Guide

User Manual:

Open the PDF directly: View PDF .
Page Count: 21

Lithium ELNES with WIEN2k

October 12, 2018

Contents

1 Tools 2

2 Path 2

3 Setup 3

3.1 LoadStructure .................................. 3

3.2 Initialization.................................... 4

4 Convergence 6

4.1 Cellparameters:.................................. 6

4.2 Setting the Muﬃn Tin Size . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

4.3 K point and RKMax convergence . . . . . . . . . . . . . . . . . . . . . . . . 8

5 Density of States 9

6 TELNES3 10

6.1 Monopoleeﬀects ................................. 12

7 Core Hole 13

8 Density Calculations 16

8.1 Xcrysden...................................... 16

8.2 Critic2....................................... 16

9 Common Errors/Issues Encountered in Wien2k 19

9.1 SettingRMT/RKMax .............................. 19

9.2 Ghostbands .................................... 19

9.3 NNinOptimization................................ 20

9.4 NRandNT .................................... 21

9.5 GMax Value less than Gmin . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

9.6 Ordering...................................... 21

1 Tools

Here’s a list of software I use to run simulations. I’ve put some installation instructions for

linux as needed.

•Wien2k - If you are reading this guide, you either already have it installed somewhere,

or should ﬁgure out how to do that before buying a license. Otherwise, read the

userguide where the installation process is well explained.

•VESTA: http://jp-minerals.org/vesta/en/download.html. Download the .rpm

ﬁle, install it with your package manager in the directory it was downloaded to, eg.

“sudo apt-get vesta...rpm”

•Xcrysden: http://www.xcrysden.org/Download.html. Download the appropriate

tar for your operating system, and make sure to select the semishared version. Unzip

it, and just run “$./xcrysden”. Add the executable to your path in .bashrc to be able

to run it anywhere.

•Critic2: https://github.com/aoterodelaroza/critic2. Download the zip from

GitHub, unzip where you want to install it (eg ∼/Programs or something), use your

package manager (dnf/apt-get) to install autoconf, automake, and your favourite ﬂavour

of fortran (if unsure, it is probably gfortran). Then run the 4 commands from the

readme: “autoreconf -i”, “./conﬁgure”, “make”, and “(sudo) make install”. You may

or may not need the sudo for the last one.

2 Path

Here is the overall list of things that need to be done to rigorously obtain ELNES spectra

with WIEN2k:

•Load a structure

•Initialize

•Volume convergence/structure optimization

•Set RMT values

•K point/RKmax convergence

•Calculate DOS

•Calculate ELNES

•Include Core Hole, if indicated by DOS, calculate PDOS

•Calculate Screening from density

•Calculate ELNES from partial hole

3 Setup

3.1 Load Structure

There a couple things required to start ELNES simulations. Foremost is a crystal struc-

ture. This can be obtained from the literature, XRD, or alternatively Materials Project:

https://materialsproject.org/. Download a cif (the primative cell typically) or enter

the coordinates directly into the wien2k struct gen tool, set the rmt and save the ﬁle.

Make a new Wien2k session:

Create/change a working directory, and change the session information for parallel cal-

culation.

You will also need to make a “.machines” ﬁle which you can either steal from one of my

directories, look in the user guide and make your own, cut and paste the one from below, or

wait for w2web to automatically generate one at some point after it inevitably crashes on

something. Sample .machines ﬁle:

#=======================================

#This i s a v a l i d . machines f i l e

granularity :1

1 : l o c a l h o s t #as many o f t h e se l i n e s as you want cpu c or e s running

1 : l o c a l h o s t #i d e a l l y pi ck a mu lt ip le of kp oint number

1 : l o c a l h o s t #so something l i k e 5 c or e s f o r 10 k po in ts .

1: localhost

Next, go make a struct ﬁle, with struct gen, either by importing the cif, or entering

the positions manually. Use VESTA (drag n’ drop the .struct ﬁle) to make sure that the

structure is what you’d expect. In the Li2O case this looks like:

3.2 Initialization

The next step is initialization. This is best done via the command line as it give a little more

ﬂexibility and intuition than in w2web, and will need to be run multiple times. The ﬁrst

time will be to set up the calculation for volume optimization, which needs smaller muﬃn

tins. Run init lapw and go through the following steps:

•setrmt. Sets the muﬃn tin size on the atoms. Reduce the sphere size by 5-10% (I

used 7%), using either old or new scheme and accept.

