manual for details.
A cutoff is applied to QM/MM interactions in QM/MM simulations with and without PBCs. This
leads to discontinuities in the potential energy surface and poor energy conservation. In the case of QM/MM
simulations without PBCs, this cutoff (qmcut variable in the qmmm namelist) can be set to a number that is larger
than the simulated system, thus effectively not applying a cutoff. This is recommended.
Also note that ORCA only supports OpenMPI for parallel calculations.
Limitations

150

9.2. Interface for ab initio and DFT methods
&orc Namelist variables

basis

Basis set type to be used in the calculation. Possible choices include ’svp’, ’6-31g’, etc. See Orca
manual for a complete list. (Default: basis = ’SV(P)’)

cbasis

Auxillary basis set for correlation fitting. See Orca manual for a complete list. (Default: basis =
’NONE’)

jbasis

Auxillary basis set for Coulomb fitting. See Orca manual for a complete list. (Default: basis =
’NONE’)

method

Method to be used in the calculation. Popular choices include ’hf’, ’pm3’, ’blyp’, and ’mp2’. (Default:
method = ’blyp’)

convkey

General SCF convergence setting for simplified Orca input. Can take values ’TIGHTSCF’, ’VERYTIGHTSCF’, etc. (Default: convkey=’VERYTIGHTSCF’)

scfconv

SCF convergence threshold for the energy. (Default: scfconv = -1, that is, not in use since we use the
general convergence settings keyword convkey. Otherwise this would lead to SCF energy convergence
of 10−N au, if set to N.)

grid

Grid type used during the SCF for the XC quadrature in DFT. (Default: grid = 4, this corresponds to
Intacc = 4.34 for the radial grid and an angular Lebedev grid with 302 points. Conservatively chosen
together with finalgrid to conserve energy.)

finalgrid

Grid type used for the energy and gradient calculation after the SCF for the XC quadrature in DFT.
(Default: finalgrid = 6, this corresponds to Intacc = 5.34 for the radial grid and an angular Lebedev
grid with 590 points. Conservatively chosen together with grid to conserve energy.)

maxiter

Maximum number of SCF iteractions. (Default maxiter = 100)

maxcore

Global scratch memory (in MB) used by Orca. You may need to increase this when running larger
jobs. See Orca manual for more information. (Default maxcore = 1024)

num_threads Number of threads (and thus CPU cores) for Orca to use. Note that Orca only supports OpenMPI.
(Default: num_threads = 1)
use_template Determine whether or not to use a user-provided template file for running external programs. (Default: use_template = 0)
ntpr

Controls frequency of printing for the dipole moment to file orc_job.dip (Defaults to &cntrl namelist
variable ntpr)

dipole

Toggles writing of the dipole moment to file orc_job.dip (Default: dipole = 0)

An input file for QM or QM/MM MD with Orca using the BLYP functional, the SVP basis set
therefore would have to contain
Example

&orc
method = ’blyp’,
basis = ’svp’,
/

151

9. QM/MM calculations
The template file for Orca should be named orc_job.tpl and must at least contain
keywords specifying the method and basis set to be used in the calculation, for example:

Template input file

# ORCA input file for BLYP/SVP simulation
! BLYP SVP

You should not include the following keywords in the template file as these are taken care of by sander (like
setting the runtype and adding coordinates):
# NOT to be included in ORCA input file
!engrad
!energy # (or any run type)
%pointcharges
*xyzfile # (or any coordinates)

9.2.6.5. AMBER/TeraChem

To use TeraChem with the interface, TeraChem must be properly installed on the system. In particular, the
terachem executable needs to be in the search path. Namelist parameters correspond to TeraChem keywords, see

the TeraChem manual for details.
A cutoff is applied to QM/MM interactions in QM/MM simulations with and without PBCs. This
leads to discontinuities in the potential energy surface and poor energy conservation. In the case of QM/MM
simulations without PBCs, this cutoff (qmcut variable in the &qmmm namelist) can be set to a number that is
larger than the simulated system, thus effectively not applying a cutoff. This is recommended.
Limitations

&tc Namelist variables

basis

Basis set type to be used in the calculation. Possible choices presently (TeraChem version 1.4) are
’STO-3G’, ’3-21G’, ’6-31G’ and ’6-311G’, ’3-21++G’ and ’6-31++G’ (Default: basis = ’6-31G’)

method

Method to be used in the calculation, can be either ’RHF’ or some supported DFT functional. Popular
choices are ’BLYP’, ’PBE’ and ’B3LYP’. (Default: method = ’BLYP’)

dftd

Determines whether dispersion corrections are applied in the case of DFT calculations. (Default: dftd
= ’no’)

precision

Precision model setting (single vs double precision). (Default: precision = ’mixed’)

dynamicgrid Use coarse grid during early SCF iterations. (Default: dynamicgrid = ’yes’)
threall

Determines a variety of thresholds. (Default: threall = 1.0E-11)

convthre

SCF convergence threshold for the wavefunction. (Default: convthre = 3.0E-05, which leads to SCF
energy convergence of approximately 10−7 au or 10−4 kcal/mol)

maxit

Maximum number of SCF iterations. (Default: maxit = 100)

dftgrid

DFT grid to be employed for the numerical XC quadrature in DFT calculations. (Default: dftgrid = 1)

ngpus

Determines how many GPUs are to be used. (Default: ngpus = 0, which uses all available GPUs)

gpuids

If ngpus has a value other than zero, this determines the IDs of the GPUs to be used for the calculation.
(Default: gpuids = 0, 1, 2, etc.)

executable Name of the TeraChem executable. (Default: executable = terachem)
use_template Determine whether or not to use a user-provided template file for running external programs. (Default: use_template = 0)

152

9.3. Adaptive solvent QM/MM simulations
ntpr

Controls frequency of printing for dipole moment and atomic charges to files tc_job.ext. (Defaults
to &cntrl namelist variable ntpr)

charge_analysis Toggles writing of atomic charges to file tc_job.chg (Options: ’none’ or ’Mulliken’. Default:
dipole = ’none’)
dipole

Toggles writing of dipole moment to file tc_job.dip (Default: dipole = 0)

An input file for QM or QM/MM MD with TeraChem using the PBE functional and the 6-31G* basis
set therefore would have to contain
Example

&tc
method = ’PBE’,
basis = ’6-31G*’,
/

Template input file

The template file for Terachem should be named tc_job.tpl and must at least contain the

following keywords:
basis
method

Any content of the template file after a line containing the end keyword will be ignored.
You should not include the following keywords in the template file as these are taken care of by sander.
Instead, specify these via the &qmmm or &tc namelist:
run
charge
spinmult
coordinates
pointcharges
amber
gpus

9.3. Adaptive solvent QM/MM simulations
Traditional QM/MM approaches are based on a static QM/MM partitioning in which atoms belonging to the
QM and MM regions are selected at the beginning of a molecular dynamics simulation. Such a static partitioning
cannot be applied if part of the bulk solvent in the vicinity of a region of interest needs to be included in the QM
region. Examples include cases in which the bulk solvent participates directly in a chemical reaction or in which
important interactions between the solute and the bulk solvent, such as polarization and charge transfer, are not
well parameterized at the QM/MM level and thus need to be described quantum mechanically. Due to molecular
diffusion, solvent molecules will constantly exchange between the QM and MM regions and thus require a special
treatment.
Several approaches have been developed that allow molecules to change their QM or MM character when crossing the boundaries between the QM and MM regions. A good overview and comparison of these approaches
is available in the work by Bulo et al.[264]. One of the most accurate approaches is the difference-based adaptive solvation (DAS) method[265], in the following simply called adaptive QM/MM (adQM/MM). This method
is available in Amber through a parallelized implementation that has been developed by Andreas Goetz (SDSC)
with help from Ross Walker (SDSC), Rosa Bulo (Utrecht University) and Kyoyeon Park (UCSD). The usefulness of this adQM/MM approach for aqueous systems has been demonstrated with a development version of this
implementation[266]. If you publish work that results from using this implementation, please cite the following
work:

153

9. QM/MM calculations
• A. W. Götz, K. Park, R. E. Bulo, F. Paesani, R. C. Walker, Efficient adaptive QM/MM implementation:
Application to ion binding by peptides in solution, in preparation.
• R. E. Bulo, B. Ensing, J. Sikkema, L. Visscher, Toward a practical method for adaptive QM/MM simulations,
J. Chem. Theory Comput. 9, 2212-2221 (2009), DOI: 10.1021/ct900148e
In what follows we will describe the theoretical background of this implementation and how to perform adQM/MM
simulations. For an alternative approach, see section 9.4.

9.3.1. Theoretical background
In adQM/MM simulations, we distinguish three different regions, an active region (A), a transition region (T),
and the environment region (E). The active region contains both the part of the system that is permanently treated
quantum mechanically (similar to the QM region in regular QM/MM simulations) and the solvent molecules in its
vicinity that are also treated quantum mechanically. The E region is the part of the system that is treated at the MM
level. Within the T region, molecules change their character from purely QM to purely MM, that is, molecules in
the T region have partial QM and MM character, depending on their position within the T region. The T region that
connects the A and E regions is required to guarantee that the potential energy surface or forces remain continuous
throughout the simulation.
9.3.1.1. System partitioning

In the adQM/MM method[265], a partial MM character λ is assigned to each molecule in the T region. The
value of λ depends on the distance of the molecule from the center of the A region according to


0

r ≤ RA


1

r ≥ RT

)2 (3RT −RA −2r)
λ (r) = (r−RA(R
3
T −RA )


RA < r < RT ,

(9.16)

where RA and RT are the inner and outer radii delimiting the T region. The switching function thus interpolates
smoothly between QM (A region) and MM (E region).
The adQM/MM energy can be constructed as a weighted average of regular QM/MM energies due to all possible
2NT partitionings in which the NT molecules in the T region are assigned either to the QM or the MM region,
QM/MM

E adQM/MM = ∑ σa Ea

(9.17)

The statistical coefficients σa for the QM/MM partitionings are defined on basis of the λ values defined above,
(
MM
0
if max({λ }QM
a ) > min({λ }a )
,
(9.18)
σa =
QM
QM
MM
min({λ }MM
a ) − max({λ }a ) if max({λ }a ) ≤ min({λ }a )
and {λ }MM
are the sets of λ values for a given QM/MM partitioning a that are assigned to the QM
where {λ }QM
a
a
and MM regions, respectively. Due to this choice of coefficients, the weight σa of a QM/MM partitioning is zero if
the partition contains one or more MM molecules closer to the A region than any of the QM molecules. The total
number of non-zero QM/MM partitionings in an adQM/MM simulations is thus NT + 1. In addition it is guaranteed
that the weight of each partition varies smoothly from 0 to 1, removing discontinuities in the system dynamics that
would appear in standard QM/MM simulations if a molecule would change its character by diffusing in or out of
the QM region.
9.3.1.2. Force interpolation

The forces resulting from the adQM/MM energy 9.17 are a weighted sum of the force from each non-zero
QM/MM partitioning and also contain a term that depends on the derivatives of the weight functions,

154

9.3. Adaptive solvent QM/MM simulations

"
FadQM/MM = − ∑
a

QM/MM

∂ Ea
σa
∂R

#
∂ σa QM/MM
+
Ea
.
∂R

(9.19)

This introduces an artificial dependence on the relative energies of the different QM/MM partitionings. Thus, in
place of the energy interpolation scheme, a force interpolation is applied in which the forces are given as
QM/MM

F̃adQM/MM = − ∑ σa
a

∂ Ea
∂R

(9.20)

The force interpolation does not conserve the energy from equation 9.17 but it is possible to define a conserved
quantity according to
Ẽ adQM/MM = E adQM/MM +W,

(9.21)

where the correction term W is defined through
∂W
∂ σa QM/MM
= −∑
Ea
.
∂R
a ∂R

(9.22)

The quantity Ẽ adQM/MM is not a potential energy since it is only defined along the path taken by the system during
the simulation. It is nevertheless useful to monitor this quantity to determine whether the simulation settings lead
to numerical stability. The correction term W can be expressed as the path integral of its force vector from equation
9.22, which can be discretized. For step n of an MD simulation it is given as
n

QM/MM

Wn = ∑ ∑ Ea
i

(i)

σa (i + 1) − σa (i − 1)
.
2

(9.23)

The Amber implementation uses exclusively the force interpolation scheme from equation 9.20 and optionally
computes the correction term W from equation 9.23 to enable monitoring of the conserved quantity Ẽ adQM/MM
from equation 9.21.
9.3.1.3. Alternative definitions of active, transition and environment regions

So far we have defined the boundaries between the A, T, and E regions with the distances RA and RT from
the center of the active region. In this case both the A and the T regions have fixed volumes but the number
of solvent molecules inside each region can vary during the simulation. Alternatively, we can fix the number of
solvent molecules NA and NT within the A and T regions, respectively. In this case the volume of the A and
T regions as well as the radii RA and RT will vary during the course of a simulation. The advantage of fixing
the number of solvent molecules in the T region is that the number of QM/MM partitionings that needs to be
considered also remains constant (NT + 1). This is useful to optimize load balancing in a parallel adQM/MM
implementation. The downside is that expression 9.23 does not strictly hold any more since the coefficients σa
depend on the λ values which in turn depend on RA and RT . However, in practice, this is usually not an issue since
the conserved quantity Ẽ adQM/MM needs monitoring only during simulation setup to choose settings that afford
sufficient numerical stability. One thus can test simulation settings with fixed radii RA and RT and then switch to
fixed molecule numbers NA and NT for production runs.

9.3.2. Running adQM/MM simulations with sander
Performing simulations with the adQM/MM approach described above requires the MPI parallelized sander
executable sander.MPI. The implementation features a dual layer parallelization in which the calculations for all
individual QM/MM partitionings are performed in parallel. Each of these QM/MM calculations can in turn be run
in parallel. The parallelization across QM/MM partitionings is based on the multisander code infrastructure which
effectively runs independent copies of sander for each QM/MM partitioning (similar to the replica exchange, path
integral and thermodynamic integration implementations).

155

9. QM/MM calculations
In order to run an adQM/MM simulation, the mdin input file needs to be set up similar to a regular QM/MM
simulation. The QM region as defined in the &qmmm namelist defines the atoms that are in the permanent QM
region. In addition, the &qmmm namelist variable vsolv needs to be set to 2 or 3 for fixed number of molecules
in the A and T region or fixed size of A and T region, respectively. The following shows the minimum additions
to the mdin input file that are required to perform an adQM/MM simulation as compared to a traditional QM/MM
simulation with fixed QM and MM regions:

# mdin file - minimum additional content for adaptive solvent QM/MM
&qmmm
...
adjust_q = 0, ! required, charge cannot be redistributed
vsolv = 2,
! switch on adQM/MM with fixed molecule numbers
!
in A and T region
/
&vsolv
nearest_qm_solvent = 6, ! number of solvent molecules in A region
/
&adqmmm
n_partition = 4, ! number of QM/MM partitionings
! = number of molecules in T region + 1
/

In this example, a fixed number of solvent molecules is contained in the A region (6) and in the T region (3, since
the number of QM/MM partitionings is NT + 1). Thus, the volume of the A and T regions changes during the
simulation. Details of all namelist variables are collected below.
In addition, a groupfile for multisander is required. This groupfile should point all sander copies to the same
mdin input file, inpcrd coordinate file and prmtop parameter and topology file:

# groupfile for adaptive solvent QM/MM run with n_partition = 4
-O -i mdin -c inpcrd -p prmtop
-O -i mdin -c inpcrd -p prmtop
-O -i mdin -c inpcrd -p prmtop
-O -i mdin -c inpcrd -p prmtop

If you explicitly specify output file names, make sure to give separate names to each group (for example
mdout.000, mdout.001 etc), see also the multisander documentation. The mutisander adQM/MM simulation
can then be executed with

mpirun -np 4 sander.MPI -rem 0 -ng 4 -groupfile groupfile

In this example, 4 MPI processes will be launched for 4 process groups (sander copies). The individual QM/MM
calculations for each partitioning would thus run in serial. To run the individual QM/MM calculations in parallel,
the number of MPI processes must be a multiple of the number of process groups.
Adaptive solvent QM/MM simulations can be performed both with the semiempirical NDDO-type and DFTB
methods that are native to sander or with QM methods that are available via the interface to external QM programs. In the latter case, each process group will launch only one instance of the external QM program and
the parallelization of the QM part of the QM/MM calculations is determined by the settings for the external QM
program.
9.3.2.1. Important notes for system preparation and adQM/MM simulations

At the time of writing (release of AMBER 14) there is only a limited body of experience with adQM/MM
simulations documented in the literature. Running adQM/MM simulations requires careful simulation setup, in
particular regarding the size of the A and T regions. The A region needs to be sufficiently large to correctly describe
the physics of the system of interest. The T region on the other hand needs to be sufficiently large to minimize force
interpolation errors between the QM and MM regions. Since the cost of an adQM/MM simulation scales linearly
with the number of molecules in the T region, a tradeoff between accuracy and cost often needs to be made. This

156

9.3. Adaptive solvent QM/MM simulations
in turn might lead to simulations that behave nicely for many time steps but eventually experience sudden, large
(unphysical) forces on atoms at the T region boundaries. Similarly, whether it is more appropriate to define the
center of the A region via an atom or the center of mass of the permanent QM region will affect the numerical
stability of a simulation, depending on the particular system. Likewise for determining the distances of the solvent
molecules via an atom or the center of mass of the solvent. In the case of water as solvent, problems can arise due
to autoprotolysis which can lead to the formation of hydroxide and hydronium ions in the A region. Since the MM
force field is not parameterized for hydroxide or hydronium ions, these will experience strong (unphysical) forces
upon entering the T region. Careful monitoring of adQM/MM simulations and a bit of patience is thus advisable.
It is a good idea to monitor the size of the A and T region and to check coordinates of atoms in the QM regions of
all partitionings.
9.3.2.2. Namelist parameters for adaptive solvent QM/MM simulations

Adaptive solvent QM/MM simulations require setting the vsolv variable in the &qmmm namelist and setting
variables in the &vsolv and &adqmmm namelists.
The &vsolv namelist contains parameters that describe which solvent molecules
are contained in the A region in addition to the permanent QM region that is defined in the &qmmm namelist. This
namelist can be used without the &adqmmm namelist in a regular QM/MM simulation with sander if the variable
vsolv in the &qmmm namelist is set to 1 instead of 2 or 3 (see 9.1.6). In this case there is no transition region and
solvent molecules entering / leaving the QM region during the simulation would switch abruptly between QM and
MM description. This is not recommended since it will results in large unphysical forces whenever such a switch
occurs. However, this option is useful for post-processing of trajectories with single point QM/MM calculations in
which the solvent molecules closest to the permanent QM region are treated quantum mechanically.
&vsolv namelist parameters

nearest_qm_solvent_resname Residue name of the solvent that can exchange between QM and MM region (Default: nearest_qm_solvent_resname = ’WAT’)
nearest_qm_solvent Number of solvent molecules in the A region (Default: nearest_qm_solvent = 0)
nearest_qm_solvent_fq Frequency of updating of the A region. Should be set to 1 (every MD step) for adQM/MM
simulations. (Default: nearest_qm_solvent_fq = 1)
nearest_qm_solvent_center_id Determines the atom(s) of the solvent molecules that is used to calculate the distance to the QM region.
= 0 Use the atom that is closest to the QM region. (default)
= -1 Use the center of mass.
> 0 Use this atom number within the solvent residue.

qm_center_atom_id Determines the atom of the permanent QM region that is used to calculate the distance to the
solvent molecules.
= 0 Use the atom of the permanent QM region that is closest to a solvent molecule. Not supported for

adQM/MM since the radii of the A and T region would remain undefined - a common point of
reference is required for all solvent molecules. Useful only for post-processing of trajectories.
(default)
= -1 Use the center of mass of the permanent QM region.
> 0 Use this atom number. Note that this is an absolute atom number - obviously, you should choose

an atom that is in the permanent QM region.
verbosity

Controls verbosity of vsolv output in the mdout file.
= 0 Standard verbosity. (default)
> 1 Increase verbosity.

157

9. QM/MM calculations
If the &qmmm namelist variable vsolv is set to 2 or 3, an adQM/MM simulation with a fixed number of molecules in the A and T regions or fixed size of the A and T regions, respectively, is
requested. Details of the adQM/MM simulation are set in the &adqmmm namelist as follows.
&adqmmm namelist parameters

n_partition Defines the number of QM/MM partitions to be used. For vsolv=2 this also determines the number of
solvent molecules in the transition region, which is n_partition - 1. For vsolv=3 it has to be set to the
largest number of QM/MM partitionings that will be encountered for the chosen values of RA and RT.
(Default: n_partition = 1)
RA

Defines the radius RA of the A region in Angstrom. Only relevant for vsolv=3. Needs to be changed
from the default value and requires setting of RT. (Default: RA = -1.0)

Defines the radius RT of the T region in Angstrom. Only relevant for vsolv=3. Needs to be changed
from the default value and requires setting of RA. (Default: RT = -1.0)

calc_wbk

Controls whether the book-keeping term W is calculated.
= 0 Do not calculate W . (default)
= 1 Calculate W via one-sided difference approximation (not recommended).
= 2 Calculate W via central-difference approximation, see equation 9.23. Requires additional compu-

tations for (dis)appearing partitionings. (recommended if W is desired).
verbosity

Controls verbosity of adQM/MM output in the mdout file.
= 0 Standard verbosity. (default)
= 1 Increase verbosity - write distances of residues in T region from center of A region to file
adqmmm_res_distances.dat and σa values to file adqmmm_weights.dat. These files get over-

written at each program start.
= 2 Increase verbosity - write distances and σa values also to the mdout file. Also write λ values.

print_qm_coords Controls whether coordinates of the QM atoms in each partitioning are written to file.
= 0 Do not write coordinates. (default)
= 1 Write QM coordinates for all QM/MM paritionings in xyz format to files QM_coords.001 etc.

Files are overwritten upon each program call.

9.4. Adaptive buffered force-mixing QM/MM
9.4.1. Introduction
In hybrid quantum mechanical – molecular mechanical (QM/MM) methods the reactive part of the system (i.e.
where a significant change of the charge density distribution is expected) is described using a quantum mechanical
model while the rest of the system is treated using molecular mechanics. Conventional (“energy-mixing”) QM/MM
methods (convQM/MM) define a unique total energy function for the whole system that consists of three terms:
the energy of the QM model applied to the atoms in the QM region, the energy MM model applied to atoms in the
MM region and the interaction energy between the two regions:
E QM/MM (QM+MM) = E QM (QM) + E MM (MM) + E QM↔MM (QM+MM),

(9.24)

where the superscript represents the level of theory, while the region to which they are applied are indicated in
parentheses. The coupling between the quantum region and the surrounding atoms (E QM↔MM (QM+MM)) can
be taken into account in several ways. For example, in the more sophisticated approaches, the effects of the
MM charges are included in the quantum mechanical SCF calculation in the form of an externally applied field.
Given a total energy, performing Hamiltonian or any other standard dynamics is straightforward. However, several
uncontrolled errors could potentially be introduced by such schemes. Representing the environment by a set

158

9.4. Adaptive buffered force-mixing QM/MM
of point charges can over-polarise the QM region, and conversely the electrostatic effect of the ever-changing
quantum mechanical charge density on atoms at the edge of the MM region is quite different from what is assumed
when the MM force field parameters are determined. The delicate balance that exists between the various nonbonded MM terms is therefore no longer maintained across the QM-MM boundary. Furthermore, if adaptivity, i.e.
transitions of atoms between the two regions, is allowed, a new problem appears: in general there can be chemical
potential differences between the QM and MM regions for various species, and this results in a net flow betwen the
regions, leading to unphysical density differences, structure and dynamics. Allowing adaptivity in this sense can
be important when the active region itself is mobile (e.g. penetration, adhesion, crack propagation), or diffusional
processes in the environment are relevant (e.g. water molecules, ions, residues enter and exit the QM region during
the dynamics). To overcome these problems the adaptive buffered “force-mixing” QM/MM (abfQM/MM) method
was introduced [267, 268]. The implementation of abfQM/MM was carried out by Letif Mones (University of
Cambridge) and Gabor Csanyi (University of Cambridge) with help from many others (see the article below).
When using this implementation in your work please cite the following papers:
• Noam Bernstein, Csilla Várnai, Iván Solt, Steven A. Winfield, Mike C. Payne, István Simon, Mónika Fuxreiter and Gábor Csányi, QM/MM simulation of liquid water with an adaptive quantum region, Phys. Chem.
Chem. Phys., 14, 646–656 (2012), DOI: 10.1039/c1cp22600b
• Csilla Várnai, Noam Bernstein, Letif Mones and Gábor Csányi, Tests of an Adaptive QM/MM Calculation
on Free Energy Profiles of Chemical Reactions in Solution, |J. Phys. Chem. B, 117, 12202−12211 (2013),
DOI: 10.1021/jp405974b
• Letif Mones, Andrew Jones, Andreas W. Götz, Teodoro Laino, Ross C. Walker, Ben Leimkuhler, Gábor
Csányi and Noam Bernstein, Implementation of the Adaptive Buffered Force QM/MM method into CP2K
and Amber program packages, in preparation.

9.4.2. Technical details of abfQM/MM
In the abfQM/MM method two independent force calculations are performed at each MD step. The first and
more time consuming calculation is an extended conventional QM/MM calculation, which is used for calculating
the forces of atoms treated quantum mechanically during the dynamics. We start with a core QM region, which
comprises atoms that will always be treated quantum mechanically throughout the simulation. This region is
enlarged (using a distance criterion, see below) to obtain the dynamical QM region which contains the atoms that
follow QM forces. The dynamical QM region is surrounded by a buffer region whose size can be determined by
simple force convergence tests [269, 270] and its construction in practice is based on geometrical considerations:
atoms or molecules that are within a specified distance from the dynamical QM region are added to the buffer
region. From this first calculation only the forces of the atoms in the dynamical QM region are kept and the rest
(namely the forces on atoms in the buffer region) are discarded. The second calculation is used for obtaining good
forces on MM atoms, especially important near the QM/MM boundary. For this, either fully MM representation
of the whole system is used or, alternatively, another QM/MM force calculation, but this time using a smaller
(reduced) QM region consisting of only the atoms in the core QM region. The abfQM/MM method is an abrupt
force mixing method, which means that the forces are not derived from a total energy expression but a simple
combination of the forces of the two calculations described above

FabfQM/MM
(QM+MM) =
i

FExtended
i
FReduced
i

if i is in the dynamical QM region,
otherwise,

(9.25)

where the superscripts Extended and Reduced denote that the forces are taken from the first and second calculations
described above, respectively. The selection of the QM and buffer atoms is controlled by distance criteria. Using a
single distance criterion measured from some key atoms in the QM region, however, would lead to rapid fluctuation
in the region definitions because atoms may cross and re-cross repeatedly. To reduce this effect, a hysteretic
algorithm can be applied using an inner (rin ) and an outer (rout ) radius [267]. Thus, an MM atom is redesignated
to be QM if its distance measured from the QM region (as defined by a set of atoms always treated quantum

159

9. QM/MM calculations
mechanically) is less than rin and a QM atom is redesignated to be MM if this distance is larger than rout . Similar
hysteretic algorithms are applied for the definition of the dynamical QM region as well as the buffer region.
The above definitions may lead to QM atoms that have covalent chemical bonds with MM atoms. This is not
necessarily a problem, as these bonded interactions can be treated in several ways from the point of view of carrying
out the the QM/MM calculation (e.g. link atoms, special pseudopotentials, frozen localized orbitals etc.). However,
none of these schemes are general, i.e. cutting some type of QM-MM bonds in this way might not yield reasonable
forces. For example, highly polarized bonds, bonds with bond order larger than 1 and delocalized bonds such as
those in aromatic rings should be protected from being cut. In the conventional, nonadaptive QM/MM scheme it is
easy to handle this problem, because the QM region is specified at the beginning of the simulation and the user can
pick a chemically sensible set of atoms. For our dynamically varying QM (and buffer) regions, chemically sensible
decisions have to be made algorithmically. Our implementation allows the user to specify a list of the breakable
types of bonds which the software then uses to build the regions automatically.
Finally, as in all force mixing schemes, the abfQM/MM scheme uses dynamical forces that are not conservative,
that is they are not the derivatives of a total energy function. This is the price we pay for adaptivity. The nonconservative nature of the dynamics necessitates the use of a thermostat to maintain the correct kinetic temperature
throughout the system. The strength of the thermostat we need to use in practice is similar to those that are conventionally used in biomolecular simulations, which suggests that no ill effects will arise purely from the use of a
thermostat – the only caveat is that since the use of a thermostat is mandatory, strictly microcanonical simulations
cannot be performed. A simple Langevin thermostat is not appropriate in the presence of net heat generation (and
would lead to a steady state temperature deviation of several tens of degrees near the QM/MM boundary), so a
special adaptive thermostat (a combination of Langevin and Nose-Hoover thermostats) that is able to maintain the
correct temperature even in the presence of intrinsic heating or cooling is used [271].