•nn. Checks for overlapping muﬃn tins. Enter “2.0”, close the ﬁrst ﬁle and use the

new NN ﬁle if suggested, run nn with 2.0 again, look at how much “wiggle room” you

have on the spheres, by comparing the “sums to” and the “nn-dist” for every atom,

see below:

•sgroup. Veriﬁes the space group. Again, accept any changes the program makes, this

ﬁrst run is all about reducing the cell to the smallest unit cell. Again, the ﬁles can be

largely ignored at this point, unless there is an indication of a Bravais lattice change, in

which case, accept the new struct ﬁle. If so, nn and sgroup will run again with “nice”

results.

•symmery. Generates all the symmetry operations. Run it and continue (enter “c”)

•lstart. Sets the spin state, selects which XC kernel, and deﬁnes cutoﬀ between core and

valence states. Accept default spins (up, the no spin case), unless there is a transition

metal involved. Select GGA PBE as the XC potential, again, unless there is reason to

suspect otherwise. For the purposes of volume optimization, the defualt energy value

(-6.0 Ry) works well.

•kgen Set the RkMax value in case.in1 st from 7 to the desired value:

and then pick a k point number. Both of these values should initially be taken for fast

convergence, in this case I chose RkMax=6.0, and 16 k points (Li2O is an insulator,

for metals 1000 k points is a good starting point).

•Dstart: make sure to pick non spin polarized, unless there is reason to believe other-

wise (is there a transition metal in the sample?)

Assuming there were no warnings in the ﬁnal run through init lapw, the case can begin

to be converged.

4 Convergence

Ideally, every variable should be converged regarding the simulation. Typically this is cell

parameters, k points and Rkmax. The ﬁrst step is to just make sure the calculation converges,

running it with a small RKmax and few kpoints and making sure it ﬁnishes without error.

Once ﬁnished altering parameters in init lapw, run $run lapw -p to start a calculation. If

errors occur, search for them in the userguide, or in the ﬁnal chapter of this guide. Otherwise

proceed with convergence.

4.1 Cell parameters:

This process works best from W2Web, as described in the tutorials and using the muﬃn

tins from setrmt reduced by a healthy percentage (∼5-10%) to avoid nn errors (you may

need to rerun init lapw). As a number (5-11+) of calculations are run in this process, a

low number of k points and RKMax values is ideal here. For Li2O, I used 16 kpoints and an

RKmax of 6. In the x “optimize” tab, choose what you want to optimize, the ﬁrst option

(volume) works well, unless there are suspicions otherwise. Enter a range of values of test

volumes, see picture:

Also make sure to edit “optimize.job” to enable parallization by moving the “#” to after

the “-p” in the run lapw line:

You can then “run optimize.job” from w2web. This job is okay to run in the background,

which means the browser can be closed without the job stopping. When it is done, plot the

“Energy vs Volume,” which should look something like this:

The important feature here is that the overall trend of the curve is clear, if too few k

points are chosen, the points will be “noisy.” In the case of Li2O, the used lattice parameter

of 4.6590˚

A results in the minimum energy, so no further changes are needed. If we needed a

structure with a diﬀerent lattice parameter, search the case directory for all the struct ﬁles

and rename the appropriate one to case.struct, eg “Li2O vol 3.0.struct →Li2O.struct”.

Alternatively, cut and paste the lattice parameters from this ﬁle into the w2web structgen

tool.

4.2 Setting the Muﬃn Tin Size

This is the ﬁrst step of the process that is speciﬁc to calculating the Lithium ELNES. In

order to have the ability to insert core holes, the lithium 1S state needs to be treated as a

core state. This requires playing with the muﬃn tin radius and the energy cutoﬀ value in

lstart. Rerun init lapw with no reduction in muﬃn tin size, and during lstart, use a cuutoﬀ

energy of -3.5 Ry (The Li 1 S state has an energy of -3.8 Ry). Investigate the case.outputst

ﬁle (it should pop up) for the following lines for the lithium atom and look at the 1S states:

E−up (Ry) E−dn (Ry) Occupancy q/ sp her e core−state

1S −3.801968 −3.785309 1 . 00 1. 0 0 0.9 90 4 T

2S −0.236711 −0.003297 1 . 00 0. 0 0 0.0 59 5 F

These indicate the core states (T/F), their energy levels (in this case ∼-3.8eV) and how

much the electrons in these states are contained in the muﬃn tins (0.9904). As this is less

than 1, it means that some 1S lithium electron is leaking out of the muﬃn tins, which is

why there should now be all kinds of warnings popping up. So go ahead and “ctrl-c” out of

init lapw.

To ﬁx the leakage problem, the Lithium muﬃn tins need to be bigger. Ideally, they should

be just big enough to hold all of the 1S electrons, without making them too diﬀerent from the

other muﬃn tins, as the larger this diﬀerence, the harder things get to calculate/converge.

In this case, try Li=2.0, O=1.8 and try init lapw again making sure to discard the suggested

muﬃn tins from setrmt. If that still didn’t work (lstart still has leakage errors), keep going

until it does. If the RMT values are too diﬀerent (eg if RMTLi ≈1.5RMTO) they will cause

errors (ghostbands) further along. The solution then is to try and minimize the leakage (and

ignore the warnings) and acknowledge that it might be unavoidably causing artifacts. Once

the spheres are set in the struct ﬁle, rerun init lapw.

4.3 K point and RKMax convergence

To converge these parameters, again start with very low values and then increase them,

checking the total energy to determine when they are converged. Generally k points are

easier to converge, so start with them and then move on to RKMax. The procedure for

converging both of these values is:

•set/increase kpoints or RKMax, either by re-running “x kgen” or editing case.in1 and

case.in1 st.

•Run the scf cycle using “run lapw -p -NI”, the NI ﬂag means it will continue from

where the previous calculation left oﬀ which can save time. The ﬁnal converged choice

should still be run from scratch though (rerun init lapw, rm *broyd* before running

run lapw)

•Check the energy in case.scf. To do this, ﬁnd it in “scf ﬁles” on w2web and use ctrl+f

in your browser to search the ﬁle for “:ene” , which should appear in a line that looks

like:

:ENE : ∗∗∗∗∗∗∗∗∗∗ TOTAL ENERGY IN Ry = −180.82588608

There will be one of these lines for each scf cycle, so ﬁnd the last one in the document

and note the energy.

•loop through the ﬁrst 3 steps until the energy no longer changes signiﬁcantly when you

increase the kpoints/RKMax. A table is useful here to track these eﬀects eg:

kpoints RKMax Energy

16 6.0 -180.82588608

32 6.0 -180.85055185

64 6.0 -180.85166466

32 7.0 -180.85723055

32 8.0 -180.85860820

32* 8.0* -180.85977640

Depending on the case, the energy will only converge so far. For the above table, I would

choose 32 k points and RKMax of 8.0. there were only minimal changes (<0.001Ry) to the

energy beyond those. The starred case indicate the calculation being run from scratch.

5 Density of States

In order to determine whether or not a core hole is necessary, the ﬁrst step is to analyze the

DOS. These are best calculated in w2web, clicking the buttons.

To start, run the optional steps and increase the upper energy limit in case.in1 from 1.5

to ∼2-3.5, see picture. This value deﬁnes how many higher energy states are included in

Ry (1 Ry ≈13.6 eV, 1.5Ry ≈20eV). Therefore, to obtain the correct DOS and ELNES for

features more than 20eV from the onset, this should be increased to match.

Then boost the k points value by an order of magnitude to pull out the ﬁner features in

the DOS, and run x lapw -p and x lpaw2 -qtl -p. Then select the total DOS of each atom

in case.int, using the w2web interface:

Run x tetra and plot the DOS with dosplot, which should give you something like this:

What we are looking for here is whether the conduction band has a large component

of unoccupied lithium states, which it does in this case, indicated by the arrows. If the

unoccupied states were purely oxygen in nature, no core hole eﬀect would be expected, as

there would be no states to alter eﬀecting the ELNES. For more insight on this behavior,

the reader is referred to Mauchamp, Jaouen and Schattschneider (2009) “Core-hole eﬀect in

the one-particle approximation revisited from density functional theory.”