9.4.3. Relation to other adaptive QM/MM methods
It is worth noting that the current implementation of abfQM/MM supports the use of several other adaptive
QM/MM methods. For example, setting r_qm_in, r_qm_out, r_buffer_in and r_buffer_out variables to 0 (for
definitions see the next section) leads to the adaptive conventional QM/MM (adconvQM/MM) technique that can be
considered also as the zero limit of the adaptive solvent QM/MM (adQM/MM) method [265] without a transition
region (see also section 9.3). In this case the extended and reduced systems are identical and the dynamics is
propagated by forces of a convQM/MM calculation whose QM region is adaptive. To save computational time for
adconvQM/MM the program first performs the corresponding convQM/MM calculation and then a dummy full
MM calculation whose forces are discarded. Another limit can be obtained when r_buffer_in and r_buffer_out
variables are set to 0 (and all other radii are not). This method can be called unbuffered force mixing QM/MM
(unbuffQM/MM). It has been observed that the applicability of both adconvQM/MM and unbuffQM/MM depends
on several factors (system, QM method, size of core / qm regions etc.) and it is advised to perform a force
convergence test [269, 270] before using them.

9.4.4. Technical glossary
9.4.4.1. Systems

• extended system: the first (QM/MM) calculation, which is used for calculating the forces on atoms in the
dynamical QM region. To get converged forces on these atoms, a buffer region is added, leading to an
extended QM region.
• reduced system: the second calculation, which is used for obtaining the MM forces. Either a full MM
representation can be used or a QM region that is smaller than the dynamical QM region.
9.4.4.2. Atom types

There are basically four regions in the abfQM/MM method depending on their role during the dynamics: the
core, the qm, the buffer and the mm regions. These sets are disjoint by definition. There are atoms which are
permanent members of a given region and there are others that can change their identity by moving from one

160

9.4. Adaptive buffered force-mixing QM/MM
region to another. This section describes the different atom types and also gives their name and id used in the
implementation. Please note the distinction between the labels “QM” and “qm” atoms: the former indicates the
QM region used in the actual extended or reduced QM/MM calculations, while the latter is a label used to describe
those atoms that, together with the core atoms, follow dynamics using quantum mechanical forces.
• core atoms (id = 1-2): those atoms that constitute the QM region for the reduced system calculation. (The
QM atoms in the extended calculation are the core, the qm and buffer atoms together.)
• user specified core atoms (id = 1, tag = CORE_USER): core atoms specified by the user. These atoms are
permanent core atoms during the whole simulation.
– core extension atoms (id = 2, tag = CORE_EXT): core atoms selected by geometrical criteria around
the user specified core atoms. These atoms belong temporarily to the core region.
atomi ∈ {core extension atoms} ⇐⇒ atomi = f (rcore_in , rcore_out , {user specified core})
{core atoms} = {user specified core atoms} {core extension atoms}
S

• qm atoms (id = 3-4): atoms, whose QM forces are used in the MD simulation similarly to core atoms but
qm atoms are excluded from the QM region in the reduced calculation. Their forces are calculated in the
extended QM/MM calculation.
– user specified qm atoms (id = 3, tag = QM_USER): qm atoms specified by the user. These atoms are
qm atoms during the whole simulation or occasionally can become core extension atoms.
– qm extension atoms (id = 4, tag = QM_EXT): qm atoms selected by geometrical criteria around the
core and user specified qm atoms. These atoms belong temporarily to the qm region.
atomi ∈ {qm extension atoms} ⇐⇒ atomi = f (rqm_in , rqm_out , {user specified qm} {core atoms})
S

{qm atoms} = {user specified qm atoms} {qm extension atoms}
S

• buffer atoms (id = 5-6): these atoms are in the buffer region. Although they are treated as QM atoms in the
extended calculation, forces on them from this calculation are discarded and they move with forces coming
from the reduced calculation in which they are treated with MM.
– user specified buffer atoms (id = 5, tag = BUFFER_USER): buffer atoms specified by the user. These
atoms are permanent buffer atoms during the whole simulation or occasionally can become qm or even
core extension atoms.
– buffer extension atoms (id = 6, tag = BUFFER_EXT): buffer atoms selected by geometrical criteria
around the qm and core atoms. These atoms belong temporarily to the buffer region.
atomi ∈ {buffer extension atoms} ⇐⇒ atomi = f (rbuffer_in , rbuffer_out , {qm atoms} {core atoms})
S

{buffer atoms} = {user specified buffer atoms} {buffer extension atoms}
S

• mm atoms (id = 7, tag = MM): they are MM atoms in both the extended and reduced calculations. For the
MD the forces are obtained from the reduced calculation.
• QM atom selections in the reduced and extended QM/MM calculations:
{QM atoms in the reduced system} = {core atoms} S
S
{QM atoms in the extended system} = {core atoms} {qm atoms} {buffer atoms}

161

9. QM/MM calculations

9.4.5. Namelist parameters for adaptive buffer-forced QM/MM simulations
The abfQM/MM implementation requires only two calculations for each MD step, which are performed sequentially (first the computationally more expensive extended then the reduced calculations are carried out). Consequently, unlike adaptive solvent QM/MM (adQM/MM, section 9.3) the subroutines of abfQM/MM are called
directly from sander and no groupfile is needed. All abfQM/MM related variables should be specified in the
&qmmm namelist. An example of an abfQM/MM dynamics is shown below:

# mdin file - example for adaptive buffered-force QM/MM dynamics
&cntrl
...
ntt=6,
! adaptive Langevin thermostat is used
...
ifqnt=1,
/
&qmmm
...
abfqmmm=1,
! activate abf QM/MM
r_core_in=3.0,
! inner radius for extended core region
r_core_out=3.5,
! outer radius for extended core region
r_qm_in=3.0,
! inner radius for extended qm region
r_qm_out=3.5,
! outer radius for extended qm region
r_buffer_in=4.0,
! inner radius for buffer region
r_buffer_out=4.5,
! outer radius for buffer region
coremask=’:1’,
! core region mask
qmmask=’:112, 1129, 1824, 2395’,
! qm region mask
buffermask=’’,
! buffer region mask
corecharge=0,
! core region charge
qmcharge=0,
! qm region charge
buffercharge=0,
! buffer region charge
/

9.4.5.1. Basic namelist parameters

abfqmmm 1 activates the adaptive buffered force-mixing method, default is 0 (no abf-QM/MM method is applied).
coremask

core atom list specification (in ambmask format). Optional, by default (when it is missing or coremask=’ ’) it is an empty set and in this case the reduced system is the full MM representation. Note
that at least one of the coremask or qmmask sets has to be specified.

qmmask

qm atom list specification (in ambmask format). Optional, by default (when it is missing or qmmask=’
’) it is an empty set and in this case only atoms in the core region will be treated as QM atoms during
the dynamics. Note that at least one of the coremask or qmmask sets has to be specified.

buffermask buffer atom list specification (in ambmask format). Optional, by default (when it is missing or buffermask=’ ’) it is an empty set.
corecharge Total charge of core atom list defined in coremask, default is 0.
qmcharge

Total charge of qm atom list defined in qmmask, default is 0.

buffercharge Total charge of buffer atom list defined in buffermask, default is 0.
r_core_in

162

Inner radius for determining core extension region around user specified core atoms. Default is 0.

9.4. Adaptive buffered force-mixing QM/MM
r_core_out Outer radius for determining core extension region around the user specified core atoms. Default is
the value specified for r_core_in. If r_core_out < r_core_in then r_core_out = r_core_in.
r_qm_in

Inner radius for determining qm extension region around the core and user specified qm atoms. Default
is 0.

r_core_out Outer radius for determining qm extension region around the core and user specified qm atoms. Default is the value specified for r_qm_in. If r_qm_out < r_qm_in then r_qm_out = r_qm_in.
r_buffer_in Inner radius for determining buffer extension region around the qm and core atoms. Default is 0.
r_core_out Outer radius for determining buffer extension region around the qm and core atoms. Default is the
value specified for r_buffer_in. If r_buffer_out < r_buffer_in then r_buffer_out = r_buffer_in.
9.4.5.2. Adaptive thermostats’ namelist parameters

ntt

Besides the original thermostats in sander, new adaptive ones are also introduced to be able to absorb
the heat production due to the nonconservative force-mixing dynamics. The corresponding thermostat
can be activated using the ntt command. In general, 5 activates the Nose–Hoover (chain)–Langevin,
6 the adaptive Langevin, 7 the adaptive Nose-Hoover chain and 8 the adaptive Nose-Hoover (chain)–
Langevin thermostat. For adaptive QM/MM ntt=6 or 8 should be used.

gamma_ln Collision frequency in ps−1
nchain

Number of thermostats in each Nose–Hoover chain of thermostats (default is 1)

9.4.5.3. Miscellaneous namelist parameters

selection_type Type of selection of the different regions. Default is the atom–atom distance based selection
(selection_type = 1). In this case a given atom is going to belong to an outer region if the distance
between the atom in question and any atom in the inner region is less or equal than the corresponding
criterion. Option 2 is the flexible sphere selection: for each inner region the radius of the region is
calculated (as the largest distance between the centre of mass of the region and any atom belonging to
that region), and the distance between the edge of the inner region and the atom in question will determine weather the atom belongs to the outer region or not. Option 3 is fixed sphere based selection:
it is the same as option 2 except that only the edge of the innermost region is calculated based on its
atoms and then all the other region’s borders are calculated geometrically as concentric spheres. For
option 2 and 3 the radii of spheres are calculated using the centre region, which is either defined by
the user (centermask) or it is the coremask if specified, otherwise it is qmmask. Note that option 2
and 3 selects significantly larger number of atoms than option 1.
initial_selection_type Type of initial selection type. This command controls the initial selection if not an
abfqmmm restart is performed (i.e. read_idrst_file is not specified). Default is 0, which is the middle
sphere selection (i.e. the mean of the corresponding inner and outer radii). Option -1 uses the inner
and option 1 applies the outer radius for the first selection.
center_type Type of calculation of center for selection_type = 2 and 3. Default is center of mass (option 1), while
option 2 is geometric center.
gamma_ln_qm Collision frequency of actual core and qm atoms in ps−1 when adaptive massive Langevin thermostat is applied. Default value is the same as gamma_ln defined in &cntrl session.
mom_cons_type Type of force correction for momentum conservation. Default is 1 when the extra force is distributed among the corresponding atoms as equal accelerations. Option 2 applies equal forces on each
atom. Options -1/-2 apply an acceleration/force proportional to the absolute value of the current acceleration/force of each atom. The region of atoms where the force correction is distributed is specified
by mom_cons_region. Option 0 does not apply momentum conservation.

163

9. QM/MM calculations
mom_cons_region Specifies the region where the force correction for momentum conservation is distributed.
Default is 1 that distributes the correction among only current core+qm atoms, option 2 distributes
it among current core+qm+buffer atoms and option 3 distributes the forces on all atoms. When
mom_cons_region = 0 the distribution is applied only among core atoms.
fix_atom_list > 0 activates the fixed atom list method, default is 0. In fixed atom list mode the different regions
are extended only by those solvent molecules that satisfy the given geometrical criteria and no solute
atoms will be selected besides the user specified ones in the coremask, qmmask and buffermask.
Useful when only solvent exchange is expected.
solvent_atom_number Number of atoms in solvent molecule for fixed atom list mode (fix_atom_list > 0), default
is 3. Defining this variable is important when the solvent is other than water and the solvent molecule
contains more (or less) than 3 atoms.
centermask Centre region atom list specification. Optional, if not defined then it is equal to coremask. If coremask is neither specified then centermask equals to qmmask.
oxidation_number_list_file File name of oxidation numbers. Each line in the file must be either a comment (starting by ’!’ or ’#’) or a triplet: RES ATOM OXID, where RES can be ’all’ (specification for all residues),
’atom’ (specification for a given atom), residue name or residue index. If RES 6= ’atom’ then ATOM is
the atom type name that can be specified completely (e.g. HE2) or partially using ’*’ (e.g. H* or HE*).
If RES = ’atom’ then ATOM is the atom index in the topology. OXID is the integer oxidation number.
Since different specifications can refer to the same atom, there is a hierarchy of the assignment and
the later step always overwrites the previous one: 1. RES = ’all’ with partial atom type specification
(in the order of X* → XY* → XYZ* ), 2. RES = ’all’ with complete atom type specification (XYZ1),
3. specified residue type with partial atom type specification, 4. specified residue type with complete
atom type specification, 5. residue index with partial atom type specification, 6. residue index with
complete atom type specification, 7. atom index specification.
ext_coremask_subset Possible core extension atom set. If specified only those atoms will be chosen according to
the corresponding geometrical criteria that can be also found in this list (in the case of fixed atom list
method solvent residues having at least one atom in the set will be chosen). If not defined then by
default it is the all atom list.
ext_qmmask_subset Possible qm extension atom set. If specified only those atoms will be chosen according to
the corresponding geometrical criteria that can be also found in this list (in the case of fixed atom list
method solvent residues having at least one atom in the set will be chosen). If not defined then by
default it is the all atom list.
ext_buffermask_subset Possible buffer extension atom set. If specified only those atoms will be chosen according
to the corresponding geometrical criteria that can be also found in this list(in the case of fixed atom
list method solvent residues having at least one atom in the set will be chosen). If not defined then by
default it is the all atom list.
cut_bond_list_file File name of breakable bonds for intelligent termination of different regions (core/qm/buffer).
Each line in the file must be either a comment (starting by ’!’ or ’#’) or a triplet: ATOM1 ARROW
ATOM2. ATOM1 and ATOM2 are both either atom types or atom indexes. ARROW specifies the
direction of bond breaking: if it is ’<=>’ then the bond can be split from both directions, if it is ’=>’
or ’<=’ then the bond can be cut only from ATOM1 or ATOM2 directions, respectively.
max_bonds_per_atom Maximum number of ligands around any atom in the system. This controls the size of
arrays for the intelligent termination. Default is 4 that is good for most biological systems. If there are
atoms having more than 4 ligands then adjustment is required.
n_max_recursive Intelligent termination scheme is a recursive subroutine to get a fast and reliable performance.
However, it may happen that according to the user specified breakable bonds a very large bond network

164

9.5. SEBOMD: SemiEmpirical Born-Oppenheimer Molecular Dynamics
will be chosen for a given region. To avoid it this variable can be used to control the maximum
number of iterations: when the number of iteration reaches the value of n_max_recursive the program
terminates. Default value is 10000.
min_heavy_mass To keep low the number of atoms in each extension region, by default the geometrical region selection algorithm measures the distances between only heavy atoms, and hydrogen atoms are assigned
in a second step according to the heavy atoms they are bonded to. To extend the distance based selection for H atoms as well, decrease the value from its default 4.0 below the atomic mass of hydrogen
(e.g. 0.0).
pdb_file

File name of a special abfQM/MM related pdb file generated during the dynamics. The first 8 columns
have the standard pdb format (’ATOM’, atom index, atom name, residue name, residue index, Cartesian coordinates of atom), 9th column is the oxidation number, 10th and 11th columns are the id
number and tag according to abfQM/MM implementation, respectively, and the possible following
columns include the atom indexes of MM atoms having direct bond to the given atom treated as QM
atom in the extended calculation. Default name is abfqmmm.pdb.

ntwpdb

Frequency of printing out abfQM/MM information into pdb_file. Default value is 0 (no printing).
Using ntwpdb < 0 allows the user to perform a selection test. In this case neither dynamical nor even
point calculations are performed, the program terminates after printing the pdb file out.

read_idrst_file Name of abfQM/MM atom id restart file used for restarting simulations. In the beginning of the
simulation besides the user specified atoms those become also member of a given region that are within
the outer radius. For a given region if the outer radius differs from the inner one, in the beginning of
the dynamics the number of atoms will change until it reaches a dynamical equilibrium fluctuation. To
avoid this natural transient period in a consecutive restart calculation one can use the read_idrst_file
generated in the previous run telling the program the abfQM/MM atom id’s of the restart configuration.
Note that the safe use of read_idrst_file requires the same region specifications as in the previous run.
write_idrst_file Name of abfQM/MM atom id restart file generated during the run. Default name of the file is
abfqmmm.idrst.
ntwidrst

Frequency of printing the abfQM/MM atom id restart file out. Default is 0 (no printing).

hot_spot

1 activates the hot spot–like adaptive calculation [272] in which the forces of atoms in the buffer
region are linear combinations of the forces obtained from the extended and reduced calculations
using a smoothing function. Default is 0 (no hot spot–like calculation is performed).

9.5. SEBOMD: SemiEmpirical Born-Oppenheimer Molecular Dynamics
The sander program provides the ability to run SEBOMD (SemiEmpirical Born-Oppenheimer Molecular Dynamics) simulations. During a SEBOMD simulation, all atoms are considered as quantum atoms within the NDDO
semiempirical approach (e.g., AM1, PM3, etc). Therefore, unlike in QM/MM methods, there is no link atom, no
frontier bond, no interaction between any QM and MM atoms (since there is no MM atom). Another consequence
of SEBOMD simulations is the computational time requested to compute energy and forces at each step of a
molecular dynamics can be (very) important. To allow for the computation of “large” systems (i.e., up to a couple
of thousands of atoms), an optional linear scaling divide and conquer strategy is implemented[273, 274]. Periodic
boundary conditions with long-range electrostatic interactions (i.e., Ewald sum) can also be applied.
The SEBOMD code implemented in sander is originated from the DivCon program developed in the Merz group
while at Pennsylvania State University:
• Steve L. Dixon, Arjan van der Vaart, Valentin Gogonea, James J. Vincent, Edward N. Brothers, Lance M.
Westerhoff and Kenneth M. Merz, Jr. DivCon99, The Pennsylvania State University, 1999.
Major contributors to the SEBOMD interface are as follows:

165

9. QM/MM calculations
• Maintenance, code refactoring, debugging, testing by Gerald Monard
• Original roar interface by Gerald Monard and Arjan van der Vaart[275]
• Original sander port by Jennifer Thomas
• Ewald and Particle Mesh Ewald summation by Laurent Teixidor
• PIF and MAIS semiempirical correction implementation, peptidic corrections by Antoine Marion[276]

9.5.1. Functionalities and limitations
The current SEBOMD implementation allows to run sander simulations with the following functionalities:
• molecular dynamics or energy minimization (imin = 0, 1, or 5)
• gas phase or periodic boundary conditions (as defined in the topology file), no support for Generalized Born
solvent effect
• For PBC runs, different long range interactions handlers are possible: none, external Particle Mesh Ewald
using MM point charges as defined in the topology file, or direct Mulliken Ewald summation.
• temperature regulation as implemented in sander (ntt flag)
• pressure regulation: only barostat = 2 is supported (Monte Carlo barostat)
• parallel implemention (sander.MPI): only the Divide & Conquer approach can be used (method > 0)
• available hamiltonians: MNDO, AM1, AM1/d-PhoT, RM1, PM3, PM3/PDDG
• available corrections to PM3 hamiltonians: MAIS and PIF
• as d-orbitals are not yet implemented in the SEBOMD code, only the following elements are implemented:
H, C, N, O, P, S, F, Cl, Br, I (except for AM1/d-PhoT for which the P element is not yet available because it
requires d-orbital implementation)
• maximum number of atoms: 2600; maximum number of residues: 1000
Note: the SEBOMD code currently uses a static memory allocation as defined in $AMBERHOME/AmberTools/src/sebomd/divcon.dim. Users wishing to simulation bigger systems will have to modify the SEBOMD
source code and recompile.

9.5.2. Sample SEBOMD input
To run a SEBOMD calculation, a specific namelist (&sebomd) must be used. It contains all the necessary
information for the run. To inform sander that a SEBOMD simulation must be run, two steps are required: 1)
switch the ifqnt keyword to 1 (as for a QM/MM calculation); 2) define the qm_theory keyword in the &qmmm
namelist to ’SEBOMD’. Here is a sample mdin file for SEBOMD:
! example input for SEBOMD
&cntrl
...
ifqnt = 1,
!
/
&qmmm
qm_theory = ’SEBOMD’, !
/
&sebomd
hamiltonian = ’AM1’,
!
charge = 0,
!
/

166

simulation

switch on QM calculation

use specific SEBOMD routines

Use the AM1 semiempirical hamiltonian
total charge on the (full) system is 0

9.5. SEBOMD: SemiEmpirical Born-Oppenheimer Molecular Dynamics

9.5.3. &sebomd namelist variables
charge

= Integer Net charge of the system (Default = 0).
Note: SEBOMD only supports closed shell molecular systems.

method

Algorithm for the SCF computation.
= 0 (Default) Standard closed-shell algorithm: the Fock matrix is diagonalized at each SCF iteration.

(Note: all subsetting parameters are ignored, only one subsystem containing all the atoms will
be generated).
= 1 Use linear scaling divide & conquer SCF algorithm. Buffer regions must be specified (dbuff1 and

dbuff2). Subsystems are built on an atom-based principle.
= 2 Use linear scaling divide & conquer SCF algorithm. Buffer regions must be specified (dbuff1

and dbuff2). Subsystems are built on an residue-based principle (recommended option over
method=1).
ncore

= Integer When using divide and conquer method (method > 0): specify the number of residues used
to build the core. (default: ncore = 1)

dbuff1

= Float When using divide and conquer method (method > 0): specify the extent of the first buffer
region from the core in Å. (default: dbuff1 = 6.0)

dbuff2

= Float When using divide and conquer method (method > 0): specify the extent of the second buffer
region from the core in Å. (default: dbuff2 = 0.0)

hamiltonian Semiempirical hamiltonian to be used for energy and force calculations. All atoms within the molecular system will be treated at this level of theory. Available semiempirical hamiltonians:
MNDO Request the use of MNDO semiempirical hamiltonian[231]
AM1 Request the use of AM1 semiempirical hamiltonian[229]
PM3 Request the use of PM3 semiempirical hamiltonian (default)[228]
PM3PDDG Request the use of PM3/PDDG semiempirical hamiltonian[232]
RM1 Request the use of RM1 semiempirical hamiltonian[230]
AM1D Request the use of AM1/d-PhoT semiempirical hamiltonian[240]

(Note: phosphorous (P) element is not yet implemented, therefore the AM1D hamiltonian is
available only for H, C, N, O, S, F, Cl, Br and I elements)
modif

Modification/corrections to the semiempirical energy. Some semiempirical methods have been extended to improve results, mostly in the case of intermolecular interactions. For the moment only
PM3 corrections to the energy are available. Possible values are:
none (default) no correction
PIF2 PM3 hamiltonian is modified for intermolecular core-core interactions according to the work of

Bernal-Uruchurtu et al. and Harb et al. [235, 277–279]. This correction can be applied when using PM3 hamiltonian with a molecular system composed of one (or more) organic molecule(s) in
interaction with explicit water molecules. Intermolecular water-water core-core interactions are
computed using specific PM3-PIF parameters for aqueous solvent, while intermolecular organicorganic and organic-water intermolecular core-core interactions are computed using another specific set of PM3-PIF parameters. The intermolecular PM3-PIF (PIF2 version) parameters are
available only for the following interactions:

167

9. QM/MM calculations
Water
Organic
Hw Ow H C N O Cl
Hw
X
X
X X X X X
Ow
X
X
X X X X X
H
X
X
X X X X X
C
X
X
X ∅ X X ∅
N
X
X
X X X X ∅
O
X
X
X X X X X
Cl
X
X
X ∅ ∅ X ∅
(X: intermolecular interaction parameters between the two considered atom types are available;
∅: no intermolecular parameter available)
PIF3 PIF3 is an extension of the PIF2 parameters in which organic hydrogens are distinguished be-

tween “hydrophylic” hydrogens and “hydrophobic” hydrogens[276]. In the case of hydrophylic
hydrogens, intermolecular interactions between the hydrogen atom and water molecules are computed using PIF2 parameters. In the case of hydrophobic hydrogens, intermolecular interactions
between these hydrogen atoms and water molecules are computed using specific parameters. The
distinction between hydrophobic and hydrophylic hydrogens is performed using the atom types
as specified in the topology file. Hydrogen atom types which are considered as hydrophylic are:
H, HO, HS, HW, hn, ho, hp, hs, hw, Ho, hO, hN, and hR. Other hydrogen atom
types are considered as hydrophobic.
MAIS1 MAIS extension of the PM3 hamiltonian in which intramolecular and intermolecular core-

core functions are replaced by specific MAIS functions. This option corresponds to the initial
work of Bernal-Uruchurtu et al.[234]. Parameters are only available for liquid water (H and O
elements).
MAIS2 Second version of the MAIS extension. Parameters are only available for H, O, and Cl

elements[235].
solute

To be used in conjunction with nsol keyword when using PIF parameters (modif = ’PIF2’ or modif =
’PIF3’). This enables the control of intermolecular interactions.
= 0 (Default) no solute.
= 1 a solute composed of nsol residue is present.

nsol

Control the definition of solute-solvent interactions.
= 0 (Default) no solute.

6= 0 number of residues which constitutes the solute (e.g., nsol = 5 means the solute is composed of
the first 5 residues in the current topology file, all other “residues” are solvent molecules).
longrange

Select the type of long range interaction when using periodic boundary conditions:
= 0 (Default) No long range interaction. Only the minimum image convention.
= 1 Perform PME (Particle Mesh Ewald) summation using constant atomic charges extracted from

the topology file.
= 2 Perform an Ewald summation using Mulliken atomic charges extracted from the semiempirical

wavefunction. Long-range Ewald Mulliken charge effects are incorporated in the Fock matrix of
the system to polarize the wavefunction.
iprec

SCF convergence criteria:
= 4 (Default) SCF is converged when the relative SCF energy between two iterations ( ∆E
E ) is less than

104 times the machine precision. This ensures the conservation of the total energy during NVE
simulations.

5
= 5 Loose criteria: SCF energy is converged when ∆E
E < 10 x the machine precision.