As there is indeed a large component of lithium states in the low conduction band, a core

hole is necessary. Before that, we calculate an initial ELNES for comparison later.

6 TELNES3

Once the calculation is converged, ELNES can be calculated. There is a good deal of useful

information and a complete description of the input ﬁle in the wien2k userguide, which

should solve most issues. Again, there is a large list of parameters that must be set for in

order to obtain reasonable results. It is also worth converging Kpoints and RKMax against

the spectra as well.

The majority of the important parameters need to be set in the case.innes ﬁle, which is

easiest through w2web. Choose the right atom (in this case Li1) for the edge, and the right

atomic numbers:

Edge n l

K 1 0

L1 2 0

L23 2 1

M45 3 2

Next, set the edge onset, edge values can be found at http://www.kayelaby.npl.co.

uk/atomic_and_nuclear_physics/4_2/4_2_1.html as well as at a number of other loca-

tions. Set the beam energy to it’s correct value, same goes for the collection and convergence

angles, although TELNES is relatively robust to these: eg 5mrad produces very similar re-

sults to 1 mrad.

Set the energy grid to a large range of values, eg -20-50eV so you can see all of the

features that might appear. The defaults for the remaining values should be ﬁne.

In addition to the case.innes parameters, increase the number of kpoints, to at least dou-

ble, or 10×, so that there is less doubt about this being converged, use x kgen for this. This

should already have been done for calculating the DOS.

Increase the upper energy limit in case.in1 from 1.5 to ∼2-3.5, see picture. This value

deﬁnes how many higher energy states are included in Ry (1 Ry ≈13.6 eV, 1.5Ry ≈20eV).

Therefore, to obtain the correct ELNES for features more than 20eV from the onset, this

should be increased to match.

x lapw1 -p, x qtl -telnes -p, and x telnes3 can then all be run in succession. These

are relatively slow commands, so to avoid waiting on them, run all three in sequence from

the command line: $x lapw1 -p && x qtl -telnes -p && x telnes3.

Once telnes3 is ﬁnished, edit case.inb and play with the spectrometer broadening on

the last line, before clicking “x broadening” to generate the ﬁnal spectra. To experiment

with diﬀerent broadening values, just repeat these steps. Sometimes w2web gets stuck and

stops re-broadening the spectra, in which case, just delete the case.broadspec ﬁle and run

“x broadening” again. Plot the spectrum in a new tab (ctrl+click on “plot”) for easy com-

parison to new spectra without having to save it. Rerun all the TELNES steps, but with

a higher/lower number of kpoints to conﬁrm that that is not eﬀecting the spectra. At this

point you can also uncheck the “calculate and write DOS” checkboxes in case.innes, which

allow you to rerun telnes without needing to rerun lapw1 each time (unless you change

the k points/case.in1 ﬁle). If everything went well, the ELNES should look something like

(broadening = 0.5):

6.1 Monopole eﬀects

Because the lithium 1S state is quite delocalized, it is susceptible to monopole eﬀects. These

cause artifacts in the spectra resulting from non orthogonal states at the muﬃn tin bound-

ary. Fortunately, it is easy to check for them. Return to the case.innes ﬁle and change the

interaction order to 0.

Telnes now only calculates the monopole component, look in case.outputelnes to conﬁrm

this. Compare the monopole contribution to the full spectra, it should be much smaller

(∼100×). If it is not, set the interaction order to 1 to enforce dipole selection and use that

for all future spectra.

7 Core Hole

To introduce a core hole, start an entirely new case, and copy only the struct and the .ma-

chines ﬁles. Because of the periodic boundary conditions, it is often necessary to use a

supercell to avoid core holes interacting with themselves in neighbouring cells. This depends

largely on the size of the original unit cell, for a single atom case (metallic lithium) a 3×3×3

cell is required, for large calcium phosphates, no cell is needed.

Li2O is a small sized cell, so a 2×2×2 supercell is a good starting point, which would

need to be compared to a 3×3×3 supercell for full rigorousness.

To generate the supercell, run x supercell in the new directory and use the copied struct

ﬁle, with no shift.