168

9.5. SEBOMD: SemiEmpirical Born-Oppenheimer Molecular Dynamics
fullscf

Option to enable pseudo-diagonalization routines
= 0 enable pseudo-diagonalization routine when possible. This can speed-up SCF calculations. (de-

fault)
= 1 turn off pseudo-diagonalization. Full diagonalization of the Fock matrix is performed at each

iteration of the SCF cycle.
ipolyn

Option to activate polynomial interpolation of the guess density matrix
= 0 Use converged density matrix of the previous step as initial (guess) density matrix for the current

step. Recommended option for minimization.
= 1 Use polynomial interpolation of the density matrix elements from the last three steps as initial

(guess) density matrix for the current step. Recommended option for molecular dynamics runs.
(default)
screen

verbosity option for SEBOMD calculations
= 0 minimum output. (default)
= 1 output semiempirical energy details at each step
= 2 output semiempirical energy details + the composition of all subsystems when using method > 0.

lambda

= Float (default 1.0) Enable the computation of a mixed energy value between SEBOMD and full
MM computations. If lambda 6= 1.0, in addition to a semiempirical calculation, the energy of the full
system is evaluated at the MM level. Then energy and forces are mixed according to:
E pot = λ E(SEBOMD) + (1 − λ )E(MM)
Since, sometimes, semiempirical potential energy surfaces are (very) different from MM surface, the
use of the lambda keyword permits to equilibrate MD more easily. For example, from an equilibrated
MM system, it is possible to run several SEBOMD simulations using different lambda values from
0.0 (full MM energy) to 1.0 (full QM energy) to obtain an equilibrated SEBOMD simulation.

charge_out Filename used to save atomic charges. Default = ’sebomd.chg’
ntwc

Every ntwc steps, the (Mulliken) atomic charges will be written to the charge_out file. If ntwc = 0, no
atomic charge file will be written. Default = 0.
The format of the charge_out file is the following: every ntwc steps, the energy of the system is first
written, then one line per atom is written, containing the x, y, z coordinates and the Mulliken atomic
charge of the atom.

peptcorr

flag to apply force field corrections on peptidic bonds
Some semiempirical methods do not correctly describe peptidic bond properties, leading to a pyramidal peptide bond nitrogen. An empirical force field correction can be applied to force the planearity
of a peptide bond[280].
= 0 no peptidic correction. (default)
= 1 apply peptidic correction (see Ludwig et al. for details[280])

peptk

= Float The force constant of the peptidic correction (in kcal/mol).
AM1 default value: peptk = 5.9864
PM3 default value: peptk = 9.8526
MNDO default value: peptk = 6.1737

169

10. paramfit
Robin Betz
The paramfit program allows specific forcefield parameters to be optimized or created by fitting to quantum
energy data. Paramfit can be used when parameters are missing in the default forcefields and antechamber cannot
find a replacement, or when existing parameters do not describe the system to the desired level of accuracy, such
as for dihedral constants on protein backbones.
Paramfit attempts to make the following statement true: With the correct AMBER parameters, calculations
performed at a quantum level over many conformations of a structures should match those calculated by
AMBER.
Paramfit can calculate the energy of each conformation and/or the force on each atom, and adjust the force field
parameters so that these values correspond to input quantum data.
For energies, Paramfit attempts to fit the AMBER energy to the quantum energy for a variety of conformations
of the input structure, minimizing the equation
N

∑ wi

i
2
EMM (n) − E QM (n) + K = 0

n=1

where K is a constant that adjusts for different origins in the QM and MM calculations so that minimization may
be done to zero and N is the number of molecular conformations that are considered.
For forces, the equation that is optimized is
N

Natoms

∑ ∑

wi F(n, atom)MM − F(n, atom)QM

n=1 atom=1

where the sum of the differences in the forces on each atom should match given the correct set of parameters.
Individual structures can be assigned weights wi to give them more or less relative importance in the fit. By default,
all weights are set to 1.
The program works by altering the parameters that AMBER uses to describe the molecule, which alter the
elements in the AMBER sum that is used to calculate the energy or forces. It is necessary to evaluate over many
conformations of the molecule because the parameters should predict how the molecule will behave dynamically
rather than statically. To get a good idea of the forces on a dihedral, for example, the energy needs to be evaluated
for multiple conformations of the dihedral to see how it changes each time. Paramfit will fit so that the energy
changes that AMBER predicts will happen when the dihedral twists match the changes predicted with quantum
methods.
In order to facilitate force field development, Paramfit supports fitting parameters across multiple molecules (for
example, fitting a single dihedral backbone term across a variety of input amino acids). Single molecule fits can
also be done to generate parameters that are missing or inadequate to describe small molecules or ligands.
Paramfit provides functionality for the majority of steps in the fitting process, including writing input files for
quantum packages, specifying which parameters are to be fit, determining the value of K for the system, and
finally conducting the fit and saving it in a force field modification file that can be used by other programs. An
external quantum program is needed to generate the energies needed for paramfit to conduct a fitting. Currently,
the program is capable of writing input files for ADF, GAMESS, and Gaussian, although if you write your own
input files instead of using paramfit’s functionality, any quantum package will work.
Paramfit has OpenMP support for parallelization of the AMBER function evaluation over the input conformations, where each core will evaluate the energy for a subset of the conformations. Enable this by adding the
-openmp option to configure and rebuilding paramfit. By default all available cores will be used. To change this,

171

10. paramfit
set the OMP_NUM_THREADS environment variable to the number of threads to be executed. You will see a speedup
directly proportional to the number of cores you are running.
Paramfit now includes several ways fitting functions to aid in parameter generation. It can fit such that the energy
of each input structure matches the single-point quantum energies inputted, or can now do the same fitting only
with the forces on each atom, which may produce a more accurate fit that is less sensitive to problems with the input
structure, and can also fit all dihedral force constants and phases simultaneously to a small set of quantum energies
using a method developed by Chad Hopkins and Adrian Roitberg. This method fits every term and requires fewer
function evaluations than running the full minimization algorithm, but requires especially good sampling of each
torsion angle of interest.
Fitting forces requires several additional options to specify the location of the output forces files in the job
control file. The easiest way to create a job control file for any of these options is to use the wizard, which runs
automatically when no job control file is specified. This will walk you through the creation of a job control file and
write it for you while prompting for all necessary options for the selected fitting function.
It is highly recommended that you fit to single-point quantum energies, as fitting to forces is considerably more
expensive in terms of required calculation and still somewhat experimental. The implementation of the dihedral
fitting method is requires a varied set of input structures, and does not allow specifying individual dihedrals to
be fit. No matter which method is used, please take care to carefully validate all parameters for reasonableness–
paramfit’s fit is dependent on the variation and quality of the input structures and the resulting parameters are not
guaranteed in ill-defined areas of the input conformation set. For example, if you fit a dihedral torsion term with
input structures sampling the 0-30 degree range of that dihedral, the resulting parameters cannot be expected to
give a valid energy of a structure with the dihedral at 90 degrees, as the algorithm merely fits to the available data
and cannot make other predictions.

10.1. Usage
Paramfit is called from the command line as follows for a single molecule fit:
paramfit -i Job_Control.in -p prmtop -c mdcrd -q QM_data.dat \
-v MEDIUM --random-seed seed

Running paramfit without any options will run a wizard that assists in the creation of a job control file. It is highly
recommended that you use the wizard to assist you in setting run options.
The following switches apply to single molecule fits only:
-p prmtop The molecular topology file for the structure.
-c mdcrd A coordinate file containing many conformations of the input structure. These may be generated by

running a short simulation in solution, or by manually specifying coordinates for each atom. It is important
that there be a good representation of the solution space for any parameters that are to be optimized– for
example, if you want a bond force constant it would be a good idea to have input structures with a good
range of values for the length of the that bond type. See Subsection 10.2.6
-q QM_data.dat A file containing the quantum energies of the structures in the coordinate file, in order, one per

line. You will have to extract the energies from the output files that the quantum package produces. An
example script to do this for Gaussian formatted output files can be found in $AMBERHOME/AmberTools/src/paramfit/scripts.
To fit multiple molecules, the following switches are used:

paramfit -i Job_Control.in -pf prmtop_list -cf mdcrd_list -v MEDIUM --random-seed seed

Here is a very brief description of the command-line arguments for a multiple molecule fit. For more information
on conducting these, fits, please see 10.3.

172

10.2. The Job Control File
-pf prmtop_list A file containing a plain-text list of input topology files and the adjustment constant K for each file

separated by a space, one per line.
-cf mdcrd_list A file containing a plain text list of input coordinate files, number of structures to read from each

file, and directory containing quantum output from each file, separated by a space. These should be specified
in the same order as the topologies in the prmtop_list.
The following switches apply to either type of fit:
-i Job_Control.in The job control file for the program. See Section 10.2 for a description of the options and format

for this file. If no job control file is specified, a wizard will be initiated that will prompt you for options and
help create the file. Use of the wizard is highly recommended when running Paramfit for the first time.
-v MEDIUM The verbosity level to run the program at, either LOW, MEDIUM, or HIGH.
–random-seed seed The integer seed for the random number generator. Only specify this parameter when exactly

reproducible results are needed for debugging.

10.2. The Job Control File
Similarly to sander and other programs, paramfit requires a job control file that specifies individual options for
each run. The options that apply to your run vary depending on the runtype and the other settings, and they are
quite numerous. To aid you in creating a job control file, a wizard has been included that will prompt you about
applicable settings and create the job control file for you. Using the wizard is highly recommended, especially
when running a fit for the first time. To use the wizard, simply run paramfit without any options. It is highly
recommended that you use the wizard to create job control files, as it prompts for all options relevant to your
run and the resulting file can then be easily edited by hand.
The format consists of variable assignments, in the format variable=value, with one assignment per line. Pound
signs (#) will comment out lines. See the following sections for a description of what to put in the job control file
for various tasks:

10.2.1. General options
paramfit requires several options be set for every run. These variables should usually appear in your job control
file.
RUNTYPE Specifies whether this run will be creating quantum input files, setting parameters, or conducting a fit.
= CREATE_INPUT The structures in the coordinate file will be written out as individual input files for a

quantum package. See 10.2.2.
= SET_PARAMS Provides an interactive prompt allowing you to specify which parameters will be fit for

this molecule. See 10.2.3.
= FIT Conducts a fitting using one of the two minimization algorithms. See 10.2.4for other options that need

to be specified.
NSTRUCTURES Specifies how many structures are in the input coordinate file. If this value is less than the total

number of structures in the file, only the first n will be read. Only applies to single molecule fits! If you
are fitting multiple molecules at once, the number of structures for each molecule should be specified in the
mdcrd_list file as described in 10.3.

173

10. paramfit

10.2.2. Creating quantum input files
Given a trajectory, Paramfit can write input files for a variety of quantum packages. This is necessary to generate
the energy values for each input conformation that Paramfit will fit to. You do not necessarily need to do this step
and can write your own input files if desired. Currently Gaussian, ADF, and GAMESS formats are supported.
Job files will be named sequentially with filename prefix and suffix specified in the job control file. Once all
the input files are written, you must run the quantum package yourself. Paramfit can read Gaussian output files
directly, but for other packages you must extract the energies yourself into a file with one energy per line in the
same order as the input structures.
Currently Paramfit only supports Gaussian if you are fitting forces, and will read the output files and extract the
force information for you. See 10.4 for more information on fitting these.
To enter this mode, set RUNTYPE=CREATE_INPUT and specify the following options in your job control file:
QMHEADER File that will be prepended to all created input files for the quantum program. This specifies things

on a per-system basis, such as choice of basis set, amount of memory to use, etc. These parameters will
vary depending on which quantum package you are using. Sample header files for all supported quantum
packages are included in example_config_files in paramfit’s source directory.
QMFILEFORMAT Specifies which quantum package the created input files should be formatted for.
= ADF Use the Amsterdam Density Functional Theory package.
= GAMESS Use the General Atomic and Molecular Electronic Structure System (GAMESS).
= GAUSSIAN Use Gaussian.
QM_SYSTEM_CHARGE The integral charge of the system. Defaults to 0. Note that some quantum packages may

require this to also be specified in your header file.
QM_SYSTEM_MULTIPLICITY The integral multiplicity of the system. Defaults to 1 (singlet).
QMFILEOUTSTART The prefix for each of the created input files. Defaults to ’Job.’ The structure number and

then the suffix will be appended to this value.
QMFILEOUTEND The suffix for each of the created input files. Defaults to ’.in’. With both default options, the

file will be named Job.n.in.

10.2.3. Specifying parameters
In order to facilitate batch runs as well as simplify the process of running paramfit on larger systems, the
parameters to be fit are saved and then loaded in during actual fitting so that they do not have to be specified
every time. The parameter setting runtype accomplishes this by prompting whether you would like to fit bond,
angle and/or dihedral parameters and then displaying a list of the specific atom types for each so that you can pick
exactly what paramfit should optimize. This saved file does not specify a value for any of the parameters, but
simply indicates which ones are to be changed during fitting.
If you do not wish to save a parameter file, you may instead fit a default set of parameters or be prompted every
time. See Subsection 10.2.4.
To enter this mode, set RUNTYPE=SET_PARAMS and the following options:
PARAMETER_FILE_NAME Specifies the name of a file in which to store the parameters. When loading these

parameters in during a fitting, this line will stay the same. Do not modify this file by hand: paramfit numbers
each bond, angle, and dihedral in a manner that is consistent but not human-readable.

174

10.2. The Job Control File

10.2.4. Fitting options
The fitting function accomplishes the actual parameter modification. It does this by minimizing the least squares
difference between the quantum energy and the energy calculated with the AMBER equation over all of the input
conformations. For a perfect fit, this means that over all structures, EMD − E QM + K = 0.
K is the intrinsic difference between the quantum and the classical energies, which is represented as a parameter
that is also fit. The value of K depends on the system, and should be fit once as the only parameter before fitting
any other parameters.
To enter this mode, set RUNTYPE=FIT and set the following additional variables:
ALGORITHM The minimization algorithm to use. paramfit currently implements a genetic algorithm and a sim-

plex algorithm for conduction minimization. Each algorithm requires several parameters and is suited to
different problems. Please see 10.2.5 for descriptions of these options and a guide on choosing the appropriate algorithm.
= GENETIC
= SIMPLEX
= BOTH Runs the hybrid genetic algorithm followed by the simplex algorithm to fine tune results
= NONE No fit is performed- useful for calculating energy of each structure with the initial parameters to

see their quality
FUNC_TO_FIT The fitting function to use in the calculation.
= SUM_SQUARES_AMBER_STANDARD Standard fit to single-point energies. Recommended selection.
= AMBER_FORCES Fit to the forces on atoms involved in fitted parameters. Currently only supports Gaus-

sian output. See Section 10.4 for details.
= DIHEDRAL_LEAST_SQUARES Use Chad Hopkins and Adrian Roitberg’s method to fit all dihedral terms

at once. This method will fit all dihedral torsion terms simultaneously with a minimal number of
function evaluations, but requires very good sampling of the relevant torsion angles.
K The intrinsic difference between the quantum and classical energies. This value needs to be determined once

for each system so that the algorithm can minimize to zero instead of to a constant. See Subsection 10.5.2
for an example.
PARAMETERS_TO_FIT Sets how paramfit determines which parameters are to be fit. paramfit does not fit elec-

trostatics, but is capable of fitting every other element of the AMBER sum, which include bond harmonic
force constant and equilibrium length, angle harmonic force constant and equilibrium angle, and proper and
improper dihedral barrier height, phase shift, and periodicity. As a general rule, the fewer parameters there
are to fit, the faster and more accurate the results will be. Avoid fitting more parameters than necessary.
= DEFAULT Fit all bond force constants and lengths, angle force constants and sizes, and dihedral force

constants. This option will usually fit a very large number of parameters, and is rarely necessary. For
most cases, only a few parameters are desired, and they should be fit individually.
= K_ONLY Do not fit any force field parameters. Only fit the value of K (the difference between quantum

and classical energies for the system). This needs to be done once per system in order to determine K
before any other parameters are fit, as attempting to fit it at the same time results in inaccurate results.
Since small changes in K produce a great change in the overall least squares sum, the algorithm will
tend to focus on changing the value of K and will neglect the parameters.
= LOAD The list of parameters to be fit is contained in a file that was previously created with the parameter

setting runtype. Set PARAMETER_FILE_NAME to the location of this file. To create this file, run
paramfit with RUNTYPE=SET_PARAMS.
SCEE The value by which to scale 1-4 electrostatics for the AMBER sum. Defaults to 1.2
SCNB The value by which to scale 1-4 van der Waals for the AMBER sum. Defaults to 2.0.

175

10. paramfit
QM_ENERGY_UNITS The unit of energy in the quantum data file if you are fitting to energies. This will depend

on your quantum package and settings used for the single point calculations.
= HARTREE Default
= KCALMOL
= KJMOL
QM_FORCE_UNITS The unit of force in the quantum data files if you are fitting to forces. This will depend on

your quantum package and settings used for the force calculations.
= HARTREE_BOHR Default
= KCALMOL_ANGSTROM
WRITE_ENERGY Saves the final AMBER energy and the quantum data for each structure to the specified file.

Plotting these data is useful in verifying the results of the fitting and identifying any problem structures. See
Subsection 10.5.3 for more on how to verify the accuracy of results.
WRITE_FRCMOD When the fitting is complete, the parameters will be saved in a force field modification file at

this location in addition to displaying them in standard output. This file may be used with leap to create a
new prmtop. If no value is specified the file will not be created.
SCATTERPLOTS Creates graphs of the bond, angles, and dihedrals found in the input files for each parameter that

is being fit. These plots can be visualized using scripts/scatterplots.sh found in paramfit’s source directory.
This can be helpful in assessing the quality of the input conformations. No need to specify anything after
the = sign for this parameter.
SORT_MDCRDS
= YES Sorts the input structures in order of increasing energy before conducting the fit. This can aid in

identification of problem regions for the initial or fitted parameters, as they may be generally worse on
structures in certain energy ranges.
= NO Default
COORDINATE_FORMAT The format of the input coordinate set. Paramfit will return an error if the file is in an

unexpected format.
= TRAJECTORY Default
= RESTART

10.2.5. Algorithm options
Paramfit implements two minimization algorithms: a simplex and a hybrid simplex-genetic algorithm (GA).
The current version of paramfit incorporates numerous refinements to the genetic algorithm that require much less
input from the user– it is no longer necessary to choose between the simplex or GA. This improved algorithm
means that iterative fits are no longer necessary, and the algorithm will converge very close to or at the global
minimum on a single run.
The genetic algorithm starts with a randomly generated solution set, which it recombines and alters in ways
similar to evolution. The GA will start with many initial randomly generated sets of parameters. It will then
determine which are the best by evaluating the AMBER sum, select them for recombination to produce a new set
of parameters, randomly alter a few parameters slightly to prevent premature convergence, and iterate. Once several
“generations” have passed without improvement, a loosely converging simplex algorithm is run on a random subset
of the population, which is then allowed to recover for several generations before further simplex iterations are
conducted. This hybrid approach dramatically speeds convergence to the global minimum, while maintaining the
strengths of the genetic algorithm in searching a large, complex solution space with low sampling.
The following options in the job control file will control the behavior of the genetic algorithm. In general the
default values for these options is sufficient to produce good results, and alterations to them will speed convergence.

176

10.2. The Job Control File
Options marked internal algorithm parameter should not need to be altered by the vast majority of users, as they
are already set to their optimum. The algorithm’s results should be independent of these values if they are within
reasonable ranges (run the wizard for suggestions).
OPTIMIZATIONS The integer number of possible optimizations the algorithm will use. Analogous to the popu-

lation size in evolution; larger values require more function evaluations and are slower but produce better
initial sampling, and smaller ones will delay convergence. Defaults to 50.
SEARCH_SPACE If positive, the algorithm will search for new parameters for everything except dihedral phases

within this percentage of the original value, where 1.0 will search within ±100% of the value found in the
input prmtop. Defaults to searching over the entire range of valid values and ignoring the original value in
the topology file. You may wish to alter this value if you know that the original parameters are good and you
wish to search in their neighborhood.
MAX_GENERATIONS The maximum number of iterations the algorithm is allowed to run before it returns the

best non-converged optimization. Defaults to 50,000. If you find that you repeatedly need to increase this
value compared to the default, there are likely significant problems with your system or insufficient input
structures.
GENERATIONS_TO_SIMPLEX The number of iterations in a row that must pass without improvement in the best

parameter set for simplex refinements to be run on a random 5% of the populations. Set to 0 for a pure genetic
algorithm. Smaller values will speed convergence but may result in retrieval of local minima. Defaults to
10.
GENERATIONS_WITHOUT_SIMPLEX The number of generations that must pass between runs of simplex refine-

ment, regardless of improvement in the best parameter set. These iterations serve as a recovery period for
the population of the genetic algorithm, and allows time for the simplex results to be incorporated. If set to
small or zero values, simplex refinement may run too often, resulting in convergence to a local minima and
eliminating the global search properties of the genetic algorithm. Defaults to 10.
GENERATIONS_TO_CONV The number of iterations in a row that must pass without improvement in the best

parameter set for the algorithm to be considered converged. Set to a larger value for a longer but potentially
more accurate run. Defaults to 50, which is too large for most systems. This counter increments along with
the counter to trigger simplex refinement, and at the global minimum simplex refinement will produce no
improvement on the population, allowing convergence.
MUTATION_RATE Internal algorithm parameter The chance an allele (potential parameter) in the genetic algo-

rithm population has to be randomly set to a new value each generation. Defaults to 0.05.
PARENT_PERCENT Internal algorithm parameter The percentage of each generation that is allowed to pass on

alleles to the next generation. Defaults to 0.25.
The simplex algorithm is excellent at refining a good set of input parameters, but can converge on physically
unreasonable values (such as negative bond force constants) if given a naive guess. For this reason, the genetic
algorithm is recommended for finding the global minimum or a close approximation thereof, and the simplex
algorithm may be run on the resulting parameters to confirm the results, if desired. The simplex algorithm starts
at an initial set of parameters and moves “downhill” iteratively while sampling neighboring areas (much like an
amoeba crawling along the function landscape), and converges when the improvement from one step to another
becomes negligible. The simplex algorithm is generally faster than the GA, and excels at well-defined systems with
a small number of dimensions. This algorithm requires a very well-defined sample space, and the input structures
should contain a good range over all the bonds, angles, and dihedrals that are to be optimized. Otherwise, the
algorithm tends to wander and will converge in badly defined areas of the sample set. In smaller, well-defined
systems with only a few parameters, this algorithm will outperform the genetic algorithm.
Choose the simplex algorithm if you wish to fit only a few parameters and have a large number of input conformations. You may specify the following options to fine-tune the step sizes taken, but for the vast majority of cases
the defaults should suffice:

177

10. paramfit
BONDFC_dx Intrinsic length of parameter space for minimization. Used to determine the size of the steps to

construct the initial simplex. Should be large enough that the steps sample a sufficiently large area but small
enough to not move outside of normal parameter range. Bond force constant step size defaults to 5.0.
BONDEQ_dx Bond equilibrium length step size. Defaults to 0.02.
ANGLEFC_dx Angle force constant step size. Defaults to 1.0.
ANGLEEQ_dx Angle equilibrium step size. Defaults to 0.05.
DIHEDRALBH_dx Dihedral force constant step size. Defaults to 0.2.
DIHEDRALN_dx Dihedral periodicity step size. Defaults to 0.01.
DIHEDRALG_dx Dihedral phase step size. Defaults to 0.05.
K_dx Step size for intrinsic difference constant. Defaults to 10.0.
CONV_LIMIT Floating point number that details the convergence limit for the minimization. The smaller the

number, the longer the algorithm will take to converge but the results may be more accurate. Defaults to
1.0E-15, which is very strict.

10.2.6. Bounds Checking
In order to ensure that the algorithms can return meaningful results, bounds checking routines are included in
paramfit. The bounds checking functionality ensures that the algorithm’s results are reasonable given the initial
sample set, and also makes sure that the sample set is well-defined.
Since bonds and angles are approximately harmonic, the algorithm’s result is reasonable if it lies within a welldefined area of the sample set. Bonds and angle values are therefore checked after the algorithm has finished
running. In order to properly fit dihedrals, sample structures should span the entire range of phases for each
dihedral that is to be fit. Dihedral checking is therefore accomplished before the algorithm begins to conduct the
fit.
Bounds checking defaults to halting execution of the program upon reaching a failing condition. It is not recommended that this behavior be disabled, since the results of the fit are most likely inaccurate. Using the fitted
parameters anyway will probably result in an inaccurate depiction of the molecule. Properly represented parameters in the input structures are crucial for a valid fit. Instead of using the parameters, fix the input structures so that
data are provided in the missing ranges, which will be stated in the error message, and rerun the program twice:
first in CREATE_INPUT mode to obtain quantum energies for the added structures and then in FIT mode to redo
the fit.
If you know that your input structures describe the parameters to be fit quite well, the selectivity of the bounds
checking can be altered by the specifying the following options in the job control file. Use these options with
caution, and verify the generated parameters carefully.
CHECK_BOUNDS
= ON The recommended and default option. This will halt execution when the bounds check fails.
= WARN Continue upon reaching a bounds failure condition, but output a warning. Do not use the pa-

rameters generated by this fit without careful verification! Use the error message and other results to
determine if they are reasonable.
BOND_LIMIT Fitting results for bond lengths that are this many Angstroms away from the closest approximation

in the input structures will result in a failing condition. Defaults to 0.1.
ANGLE_LIMIT Fitting results for angles that are more than this many radians away from the closes approximation

in the input structures will result in a failing condition. Defaults to 0.05π.
DIHEDRAL_SPAN The entire range of valid dihedral angles, 0 to π, for each dihedral that is to be fit should

be spanned by this many input structure values, otherwise a failing condition will result. Defaults to 12,
π
radian interval of the valid range.
meaning that there needs to be a dihedral in every 12

178

10.3. Multiple molecule fits

10.3. Multiple molecule fits
Paramfit supports fitting one or more parameters across multiple molecules, and contains several features to aid
in force field development. The program is invoked differently, using a prmtop list and mdcrd list that specify
topology and structures for each molecule to fit. Since the value of K is also system-dependent, you will need to
fit K for each molecule individually.
Input topologies are specified in a prmtop list, which contains the filename of each topology and the value of K
for that system, separated by a space. There are no comments permitted in this file. For example:
molecule1.prmtop 50.0
molecule2.prmtop 100.0

To obtain the value of K for each topology file, conduct a single-molecule fit using all the structures corresponding
to that topology and put the resulting value in this file. This enables fitting to zero over multiple molecules.
Input coordinate files are stated in the coordinate list, which contains the filename of each coordinate set, the
number of structures contained in it, and the filename containing the energy of each structure, separated by a
space. Each energy file is exactly the same as single-molecule fits, containing the energy of each structure, one
per line, in the same order as the corresponding coordinate file. If there are more structures available in the
coordinate file than the number N specified, the first N structures will be used in the fit. An example coordinate
list would be:
molecule1.mdcrd 200 energy1.dat
molecule2.mdcrd 100 energy2.dat

Parameters to fit must be present in all of the available topologies, and the parameter specification file
(PARAMETER_FILE_NAME) should be created using a single-molecule invocation of paramfit. Saved output
files such as energy profile will be named according to the input file name, and a single frcmod will be written if
specified. A multiple molecule invocation of paramfit uses the following command line options:
paramfit -i Job_Control.in -pf prmtop_list -cf mdcrd_list [-v MEDIUM] [--random-seed seed]

The only alteration to the job control file necessary for multiple molecule fits is the deletion of the NSTRUCTURES
parameter. NSTRUCTURES should not be specified as it is now ambiguous and will result in a program error.

10.4. Fitting Forces
Paramfit can fit to the forces on each atom within an input structure rather than to single point energies. In
theory, this provides more data to the fitting algorithm and reduces noise by considering only the forces on atoms
involved in a fitted parameter in the function evaluation. This section will walk you through the process of fitting
forces using paramfit.
Currently, force fitting can only read in Gaussian output files, so input files will be created in the format
accepted by that program. Specify in the QMHEADER file the “force” keyword, so Gaussian will print out the
forces on each atom, and run paramfit in the CREATE_INPUT mode as normal. Then run Gaussian on those
input files, keeping the resulting output with the same naming scheme, for example appending “.out” to the name
of an input file to indicate its input. For example, in bash:
for i in ‘ls output/Job.*.gjf‘; do g09 < $i > $i.out; done

To run a fit with forces, you must specify the following options in the job control file, or use the wizard. Paramfit
will read in the output files from the Gaussian job using the same order and naming scheme, so alter the QM
filename parameters so that they match the suffix you appended to Gaussian output files.
# Enable force fitting function
FUNC_TO_FIT=AMBER_FORCES
# K irrelevant for force fitting

179

10. paramfit
K=0.0
# Force units used by Gaussian
QM_FORCE_UNITS=HARTREE_BOHR
# Naming scheme of gaussian output files
QMFILEOUTSTART=output/Job.
QMFILEOUTEND=.gjf.out

Specify parameters to fit, algorithm and output options as described previously for fits to energy.
As forces fitting is still experimental, take care to evaluate the resulting parameters.

10.5. Examples
10.5.1. Setting up to fit
The fitting process with paramfit follows a specific order. Example job control files for each step and a description of the step follow.
First, write a job control file to create the input structures and run paramfit:
RUNTYPE=CREATE_INPUT
# Trajectory has 50 structures
NSTRUCTURES=50
# Write in Gaussian format
QMFILEFORMAT=GAUSSIAN
# Prepend this file to QM inputs
QMHEADER=Gaussian.header
$AMBERHOME/bin/paramfit -i Job_Control.in -p prmtop -c mdcrd

After all 50 input files have been created, run the quantum program on them. Once it’s finished, extract the
quantum energies from the output files using the provided script, or write your own. Since the example used
Gaussian:
$AMBERHOME/AmberTools/src/paramfit/scripts/process_gaussian.x \
output_directory energies.dat

Now, or while the quantum jobs are running, since neither the energies nor the structures are needed yet,
determine which parameters are to be fit and save them.
RUNTYPE=SET_PARAMS
# File to be created
PARAMETER_FILE_NAME=saved_params
$AMBERHOME/bin/paramfit -i Job_Control.in -p prmtop

Now the quantum energies to fit have been obtained and the parameters to fit have been set, and the fitting process
may begin.

10.5.2. Fitting K
The first step in fitting is determining the value of K for a system. A job control file that will only fit K follows:
RUNTYPE=FIT
PARAMETERS_TO_FIT=K_ONLY
# Use the simplex function
FITTING_FUNCTION=SIMPLEX

Then,

180

10.5. Examples
$AMBERHOME/bin/paramfit -i Job_Control.in -p prmtop -c mdcrd -q energies.dat

Take this value of K and put it back in the job control file when conducting the actual fit.
RUNTYPE=FIT
# Use the parameters specified earlier
PARAMETERS_TO_FIT=LOAD
PARAMETER_FILE_NAME=saved_params
# Genetic algorithm options
FITTING_FUNCTION=GENETIC
OPTIMIZATIONS=500
GENERATIONS_TO_CONV=10
GENERATIONS_TO_SIMPLEX=2
GENERATIONS_WITHOUT_SIMPLEX=5
# Save parameters so they can be read into leap
WRITE_FRCMOD=fitted_params.frcmod

And call paramfit just as before. This example fit will create a force field modification file that can later be read
into leap to create a new prmtop with the modified parameters for the molecule.

10.5.3. Evaluating Results
When using paramfit, it is important to verify the accuracy of the fitted parameters for your input structures. The
WRITE_ENERGY option in the Job Control file is useful for this. Set it to a filename and paramfit will write the
final AMBER energy of each structure next to the quantum energy for the same structure in a file that can be easily
graphed.
If you have gnuplot, a script has been provided to quickly show each structure’s energies. Assuming your
energy file is named energy.dat:
$AMBERHOME/AmberTools/src/paramfit/scripts/plot_energy.x energy.dat

The resulting graph makes the identification of problem structures much easier, and gives a good visualization of
the fit. In general, carefully validate parameters generated by paramfit against other data before conducting large
simulations.
The SCATTERPLOT option in the job control file can also be useful in assessing the quality of the input
structures. If this option is set, paramfit will dump a variety of data files indicating the value for all fitted bonds,
angles, and dihedrals in the input conformations. These data may be visualized if you have the program gnuplot
by running the following command in the directory where paramfit was run:
$AMBERHOME/AmberTools/src/paramfit/scripts/scatterplots.sh

The resulting graphs feature different colored points for each bond, angle, and dihedral type that is being fit for
each of the input structures. This is useful in evaluating if the results of the fit are reasonable– for example, if the
algorithm converges with an equilibrium bond length that is not similar to any of the structures, that parameter may
not be accurate.

181

Part III.

System preparation

183

11. Preparing PDB Files
The only required or useful data in a PDB file to set up AMBER simulations are: atom names, residue names,
and maybe chain identifiers (if more than one chain is present), and the coordinates of heavy atoms. Non-protein
structures (especially low-molecular-weight ligands) will cause problems unless extra libraries are loaded;water
and monatomic ions are generally recognized if their names in the PDB file correspond to the internal names in the
AMBER libraries.
The upshot is that most PDB files require some modification before being used in Amber. Most of the
recommended steps given below can be achieved with the pdb4amber and reduce programs:
pdb4amber -i orig.pdb -o orig1.pdb --nohyd --dry
reduce -build -nuclear orig1.pdb > orig2.pdb

This converts the original pdb file into one likely to be more suitable for input into LEaP. But these programs
(which are described in Sections 11.4 and 11.5 below) cannot anticipate all situations, so you should still examine
the output pdb file to consider the points below.

11.1. Cleaning up Protein PDB Files for AMBER
This is a crucial step in the preparation and many potential problems and subsequent errors arise from omiting
this step! (But also note that these are guidelines for beginners: there are certainly circumstances where you may
wish to modify the ideas presented here.)
• Analyze the PDB file visually in any viewer that can represent (and maybe modify) the file. Alternatively,
use a text editor. Delete all parts which are judged irrelevant for the simulation. Be aware that anything not
protein or water will require you to prepare and load extra library files.
• If the x-ray unit cell in the PDB file contains more than one image, choose the entity you want to use and
delete the other(s).
• If there is a ligand, save it as an MDL standard data file (SDF). Many software packages are able to do this
directly. You may also save the ligand in PDB format and then use some other tools later to convert it into a
decent SDF file (including correct bond order and all hydrogens). It is crucial to keep the coordinates
of its heavy atoms at their original location. Then delete it from the PDB file. The ligand must treated
separately later.
• Delete all water molecules that are not considered relevant. Some waters might be essential for ligand
binding. If those waters are kept, they should be made part of the receptor (as distinct "residues"), not of the
ligand. leap recognizes water if the residue name is WAT or HOH. In later simulations, they may have to be
tethered (more or less strongly) to their original positions to prevent them from "evaporating".
• Apply the same delete procedure to ions, co-factors, and other stuff that has no special relevance for the
planned simulation.
• Get rid off all protein (or peptide) hydrogens that are explicitly expressed in the PDB file. The reduce
and leap utilities adds hydrogens automatically with predefined names. Having hydrogens in PDB files with
names that leap does not recognize within its residue libraries leads to a total mess.
• Eventually, remove also all connectivity records. These are mostly referring to ligands, or, in some cases,
to disulfide links. The latter should be explicitly re-connected (see later) without relying on connectivity
records in the PDB file.

185

11. Preparing PDB Files
• The final PDB file of the protein should only contain unique locations for heavy atoms of amino acids (and
maybe oxygens of specific water molecules). (In some PDB files, the same amino acid may be represented
by different states (conformations). You must decide which unique location you want to use later in the
simulations. If you don’t do anything, Amber will use the “A” conformation, which is generally the most
highly occupied one.) Missing atoms in amino acids are mostly allowed since leap can rebuild them if the
residue names are correct and if the atoms already present have correct names also.
• Make use of "TER" records to separate parts in the PDB file which are not connected covalently. This
is especially important in protein structures in which parts are missing (gaps). Not separating the loose ends
by a "TER" record may lead to strange (and wrong) behavior in leap or later in the simulations. Apply the
same rule to individual water molecules which you want to keep and separate each water by a "TER" record.

11.2. Residue naming conventions
Tautomeric and protonation states are not rendered in PDB files. If a defined state for a residue is required, its
name in the PDB file must reflect the choice. The following subsections deal with these cases. Important: if you
change a residue name in a PDB file, make sure to change it for all atoms of that residue!
Note also that PDB files written out by leap will keep the "special" names, which sometimes leads to annoying
effects in software packages which are not prepared for amino acids called HIE, HIP, CYX, and alike. You might
consider to change these names back to the standard prior to using these PDB files in other software packages. You
can also use the “-bres” option in ambpdb to do that.
Histidine can exist in three forms (δ , ε, and protonated). The PDB file must reflect the choice of the user. In the

current versions of leap command files included with AMBER, ε-histidine is the default, i.e., a "HIS" residue
in a PDB file will be translated automatically to HIE (for ε-histidine). If the residue is called "HID" in the
PDB file, the resulting residue for AMBER will become δ -histidine, while "HIP" will yield the protonated
form.
Cysteine can exist in free form or as part of a disulfide bridge. PDB residues named "CYS" are automatically

converted into a free cysteine with a SH side chain end. If the cysteine is known to be in a S-S bridge, the
residue name in the PDB file must be "CYX". In that case, no hydrogen is automatically added to the side
chain which ends in a bare sulfur. However, S-S bonds to pairing cysteines are not automatically made but
must be specified by the user. The pytleap Python script described in section 39.3 takes care of this through
a special command line option and a file specifying which residues are to be connected (page 755).
Asp,Glu,Lys Sometimes the usually charged residues aspartate "ASP", glutamate "GLU", and lysine "LYS" might

have to be used in their uncharged form. The residue names must then be changed to "ASH", "GLH", and
"LYN", respectively. A neutral form of arginine is not foreseen in AMBER (as the pKa of arginine is around
12, it is always considered protonated).
Terminals: ACE, NHE, NME There are special N- and C-terminal cap residues which can be used to neutralize

the N- and C-terminal in peptide chains when the defaults (NH3+ for the N-terminal and COO− for the
C-terminal) are not appropriate.
The "ACE" residue [−C(= O) − CH3 ] can be used to cap the N-terminal. The PDB entry of the capping
residue ACE must be:

ATOM
ATOM
ATOM

1
2
3

CH3 ACE
C
ACE
O
ACE

resnumber
resnumber
resnumber

x
x
x

y
y
y

z
z
z

Note the atom name "CH3" for this special carbon: another name is not allowed. Hydrogens should be
omitted. They are automatically added if the residue name and the heavy atoThe version bundled with
AmberTools 1.4 is reduce.3.14.080821. See the files in $AMBERHOME/AmberTools/src/reduce for more
information. The information below is taken from the README.usingReduce.txt file.m names are correct.

186

11.3. Chains, Residue Numbering, Missing Residues
For capping the C-terminus, two possibilities are given. The first one is a simple NH2 termination giving
[C(= O) − NH2 ]. This residue is called "NHE" in the PDB file and consists of a single atom to be named N:

ATOM

NHE

resnumber

The second possible C-terminal cap is NH −CH3 , resulting in [C(= O) − NH −CH3 ] at the C-terminal. Its
entry in the PDB file must have the residue name "NME" and has the following PDB entry:

ATOM
ATOM

1
2

N
NME
CH3 NME

resnumber
resnumber

x
x

y
y

z
z

As above for "ACE", the atom name for the carbon must be "CH3". "NHE" and "NME" residues are
automatically completed with hydrogens. Do not enter them explicitly.
The "ACE" residue should be the first residue in a chain (strand) while "NHE" or "NME" should be the last.
If cap residues are used to terminate gaps in incomplete protein chains, they must appear at the exact gap
location, respecting N-terminal and C-terminal order. Gaps must be separated by a "TER" record in the PDB
file. See section 11.3.

11.3. Chains, Residue Numbering, Missing Residues
• AMBER preparation modules assume that residues in a PDB file are connected and appear sequentially in
the file. If not covalently connected (i.e., linked by an amide bond), the residues must be separated by "TER"
records in the PDB file. (Alternatively, the chainid must change on going from one chain to the next chain.)
Thus for example, a protein consisting of two chains should have a "TER" record after the final residue of
the first chain. Similarly, if residues are missing (e.g., not detected in x-ray, or cut by the user), the gap
should also be separated by a "TER" record. Terminal residues will be charged by default. If the user wants
to avoid this (especially for gaps), these residues should be capped (by ACE and NHE or NME).
• In general, leap and tools calling it refer to the original input residue numbers. Thus, residues are numbered
(rather "named") according to the original PDB file for special commands like the disulfide connections.
• In output files from leap, residues will always be numbered starting from 1, irrespective of the original
numbering. Gaps are not considered either. Thus if a protein chain runs from 21 to 80, with residues 31 to
40 (i.e., 10 residues) missing, the final numbering of residues will run from 1 to 50.
The final residue numbers are the ones that must be used in later simulations to refer to individual residues
via AMBER masks or NAB atom expressions. For example, if a protein chain with residues from 30 to 110
is prepared for AMBER simulations, the final numbering will go from 1 to 81. If the original residues 35
to 40 should be fixed or tethered, the actual residues to be specified are 6 to 11. This can lead to serious
errors. So be careful about residue numbers. The script pytleap described later will always generate a new
PDB file with exact AMBER residue numbering and atom names. This PDB file should be used as reference
throughout all subsequent AMBER simulations. Above all, when using atom masks or atom expressions (see
Appendix 19), always check that they really refer to the desired atoms before running lengthy simulations.
Fixing or tethering wrong atoms are a common error which may easily go unnoticed.

11.4. pdb4amber
pdb4amber analyses PDB files and cleans them for further usage, especially with the LeaP programs of Amber.
It does NOT use any information in the original PDB file other than that contained in ATOM and HETATM records.
The final output files are stripped of everything not directly related to ATOM or HETATM records.

187

11. Preparing PDB Files

11.4.1. Usage
Typing pdb4amber on the command line without options (or followed by -h) produces the following help message:

Usage: pdb4amber [options]
Options:
--version
-h, --help
-i FILE, --in=FILE
-o FILE, --out=FILE
-y, --nohyd
-d, --dry
-p, --prot
--noter
--constantph

show program’s version number and exit
show this help message and exit
PDB input file
(default:
PDB output file
(default:
remove all hydrogen atoms
(default:
remove all water molecules
(default:
keep only Amber-compatible residues (default:
remove TER, MODEL, ENDMDL cards
(default:
rename GLU,ASP,HIS for constant pH simulation

stdin)
stdout)
no)
no)
no)
no)

--version prints current version and quits
-i or --in specifies the PDB input file
-o or --out specifies the name of the new PDB output file

NOTE:

Input and output file names cannot be the same. Be aware that on some systems with non
case-sensitive formats, the automatic check of this requirement might fail if input and
output files just differ in upper- and lower-case.

-y or --nohyd requires that all hydrogens are removed (NOT default). Usually PDB standard PDB files from

x-ray have no hydrogens, but already processed PDB files might have hydrogens attached. Since LeaP
is strict on hydrogen atom names, it is better to remove all hydrogens prior to running a PDB file
through LeaP.
-d or --dry removes all water molecules (NOT default). Water molecules in PDB files are useful in many occa-

sions. However, you might want to remove them prior to explicit solvent simulations or when running
3D-RISM. Removed waters are stored in a separate file by pdb4amber (see under OUTPUT below).
-p or --prot keeps only the actual protein + water + recognized CAPS like ACE, NME, and NHE (NOT default).

It is advised to separate pure protein and ligands prior to preparing files for Amber simulations. With
this option, anything not pure protein is removed by pdb4amber and stored in a separate file for further
processing later (see under OUTPUT below). Without this option, everything is kept also in the newly
generated PDB file.
–noter

option to remove all TER, MODEL, ENDMDL cards from output. The default behavior is to keep
them.

–constantph option to rename GLU, ASP and HIS residues to GL4,AS4 and HIP which is required for constant
pH simulations 22
pdb4amber also command line redirect operators (< for input, > for output) so that it can be used effectively in
pipes. Eg. usage:
pdb4amber -i pdbin.pdb -o pdbout.pdb
pdb4amber < pdbin.pdb >pdbout.pdb
cat pdbin.pdb | pdb4amber -o pdbout.pdb

In general, you should use the ’-ypd’ combination of options to generate a protein-only file.

188

11.4. pdb4amber

11.4.2. Output
The new output file (specified with -o or –out) is a standard PDB file with all residues sequentially re-numbered
from 1 to N. In addition, several other files are created automatically:
• A text file with the output PDB file name and _renum.txt added. This is a table to help convert the renumbered residues into the original ones.
• A PDB file with the output PDB file name and _nonprot.pdb appended. This is a PDB file that contains only
non-protein residues (apart from water), i.e., mainly ligands and other stuff.
• When using -d (--dry), a PDB file with the output file name plus _water.pdb added. This file contains
exclusively the water that has been stripped from the original PDB file.
• A text file with the output PDB file name and _sslink attached, if disulfide bonds have been detected by
pdb4amber. This file might be used by the pytleap script to generate the correct disulfide bonds between
cysteines.
The following information is written to the screen (but can also be captured into a text file by ending the
command line with ’2>’, e.g.:
pdb4amber -i pdbin.pdb -o pdbout.pdb [-options] 2> some_file_name.log

Chains: All chain indicators in the PDB file are listed. This is useful especially in cases where the x-ray unit cell

contains more than one image of a protein (or complex). In many cases, one is only interested in one main
peptide chain. A long list of different chains may indicate that the PDB file should cleaned manually prior
to using pdb4amber.
Insertions: Insertions are mostly ’artificial’ residue numbers to keep specific key residue numbers in large protein

families constant. pdb4amber discards insertion codes and re-numbers all residues from 1 to N. But the
insertions are listed to the screen and also included in the _renum.txt file.
Histidines: By default, Amber routines assume that all histidines are epsilon tautomers (HIE). pdb4amber lists

all HIS residues (with the final renumbered residue numbers) to allow users to easily locate HIS and change
some to delta (HID), protonated (HIP), or histidinium bound to zinc (HIN) if required by the local environment in the protein. The residue numbers refer to the renumbered scheme!
Non-standard residues: Non-standard residues (i.e., residues not automatically recognized by Amber) are listed.

Mostly they are ligands (sometimes co-factors, detergent, buffer components, etc.). The user must take
care of these separately. These residues are also found in the _nonprot.pdb file mentioned above. They are
removed from the final output PDB file if the -p (--prot) option was chosen. Otherwise they are left also
in the output PDB file.
Cysteines in disulfide bonds: pdb4amber locates possible (most probable) disulfide bonds by checking the dis-

tance between SG (gamma sulfur) atoms in cysteines. If a distance SG-SG less than 2.5 Angstroem is found
between the SG atoms of two CYS, a disulfide bond is assumed. The respective CYS residues are renamed
to CYX (required for Amber) in the final PDB output file. CONECT records are also printed in the final
PDB output file which are then automatically recognized by tleap. The residue numbers of the CYX residues
refer to the renumbered scheme!
Gaps: pdb4amber tries hard (and mostly succeeds) in locating ’gaps’, i.e., missing residues in the PDB file. This

is done by checking distances of consecutive C-alpha atoms. If such a distance is larger than 5 Angstroem,
pdb4amber considers that there is a gap between the two residues and reports the gap to the screen. The
listed residue numbers refer to the renumbered scheme! It is up to user to decide how to handle the gaps.
Doing nothing at all will most probably lead to trouble later! By simply introducing a TER record at the
gap, Amber (LeaP) will later introduce the charged N (NH3+) or C (COO-) terminals at the gap borders. If
far from the binding site, this might be OK (except in long and unconstrained MD, where such unnatural

189

11. Preparing PDB Files
charges will inevitably lead to unrealistic behavior). The better solution is to introduce ACE or NME caps at
the correct positions (in addition to a TER record separating the gap residues). This can be done in various
ways (e.g. with PyMol). The correct names of the newly introduced residues (ACE or NME) and atoms
(CH3 for the methyl carbon, C, N, O for the others) must be observed!
Missing atoms: pdb4amber tries to determine missing heavy atoms atoms in standard amino acids and reports

these. Residue numbers refer to the renumbered sequence. Note that this has no implcations on further usage
of the file with LeaP since missing atoms are added automatically anyway. In some cases, this addition may
lead to clashes however and it might be useful to know which residues are actually affected by LeaP.

11.4.3. Final remark
The PDB file format and the non-respect of the (in principle well-defined rules) is a constant source of trouble.
There is no guarantee that pdb4amber works for all possible variants and format violations in PDB files. If you
encounter problems with this routine, please report them to me so that I can find a solution. You will never loose
(overwrite) your original PDB file. The worst that can happen is that the resulting output is flawed and not usable.
Romain M. Wolf, Basel, December 2013 romain.wolf (at) gmail.com

11.5. reduce
Reduce is a program for adding hydrogens to a Protein DataBank (PDB) molecular structure file. It was developed by J. Michael Word at Duke University in the lab of David and Jane Richardson. Reduce is described
in: Word, et. al. (1999) Asparagine and Glutamine: Using Hydrogen Atom Contacts in the Choice of Side-chain
Amide Orientation, J. Mol. Biol. 285, 1733-1747.
Both proteins and nucleic acids can have hydrogens added. HET groups can also be processed as long as
the atom connectivity is provided. A slightly modified version of the connectivity table provided by the PDB is
included. The latest version of reduce is available at http://kinemage.biochem.duke.edu/.
In most circumstances, the recommended command when using reduce to add hydrogens to a PDB file and
standardize the bond lengths of existing hydrogens is
reduce -build -nuclear coordfile.pdb > coordfileH.pdb

which includes the optimization of adjustable groups (OH, SH, NH3+, Met-CH3, and Asn, Gln and His sidechain
orientation). Disulfides, covalent modifications, and connection of the ribose-phosphate nucleic acid backbone,
are recognized and any hydrogens eliminated by bonding are skipped. When an amino acid main-chain nitrogen
is not connected to the preceding residue or some other group, reduce treats it as the N-terminus and constructs an
NH3+ only if the residue number is less than or equal to an adjustable limit (1, by default). Otherwise, it considers
the residue to be the observable beginning of an actually-connected fragment and does not protonate the nitrogen.
Reduce does not protonate carboxylates (including the C-terminus) because it does not specifically consider pH,
instead modeling a neutral environment.
Hydrogens are positioned with respect to the covalently bonded neighbors and these are identified by name.
Nonstandard atom names are the primary cause of missing or misplaced hydrogens. If reduce tries to process a
file which contains hydrogens with nonstandard names, the existing hydrogens may not be recognized and may
interfere with the generation of new hydrogens. The solution may be to remove existing hydrogens before further
processing.
There are a number of other, more advance, options for reduce, which can be viewed by running:
reduce -h

190

12. LEaP
12.1. Introduction
LEaP is the generic name given to the programs teLeap and xaLeap, which are generally run via the tleap
and xleap shell scripts. These two programs share a common command language but the xleap program has
been enhanced through the addition of an X-windows graphical user interface. The name LEaP is an acronym
constructed from the names of the older AMBER software modules it replaces: link, edit, and parm. Thus, LEaP
can be used to prepare input for the AMBER molecular mechanics programs.
LEaP is the basic tool to construct force field files (see Fig. 1.1). Using tleap, the user can:
Read AMBER PREP input files
Read Amber PARM format parameter sets
Read and write Object File Format files (OFF)
Read and write PDB files
Construct new residues and molecules using simple commands
Link together residues and create nonbonded complexes of molecules
Modify internal coordinates within a molecule
Generate files that contain topology and parameters for AMBER and NAB
usage: tleap [ -I ] [ -f |- ]

The command tleap is a simple shell script that calls teLeap with a number of standard arguments. Directories to
be searched are indicated by one or more “-I” flags; standard locations are provided in the tleap script. The “-f”
flag is used to tell tleap to take its input from a file (or from stdin if “-f -” is specified). If there is no “-f” flag,
input is taken interactively from the terminal.
A key command for LEaP is loadPdb, which inputs sequence and structure information from Protein Databank
Files. Be sure to read Section 11 for information on how to “clean up” PDB files before loading them.

12.2. Concepts
In order to effectively use LEaP it is necessary to understand the philosophy behind the program, especially
the concepts of LEaP commands, variables, and objects. In addition to exploring these concepts, this section also
addresses the use of external files and libraries with the program.

12.2.1. Commands
A researcher uses LEaP by entering commands that manipulate objects. An object is just a basic building block;
some examples of objects are ATOMs, RESIDUEs, UNITs, and PARMSETs. The commands that are supported
within LEaP are described throughout the manual and are defined in detail in the “Command Reference” section.
The heart of LEaP is a command-line interface that accepts text commands which direct the program to
perform operations on objects. All LEaP commands have one of the following two forms:
command argument1 argument2 argument3 ...
variable = command argument1 argument2 ...

For example:
edit ALA trypsin = loadPdb trypsin.pdb

191

12. LEaP
Each command is followed by zero or more arguments that are separated by whitespace. Some commands return
objects which are then associated with a variable using an assignment (=) statement. Each command acts upon
its arguments, and some of the commands modify their arguments’ contents. The commands themselves are caseinsensitive. That is, in the above example, edit could have been entered as Edit, eDiT, or any combination
of upper and lower case characters. Similarly, loadPdb could have been entered a number of different ways,
including loadpdb. In this manual, we frequently use a mixed case for commands. We do this to enhance the
differences between commands and as a mnemonic device. Thus, while we write createAtom, createResidue,
and createUnit in the manual, the user can use any case when entering these commands into the program.
The arguments in the command text may be objects such as NUMBERs, STRINGs, or LISTs, or they may be
variables. These two subjects are discussed next.

12.2.2. Variables
A variable is a handle for accessing an object. A variable name can be any alphanumeric string whose first
character is an alphabetic character. Alphanumeric means that the characters of the name may be letters, numbers,
or special symbols such as “*”. The following special symbols should not be used in variable names: dollar sign,
comma, period (full stop), pound sign (hash), equals sign, space, semicolon, double quote, or the curly braces {
and }. LEaP commands should not be used as variable names. Unlike commands, variable names are
case-sensitive: “ARG” and “arg” are different variables. Variables are associated with objects using an
assignment statement not unlike that found in conventional programming languages such as Fortran or C.
mole = 6.02E23
MOLE = 6.02E23
myName = "Joe Smith"
listOf7Numbers = { 1.2 2.3 3.4 4.5 6 7 8 }

In the above examples, both mole and MOLE are variable names, whose contents are the same (6.02 × 10 23 ).
Despite the fact that both mole and MOLE have the same contents, they are not the same variable. This is due to
the fact that variable names are case-sensitive. LEaP maintains a list of variables that are currently defined. This
list can be displayed using the list command. The contents of a variable can be printed using the desc command.