With all core holes, it is essential to remove the symmetry in the unit cell, so there is

only a single hole per cell:

To achieve this, the cell must be reduced to a P1 space group. This can be done in

VESTA. Import the struct ﬁle and “remove symmetry”, under “Edit →Edit Data →Unit

Cell.” The spacegroup should switch to P-1, then select “apply”

To save these changes select “File →Export data” and save the structure as a .cif.

Running Cif2struct name of cif.cif should then be able to generate a valid struct ﬁle,

which should be renamed as case.struct. For some reason, I am only able to do this with my

local install of wien2k/cif2struct, the remote version complains about “ Space group name

nor symmetry operations are not given!” I have attached the ﬁnal struct ﬁle that I used to

the github directory. It might be necessary to install wien2k after all...

In the new struct ﬁle, add a “1” to the end of a lithium atom (AND DELETE A SPACE).

The ﬁle should look like this:

Now run init lapw and accept all of the defaults proposed by nn, which may require

cycling through nn a couple times. sgroup should not have any suggestions here, if it does,

something has gone wrong and it is trying to restore all of the symmetry. At the end of

init lapw, make sure to double check that the struct ﬁle is still what it should be in VESTA.

When the struct ﬁle is ready, initialize use the converged parameters (RMT, RKMax)

from the single cell case. (divide the number of k points appropriately, in this case by 8)

After init lapw, insert a core hole into the cell by editing case.inc and case.inm:

In case.inc, a hole is inserted by changing the occupancy of an orbital, the trick in this

case is picking out one of the isolated lithium atoms (with mult=1) which depends on where

you added the 1 to the struct ﬁle. Each set of lines corresponds to the core states of each atom

deﬁned in the struct ﬁle. This is why it is important to be able to treat the lithium states as

core states, otherwise it would not be possible to insert a hole. This is also why it is impor-

tant to break the symmetry in the unit cell as if not, this would insert holes into multiple

atoms. In the case.inm ﬁle, the excited electrons need to be added to the background charge:

The value is negative as this operation adds charge, not electron number. Once the

background charge is added, execute run lapw -p and calculate the ELNES. This might

take some time, so you may want to run it remotely using tmux so you can leave it overnight

if need be. Once done, calculate the DOS of the Lithium p states and compare to the no

hole case:

Figure 1: Li P states from no-hole case (left) and hole case (right). Note the dramatic peak

growth at 4eV and 7 eV in the hole case, indicating excitonic eﬀects.

The dramatic diﬀerences in the unoccupied band when a core hole is introduced, combined

with the signiﬁcant fraction of total unoccupied states represented by the lithium band,

indicates core hole eﬀects in the ELNES. Calculating the ELNES, should look something

like this (broadening = 0.9), if it all went well:

Having determined the need to include a core hole in the calculations, the next step is

determining if the hole is shielded, which can be done using density calculations.

8 Density Calculations

Density calculations can be done using either xcrysden (qualitative) or Critic2 (quantitative).

8.1 Xcrysden

Xcyrsden density rendering is well explained in the online wien2k tutorials to which the

reader is referred to. Depending on your particular setup, density rendering with xcrysden

may need to be performed locally, which will also require install wien2k locally. For Li2O,

the density looks something like:

Qualitative investigation of plot shows the excited lithium atom (atom with a hole)

distorting the oxygen electron clouds. The atomic basin appears smaller, deﬁned by the ring

of yellow surrounding the excited atom, although the entire region around the excited atom

has experienced a net gain in electron count (more green than red). These features suggest

that there is a non negligible response from the material to the core hole. The regularity

of the ground state lithium’s appear to indicate that the distance between excited lithium

atom is suﬃcient to isolate the core holes. In this case this distance is more than 9˚

A which

is considered suitable for this purpose.

8.2 Critic2

While Xcrysden is ideal for qualitative analysis, Critic2 is better suited to quantifying elec-

tron density features. The documentation for critic, while very complete, is in a giant txt

ﬁle. The code takes three inputs: the case.struct ﬁle, the case.clmsum ﬁle, and an input

parameters ﬁle “more.cri”. A sample .cri ﬁle is below:

c r y s t a l . / Li2O . s t r u c t

load . / Li2O . clmsum . / Li2O . s t r u c t

auto

i n t e g r a l s ga u le g 50 50 cp 2 v erb os e

s p h e r e i n t e g r a l s gau le g 50 50 cp 2 R0 1d−1 REND 2 . 0

The ﬁrst line inputs the crystal structure, and the second line tells the code where to ﬁnd

the clmsum/struct ﬁles in the second line.