12.2.3. Objects
The object is the fundamental entity in LEaP. Objects range from the simple, such as NUMBERs and STRINGs,
to the complex, such as UNITs, RESIDUEs and ATOMs. Complex objects have properties that can be altered using
the set command, and some complex objects can contain other objects. For example, RESIDUEs are complex
objects that can contain ATOMs and have the properties: residue name, connect atoms, and residue type.
NUMBERs

NUMBERs are simple objects holding double-precision floating point numbers. They serve the same function
as “double precision” variables in Fortran and “double” variables in C.
STRINGs

STRINGs are simple objects that are identical to character arrays in C and similar to character strings in
Fortran. STRINGs store sequences of characters which may be delimited by double quote characters. Example
strings are:
"Hello there"
"String with a "" (quote) character"
"Strings contain letters and numbers:1231232"

192

12.2. Concepts
LISTs

LISTs are made up of sequences of other objects delimited by LIST open and close characters. The LIST open
character is an open curly bracket ({) and the LIST close character is a close curly bracket (}). LISTs can contain
other LISTs and be nested arbitrarily deep. Example LISTs are:
{ 1 2 3 4 }
{ 1.2 "string" }
{ 1 2 3 { 1 2 } { 3 4 } }

LISTs are used by many commands to provide a more flexible way of passing data to the commands. The zMatrix
command has two arguments, one of which is a LIST of LISTs where each subLIST contains between three and
eight objects.
PARMSETs (Parameter Sets)

PARMSETs are objects that contain bond, angle, torsion, and non-bonding parameters for AMBER force field
calculations. They are normally loaded from force field data files, such as parm94.dat, and frcmod files.
ATOMs

ATOMs are complex objects that do not contain any other objects. The ATOM object corresponds to the chemical concept of an atom. Thus, it is a single entity that may be bonded to other ATOMs and used as a building
block for creating molecules. ATOMs have many properties that can be changed using the set command. These
properties are defined below.
name This is a case-sensitive STRING property and it is the ATOM’s name. The names for all ATOMs in a

RESIDUE should be unique. The name has no relevance to molecular mechanics force field parameters; it is
chosen arbitrarily as a means to identify ATOMs. Ideally, the name should correspond to the PDB standard,
being 3 characters long except for hydrogens, which can have an extra digit as a 4th character.
type This is a STRING property. It defines the AMBER force field atom type. It is important that the charac-

ter case match the canonical type definition used in the appropriate force field data (*.dat) or frcmod file.
For smooth operation, all atom types must have element and hybridization defined by the addAtomTypes
command. The standard AMBER force field atom types are added by the selected leaprc file.
charge The charge property is a NUMBER that represents the ATOM’s electrostatic point charge to be used in a

molecular mechanics force field.
element The atomic element provides a simpler description of the atom than the type, and is used only for LEaP’s

internal purposes (typically when force field information is not available). The element names correspond to
standard nomenclature; the character “?” is used for special cases.
position This property is a LIST of NUMBERs. The LIST must contain three values: the (X, Y, Z) Cartesian

coordinates of the ATOM.
RESIDUEs

RESIDUEs are complex objects that contain ATOMs. RESIDUEs are collections of ATOMs, and are either
molecules (e.g., formaldehyde) or are linked together to form molecules (e.g., amino acid monomers). RESIDUEs
have several properties that can be changed using the set command. (Note that database RESIDUEs are each
contained within a UNIT having the same name; the residue GLY is referred to as GLY.1 when setting properties.
When two of these single-UNIT residues are joined, the result is a single UNIT containing the two RESIDUEs.)
One property of RESIDUEs is connection ATOMs. Connection ATOMs are ATOMs that are used to make
linkages between RESIDUEs. For example, in order to create a protein, the N-terminus of one amino acid residue
must be linked to the C-terminus of the next residue. This linkage can be made within LEaP by setting the N

193

12. LEaP
ATOM to be a connection ATOM at the N-terminus and the C ATOM to be a connection ATOM at the C-terminus.
As another example, two CYX amino acid residues may form a disulfide bridge by crosslinking a connection atom
on each residue.
There are several properties of RESIDUEs that can be modified using the set command. The properties are
described below:
connect0 This defines the first of up to three ATOMs that are used to make links to other RESIDUEs. In

UNITs containing single RESIDUEs, the RESIDUE’s connect0 ATOM is usually defined as the UNIT’s
head ATOM. (This is how the standard library UNITs are defined.) For amino acids, the convention is to
make the N-terminal nitrogen the connect0 ATOM.
connect1 This defines the second of up to three ATOMs that are used to make links to other RESIDUEs. In

UNITs containing single RESIDUEs, the RESIDUE’s connect1 ATOM is usually defined as the UNIT’s
tail ATOM. (This is done in the standard library UNITs.) For amino acids, the convention is to make the
C-terminal oxygen the connect1 ATOM.
connect2 This defines the third of up to three ATOMs that are used to make links to other RESIDUEs. In amino

acids, the convention is that this is the ATOM to which disulfide bridges are made.
restype This property is a STRING that represents the type of the RESIDUE. Currently, it can have one of

the following values: “undefined”, “solvent”, “protein”, “nucleic”, or “saccharide”. Some of the LEaP
commands behave in different ways depending on the type of a residue. For example, the solvate commands
require that the solvent residues be of type “solvent”. It is important that the proper character case be used
when defining this property.
name The RESIDUE name is a STRING property. It is important that the proper character case be used when

defining this property.
UNITs

UNITs are the most complex objects within LEaP, and the most important. They may contain RESIDUEs and
ATOMs. UNITs, when paired with one or more PARMSETs, contain all of the information required to perform a
calculation using AMBER. UNITs can be created using the createUnit command. RESIDUEs and ATOMs can
be added or deleted from a UNIT using the add and remove commands. UNITs have the following properties,
which can be changed using the set command:
head
tail These define the ATOMs within the UNIT that are connected when UNITs are joined together using the

sequence command or when UNITs are joined together with the PDB or PREP file reading commands. The
tail ATOM of one UNIT is connected to the head ATOM of the next UNIT in any sequence. (Note: a TER
card in a PDB file causes a new UNIT to be started.)
box This property can either be null, a NUMBER, or a LIST. The property defines the bounding box of the UNIT.

If it is defined as null then no bounding box is defined. If the value is a single NUMBER, the bounding box
will be defined to be a cube with each side being box Å across. If the value is a LIST, it must contain three
NUMBERs, the lengths of the three sides of the bounding box.
cap This property can either be null or a LIST. The property defines the solvent cap of the UNIT. If it is defined

as null, no solvent cap is defined. If it is a LIST, it must contain four NUMBERs. The first three define the
Cartesian coordinates (X, Y, Z) of the origin of the solvent cap in Å, while the fourth defines the radius of
the solvent cap, also in Å.
Examples of setting the above properties are:
set dipeptide head dipeptide.1.N
set dipeptide box { 5.0 10.0 15.0 }
set dipeptide cap { 15.0 10.0 5.0 8.0 }

194

12.3. Running LEaP
The first example makes the amide nitrogen in the first RESIDUE within “dipeptide” the head ATOM. The second
example places a rectangular bounding box around the origin with the (X, Y, Z) dimensions of ( 5.0, 10.0, 15.0 )
in Å. The third example defines a solvent cap centered at ( 15.0, 10.0, 5.0 ) Å with a radius of 8.0 Å. Note: the
set cap command does not actually solvate, it just sets an attribute. See the solvateCap command for a more
practical case.
Complex objects and accessing subobjects

UNITs and RESIDUEs are complex objects. Among other things, this means that they can contain other objects.
There is a loose hierarchy of complex objects and what they are allowed to contain. The hierarchy is as follows:
• UNITs can contain RESIDUEs and ATOMs.
• RESIDUEs can contain ATOMs.
The hierarchy is loose because it does not forbid UNITs from containing ATOMs directly. However, the convention
that has evolved within LEaP is to have UNITs directly contain RESIDUEs which directly contain ATOMs.
Objects that are contained within other objects can be accessed using dot “.” notation. An example would be a
UNIT which describes a dipeptide ALA-PHE. The UNIT contains two RESIDUEs each of which contain several
ATOMs. If the UNIT is referenced (named) by the variable dipeptide, then the RESIDUE named ALA can be
accessed in two ways. The user may type one of the following commands to display the contents of the
RESIDUE:
desc dipeptide.ALA
desc dipeptide.1

The first command translates to “describe some RESIDUE named ALA within the UNIT named dipeptide”. The
second form translates as “describe the RESIDUE with sequence number 1 within the UNIT named dipeptide”.
The second form is more useful because every subobject within an object is guaranteed to have a unique sequence
number. If the first form is used and there is more than one RESIDUE with the name ALA, then an arbitrary
residue with the name ALA is returned. To access ATOMs within RESIDUEs, either of the following forms of
command may be used:
desc dipeptide.1.CA
desc dipeptide.1.3

Assuming that the ATOM with the name CA has a sequence number 3 within RESIDUE 1, then both of the above
commands will print a description of the $alpha$-carbon of RESIDUE dipeptide.ALA or dipeptide.1. The reader
should keep in mind that dipeptide.1.CA is the ATOM, an object, contained within the RESIDUE named ALA
within the variable dipeptide. This means that dipeptide.1.CA can be used as an argument to any command that
requires an ATOM as an argument. However dipeptide.1.CA is not a variable and cannot be used on the left hand
side of an assignment statement.

12.3. Running LEaP
xleap -h or tleap -h

will give a list of command-line arguments (which are very simple). Once you have started either program, typing
“help” will bring up a lot of useful information about possible actions.
A file called leaprc is executed as a script file at the start of the LEaP session unless the user suppresses it with
a command line option. Sample files are in $AMBERHOME/dat/leap/cmd, and you may wish to copy one of these
to become "your" default file. LEaP will look first for a learpc file in the user’s current directory, then in any
directories included with -I flags.
The command line interface allows the user to specify a log file that is used to log all input and output within
the command line environment. The log file is named using the logFile command. The file has two purposes: to

195

12. LEaP
allow the user to see a complete record of operations performed by LEaP, and to help recover from (and recreate)
program crashes. Output from LEaP commands is written to the log file at a verbosity level of 2 regardless of the
verbosity level set by the user using the verbosity command. Each line in the log file that was typed in by the user
begins with the two characters "> " (a greater-than sign followed by a space). This allows the user to extract the
commands typed into LEaP from the log file to create a script file that can be executed using the source command.
This provides a type of insurance against program crashes by allowing the user to regenerate their interactive
sessions. An example of a command that will create a script to reenact a LEaP session is:
cat LOGFILE | grep "^> " | sed "s/^> //" > SOURCEFILE.x

Note that changes via graphical and table interfaces (xleap) are not captured by command-line traces.
tleap (terminal LEaP) is the non-graphical, command-line-only interface to LEaP. It has the same functionality
as the xleap main window (Universe Editor Command Window, described below), and uses standard text control
keys. xleap is a windowing interface to LEaP. In addition to the command-line interface contained in the Universe
Editor window, it has a Unit Editor (graphical molecule editor), an Atom Properties Editor, and a Parmset Editor.
These editors are discussed in subsequent subsections.

12.3.1. Universe Editor
The window that first appears when the user starts xleap is called the Universe Editor. The Universe Editor is the
most basic way in which users can interact with xleap. It has two parts, the "command window," which corresponds
to the tleap command interface, and the "pulldown" items above the window, which provide mouse-driven methods
to generate specific commands for the command window, either directly or via popped-up dialog boxes. The items
in the pulldowns allow the user to generate commands using dialog boxes. To display the "File" pulldown, for
example, press the left mouse button on "File;" to select an item in the pulldown, keep the button down, move the
mouse to highlight the item, then release the mouse button. A dialog box will then pop up containing fields which
the user can fill in, and lists from which values can be chosen; these will be used to generate commands for the
command window interface.

12.3.2. Unit Editor
When the user enters the \fCedit\fR command from the Universe Editor Command Window, the Unit Editor will
be displayed if the argument to the \fCedit\fR command is an existing UNIT or a nonexistent (i.e. new) object.
The Parmset Editor will be activated if the argument is a PARMSET. The Parmset Editor is discussed later in this
subsection.
The Unit Editor has five parts. At the top of the window is a pulldown menu bar; below it is a set of buttons titled
"Manipulation" that define the mode of mouse activity in the graphics window, and below that, a list of elements
to select for the manipulation "Draw" mode (selecting one automatically selects "Draw" mode). Then comes the
graphical molecule-editing ("viewing") window itself, and at the very bottom a text window where status and errors
are reported.
Unit Editor Menu Bar

The menu bar has three pulldowns: "Unit," "Edit," and "Display."
Unit pulldown The Unit pulldown contains commands affecting the whole UNIT.

• "Check unit" – checks the UNIT in the viewing window for improbable bond lengths, missing force
field atom types, close nonbonded contacts, and a non-integral and non-zero total charge. Information
is printed in the text window at the bottom of the Unit Editor.
• "Calculate charge" – the total electrostatic charge for the UNIT is displayed in the text window at the
bottom of the Unit Editor.
• "Build," "Add H & Build" – the coordinates of new atoms are adjusted according to hybridization
(inferred from bonds) and standard geometries. (See also the Edit pulldown’s "Relax” selection.)

196

12.3. Running LEaP
Newly-drawn ATOMs are marked as "unbuilt" until they are marked otherwise by one of the Build
commands or by the Edit pulldown’s "Mark selection (un)built." The builder only builds coordinates
for unbuilt ATOMs. This allows users to draw molecules piecemeal and make adjustments as they
draw, without worrying that the builder is going to undo their work. "Add H & Build" adds hydrogens
to the ATOMs that do not have a full valence and builds coordinates for the hydrogens and any other
ATOMs that are marked "unbuilt." The number of hydrogens added to each ATOM is determined by
the hybridization and element type of each ATOM.
• "Import unit" – a selection window pops up for the user to incorporate a copy of another unit in the
current one. The imported unit will generally superimpose on the existing one. (Hint: select all atoms
in the current unit before doing this to simplify dragging them apart using the Manipulation Move
mode.)
• "Close" – Exit the Editor.
Edit pulldown The Edit pulldown contains commands relating to the currently- selected ATOMs in the viewer

window. Selection is described below in the "Manipulation buttons" section.
• "Relax selection" – performs a limited energy minimization of all selected ATOMs, leaving unselected
ATOMs fixed in place, by relaxing strained bonds, angles, and torsions. If atom types have been
assigned and can be found in the currently-loaded force field, force field parameters are used. If no
types are available then default parameters are used that are based on ATOM hybridization. This
command invokes an iterative algorithm that can take some time to converge for large systems. As the
algorithm proceeds, the modified UNIT will be continuously updated within the viewing window. The
user can stop the process at any time by placing the mouse pointer within the viewing window and
typing control-C. Since only internal coordinates are energy minimized, steric overlap can result.
• "Edit selected atoms" – pops up an Atom Properties Editor, a tool for examining/setting the properties
of the selected ATOMs. The Atom Properties Editor allows the user to edit the ATOM names, types
and charges in a convenient table format. It is described in a separate subsection below.
• "Flip chirality" – This command inverts the chirality of all selected ATOMs. In order for the chirality
to be inverted, the ATOM cannot be in more than one ring. The operation causes the lightest chains
leaving the ATOM to be moved so as to invert the chirality. If the ATOM has only three chains attached
to it, then only one of the chains will be moved.
• "Select Rings/Residues/Molecules" – expands the currently selected group of atoms to include all
partially-contained rings, residues, or molecules.
• "Show everything" – causes all ATOMs to become visible.
• "Hide selection" – makes all selected ATOMs invisible.
• "Show selection only" – makes only selected ATOMs visible.
• "Mark selection unbuilt/built" - see "Unit/Build," above.
Display pulldown The Display pulldown contains commands that determine what information is displayed within

the viewing window.
• "Names" – toggles display of ATOM names at each ATOM position.
• "Types" – toggles display of molecular mechanics atom types. The ATOM types are displayed within
parentheses "()".
• "Charges" – toggles display of the atomic charges.
• "Residue names" – toggles display of residue names. These are displayed at the position of the first
ATOM, before any of that ATOM’s information that may be displayed. The residue names are displayed within angled brackets "<>".
• "Axes" – toggles display of the Cartesian coordinate axes. The origin of the axes coincides with the
origin of Cartesian space.
• "Periodic box" – toggles display of the periodic box, if the UNIT has one.

197

12. LEaP
Unit Editor manipulation buttons

The Manipulation buttons are Select, Twist, Move, Erase, and Draw. They determine the behavior of the mouse
left-button when the mouse pointer is in the Viewing Window.
Select This button allows one to select part or all of a UNIT in anticipation of a subsequent operation or action.

In the Select mode, the user can highlight ATOMs within the viewing window for special operations. The
mouse pointer becomes a pointing hand in the viewing window in this mode. Selected ATOMs are displayed
in a different color (or different line styles on monochrome systems) from all other ATOMs. Atoms can
be selected with the left-button in several ways: first, clicking on an atom and releasing selects that atom.
Clicking twice in a row on an atom (at any speed) selects all atoms (this is a bug – only the residue should be
selected). Keeping the button down and moving to release on another atom selects all ATOMs in the shortest
chain between the two ATOMs, if such a chain exists. Finally, by first pressing the button in empty space,
and holding it down as the mouse is moved, one can "drag a box" enclosing atoms of interest. Note that
a current selection can be expanded by using the "Edit" menubar pulldown select option to complete any
partial selection of rings, residues or molecules.
If the user holds down the SHIFT key while performing any of the above actions, the same effect will be
seen, except ATOMs will be unselected.
Twist Twist mode operates on previously-Selected atoms. The intention is to allow rotation about dihedrals; if too

many atoms are selected, odd transformations can occur. While in the Twist mode, the mouse pointer looks
like a curved arrow. Twisting is driven by holding down the left-button anywhere in the viewing window and
moving the mouse up and down. It is important to select a complete torsion (all four atoms) before trying to
"twist" it.
Move Like Twist, Move mode operates on previously-Selected atoms. While in the Move mode, the mouse pointer

looks like four arrows coming out of one central point. Holding down the left-button anywhere allows
movement of these atoms by dragging in any direction in the viewing plane. (The view can be rotated by
holding down the middle-button to allow any movement desired.) This option allows the user to move the
selected ATOMs relative to the unselected ATOMs.
To rotate the selected ATOMs relative to the unselected ones, press and drag the mode (left) button while
holding down the SHIFT key. The selected ATOMs will rotate around a central ATOM on a "virtual sphere"
(see the subsubsection below on the rotate (middle) button for more information on the "virtual sphere").
The user can change which ATOM is used as the center of rotation by clicking the mode (left) button on any
of the ATOMs in the window.
Erase Erase mode causes the mouse pointer to resemble a chalkboard eraser when it is in the viewing window.

Clicking the left-button will delete any atoms or bonds under this mouse pointer, one atom or bond per click.
Draw Choosing Draw is equivalent to choosing the default "Elements" atom in the next array of buttons; the initial

default is carbon. While in the Draw mode, the mouse pointer is a pencil when in the viewing window.
Clicking the left-button deposits an atom of the current element, while dragging the mouse pointer with the
left-button held down draws a bond: if no atom is found where the button is released, one is created.
When the mouse pointer approaches an ATOM, the end of the line connected to the pointer will "snap" to
the nearest ATOM. This is to facilitate drawing of bonds between ATOMs. Any bonds that are drawn will by
default be single bonds. To change the order of a bond, the user would move the mouse to any point along
the bond and click the mode (left) button. This will cause the order of the bond to increase until it is reset
back to a single bond. The user can cycle through the following bond order choices: single, double, triple,
and aromatic.
If the user rotates a structure as it is being drawn, she will notice that all of the ATOMs that have been
drawn lie in the same plane. New ATOMs are automatically placed in the plane of the screen. The fact that
LEaP places the new ATOMs in the same plane is not a handicap because once a rough sketch of part of
the structure is compete, the user can invoke one of LEaP’s two model building facilities ("Unit/Build" and
"Edit/Relax Selection" in the Unit Editor Menu bar) to build full three dimensional coordinates.

198

12.3. Running LEaP
Unit Editor Elements Buttons" "C, H, O, ..." These buttons put the viewing window in Draw mode if it is not in

that mode already, and select the drawing element. The more common elements have their own buttons, and
all elements are also found by pulling down the other elements button.
Unit Editor Viewing Window

The viewing window displays a projection of the UNIT currently being edited. The user can manipulate the
structure within the viewing window with the mouse. By moving the mouse and holding down the mouse buttons,
the user can rotate, scale, and translate the UNIT within the window. The functions attached to the mouse buttons
are:
Rotate (Middle button) By pressing the rotate (middle) button within the viewing window and dragging the

mouse, the user can rotate the UNIT around the center of the viewing window. While the rotate (middle) button is down, a circle appears within the viewing window, representing a "virtual sphere trackball."
As the user drags the mouse around the outside of the circle, the UNIT will spin around the axis normal to
the screen. As the user drags the mouse within the circle, the UNIT will spin around the axis in the screen,
perpendicular to the movement of the mouse. The structures that are being viewed can be considered to be
embedded within a sphere of glass. The circle is the projection of the edge of the sphere onto the screen.
Rotating a UNIT while the mouse is within the circle is akin to placing a hand on a glass sphere and turning
the sphere by pulling the hand. The rotate operation does not modify the coordinates of the ATOMs; rather,
it simply changes the user’s point of view.
Translate (Right button) By pressing the translate (right) button within the viewing window and dragging the

mouse around the viewing window, the user can translate the UNIT within the plane of the screen. The
structures will follow the mouse as it moves around the window. This operation does not modify the coordinates of the UNIT.
Scale (middle plus right button) If the scale "button" (holding the middle and right buttons down at the same

time) is depressed, the user will change the size of the structures within the viewing window. Pressing the
scale (middle plus right) button and dragging the mouse up and down the screen will increase and decrease
the scale of the structures. This operation does not modify the coordinates of the UNIT.
Mode (left button) The function of the left button is determined by the current mode of the viewing window as

described in the "Manipulation" section, above. When the mouse enters the viewing window it changes
shape to reflect the current mode of the viewing window.
Spacebar Another always-available operation when the mouse pointer is in the viewing window is the keyboard

spacebar. It centers and normalizes the size of the molecule in the viewing window. This is especially useful
if the UNIT becomes "lost" due to some operation.
The functions of the middle and right buttons are fixed and always available to the user. This allows the user
to change the viewpoint of the UNIT within the viewing window regardless of its current mode. The user
might ask why there are controls to translate in the plane of the screen, but not out of the plane of the screen.
This is because LEaP does not have depth-cueing or stereo projection and this makes it difficult for users to
perceive changes in the depth of a structure. However, the user can rotate the entire UNIT by 90 degrees
which will orient everything so that the direction that was coming out of the screen becomes a direction
lying in the plane of the screen. Once the UNIT has been rotated using the rotate (middle) button, the user
can translate the structure anywhere in space. While it does take some getting used to, users can become
very adept at the combination of rotations and translations.

12.3.3. Atom Properties Editor
The Atom Properties Editor is popped up by the Unit Editor when the user selects the Edit selected atoms
command from the Edit pulldown. The Atom Properties Editor allows the user to edit the properties of ATOMs
using a convenient table format. ATOM properties are: name, type, charge, and element.

199

12. LEaP

12.3.4. Parmset Editor
If the user enters the command edit Foo in the Universe Editor and Foo is a PARMSET, then a Parmset Editor is
popped up. First, a window appears which contains a number of buttons. The buttons list the parameters that can
be edited – Atom, Bond, Angle, Proper Torsion, Improper Torsion, and Hydrogen Bond – and an option to close
the editor. Choosing one of the parameter buttons will pop up a Table Editor. This editor resembles that of the
Atom Properties Editor, having three parts: the Menu Bar, Status Window, and Table Window.

12.4. Basic instructions for using LEaP to build molecules
This section gives an overview of how LEaP is most commonly used. Detailed descriptions of all the commands
are given in the following section.

12.4.1. Building a Molecule For Molecular Mechanics
In order to prepare a molecule within LEaP for AMBER, three basic tasks need to be completed.
1. Any needed UNIT or PARMSET objects must be loaded;
2. The molecule must be constructed within LEaP;
3. The user must output topology and coordinate files from LEaP to use in AMBER.
The most typical command sequence is the following:
source leaprc.ff99SB (load a force field)
x = loadPdb trypsin.pdb (load in a structure)
.... add in cross-links, solvate, etc.
saveAmberParm x prmtop prmcrd (save files)

There are a number of variants of this:
1. Although loadPdb is by far the most common way to enter a structure, one might use loadOff, or
loadAmberPrep, or use the zmat command to build a molecule from a Z-matrix. See the Commands section
below for descriptions of these options. If you do not have a starting structure (in the form of a PDB file),
LEaP can be used to build the molecule; you will find, however, that this is not always a straightforward
process. Many experienced Amber users turn to other (commercial and non-commercial) programs to create
their initial structures.
2. Be very attentive to any errors produced in the loadPdb step; these generally mean that LEaP has misread
the file. A general rule of thumb is to keep editing your input PDB file until LEaP stops complaining. It is
often convenient to use the addPdbAtomMap or addPdbResMap commands to make systematic changes from
the names in your PDB files to those in the Amber topology files; see the leaprc files in $AMBERHOME/dat/leap/cmd for examples of this. Be sure to read Section 11 for information on how to “clean up” PDB
files before loading them.
3. The saveAmberParm command cited above is appropriate for most force fields; for polarizable calculations
you will need to use saveAmberParmPol.

12.4.2. Amino Acid Residues
For each of the amino acids found in the LEaP libraries, there has been created an N-terminal and a C-terminal
analog. The N-terminal amino acid UNIT/RESIDUE names and aliases are prefaced by the letter N (e.g., NALA)
and the C-terminal amino acids by the letter C (e.g., CALA). If the user models a peptide or protein within LEaP,
they may choose one of three ways to represent the terminal amino acids. The user may use (1) standard amino
acids, (2) protecting groups (ACE/NME), or (3) the charged C- and N-terminal amino acid UNITs/RESIDUEs. If
the standard amino acids are used for the terminal residues, then these residues will have incomplete valences.
These three options are illustrated below:

200

12.5. Commands
{ ALA VAL SER PHE }
{ ACE ALA VAL SER PHE NME }
{ NALA VAL SER CPHE }

The default for loading from PDB files is to use N- and C-terminal residues; this is established by the
addPdbResMap command in the default leaprc files. To force incomplete valences with the standard residues,
one would have to define a sequence (“ x = { ALA VAL SER PHE }”) and use loadPdbUsingSeq, or use
clearPdbResMap to completely remove the mapping feature.
Histidine can exist either as the protonated species or as a neutral species with a hydrogen at the δ or ε position.
For this reason, the histidine UNIT/RESIDUE name is either HIP, HID, or HIE (but not HIS). The default “leaprc”
file assigns the name HIS to HIE. Thus, if a PDB file is read that contains the residue HIS, the residue will be
assigned to the HIE UNIT object. This feature can be changed within one’s own leaprc file.
The AMBER force fields also differentiate between the residue cysteine (CYS) and the similar residue which
participates in disulfide bridges, cystine (CYX). The user will have to explicitly define, using the bond command,
the disulfide bond for a pair of cystines, as this information is not read from the PDB file. In addition, the user
will need to load the PDB file using the loadPdbUsingSeq command, substituting CYX for CYS in the sequence
wherever a disulfide bond will be created.

12.4.3. Nucleic Acid Residues
The “D” prefix can be used to distinguish between deoxyribose and ribose units. Residue names like “A” or
“DA” can be followed by a “5” or “3” (“DA5”, “DA3”) for residues at the ends of chains; this is also the default
established by addPdbResMap, even if the “5” or “3” are not added in the PDB file. The “5” and “3” residues
are “capped” by a hydrogen; the plain and “3” residues include a “leading” phosphate group. Neutral residues
(nucleotides) capped by hydrogens end their names with “N”, as in “DAN”.

12.5. Commands
The following is a description of the commands that can be accessed using the command line interface in tleap,
or through the command line editor in xleap. Whenever an argument in a command line definition is enclosed
in square brackets (e.g., [arg]), then that argument is optional. When examples are shown, the command line is
prefaced by “> ”, and the program output is shown without this character preface.
Some commands that are almost never used have been removed from this description to save space. You can use
the “help” facility to obtain information about these commands; most only make sense if you understand what the
program is doing behind the scenes.

12.5.1. add
add a b

UNIT/RESIDUE/ATOM a,b
Add the object b to the object a. This command is used to place ATOMs within RESIDUEs, and RESIDUEs
within UNITs. This command will work only if b is not contained by any other object.
The following example illustrates both the add command and the way the TIP3P water molecule is created for
the LEaP distribution.
>
>
>
>
>
>
>

h1 = createAtom H1 HW 0.417
h2 = createAtom H2 HW 0.417
o = createAtom O OW -0.834
set h1 element H
set h2 element H
set o element O

201

12. LEaP
>
> r = createResidue TIP3
> add r h1
> add r h2
> add r o
>
> bond h1 o
> bond h2 o
> bond h1 h2
>
> TIP3 = createUnit TIP3
>
> add TIP3 r
> set TIP3.1 restype solvent
> set TIP3.1 imagingAtom TIP3.1.O
>
> zMatrix TIP3 {
> { H1 O 0.9572 }
> { H2 O H1 0.9572 104.52 }
> }
>
> saveOff TIP3 water.lib
Saving TIP3.
Building topology.
Building atom parameters.

12.5.2. addAtomTypes
addAtomTypes { { type element hybrid } { ... } ... }

Define element and hybridization for force field atom types. This command for the standard force fields can be
seen in the default leaprc files. The STRINGs are most safely rendered using quotation marks. If atom types are
not defined, confusing messages about hybridization can result when loading PDB files.

12.5.3. addIons and addIons2
addIons unit ion1 numIon1 [ion2 numIon2]
addIons2 unit ion1 numIon1 [ion2 numIon2]

Adds counterions in a shell around unit using a Coulombic potential on a grid. If numIon1 is 0, then the unit
is neutralized. In this case, numIon1 must be opposite in charge to unit and numIon2 must not be specified. If
solvent is present, it is ignored in the charge and steric calculations, and if an ion has a steric conflict with a solvent
molecule, the ion is moved to the center of that solvent molecule, and the latter is deleted. (To avoid this behavior,
either solvate _after_ addions, or use addIons2.) Ions must be monatomic. This procedure is not guaranteed
to globally minimize the electrostatic energy. When neutralizing regular-backbone nucleic acids, the first cations
will generally be placed between phosphates, leaving the final two ions to be placed somewhere around the middle
of the molecule. The default grid resolution is 1 Å, extending from an inner radius of (maxIonVdwRadius +
maxSoluteAtomVdwRadius) to an outer radius 4 Å beyond. A distance-dependent dielectric is used for speed.
addIons2 is the same as addIons, except solvent and solute are treated the same.

12.5.4. addIonsRand
addIonsRand unit ion1 #ion1 [ion2 #ion2] [separation]

202

12.5. Commands
Adds counterions in a shell around unit by replacing random solvent molecules. If #ion1 is 0, the unit is neutralized
(ion1 must be opposite in charge to unit, and ion2 cannot be specified). Otherwise, the specified numbers of ion1
[ion2] are added [in alternating order]. If separation is specified, ions will be guaranteed to be more than that
distance apart in Angstroms.
Ions must be monoatomic. This procedure is much faster than addIons, as it does not calculate charges. Solvent
must be present. It must be possible to position the requested number of ions with the given separation in the
solvent.

12.5.5. addPath
addPath path

Add the directory in path to the list of directories that are searched for files specified by other commands. The
following example illustrates this command.
> addPath /disk/howard
/disk/howard added to file search path.

After the above command is entered, the program will search for a file in this directory if a file is specified in a
command. Thus, if a user has a library named “/disk/howard/rings.lib” and the user wants to load that library, one
only needs to enter load rings.lib and not load /disk/howard/rings.lib.

12.5.6. addPdbAtomMap
addPdbAtomMap list

The atom Name Map is used to try to map atom names read from PDB files to atoms within residue UNITs when
the atom name in the PDB file does not match an atom in the residue. This enables PDB files to be read in without
extensive editing of atom names. Typically, this command is placed in the LEaP startup file, “leaprc”, so that
assignments are made at the beginning of the session. list should be a LIST of LISTs. Each sublist should contain
two entries to add to the Name Map. Each entry has the form:
{ string string }

where the first string is the name within the PDB file, and the second string is the name in the residue UNIT.

12.5.7. addPdbResMap
addPdbResMap list

The Name Map is used to map RESIDUE names read from PDB files to variable names within LEaP. Typically,
this command is placed in the LEaP startup file, “leaprc”, so that assignments are made at the beginning of the
session. The LIST is a LIST of LISTs. Each sublist contains two or three entries to add to the Name Map. Each
entry has the form:
{ double string1 string2 }

where double can be 0 or 1, string1 is the name within the PDB file, and string2 is the variable name to which
string1 will be mapped. To illustrate, the following is part of the Name Map that exists when LEaP is started from
the “leaprc” file included in the distribution:
ADE --> DADE
: :
0 ALA --> NALA
0 ARG --> NARG
: :

203

12. LEaP
1
1
:
1

ALA --> CALA
ARG --> CARG
:
VAL --> CVAL

Thus, the residue ALA will be mapped to NALA if it is the N-terminal residue and CALA if it is found at the
C-terminus. The above Name Map was produced using the following (edited) command line:
>
>
>
>
>
>
>

addPdbResMap
{ 0 ALA NALA
{ 0 ARG NARG
{ 0 VAL NVAL
: :
{ ADE DADE }
}

{
} { 1 ALA CALA }
} { 1 ARG CARG } : :
} { 1 VAL CVAL }
: :

12.5.8. alias
alias [ string1 [ string2 ] ]

This command will add or remove an entry to the Alias Table or list entries in the Alias Table. If both strings are
present, then string1 becomes the alias to string2, the original command. If only one string is used as an argument,
then that string will be removed from the Alias Table. If no arguments are given to the command, the current
aliases stored in the Alias Table will be listed.
The proposed alias is first checked for conflict with the LEaP commands and rejected if a conflict is found. A
proposed alias will replace an existing alias with a warning being issued. The alias can stand for more than a single
word, but also as an entire string so the user can quickly repeat entire lines of input.

12.5.9. bond
bond atom1 atom2 [ order ]

Create a bond between atom1 and atom2. Both of these ATOMs must be contained by the same UNIT. By
default, the bond will be a single bond. By specifying “-”, “=”, “#”, or “:” as the optional argument, order, the
user can specify a single, double, triple, or aromatic bond, respectively. Example:
bond trx.32.SG trx.35.SG

12.5.10. bondByDistance
bondByDistance container [ maxBond ]

Create single bonds between all ATOMs in the UNIT container that are within maxBond Å of each other. If
maxBond is not specified, a default distance will be used. This command is especially useful in building
molecules. Example:
bondByDistance alkylChain

12.5.11. check
check unit [ parms ]

This command can be used to check unit for internal inconsistencies that could cause problems when performing
calculations. This is a very useful command that should be used before a UNIT is saved with saveAmberParm or
its variants. Currently it checks for the following possible problems:

204

12.5. Commands
• long bonds
• short bonds
• non-integral total charge of the UNIT
• missing force field atom types
• close contacts (< 1.5 Å) between nonbonded ATOMs
The user may collect any missing molecular mechanics parameters in a PARMSET for subsequent editing. In the
following example, the alanine UNIT found in the amino acid library has been examined by the check command:
> check ALA
Checking ’ALA’....
Checking parameters for unit ’ALA’.
Checking for bond parameters.
Checking for angle parameters.
Unit is OK.

12.5.12. combine
variable = combine list

Combine the contents of the UNITs within list into a single UNIT. The new UNIT is placed in variable. This
command is similar to the sequence command except it does not link the ATOMs of the UNITs together. In the
following example, the input and output should be compared with the example given for the sequence command.
> tripeptide = combine { ALA GLY PRO }
Sequence: ALA
Sequence: GLY
Sequence: PRO
> desc tripeptide
UNIT name: ALA !! bug: this should be tripeptide!
Head atom: .R.A
Tail atom: .R.A
Contents:
R
R
R

12.5.13. copy
newvariable = copy variable

In most cases, creates an exact duplicate of the object variable. Since newvariable is not pointing to the same
object as variable, changing the contents of one object will not alter the other object. Example:
> tripeptide = sequence { ALA GLY PRO }
> tripeptideSol = copy tripeptide
> solvateBox tripeptideSol WATBOX216 8 2

In the above example, tripeptide is a separate object from tripeptideSol and is not solvated. Had the user instead
entered

205

12. LEaP
> tripeptide = sequence { ALA GLY PRO }
> tripeptideSol = tripeptide
> solvateBox tripeptideSol WATBOX216 8 2

then both tripeptide and tripeptideSol would be solvated since they would both refer to the same object.
Note that in a few instances, the copy command does not produce an exact copy. This is particularly relevant
when making copies of oligosaccharide residues. In these, the copy command invariably inverts chirality at the
anomeric carbon. The workaround for this is to use the copy command twice, where the second call inverts the
chirality back.

12.5.14. createAtom
variable = createAtom name type charge

Return a new and empty ATOM with name, type, and charge as its atom name, atom type, and electrostatic point
charge. (See the add command for an example of the createAtom command.)

12.5.15. createResidue
variable = createResidue name

Return a new and empty RESIDUE with the name name. (See the add command for an example of the
createResidue command.)

12.5.16. createUnit
variable = createUnit name

Return a new and empty UNIT with the name name. (See the add command for an example of the createUnit
command.)

12.5.17. deleteBond
deleteBond atom1 atom2

Delete the bond between the ATOMs atom1 and atom2. If no bond exists, an error will be displayed.

12.5.18. desc
desc variable

Print a description of the object variable. In the following example, the alanine UNIT found in the amino acid
library has been examined by the desc command:
> desc ALA
UNIT name: ALA
Head atom: .R.A
Tail atom: .R.A
Contents: R

Now, the desc command is used to examine the first residue (1) of the alanine UNIT:
> desc ALA.1
RESIDUE name: ALA
RESIDUE sequence number: 1
Type: protein

206

12.5. Commands
Connection atoms:
Connect atom 0: A
Connect atom 1: A
Contents:
A
A
A
A
A
A
A
A
A
A

Next, we illustrate the desc command by examining the ATOM N of the first residue (1) of the alanine UNIT:
> desc ALA.1.N
ATOM Name: N
Type: N
Charge: -0.463
Element: N
Atom flags: 20000|posfxd- posblt- posdrn- sel- pert- notdisp- tchdposknwn+ int - nmin- nbldAtom position: 3.325770, 1.547909, -0.000002
Atom velocity: 0.000000, 0.000000, 0.000000
Bonded to .R.A by a single bond.
Bonded to .R.A by a single bond.

Since the N ATOM is also the first atom of the ALA residue, the following command will give the same output as
the previous example:
> desc ALA.1.1

12.5.19. groupSelectedAtoms
groupSelectedAtoms unit name

Create a group within unit with the name name, using all of the ATOMs within unit that are selected. If the group
has already been defined then overwrite the old group. The desc command can be used to list groups. Example:
groupSelectedAtoms TRP sideChain

An expression like “TRP@sideChain” returns a LIST, so any commands that require LISTs can take advantage of
this notation. After assignment, one can access groups using the “@” notation. Examples:
select TRP@sideChain
center TRP@sideChain

The latter example will calculate the center of the atoms in the “sideChain” group. (See the select command for
a more detailed example.)

12.5.20. help
help [string]

This command prints a description of the command in string. If no argument is given, a list of help topics is
provided.

207

12. LEaP

12.5.21. impose
impose unit seqlist internals

The impose command allows the user to impose internal coordinates on unit. The list of RESIDUEs to impose the
internal coordinates upon is in seqlist. The internal coordinates to impose are in internals, which is an object of
type LIST.
The command works by looking into each RESIDUE within unit that is listed in seqlist and attempts to apply
each of the internal coordinates within internals. The seqlist argument is a LIST of NUMBERS that represent
sequence numbers or ranges of sequence numbers. A range of sequence numbers is represented by two element
LISTs that contain the first and last sequence number in the range. The user can specify sequence number ranges
that are larger than what is found in unit, in which case the range will stop at the beginning or end of unit as
appropriate. For example, the range { 1 999 } will include all RESIDUEs in a 200 RESIDUE UNIT.
The internals argument is a LIST of LISTs. Each sublist contains a sequence of ATOM names which are of
type STRING followed by the value of the internal coordinate. An example of the impose command would be:
impose peptide { 1 2 3 } { { “N” “CA” “C” “N” -40.0 } { “C” “N” “CA” “C” -60.0 } }

This would cause the RESIDUE with sequence numbers 1, 2, and 3 within the UNIT peptide to assume an
α-helical conformation. The command
impose peptide { 1 2 { 5 10 } 12 } { { “CA” “CB” 5.0 } }

will impose on the residues with sequence numbers 1, 2, 5, 6, 7, 8, 9, 10, and 12 within the UNIT peptide a bond
length of 5.0 Å between the α and β carbon atoms. RESIDUEs without an ATOM named CB, such as glycine,
will be unaffected.
It is important to understand that the impose command attempts to perform the intended action on all residues in
the seqlist, but does not necessarily limit itself to acting only upon internals contained within those residues. That
is, the list does not limit the residues to consider. Rather, it is a list of all starting points to consider. In other words,
to specify a seqlist of { 3 4 } tells impose to attempt to set two torsions, one starting in residue 3 and the other
starting in residue 4. It does not specify that the torsion should only be set if the atoms are found within residues 3
and/or 4.
Because of this, one must be careful when setting torsions between two residues. It is necessary to know which
atoms are contained in which residues. Consider the following trisaccharide:
α-D-Glcp-(1-6)-β -D-Manp-(1-6)-β -D-Galp-OH
To build it most simply in leap requires the following directive. Note that the build order in leap is the reverse
of the standard order in which the residues are written above.
glycan = sequence { ROH 6LB 6MB 0GA }

A proper build of a 1-6 oligosaccharide linkage often requires setting three torsions. In the manner that residues
are defined in the Glycam force fields, the atoms describing two of those torsions, φ and ψ, span two residues.
However, the atoms in the third, ω, exist entirely within one residue. In fact, they exist within all three glycan
residues in the example above. The following commands will set only the three torsions in the glycosidic linkage
between residues 4 (0GA) and 3 (6MB).
impose glycan { 4 } { { “H1” “C1” “O6” “C6” -60.0 } } # O6 & C6 are in residue 3
impose glycan { 4 } { { “C1” “O6” “C6” “C5” 180.0 } } # only C1 is in residue 4
impose glycan { 3 } { { “O6” “C6” “C5” “O5” 60.0 } } # all are in residue 3

The common misconception that the seqlist sets a limit on the residues affected can cause trouble in this case. For
example, this command
impose glycan { 4 3 } { { “H1” “C1” “O6” “C6” -60.0 } }

208

12.5. Commands
will find all sequences beginning in residue 4 and in residue 3 that contain the serially bonded atoms H1 C1 O6
and C6. Therefore, in this case, it will set the specified torsions between residues 4 and 3 as well as between 3 and
2. Similarly, this command
impose peptide { 4 } { { “O6” “C6” “C5” “O5” 60.0 } }

will not affect any inter-residue linkage, but instead will set the C5-C6 torsion in the glucopyranoside (0GA) at the
non-reducing end of the oligosaccharide.
The ordering and content within the internals list is important as well. For these examples, consider the simple
peptide sequence:
peptide = sequence { ALA ALA ALA ALA }

The ordering of the internals specifies the atoms to which the torsion set is applied. The impose command will
find the first atom in the internals list, check for the presence of a bonded second atom, and so forth. It will then
apply the action, here a torsion, to those four atoms. For example, this command:
impose peptide { 3 } { { “N” “CA” “C” “N” -40.0 } }

# between 3 and 4

will set the torsion between residues 3 and 4. However, this one:
impose peptide { 3 } { { “N” “C” “CA” “N” -40.0 } }

# between 3 and 2

will set the torsion between residues 3 and 2.
If at any point, the impose command does not find an atom bonded to a previous atom in an internals list, it
will silently ignore the command. This is likely to occur in two instances. One, the atom simply might not exist in
the residue:
impose peptide { 3 } { { “N” “CA” “CB” “HB4” 10.0 } }

# no effect, silent

Here, of course, there is no atom named HB4 in alanine. Similarly, improper torsions are ignored. For example,
this command also has no effect:
impose peptide { 3 } { { “N” “HB1” “CA” “CB” 10.0 } }

# no effect, silent

because HB1 is not bonded to N.
Three types of conformational change are supported: Bond length changes, bond angle changes, and torsion
angle changes. If the conformational change involves a torsion angle, then all dihedrals around the central pair of
atoms are rotated. The entire list of internals is applied to each RESIDUE.
It is also important to note that the impose command performs its actions entirely using internal coordinates.
Because of this, it is difficult to predict the resulting behavior when the coordinates are translated back to cartesian,
for example when writing a PDB file.

12.5.22. list
List all of the variables currently defined. To illustrate, the following (edited) output shows the variables
defined when LEaP is started from the leaprc file included in the distribution:
> list A ACE ALA ARG ASN : : VAL W WAT Y

12.5.23. loadAmberParams
variable = loadAmberParams filename

Load an AMBER format parameter set file and place it in variable. All interactions defined in the parameter set
will be contained within variable. This command causes the loaded parameter set to be included in LEaP’s list of
parameter sets that are searched when parameters are required. General proper and improper torsion parameters
are modified during the command execution with the LEaP general type “?” replacing the AMBER general type
“X”
> parm91 = loadAmberParams parm91X.dat
> saveOff parm91 parm91.lib

209

12. LEaP

12.5.24. loadAmberPrep
loadAmberPrep filename [ prefix ]

This command loads an AMBER PREP input file. For each residue that is loaded, a new UNIT is constructed that
contains a single RESIDUE and a variable is created with the same name as the name of the residue within the
PREP file. If the optional argument prefix (a STRING) is provided, its contents will be prefixed to each variable
name; this feature is used to prefix UATOM residues, which have the same names as AATOM residues with the
string “U” to distinguish them.
> loadAmberPrep cra.in
Loaded UNIT: CRA

12.5.25. loadOff
loadOff filename

This command loads the OFF library within the file named filename. All UNITs and PARMSETs within the
library will be loaded. The objects are loaded into LEaP under the variable names the objects had when they were
saved. Variables already in existence that have the same names as the objects being loaded will be overwritten.
Any PARMSETs loaded using this command are included in LEaP’s library of PARMSETs that is searched
whenever parameters are required (the old AMBER format is used for PARMSETs rather than the OFF format in
the default configuration). Example command line:
> loadOff parm91.lib
Loading library: parm91.lib
Loading: PARAMETERS

12.5.26. loadMol2
variable = loadMol2 filename

Load a Sybyl MOL2 format file into variable, a UNIT. This command is very much like loadOff, except that it
only creates a single UNIT.

12.5.27. loadPdb
variable = loadPdb filename

Load a Protein Data Bank (PDB) format file with the file name filename into variable, a UNIT. The sequence
numbers of the RESIDUEs will be determined from the order of residues within the PDB file ATOM records.
This function will search the variables currently defined within LEaP for variable names that map to residue
names within the ATOM records of the PDB file. If a matching variable name is found then the contents of the
variable are added to the UNIT that will contain the structure being loaded from the PDB file. Adding the
contents of the matching UNIT into the UNIT being constructed means that the contents of the matching UNIT
are copied into the UNIT being built and that a bond is created between the connect0 ATOM of the matching
UNIT and the connect1 ATOM of the UNIT being built. The UNITs are combined in the same way UNITs are
combined using the sequence command. As atoms are read from the ATOM records their coordinates are written
into the correspondingly named ATOMs within the UNIT being built. If the entire residue is read and it is found
that ATOM coordinates are missing, then external coordinates are built from the internal coordinates that were
defined in the matching UNIT. This allows LEaP to build coordinates for hydrogens and lone-pairs which are not
specified in PDB files.
> crambin = loadPdb 1crn

210

12.5. Commands

12.5.28. loadPdbUsingSeq
loadPdbUsingSeq filename unitlist

This command reads a PDB format file named filename. This command is identical to loadPdb except it does not
use the residue names within the PDB file. Instead, the sequence is defined by the user in unitlist. For more
details see loadPdb.
> peptSeq = { UALA UASN UILE UVAL UGLY }
> pept = loadPdbUsingSeq pept.pdb peptSeq

In the above example, a variable is first defined as a LIST of united atom RESIDUEs. A PDB file is then loaded,
in this sequence order, from the file “pept.pdb”.

12.5.29. logFile
logFile filename

This command opens the file with the file name filename as a log file. User input and all output is written to the
log file. Output is written to the log file as if the verbosity level were set to 2. An example of this command is
> logfile /disk/howard/leapTrpSolvate.log

12.5.30. measureGeom
measureGeom atom1 atom2 [ atom3 [ atom4 ] ]

Measure the distance, angle, or torsion between two, three, or four ATOMs, respectively.
In the following example, we first describe the RESIDUE ALA of the ALA UNIT in order to find the identity
of the ATOMs. Next, the measureGeom command is used to determine a distance, simple angle, and a dihedral
angle. As shown in the example, the ATOMs may be identified using atom names or numbers.
> desc ALA.ALA
RESIDUE name: ALA
RESIDUE sequence number: 1
Type: protein ....

12.5.31. quit
Quit the LEaP program.

12.5.32. remove
remove a b

Remove the object b from the object a. If a does not contain b, an error message will be displayed. This
command is used to remove ATOMs from RESIDUEs, and RESIDUEs from UNITs. If the object represented by
b is not referenced by any other variable name, it will be destroyed.
> dipeptide = combine { ALA GLY }
Sequence: ALA
Sequence: GLY
> desc dipeptide
UNIT name: ALA !! bug: this should be dipeptide!
Head atom: .R.A
Tail atom: .R.A

211

12. LEaP
Contents: R R
> remove dipeptide dipeptide.2
> desc dipeptide UNIT name: ALA !! bug: this should be dipeptide!
Head atom: .R.A
Tail atom: null
Contents: R

12.5.33. saveAmberParm
saveAmberParm unit topologyfilename coordinatefilename

Save the Amber/NAB topology and coordinate files for unit into the files named topologyfilename and coordinatefilename respectively. This command will cause LEaP to search its list of PARMSETs for parameters defining all
of the interactions between the ATOMs within unit. It produces topology files and coordinate files that are identical
in format to those produced by Amber PARM and can be read into Amber and NAB for calculations. The output
of this operation can be used for minimizations, dynamics, and thermodynamic perturbation calculations.
In the following example, the topology and coordinates from the all_amino94.lib UNIT ALA are generated:
> saveamberparm ALA ala.top ala.crd

12.5.34. saveMol2
saveMol2 unit filename type-flag

Write unit to the file filename as a Tripos mol2 format file. If type-flag is 0, the Tripos (Sybyl) atom types will
be used; if type-flag is 1, the Amber atom types present in unit will be used. Generally, you would want to set
type-flag to 1, unless you need the Sybyl atom types for use in some program outside Amber; Amber itself has no
force fields that use Sybyl atom types.

12.5.35. saveOff
saveOff object filename

The saveOff command allows the user to save UNITs and PARMSETs to a file named filename. The file is written
using the Object File Format (off) and can accommodate an unlimited number of uniquely named objects. The
names by which the objects are stored are the variable names specified within the object argument. If the file
filename already exists, the new objects will be added to it. If there are objects within the file with the same names
as objects being saved then the old objects will be overwritten. The argument object can be a single UNIT, a single
PARMSET, or a LIST of mixed UNITs and PARMSETs. (See the add command for an example of the saveOff
command.)

12.5.36. savePdb
savePdb unit filename

Write unit to the file filename as a PDB format file. In the following example, the PDB file from the
“all_amino94.lib” UNIT ALA is generated:
> savepdb ALA ala.pdb

12.5.37. sequence
variable = sequence list

212

12.5. Commands
The sequence command is used to combine the contents of list, which should be a LIST of UNITs, into a new,
single UNIT. This new UNIT is constructed by taking each UNIT in list in turn and copying its contents into the
UNIT being constructed. As each new UNIT is copied, a bond is created between the tail ATOM of the UNIT
being constructed and the head ATOM of the UNIT being copied, if both connect ATOMs are defined. If only one
is defined, a warning is generated and no bond is created. If neither connection ATOM is defined then no bond is
created. As each RESIDUE is copied into the UNIT being constructed it is assigned a sequence number which
represents the order the RESIDUEs are added. Sequence numbers are assigned to the RESIDUEs so as to
maintain the same order as was in the UNIT before it was copied into the UNIT being constructed. This
command builds reasonable starting coordinates for all ATOMs within the UNIT; it does this by assigning internal
coordinates to the linkages between the RESIDUEs and building the external coordinates from the internal
coordinates from the linkages and the internal coordinates that were defined for the individual UNITs in the
sequence.
> tripeptide = sequence { ALA GLY PRO }

12.5.38. set
This command operates in two modes. In the first, it sets default values for some parameters. In the second, it
sets specific properties to containers (for example, UNITs).
Defaults can be set in LEaP for the global parameters below with this usage:
set default parameter value

For example:
set default PBRadii mbondi

OldPrmtopFormat If set to “on”, the saveAmberParm command will write a prmtop file in the format used in

Amber 6 and earlier versions; if set to “off” (the default), it will use the new format. This is discouraged for
general use and is available mainly for backwards compatibility with programs that expect old-style topology
files or for testing.
Dielectric If set to “distance” (the default), electrostatic calculations in LEaP will use a distance-dependent di-

electric; if set to “constant”, a constant dielectric will be used.
PdbWriteCharges If set to “on”, atomic charges will be placed in the “B-factor” field of PDB files saved with the
savePdb command; if set to “off” (the default), no such charges will be written.

PBRadii Used to choose various sets of atomic radii for generalized Born or Poisson-Boltzmann calculations.

Options are: “bondi”, which gives values from Ref. [249], which should be used with igb = 7; “mbondi”,
which is the default, and the recommended parameter set for igb = 1 [134]; “mbondi2”, which is a second
modification of the Bondi radii set [120], and should be used with igb = 2 or 5; “mbondi3”, which is a third
modification of the Bondi radii set [21] recommended for use with igb = 8; and “amber6”, which is only to
be used for reproducing very early calculations that used igb = 1 [118].
nocenter If set to “on”, LEaP will not center the coordinates inside the box for a periodic simulation, but will

leave them unchanged (as it does for non-periodic simulations); if set to “off” (the default), centering of
coordinates will take place (as it always has, in previous versions of LEaP). Avoiding coordinate translations
can be useful to avoid changing reference (perhaps experimental) coordinates. This option may be especially
helpful for crystal simulations.
The parameters listed below can be set for the specified containers within LEaP using the following syntax:
set container parameter object

Some examples:

213

12. LEaP
set ATOM name "name"
set RESIDUE connect0 ATOM
my_system = loadPDB file.pdb
set my_system box {25 30 32}

For ATOMs:
name A unique STRING descriptor used to identify ATOMs.
type This is a STRING property that defines the AMBER force field atom type.
charge The charge property is a NUMBER that represents the ATOM’s electrostatic point charge to be used in a

molecular mechanics force field.
position This property is a LIST of NUMBERs containing three values: the (X, Y, Z) Cartesian coordinates of

the ATOM.
pertName This STRING is a unique identifier for an ATOM in its final state during a Free Energy Perturbation

calculation. This functionality is no longer implemented in Amber.
pertType This STRING is the AMBER force field atom type of a perturbed ATOM. This functionality is no longer

implemented in Amber.
pertCharge This NUMBER represents the final electrostatic point charge on an ATOM during a Free Energy

Perturbation. This function is no longer implemented in Amber.
For RESIDUEs:
connect0 This identifies the first of up to three ATOMs that will be used to make links to other RESIDUEs. In

a UNIT containing a single RESIDUE, the RESIDUE’s connect0 ATOM is usually defined as the UNIT’s
head ATOM.
connect1 This identifies the second of up to three ATOMs that will be used to make links to other RESIDUEs. In

a UNIT containing a single RESIDUE, the RESIDUE’s connect1 ATOM is usually defined as the UNIT’s
tail ATOM.
connect2 This identifies the third of up to three ATOMs that will be used to make links to other RESIDUEs. In

amino acids, the convention is that this is the ATOM to which disulfide bridges are made.
restype This property is a STRING that represents the type of the RESIDUE. Currently, it can have one of the

following values: “undefined”, “solvent”, “protein”, “nucleic”, or “saccharide”.
name This STRING property is the RESIDUE name.

For UNITs:
head Defines the ATOM within the UNIT that is connected when UNITs are joined together: the tail ATOM of

one UNIT is connected to the head ATOM of the subsequent UNIT in any sequence.
tail Defines the ATOM within the UNIT that is connected when UNITs are joined together: the tail ATOM of one

UNIT is connected to the head ATOM of the subsequent UNIT in any sequence.
box This property defines the bounding box of the UNIT (container). If object is set to null then no bounding box

is defined. If it is a single NUMBER, the bounding box will be defined to be a cube with each side being
NUMBER Å across. If it is a LIST, it must contain three NUMBERs, the lengths (in Å) of the three sides of
the bounding box.
cap This property defines the solvent cap of the UNIT. If it is set to null then no solvent cap is defined. Otherwise,

it should be a LIST of four NUMBERs; the first three NUMBERs define the Cartesian coordinates (X, Y, Z)
of the origin of the solvent cap in Å, while the fourth defines the radius of the solvent cap, also in Å.

214

12.5. Commands

12.5.39. solvateBox and solvateOct
solvateBox solute solvent distance [ closeness ]
solvateOct solute solvent distance [ closeness ]

These two commands create periodic solvent boxes around solute, which should be a UNIT. solvateBox creates
a cuboid box, while solvateOct creates a truncated octahedron. solute is modified by the addition of copies of
the RESIDUEs found within solvent, which should also be a UNIT, such that the closest distance between any
atom originally present in solute and the edge of the periodic box is given by the distance parameter. The resulting
solvent box will be repeated in all three spatial directions.
The optional closeness parameter can be used to control how close, in Å, solvent ATOMs may come to solute
ATOMs. The default value of closeness is 1.0. Smaller values allow solvent ATOMs to come closer to solute
ATOMs. The criterion for rejection of overlapping solvent RESIDUEs is if the distance between any solvent
ATOM and its nearest solute ATOM is less than the sum of the two ATOMs’ van der Waals radii multiplied by
closeness.
> mol = loadpdb my.pdb
> solvateOct mol TIP3PBOX 12.0 0.75

12.5.40. solvateCap
solvateCap solute solvent position radius [ closeness ]

The solvateCap command creates a solvent cap around solute, which is a UNIT. solute is modified by the addition
of copies of the RESIDUEs found within solvent, which should also be a UNIT. The solvent box will be repeated
in all three spatial directions to create a large solvent sphere with a radius of radius Å.
The position argument defines where the center of the solvent cap is to be placed. If position is a UNIT, a
RESIDUE, an ATOM, or a LIST of UNITs, RESIDUEs, or ATOMs, then the geometric center of the ATOM or
ATOMs within the object will be used as the center of the solvent cap sphere. If position is a LIST containing three
NUMBERs, then it will be treated as a vector describing the position of the solvent cap sphere center.
The optional closeness parameter can be used to control how close, in Å, solvent ATOMs may come to solute
ATOMs. The default value of closeness is 1.0. Smaller values allow solvent ATOMs to come closer to solute
ATOMs. The criterion for rejection of overlapping solvent RESIDUEs is if the distance between any solvent
ATOM and its nearest solute ATOM is less than the sum of the two ATOMs’ van der Waals radii multiplied by
closeness.
This command modifies solute in several ways. First, the UNIT is modified by the addition of solvent
RESIDUEs copied from solvent. Secondly, the “cap” parameter of solute is modified to reflect the fact that a
solvent cap has been created around the solute.
> mol = loadpdb my.pdb
> solvateCap mol WATBOX216 mol.2.CA 12.0 0.75

12.5.41. solvateShell
solvateShell solute solvent thickness [ closeness ]

The solvateShell command adds a solvent shell to solute, which should be a UNIT. solute is modified by the
addition of copies of the RESIDUEs found within solvent, which should also be a UNIT. The resulting
solute/solvent UNIT will be irregular in shape since it will reflect the contours of the original solute molecule.
The solvent box will be repeated in three directions to create a large solvent box that can contain the entire solute
and a shell thickness Å thick. Solvent RESIDUEs are then added to solute if they lie within the shell defined by
thickness and do not overlap with any ATOM originally present in solute. The optional closeness parameter can
be used to control how close solvent ATOMs can come to solute ATOMs. The default value of the closeness
argument is 1.0. Please see the solvateBox command for more details on the closeness parameter.
> mol = loadpdb my.pdb
> solvateShell mol WATBOX216 12.0 0.8

215

12. LEaP

12.5.42. source
source filename

This command executes the contents of the file given by filename, treating them as LEaP commands. To display
the commands as they are read, see the verbosity command.

12.5.43. transform
transform atoms, matrix

Transform all of the ATOMs within atoms by a symmetry operation. The symmetry operation is represented as a
(3 × 3) or (4 × 4) matrix, and given as nine or sixteen NUMBERs in matrix, a LIST of LISTs. The general matrix
looks like:
r11 r12 r13 -tx r21 r22 r23 -ty r31 r32 r33 -tz 0 0 0 1
The matrix elements represent the intended symmetry operation. For example, a reflection in the (x,y) plane
would be produced by the matrix:
1 0 0 0 1 0 0 0 -1

This reflection could be combined with a 6 Å translation along the x-axis by using the following matrix:
1 0 0 6 0 1 0 0 0 0 -1 0 0 0 0 1

In the following example, wrB is transformed by an inversion operation:
transform wrpB { { -1 0 0 } { 0 -1 0 } { 0 0 -1 } }

12.5.44. translate
translate atoms direction

Translate all of the ATOMs within atoms by the vector given by direction, a LIST of three NUMBERs.
Example:
translate wrpB { 0 0 -24.53333 }

12.5.45. verbosity
verbosity level

This command sets the level of output that LEaP provides the user. A value of 0 is the default, providing the
minimum of messages. A value of 1 will produce more output, and a value of 2 will produce all of the output of
level 1 and display the text of the script lines executed with the source command. The following line is an
example of this command:
> verbosity 2
Verbosity level: 2

12.5.46. zMatrix
zMatrix object zmatrix

The zMatrix command is quite complicated. It is used to define the external coordinates of ATOMs within object
using internal coordinates. The second parameter of the zMatrix command is a LIST of LISTs; each sub-list has
several arguments:
{ a1 a2 bond12 }

216

12.6. Building oligosaccharides, lipids and glycoproteins
This entry defines the coordinate of a1, an ATOM, by placing it bond12 Å along the x-axis from ATOM a2. a2 is
placed at the origin if its coordinates are not defined.
{ a1 a2 a3 bond12 angle123 }

This entry defines the coordinate of a1 by placing it bond12 Å away from a2 making an angle of angle123
degrees between a1, a2 and a3. The angle is measured in a right-hand sense and in the xy plane. ATOMs a2 and
a3 must have coordinates defined.
{ a1 a2 a3 a4 bond12 angle123 torsion1234 }

This entry defines the coordinate of a1 by placing it bond12 Å away from a2, creating an angle of angle123
degrees between a1, a2, and a3, and making a torsion angle of torsion1234 degrees between a1, a2, a3, and a4.
{ a1 a2 a3 a4 bond12 angle123 angle124 orientation }

This entry defines the coordinate of a1 by placing it bond12 Å away from a2, and making angles angle123 degrees
between a1, a2, and a3, and angle124 degrees between a1, a2, and a4. The argument orientation defines whether
a1 is above or below a plane defined by a2, a3 and a4. If orientation is positive, a1 will be placed so that the triple
product ((a3−a2) × (a4−a2)) · (a1−a2) is positive. Otherwise, a1 will be placed on the other side of the plane. This
allows the coordinates of a molecule like fluoro-chloro-bromo-methane to be defined without having to resort to
dummy atoms.
The first arguments within the zMatrix entries (a1, a2, a3 and a4) are either ATOMs, or STRINGs containing
names of ATOMs that already exist within object. The subsequent arguments (bond12, angle123, torsion1234 or
angle124, and orientation) are all NUMBERs. Any ATOM can be placed at the a1 position, even one that has
coordinates defined. This feature can be used to provide an endless supply of dummy atoms, if they are required.
A predefined dummy atom with the name “*” (a single asterisk, no quotes) can also be used.
There is no order imposed in the sub-lists. The user can place sub-lists in arbitrary order, as long as they
maintain the requirement that all ATOMs a2, a3, and a4 must have external coordinates defined, except for entries
that define the coordinate of an ATOM using only a bond length. (See the add command for an example of the
zMatrix command.)

12.6. Building oligosaccharides, lipids and glycoproteins
Build assistance available at GLYCAM-Web:
The approaches presented below have been automated, with many additional options available, at the
GLYCAM-Web site: www.glycam.org. The capabilities of the website are being expanded. Currently, the
available functionalities include:
Oligosaccharides, linear and branched
Glycoproteins, O- or N-linked, with multiple glycans
Builds of oligosaccharides via URL directive

Build assistance available in the AmberTools tests:
Examples in addition to those described below can be found in the AmberTools tests. The relevant files are
located in:
$AMBERHOME/AmberTools/test/leap/glycam # main test directory
$AMBERHOME/AmberTools/test/leap/glycam/06EPb # extra points oligosaccharides
$AMBERHOME/AmberTools/test/leap/glycam/06j # main oligosaccharides
$AMBERHOME/AmberTools/test/leap/glycam/06j_10 # glycoprotein with ff10
$AMBERHOME/AmberTools/test/leap/glycam/06j_12SB # glycoprotein with ff12SB

217

12. LEaP
HOH2C

HOH2C

HOH2C
O

HOH2C

OH H

3GB

HOH2C

HOH2C
O

H
HO

OH H
OH H

ROH

HOH2C
O

OH
O

OH H

0GB

O
HO

OH H

H
HOH2C

HO
O

CH3

O
O

0GA

4GB

OME

OCH3

HO
OH H

Figure 12.1.: Schematic representation of disaccharide formation, indicating the need for open valences on carbon
and oxygen atoms at linkage positions.
PLEASE NOTE: The molecules in the test directories were constructed for the purpose of testing functionality in
AmberTools. They might not be ready for simulations as they are. Some might be in configurations with severe
clashes. Most structural issues can be resolved by manipulating appropriate torsions. The glycoprotein tests
contain usage examples for torsion manipulations using the impose command.
Each sub-directory below "glycam" contains tests relevant to specific force fields. To run an individual test,
saving all relevant output and intermediate files, change to the sub-directory and issue the command:
./Run.glycam evaluate

To return the directory to its previous state, run:
./Run.glycam clean

The 00_README file in the main directory contains more information about using the tests.
Additional notes about this section:
Before continuing in this section, you should review the GLYCAM naming conventions covered in Section 3.7.
After that, there are two important things to keep in mind. The first is that GLYCAM is designed to build oligosaccharides, not just monosaccharides. In order to link the monosaccharides together, each residue in GLYCAM will
have at least one open valence position. That is, each GLYCAM residue lacks either a hydroxyl group or a hydroxyl proton, and may be lacking more than one proton depending on the number of branching locations. Thus,
none of the residues is a complete molecule unto itself. For example, if you wish to build α-D-glucopyranose, you
must explicitly specify the anomeric -OH group (see Figure 12.1 for two examples).
The second thing to keep in mind is that when the sequence command is used in LEaP to link
monosaccharides together to form a linear oligosaccharide (analogous to peptide generation), the residue ordering
is opposite to the standard convention for writing the sequence. For example, to build the disaccharides illustrated
in Figure 12.1, using the sequence command in LEaP, the format would be:
upperdisacc = sequence { ROH 3GB 0GB }
lowerdisacc = sequence { OME 4GB 0GA }

While the sequence command is the most direct method to build a linear glycan, it is not the only method. Alternatives that facilitate building more complex glycans and glycoproteins are presented below. For those who need
to build structures (and generate topology and coordinate files) that are more complex, a convenient interface that
uses GLYCAM is available on the internet (http://glycam.ccrc.uga.edu or http://www.glycam.org).
Throughout this section, sequences of LEaP commands will be entered in the following format:
command argument(s) # descriptive comment

218

12.6. Building oligosaccharides, lipids and glycoproteins
This format was chosen so that the lines can be copied directly into a file to be read into LEaP. The number sign
(#) signifies a comment. Comments following commands may be left in place for future reference and will be
ignored by LEaP. Files may be read into LEaP either by sourcing the file or by specifying it on the command line
at the time that LEaP is invoked, e.g.:
tleap -f leap_input_file

Note that any GLYCAM parameter set shipped with Amber is likely to be updated in the future. The current version is GLYCAM_06j.dat. This file and GLYCAM_06j-1.prep are automatically loaded with the default
leaprc.GLYCAM_06j-1. The user is encouraged to check www.glycam.org for updated versions of these files.

12.6.1. Procedures for building oligosaccharides using the GLYCAM-06 parameters
12.6.1.1. Example: Linear oligosaccharides

This section contains instructions for building a simple, straight-chain tetrasaccharide:
α-D-Manp-(1-3)-β -D-Manp-(1-4)-β -D-GlcpNAc-(1-4)-β -D-GlcpNAc-OH
First, it is necessary to determine the GLYCAM residues that will be used to build it. Since the initial α-D-Manp
residue links only at its anomeric site, the first character in its name is 0 (zero), indicating that it has no branches
or other connections, i.e., it is terminal. Since it is a D-mannose, the second character, the one-letter code, is M
(capital). Since it is an α-pyranose, the third character is A. Therefore, the first residue in the sequence above is
0MA. Since the second residue links at its 3-position as well as at the anomeric position, the first character in its
name is 3, and, being a β -pyranose, it is 3MB. Similarly, residues three and four are both 4YB. It will also be
necessary to add an OH residue at the end to generate a complete molecule. Note that in Section 12.6.3, below,
the terminal OH must be omitted in order to allow subsequent linking to a protein or lipid. Note also that when
present, a terminal OH (or OME etc) is assigned its own residue number.
Converting the order for use with the sequence command in LEaP, gives:
Residue name sequence: ROH 4YB 4YB 3MB 0MA
Residue number:
1
2
3
4
5

Here is a set of LEaP instructions that will build the sequence (there are, of course, other ways to do this):
source leaprc.GLYCAM_06j-1 # load leaprc
glycan = sequence { ROH 4YB 4YB 3MB 0MA } # build oligosaccharide

Using the sequence command, the φ angles are automatically set to the orientation that is expected on the basis
of the exo-anomeric effect (± 60°). If you wish to change the torsion angle between two residues, the impose
command may be used. In the following example, the ψ angles between the two 4YB residues and between the
4YB and the 3MB are being set to the standard value of zero.
impose glycan {3} { {C1 O4 C4 H4 0.0} } # set psi between 4YB (3) & 4YB (2)
impose glycan {4} { {C1 O4 C4 H4 0.0} } # set psi between 3MB (4) & 4YB (3)

You may now generate coordinate, topology and PDB files, for example:
saveamberparm glycan glycan.top glycan.crd # save top & crd
savepdb glycan glycan.pdb # save pdb file

12.6.1.2. Example: Branched oligosaccharides

This section contains instructions for building a simple branched oligosaccharide. The example used here builds
on the previous one. Again, it will be assumed that the carbohydrate is not destined to be linked to a protein or a
lipid. If it were, one should omit the ROH residue from the structure. The branched oligosaccharide is

219

12. LEaP
α-D-Manp-(1–3)-β -D-Manp-(1–4)-β -D-GlcpNAc-(1–4)-β -D-GlcpNAc-OH
6
|
α-D-Manp-1
Note that the β -D-mannopyranose is now branched at the 3- and 6-positions. Consulting Tables 3.4 to 3.7
informs us that the first character assigned to a carbohydrate linked at the 3- and 6-positions is V. Thus, the name
of the residue called 3MB in the previous section must change to VMB.
Thus, when rewritten for LEaP this glycan becomes:
Residue name sequence: ROH 4YB 4YB VMB 0MA 0MA
Residue number:
1
2
3
4
5
6

To ensure that the correct residues are linked at the 3- and 6-positions in VMB, it is safest to specify these
linkages explicitly in LEaP. In the current example, the two terminal residues are the same (0MA), but that need
not be the case.
source leaprc.GLYCAM_06j-1 # load leaprc
glycan = sequence { ROH 4YB 4YB VMB } # linear sequence to branch

The longest linear sequence is built first, ending at the branch point “VMB” in order to explicitly specify
subsequent linkages. The following commands will place a terminal, 0MA residue at the number three position:
set glycan tail glycan.4.O3 # set attachment point to the O3 in VMB
glycan = sequence { glycan 0MA } # add one of the 0MA’s

The following commands will link the other 0MA to the 6-position. Note that the name of the molecule changes
from “glycan” to “branch”. This change is not necessary, but makes such command sequences easier to read,
particularly with complex structures.
set glycan tail glycan.4.O6 # set attachment point to the O6 in VMB
branch = sequence { glycan 0MA } # add the other 0MA

It can be especially important to reset torsion angles when building branched oligosaccharides. The following set
of commands cleans up the geometry considerably and then generates a set of output files:
impose branch {4} { {H1 C1 O6 C6 -60.0} } # set phi torsion and
impose branch {4} { {C1 O6 C6 H6 0.0} } # set psi 0MA(6) & VMB
impose branch {4} { {H1 C1 O4 C4 60.0} } # set phi torsion and
impose branch {4} { {C1 O4 C4 H4 0.0} } # set psi 3MB & 4YB
impose branch {3} { {H1 C1 O4 C4 60.0} } # set phi torsion and
impose branch {3} { {C1 O4 C4 H4 0.0} } # set psi 4YB & 4YB
impose branch {5} { {H1 C1 O3 C3 -60.0} } # set phi torsion and
impose branch {5} { {C1 O3 C3 H3 0.0} } # set psi 0MA(3) & VMB
saveamberparm branch branch.top branch.crd # save top & crd
savepdb branch branch.pdb # save pdb

12.6.1.3. Example: Complex branched oligosaccharides

The following example builds a highly branched, high-mannose structure shown in Figure 12.2 . In this example,
it is especially important to note that when the branching is ambiguous, LEaP might not choose the attachment
point one wants or expects. For this reason, connectivity should be specified explicitly whenever the structure
branches. That is, one cannot specify the longest linear sequence and add branches later. The sequence command
must be interrupted at each branch point. Otherwise, the connectivity is not assured. In this example, a branch
occurs at each VMA (-3,6-D-Manp ) residue.
The following set of commands, given to tleap, will safely produce the structure represented in Figure12.2 .

220

12.6. Building oligosaccharides, lipids and glycoproteins

α2
α6
α3
α2

α6
β4

β4

β ΟΗ

α3
α2

α2

Figure 12.2.: Structure of Man-9, represented in the symbolic notation used by the Consortium for Functional
Glycomics. Here, =D-Manp and =D-GlcpNAc

source leaprc.GLYCAM_06j-1
glycan = sequence { ROH 4YB 4YB VMB }
set glycan tail glycan.4.O6
glycan=sequence { glycan VMA }
set glycan tail glycan.5.O6
glycan=sequence { glycan 2MA 0MA }
set glycan tail glycan.5.O3
glycan=sequence { glycan 2MA 0MA }
set glycan tail glycan.4.O3
glycan=sequence { glycan 2MA 2MA 0MA }
impose glycan {3} { {H1 C1 O4 C4 60.0} }
impose glycan {3} { {C1 O4 C4 H4 0.0} }
impose glycan {4} { {H1 C1 O4 C4 60.0} }
impose glycan {4} { {C1 O4 C4 H4 0.0} }
impose glycan {5} { {H1 C1 O6 C6 -60.0} } # 1-6 Link from (5) to (4), Phi
impose glycan {5} { {C1 O6 C6 C5 180.0} } # 1-6 Link from (5) to (4), Psi
impose glycan {4} { {O6 C6 C5 O5 60.0} } # 1-6 Link from (5) to (4), Chi
impose glycan {10} { {H1 C1 O3 C3 -60.0} }
impose glycan {10} { {C1 O3 C3 H3 0.0} }
impose glycan {6} { {H1 C1 O6 C6 -60.0} }
impose glycan {6} { {C1 O6 C6 C5 180.0} }
impose glycan {5} { {O6 C6 C5 O5 -60.0} }
impose glycan {8} { {H1 C1 O3 C3 -60.0} }
impose glycan {8} { {C1 O3 C3 H3 0.0} }
impose glycan {7} { {H1 C1 O2 C2 -60.0} }
impose glycan {7} { {C1 O2 C2 H2 0.0} }
impose glycan {9} { {H1 C1 O2 C2 -60.0} }
impose glycan {9} { {C1 O2 C2 H2 0.0} }
impose glycan {11} { {H1 C1 O2 C2 -60.0} }
impose glycan {11} { {C1 O2 C2 H2 0.0} }
impose glycan {12} { {H1 C1 O2 C2 -60.0} }
impose glycan {12} { {C1 O2 C2 H2 0.0} }
saveamberparm glycan glycan.prmtop glycan.restrt

221

12. LEaP
O2

MYR

PGL
O2

C3
C2

C5
C4

C7
C6

C9
C8

C11
C10

C13
C12

C14

O1
O2

MY2

C5
C2

C4
C3

C6
C5

C8
C7

C10
C9

C12
C11

C14
C13

C4
N
C1

CHO
C3

Figure 12.3.: DMPC

12.6.2. Procedures for building a lipid using GLYCAM-06 parameters
The procedure described here allows a user to produce a single lipid molecule without consideration for axial
alignment. Lipid bilayers are typically built in the (x,y) plane of a Cartesian coordinate system, which requires
the individual lipids to be aligned hydrophilic “head” to hydrophobic “tail” along the z-axis. This can be done
relatively easily by loading a template PDB file that has been appropriately aligned on the z-axis.
The lipid described in this example is 1,2-dimyristoyl-sn-glycero-3-phosphocholine or DMPC. For this example,
DMPC will be composed of four fragments: CHO, the choline “head” group; PGL, the phospho-glycerol “head”
group; MYR, the sn-1 chain myristic acid “tail” group; and MY2, the sn-2 chain myristic acid “tail” group. See
the molecular diagram in12.3 for atom labels (hydrogens and atomic charges are removed for clarity) and bonding
points between each residue (dashed lines). This tutorial will use only prep files for each of the four fragments.
These prep files were initially built as PDB files and formatted as prep files using antechamber. GLYCAMcompatible charges were added to the prep files and a prep file database (GLYCAM_lipids_06h.prep) was created
containing all four files.
12.6.2.1. Example: Building a lipid with LEaP.

One need not load the main GLYCAM prep files in order to build a lipid using the GLYCAM-06 parameter set,
but it is automatically loaded with the default leaprc.GLYCAM_06j-1. Note that the lipid generated by this set of
commands is not necessarily aligned appropriately to create a bilayer along an axis. The commands to use are:
source leaprc.GLYCAM_06j-1 # source the leaprc for GLYCAM-06
loadamberprep GLYCAM_06_lipids.prep # load the lipid prep file
set CHO tail CHO.1.C5 # set the tail atom of CHO as C5.
set PGL head PGL.1.O1 # set the head atom of PGL to O1
set PGL tail PGL.1.C3 # set the tail atom of PGL to C3
lipid = sequence { CHO PGL MYR } # generate the straight-chain
# portion of the lipid
set lipid tail lipid.2.C2 # set the tail atom of PGL to C2
lipid = sequence { lipid MY2 } # add MY2 to the "lipid" unit
impose lipid {2} { {C1 C2 C3 O1 163} } # set torsions for
impose lipid {2} { {C2 C3 O1 C1 -180} } # PGL & MYR
impose lipid {2} { {C3 O1 C1 C2 180} }
impose lipid {2} { {O4 C1 C2 O1 -60} } # set torsions for
impose lipid {2} { {C1 C2 O1 C1 -180} } # PGL & MY2
impose lipid {2} { {C2 O1 C1 C2 180} }
# Note that the values here may not necessarily
# reflect the best choice of torsions.

222

12.6. Building oligosaccharides, lipids and glycoproteins
savepdb lipid DMPC.pdb # save pdb file
saveamberparm lipid DMPC.top DMPC.crd # save top and crd files

12.6.3. Procedures for building a glycoprotein in LEaP.
The LEaP commands given in this section assume that you already have a PDB file containing a glycan and
a protein in an appropriate relative configuration. Thorough knowledge of the commands in LEaP is required in
order to successfully link any but the simplest glycans to the simplest proteins, and is beyond the scope of this
discussion. Several options for generating the relevant PDB file are given below (see Items 5a-5c).
The protein employed in this example is bovine ribonuclease A (PDBID: 3RN3). Here the branched oligosaccharide assembled in the second example will be attached (N-linked) to ASN 34 to generate ribonuclease B.
12.6.3.1. Setting up protein pdb files for glycosylation in LEaP.

1. Delete any atoms with the “HETATM” card from the PDB file. These would typically include bound ligands, non-crystallographic water molecules and non-coordinating metal ions. Delete any hydrogen atoms if
present.
2. In general, check the protein to make sure there are no duplicate atoms in the file. This can be quickly done
by loading the protein in LEaP and checking for such warnings. In this particular example, residue 119
(HIS) contained duplicate side chain atoms. Delete all but one set of duplicate atoms.
3. Check for the presence of disulfide bonds (SSBOND) by looking at the header section of the PDB file. 3RN3
has four disulfide bonds, between the following pairs of cysteine residues: 26—84, 40—95, 58—110, and
65—72. Change the names of these eight cysteine residues from CYS to CYX.
4. At present, it is possible to link glycans to serine, threonine, hydroxyproline and asparagine. You must rename the amino acid in the protein PDB file manually prior to loading it into LEaP. The modified residue
names are OLS (for O-linkages to SER), OLT (for O-linkages to THR), OLP (for O-linkages to hydroxyproline, HYP) and NLN (for N-linkages to ASN). Libraries containing amino acid residues that have been
modified for the purpose are automatically loaded when leaprc.GLYCAM_06j-1 is sourced. See the lists of
library files in3.7 for more information.
5. Prepare a PDB file containing the protein and the glycan, with the glycan correctly aligned relative to the
protein surface. There are several approaches to performing this including:
a) It is often the case that one or more glycan residues are present in the experimental PDB file. In this case,

a reasonable method is to superimpose the linking sugar residue in the GLYCAM-generated glycan
upon that present in the experimental PDB file, and to then save the altered coordinates. If you use this
method, remember to delete the experimental glycan from the PDB file! It is also essential to ensure
that each carbohydrate residue is separated from other residues by a TER card in the PDB file. Also
remember to delete the terminal OH or OMe from the glycan. Alternately, the experimental glycan may
be retained in the PDB file, provided that it is renamed according to the GLYCAM 3-letter code, and
that the atom names and order in the PDB file match the GLYCAM standard. This is tedious, but will
work. Again, be sure to insert TER cards if they are missing between the protein and the carbohydrate
and between the carbohydrate residues themselves.
b) Use a molecular modeling package to align the GLYCAM-generated glycan with the protein and save

the coordinates in a single file. Remember to delete the terminal OH or OMe from the glycan.
c) Use the Glycoprotein Builder tool at http://www.glycam.org. This tool allows the user to upload protein

coordinates, build a glycan (or select it from a library), and attach it to the protein. All necessary
AMBER files may then be downloaded. This site is also convenient for preprocessing protein-only
files for subsequent uploading to the glycoprotein builder.

223

12. LEaP
12.6.3.2. Example: Adding a branched glycan to 3RN3 (N-linked glycosylation).

In this example we will assume that the glycan generated above (“branch.pdb”) has been aligned relative to the
ASN 34 in the protein file and that the complex has been saved as a new PDB file (e.g., as “3rn3_nlink.pdb”). The
last amino acid residue should be VAL 124, and the glycan should be present as 4YB 125, 4YB 126, VMB 127,
OMA 128 and OMA 129.
Remember to change the name of ASN 34 from ASN to NLN. For the glycan structure, ensure that each residue
in the PDB file is separated by a “TER” card. The sequence command is not to be used here, and all linkages
(within the glycan and to the protein) will be specified individually.
Enter the following commands into xleap (or tleap if a graphical representation is not desired). Alternately,
copy the commands into a file to be sourced.
source leaprc.GLYCAM_06j-1 # load the GLYCAM-06 leaprc for ff14SB
source leaprc.ff14SB # load the modified ff12 force field
glyprot = loadpdb 3rn3_nlink.pdb # load protein and glycan pdb file
bond glyprot.125.O4 glyprot.126.C1 # make inter glycan bonds
bond glyprot.126.O4 glyprot.127.C1
bond glyprot.127.O6 glyprot.128.C1
bond glyprot.127.O3 glyprot.129.C1
bond glyprot.34.SG glyprot.125.C1 # make glycan -- protein bond
bond glyprot.26.SG glyprot.84.SG # make disulfide bonds
bond glyprot.40.SG glyprot.95.SG
bond glyprot.58.SG glyprot.110.SG
bond glyprot.65.SG glyprot.72.SG
addions glyprot CL 0 # neutralize appropriately
solvateBox glyprot TIP3P BOX 8 # solvate the solute
savepdb glyprot 3nr3_glycan.pdb # save pdb file
saveamberparm glyprot 3nr3_glycan.top 3nr3_glycan.crd # save top, crd
quit # exit leap

224

13. Reading and modifying Amber parameter files
This chapter describes the content of Amber parameter files, along with details about ParmEd (which can be
used to examine and modify prmtop files) and paramfit (which can be used to fit force fields to quantum mechanical
and other target data).

13.1. Understanding Amber parameter files
Romain M. Wolf, Jason Swails, and David A. Case
This chapter provides a short description of Amber-compatible force field parameter files is given. Only the
actual data in parameter (*.dat) files are discussed. The special issue of deriving partial charges is not addressed.
Also, more complex subjects dealing with parameters for implicit solvent (GB or PB) or polarisability computations are skipped. This text is meant as a documentation for users who want to understand parameter files, and in
some cases might be tempted to change or add some parameters. Most of the following documentation is found in
bits and pieces at various Amber-related sites and in tutorials or original Amber manuals and these various sources
have been helpful to put together this hopefully concise documentation.

13.1.1. Parameter Transfers between Force Fields
Transferring parameters from one force field to another must respect the underlying functional form, the units in
which parameters are expressed in the parameter files, and also the exact procedures on how individual parameters
were obtained. In addition, attention must be paid to the methods used to deduce partial charges. Force fields
are self-consistent, i.e., all terms are interrelated and their actual values depend on the way they were derived.
Therefore, any parameter transfer between different force fields is dangerous, even when the functional form is the
same (or looks as if it were...).
Torsion terms are the most critical. Many torsion barriers and profiles are not easily assessed experimentally
and are often deduced from ab initio quantum mechanical (QM) computations on small fragments. Since QM
calculations offer many possibilities, the exact nature of these calculations (basis sets, Hartree-Fock and/or density
functionals, etc.) used to derive parameters should be known.
Special care must also be applied to 1-4 interactions, i.e., interactions between atoms separated by exactly three
consecutive bonds. Most Amber force fields for example assume that 1-4 interactions get a special treatment. See
section 13.1.6 for details. In many other force fields, the special treatment of 1-4 interactions is either different or
non-existent. This has an immediate influence on the torsion terms and resulting conformation energies. Therefore,
before transferring torsion terms, van der Waals parameters and partial charges from other force fields, check the
special treatment of 1-4 interactions in the source and the target force field.

13.1.2. How Amber Routines Use the Parameter Files
Amber routines that perform actual calculations (sander, pmemd, etc...) do not read parameter files directly.
They use a special file type, the parameter-topology file (parmtop from now on), which contains all the information
required by the various energy functions in the computation routines. The parmtop file is specific to the molecular
system for which it was created and is directly related to the second required file, the coordinate file.1 Smallest
changes to the system (adding or removing atoms, or even changing the order of atoms in the coordinate file)
render the parmtop useless.
1 This

file can be in the Amber coordinate ’crd’ file format or, for some applications, also in PDB format.

225

13. Reading and modifying Amber parameter files
Although parmtop files are pure ASCII files, changing parameters directly in them by standard text editors is
strongly discouraged. In the worst case, computations will run without any warnings, but results might be totally
flawed. The safest way to generate parmtop files is to use an Amber tool like tleap that has been used, tested, and
enhanced over a number of years and usually generates correct parmtop files, provided that the input is correct
and that all required information is available via fragment libraries and parameter files. The latest AmberTools
12.0 version (April 2012) includes the ParmEd python script of Jason Swails which is very useful to examine or
post-process parmtop files. However, only users with detailed knowledge on the exact format of parmtop files
should dare fiddling around with this data type.

13.1.3. "*.dat" and "frcmod.*" Files
The standard parameter files with the .dat extension are located in the folder $AMBERHOME/dat/leap/parm.
Adding or changing parameters directly in the parameter files delivered with an Amber distribution is not a good
idea for the following reasons: (a) you might mess up the parameter file, (b) you might have trouble to remember
and find your changes later and add confusion when publishing results, (c) subsequent updates or patches might
overwrite your changes.
In the above mentioned folder, there are also various frcmod.* files. They have basically the same format as
the parameter *.dat files. See some of the examples provided in the Amber distributions. These files can be read
into tleap exactly like the standard *.dat files. They merge the default parameters in the *.dat file with the new
parameters in the frcmod.* files. More important, if the same parameters already exist in the *.dat files, the
parameters in the frcmod.* files overwrite the default *.dat parameters. This offers a handy way to add new or to
change original parameters without ever touching the default parameter files. Just make sure to read the respective
frcmod.* files in tleap when the new or altered parameters should be used.

13.1.4. Parameters Required for Amber Force Fields
The simplest form of the Amber force field (neglecting implicit solvent or polarisation terms) uses the following
Hamiltonian:

Etotal

∑

kb (r − r0 )2

bonds

∑

kθ (θ − θ0 )2

angles

Vn [1 + cos(nφ − γ)]

∑
dihedrals
N−1

∑ ∑

i=1 j=i+1

Ai j Bi j qi q j
− 6 +
R12
Ri j εRi j
ij

(13.1)
In this equation, the terms kb , r0 , kθ , θ0 ,Vn , γ, Ai j , Bi j are parameters to be specified in the parameter files mentioned in section 13.1.3 for the various Amber force fields.2 The meaning of these different parameters is outlined
in the following sections.
Equation 13.1 does not have a special term for out-of-plane motions. Amber routines handle these terms through
the same formulation as the torsion terms (see section 13.1.6).
Partial charges (qi , q j in equation 13.1), although parameters also, do not appear in parameter files, but are
assigned differently (see 13.1.7).
2 Note

that equation 13.1 does not use the (physically more correct)
in the actual parameter files.

226

kb kθ
2 , 2

, and

Vn
2

notations because it refers to the constants as they appear

13.1. Understanding Amber parameter files

13.1.5. Atom Types
Amber atom types can be one or two characters long. Uppercase (standard protein and nucleotide force fields),
lowercase (GAFF General Amber Force Field ), and mixed upper-lowercase (GLYCAM sugar force field) are
allowed. Obviously, atom types must have a single, unique, definition.
If considering the definition a new atom type, think about the consequences. Of course, an atom type with an
identical name must not already exist in one of the standard force fields used in the Amber community. Depending
on how often and in how many combinations the atom type might occur, be also aware of the rather large number
of additional parameters that might be required. Especially for bond angles, this number can grow very rapidly.
A new atom type definition, if required, must be clear and precise. It should also be possible to treat the definition
in an automatic atom-type assignment procedure. Requiring users to visually verify and to change atom types by
hand will cause trouble and will make it impossible to use the force field in automatic procedures that should not
require user intervention for this task.

13.1.6. Bonded Interaction Terms
Bond Stretching Terms

The first row in equation 13.1 (page 226) is the harmonic term for bond stretching. In Amber-type parameter
files, the force constant kb is given for energy values in kcal/mol, with bond lengths in Å. The following line shows
an example from the GAFF force field file gaff.dat.
The bond between a sp3 carbon (c3) and a hydroxyl oxygen (oh) has a default (equilibrium) value of 1.426 Å
and a force constant of 314.1 kcal/mol/Å2 .
c3-oh 314.1 1.4260 SOURCE1 914 0.0129

The entrance in the parameter file starts with the definition of the bond (atomtype1 hyphen atomtype2), followed by the force constant kb (in kcal/mol/Å2 ) and the equilibrium bond length r0 (in Å). Only the first three
fields are relevant for computations. The other fields on the line above are mainly documentation.
As stated before, atom types in Amber FFs cannot have more than two characters. But if they have only one
character (e.g., a carbonyl carbon atom c), entries with a one-letter atom type must look like this:
c -oh 466.4 1.3060 SOURCE1 271 0.0041

i.e., the space is after the atom type, before the hyphen.
Starting with a space like on the next line might lead to problems.
c-oh 466.4 1.3060 SOURCE1 271 0.0041

This holds for all parameter file entries that use hyphens to separate atom types, i.e., also angle and torsion terms
(see following sections).
Angle Bending Terms

Angle bending terms are parameterised by a force constant kθ in kcal/mol/radian2 and an equilibrium angle
value θ0 in degrees. They have the format as shown below:
c3-c3-oh 67.720 109.430 SOURCE3 48 1.5023

The middle atom c3 is bonded to another c3 and to a hydroxyl oxygen oh. The equilibrium bond angle θ0 is 109.43
degrees and the force constant is 67.720 kcal/mol/radian2 . Note that internally, angle deviations are computed in
π-radian2 . The parmtop files also express the default ’equilibrium’ bond angles in radians. For example, the angle
of 109.43 degrees is internally represented as 1.9099 π-radians. Using degrees in the original parameter files is
obviously more convenient. Anything after the third field, the equilibrium angle, is mainly documentation and not
required.

227

13. Reading and modifying Amber parameter files
Torsion Terms

The third row in equation 13.1 is the usual Fourier-series expansion for torsional terms. In Amber parameter
files, these entries require a careful explanation:
First, many torsion terms contain generic entries, using the notation ’X’ for ’any atom’. These terms are used
when the parameter file does not contain more specific terms for the same torsion. They are combined with
explicit terms when present. Entries with generic ’X’ atoms must always come before the more specific ones in the
parameter files.
Second, Amber parameter files use a special notation for torsions that require more than one torsional term (see
example towards the end of section 13.1.6).
Third, the parameter file entry not only contains the torsion barrier term Vn (in kcal/mol), the phase γ (degrees)
and the periodicity n, but also a divider (integer) which splits the torsion term into individual contributions for
each pair of atoms involved in the torsion.
Fourth, torsion entries can also contain information about the special scaling of 1-4 non-bonded interactions
(see section 13.1.6 on page 230).
Consider the following example, the default term for the torsion around a Csp3 -Csp3 single bond:
X -c3-c3-X 9 1.400 0.000 3.000 JCC,7,(1986),230

The five relevant terms on this line are:
1. the definition (X -c3-c3-X)
2. the divider (9)
3. the barrier term (1.400)
4. the phase (0.000)
5. the periodicity (3.000)
Fields after the periodicity are mainly comment, except for the special flags SCNB and SCEE, that, if present,
govern the special treatment of 1-4 non-bonded interaction (see section 13.1.6)
The torsional barrier term (the actual barrier divided by two) is 1.400 and the periodicity is 3. The phase is zero
in this example, meaning that a maximum energy is encountered at zero degrees. A phase of 180 degrees on the
other hand means that there is a minimum at 180 degrees. The divider is 9 because each Csp3 has three X attached
to it and each X ’sees’ three X attached to the other Csp3 (3 × 3 = 9).
For a torsion angle φ (defined as X-c3-c3-X) of -60, 60, or 180 degrees, the torsion energy term would be zero:
1.4
× [1 + cos(3 × φ − 0.0)] = 0
9

(13.2)

This corresponds to the staggered conformation, i.e., the lowest energy state in a X3 C-CX3 connectivity like for
example ethane (H3 C-CH3 )
By rotating around the C-C bond, an eclipsed conformation where the X are exactly opposed is encountered
three times (periodicity = 3), namely at φ = 0, 120, or 240 (-120) degrees.
1.4
× [1 + cos(3 × φ − 0.0)] = 0.3111
9

(13.3)

.
Since the divider is 9, we have to multiply the value of 0.3111 by 9 to get the full torsional barrier, i.e., 9 ×
0.3111 = 2.8 kcal/mol.3 This might be used for ethane for example and would be close to the experimental torsion
barrier (ca. 3 kcal/mol).
In GAFF however, there is also a specific term for hc-c3-c3-hc that would come into play for ethane. In this
case, the divider is 1, because the term is fully defined.
3 The

228

actual barrier value of 2.8 kcal/mol here is twice the barrier term of 1.4 in the parameter file.

13.1. Understanding Amber parameter files
hc-c3-c3-hc 1 0.15 0.0 3. Junmei et al, 1999

Thus, using GAFF for ethane, this term counts 9 times because there are nine [hc,hc] pairs seeing each other.
Instead of equation 13.3, one would use
0.15 × [1 + cos(3 × φ − 0.0)] = 0.3000

(13.4)

i.e., the total torsional term in ethane would be 9 × 0.3 = 2.7 kcal/mol. The experimental torsional barrier value
of ca. 3 kcal/mol would be reached because of the additional van der Waals and Coulomb repulsion terms between
the staggered hydrogens.
Assume a connectivity for which some terms are fully defined (all four atom types are specified) while no
specific entry is given for others. In that case, the equations are combined. The specific terms are counted once
(divider = 1) and the remaining general terms are added according to
Vbarrier
× [1 + cos(periodicity × φ − phase)]
(13.5)
divider
Things get more complex when the Fourier series has more than one term. A typical example would be the
rotation around an amide bond R1-NH-C(=O)-R2. In this case, the trans amide (H and O on opposite sides,
φ = 180◦ ) is preferred over the cis-amide (H and O on the same side, φ = 0). The entry in the GAFF parameter
file for this torsion is
hn-n -c -o 1 2.50 180.0 -2. JCC,7,(1986),230
hn-n -c -o 1 2.00
0.0 1. J.C.cistrans-NMA

If the torsion definition has a "negative" periodicity (-2 in the case above), it tells programs reading the parameter
file that additional terms are present for that particular connectivity. The equation to be applied for hn-n -c -o is:
Etorsion = 2.00 × [1 + cos(1 × φ − 0.0)] + 2.50 × [1 + cos(2 × φ − 180.0)]

(13.6)

Equation 13.6 prefers the trans amide (φ = 180◦ ) over the cis amide (φ = 0) by 4 kcal/mol considering the
torsion term alone. However the more favourable Coulomb term (the 1-4 attractive interaction between the negative
carbonyl oxygen and the positive amide hydrogen) reduces the overall preference for the trans conformation close
to the experimental value of ca. 2 kcal/mol.
In addition, the following general terms have to be applied for the torsions involving R1 and R2 in the peptide
bond R1-NH-C(=O)-R2, in order to compute the high torsional barrier of an amide bond:
X -c -n -X 4 10.000 180.000 2.000

Torsional terms are obviously the most difficult part to parametrize in a force field. They are in a way the last
rescue to get torsional barriers right, after all other terms have been adjusted. Therefore, their transfer from one
force field to the other is always most risky and acceptable only if all other involved terms in two force fields are
very similar. Transferability must always be validated.
Out-of-Plane Terms

Out-of-plane terms are handled via a Fourier term, similar to the torsion terms. But the four involved atoms are
not serially (linearly) bonded, they are "branched". The "central" atom is the atom that is forced into the plane of
the other three. For example, to keep a carbonyl group R1-C(=O)-R2 planar, the central C atom must be forced
into the plane of the other three connected items R1, R2, and O. The entry in the GAFF parameter file for this
term is
X -X -c -o 10.5 180. 2. JCC,7,(1986),230

Note that in Amber the central atom type (here c) is the third in the definition.The order of the remaining atoms
should (by definition) be alphabetic in atom type. The phase is always 180◦ . In all-atom force fields, the periodicity
is always 2.
Out-of-plane terms are the only terms that are allowed to be "missing" in Amber parameter files. Common ones
are added automatically by tools like tleap. In many cases, these terms are "cosmetics" that avoid "in principle"

229

13. Reading and modifying Amber parameter files
planar structures from getting distorted under the influence of other forces (e.g., fused rings, planar nitrogens
with three substituents, etc...). The actual parameterisation is often intuitive and for many entries, the ("generic")
parameters are identical.
1-4 Non-Bonded Interaction Scaling

Figure 13.1.: 1-4 Interactions between atoms "1" and "4".
Non-bonded interactions between atoms separated by three consecutive bonds (as schematically shown in Figure
??) require a special treatment in Amber force fields. Although referring to non-bonded interactions, scaling
information is included in the torsion terms part of the parameter files.
By default, vdW 1-4 interactions are divided (scaled down) by a factor of 2.0, electrostatic 1-4 terms by a
factor of 1.2. These are default values for the protein force fields and GAFF, but not for sugar force fields GLYCAM_06EP and GLYCAM_06, for example, in which these interactions are not scaled at all.
Without any additional information, programs like tleap, used to prepare parmtop files, assume that the
standard scaling mentioned above is to be applied. However, this default can be overwritten in the torsion section
of the parameter file. An example is shown below for torsional terms in the GLYCAM_06j force field:
S -Ng-Cg-H1 1 2.00 0.0 1. SCEE=1.0 SCNB=1.0 N-Sulfates
S -Ng-Cg-Cg 1 0.0 0.0 -3. SCEE=1.0 SCNB=1.0 N-Sulfates

The special notation SCEE=1.0 SCNB=1.0 following the standard torsion terms4 will tell tleap to prepare a
parmtop file which transfers these data into a special section, as shown below:
%FLAG SCEE_SCALE_FACTOR
%FORMAT(5E16.8)
scaling factors are entered here....
%FLAG SCNB_SCALE_FACTOR
%FORMAT(5E16.8)
scaling factors are entered here....

When using standard Amber force field parameter files as delivered with AmberTools, the user does not need to
care about this. However, when adding additional parameters, especially torsion terms, one should be aware of
these scaling factors and decide if they should be default or altered.

13.1.7. Non-Bonded Terms
Van der Waals Parameters

The standard formulation of the 6-12 Lennard-Jones potential Vi, j between two atoms i and j is:
4 In

230

this case, the fields coming after the periodicity (field 5), i.e., fields 6 and 7 are also read and are not ’just’ comment!

13.1. Understanding Amber parameter files

"
Vi, j = 4εi, j

σi, j
ri, j

σi, j
−
ri, j

6 #
(13.7)

Here, ri, j is the distance separating the two atoms, εi, j is the depth of the potential well for the interaction of
atoms i and j, and σi, j is the distance where the potential is exactly zero, i.e., where ’repulsion’ starts for the two
atoms. Both εi, j and σi, j are specific for the pair of atoms (or more precisely, ’atom types’).
Another possible formulation of Vi, j , relating to the concept of van der Waals radii, is:
"

#
Rmin 12
Rmin 6
Vi, j = εi, j
−2
(13.8)
ri, j
ri, j
In this case, Rmin is the sum of the van der Waals radii, Ri + R j of atoms i and j, the contact distance at which
the potential is at its minimum, i.e., at a value of −ε.

Figure 13.2.: Example of Lennard-Jones potential: the used data are those for the c3 atom type in the gaff force
field (vdW radius Rmin = 1.908 Å, ε = 0.1094 kcal/mol)
Combining equations (13.7) and (13.8) gives for the relation between σ and Rmin :
Rmin = 21/6 σ or σ = 2−1/6 Rmin

(13.9)

In force fields, the ’A,B’ notation of the Lennard-Jones potential is commonly used:
Vi, j =

Ai, j Bi, j
− 6
ri,12j
ri, j

(13.10)

where Ai, j and Bi, j are specific parameters for atom type pairs i and j. The meaning of Ai, j and Bi, j are easily
deduced from equation (13.7):
A = 4εσ 12 and B = 4εσ 6

(13.11)

or, in terms of Rmin , using equation (13.8):
6
A = εR12
min and B = 2εRmin

(13.12)

Van der Waals data in Amber force fields are given for each atom as a single data pair, a radius Rmin (’van der
Waals’ radius in Å) and an energy ε (kcal/mol) representing the depth of the potential well.

231

13. Reading and modifying Amber parameter files
These values are given at the end of the force field parameter files. In protein force fields, lines above these data
show equivalences. For example the line
N NA N2 N* NC NB NT NY

indicates that all atom types following N (the amide nitrogen) inherit the same Lennard-Jones parameters. Thus,
no entry for NA, N2, ... has to be given explicitly.
For Amber force fields, cross terms involving different atom types i and j are evaluated according to the
Lorentz/Berthelot mixing rules:
σi, j = 0.5(σi,i + σ j, j ) or Rmin,i, j = 0.5(Rmin,i + Rmin, j )

(13.13)

(13.14)

εi, j =

εi,i · ε j, j

The parmtop file entries are in ’A’ and ’B’ terms to be used directly with equation 13.10, transforming the
[Rmin ,ε] data pairs from the parameter files.
As an example, consider ethanol (CH3 CH2 OH) with the GAFF force field. There are five different GAFF atom
types. Below are shown the corresponding [Rmin ,ε] data pairs, as found in the gaff.dat parameter file:
h1
hc
ho
oh

1.3870
1.4870
0.0000
1.7210

0.0157
0.0157
0.0000
0.2104

Veenstra et al JCC,8,(1992),963
OPLS
OPLS Jorgensen, JACS,110,(1988),1657
OPLS c3 1.9080 0.1094 OPLS

Note that there are three different hydrogen types: hc, the default H atom connected to an aliphatic carbon, h1, a
hydrogen type connected to an aliphatic carbon with one electronegative substituent (the oxygen in this case), and
the hydroxyl hydrogen ho (for which van der Waals interactions are neglected in Amber).
Partial Charges

For Amber force fields, partial charges do not appear in parameter files. For proteins and nucleic acid force fields
that use fragment (residue) libraries, partial charges are pre-defined and have been computed from electrostaticpotential fitting of high-level an initio QM. They are automatically assigned by tools like tleap. Library files are
found the folder
$AMBERHOME/dat/leap/lib.

Below is shown the alanine (ALA) residue of the library file all_amino94.lib:
"N" "N" 0 1 131072 1 7 -0.415700
"H" "H" 0 1 131072 2 1 0.271900
"CA" "CT" 0 1 131072 3 6 0.033700
"HA" "H1" 0 1 131072 4 1 0.082300
"CB" "CT" 0 1 131072 5 6 -0.182500
"HB1" "HC" 0 1 131072 6 1 0.060300
"HB2" "HC" 0 1 131072 7 1 0.060300
"HB3" "HC" 0 1 131072 8 1 0.060300
"C" "C" 0 1 131072 9 6 0.597300
"O" "O" 0 1 131072 10 8 -0.567900

The partial charges for each atom are given in the last field of each line.
For the GAFF force field, there are various options to compute partial charges, the AM1-BBC method being
probably the best trade-off between quality and speed. There are other file types that can contain user-specified
partial charges, e.g., SYBYL mol2 files. See the antechamber documentation for details.
In parmtop files, partial charges are not entered as fragments of the electron charge, but are multiplied by the
square-root of 332.05 (= 18.22), because the factor 332.05 converts the Coulomb energy into kcal/mol when using
fragments of the electron charge in the Coulomb term of equation 13.1.

232

13.2. ParmEd

13.1.8. Final Remarks
Most parameters in Amber force fields have been tested on a large variety of structures. In rare cases, situations
are encountered where structures look "strange" or where results are obviously wrong. One should first look
into details of the simulation conditions and settings before blaming the problem on actually flawed force field
parameters. Simple test cases are often helpful to resolve the enigma.
When changing or adding parameters and later publishing results, new parameter should be mentioned. Also,
the Amber developers team should be notified about possibly problematic parameters. This ensures that potential
errors are corrected via patches in later versions and it will help the entire user community.

13.2. ParmEd
ParmEd (parmed.py) is a topology file editor written in Python that enables high level control of the primary
force field file in Amber: the prmtop file. ParmEd will modify the topology file and produce a new topology file
that will work with sander, pmemd, and NAB programs, and provides options unavailable otherwise. ParmEd
currently supports topology files created with both tleap and chamber (but support is very limited for those created
with tinker_to_amber).

13.2.1. Running parmed.py
parmed.py is used in a manner very similarly to cpptraj.
usage: parmed.py [-h] [-v] [-i FILE] [-p ] [-O] [-l FILE]
[--prompt PROMPT] [-n] [-e] [-s] [-r]
[] [

Amber14

Navigation menu

Versions of this User Manual:

Views

Navigation