The third line determines the location of all the critical points.

The 4th line calculates the electron population inside the atomic basin surrounding crit-

ical point 1. “Gauleg” is an integration option, look in the critic documentation for more

options. The two 50’s are sampling parameters, increase/decrease these for more/less accu-

rate calculations.

The 5th line calculates the electron population in spheres surrounding critical point 2.

“gualeg 50 50” mean the exact same as on the previous line, R0 is the starting radius, 1d-1

sets the radius step size, REND is the ﬁnal radius (in bohr).

The program is run using $critic2 <more.cri >case.cro in the command line in

the case directory. The ﬁnal ﬁle (case.cro) is just where the output is stored and can be

called anything. To determine which critical points are needed for the integrals, look at the

complete cp list in the output ﬁle:

There are a number of similar lists which can be used to identify which atom is of interest.

The critical point should be a nuclear critical point (typ = n), and as a double check, the

coordinates should match up to those in the struct ﬁle. The name should also match, in this

case, the ncp 1 corresponds to “Li1.”

The electron populations are located towards the bottom of the document. The ﬁrst case

looks like this:

The two values indicated by the arrows are the volume of the basin and how many elec-

trons are inside. The second population is calculated directly below this in a table:

By matching to the volume value, relatively comparable values should be attainable (2.14

vs 2.17), in the no hole case. The reason for performing two types of population calculation

is due to the fact that the atomic basins can become ill deﬁned when a core hole is inserted.

This is revealed when the volume of the basin is dramatically larger (>20%) than the no

hole case. In these situations, a comparison of the sphere integrals at the same volume (set

by the basin volume in the no hole case) is more appropriate.

In the case of Li2O, the no hole case had a population of 2.135 and the no hole case had a

population of 1.283 (and a smaller basin volume). This indicates that the core hole is slightly

screened and has an eﬀective strength of 2.135-1.283 = 0.852. To implement this eﬀect, a

third, ﬁnal supercell calculation is required, with a reduced hole inserted. case.inc would

need an occupancy of 1.15 and case.inm would need a background charge of -0.85. This can

be rapidly set up by copying the struct ﬁle from the full hole calculation and initializing

from there.

For this partial hole calculation, the only relevant property is the ELNES calculation as

the DOS are somewhat unphysical. This ELNES should represent the best agreement with

experiment, if all has gone well and the calculations are converged with all aspects. If all

goes well, it should like this:

9 Common Errors/Issues Encountered in Wien2k

9.1 Setting RMT/RKMax

The RKMax value is a little obscure and takes some getting used to. It is deﬁned as the

Muﬃn Tin Radii ×Maximum K point vector. The maximum k point corresponds to the

highest frequency plane wave used in the calculation. RKMax deﬁnes a cutoﬀ for sampling

k space, based on the size of each atom. A visualization of this eﬀect is depicted below:

Two case, both with 5 kpoints and diﬀerent RKMax’s. The plot on the left depicts a smaller

RKMax relative to the one on the right.

Increasing RKMax allows for higher frequency (more precise) plane waves to be used in

the basis set. These high frequency terms are however more computationally costly (CPU

requirements scale as RKMax3) and less essential for describing large features. This is why

the RKMax includes the Muﬃn Tin radius: large atoms only require lower frequency plane

waves. Each atom has an eﬀective RKMax which is relative to the ratio of its radius and

that of the smallest atom. Choosing very diﬀerent muﬃn tin radii for diﬀerent elements

leads to issues as it becomes more diﬃcult to align the surface features of each atom.

Increasing RKMax arbitrarily does not solve the problem either, as: 1. “approximate

numerical linear dependency” occurs at large RKMax values and 2. Calculations become

prohibitively expensive. RKMax should at most be between 9-10.

9.2 Ghostbands

Ghostbands are an inevitable consequence of investigating lithium with Wien2k. They man-

ifest as an LAPW2 qtl error during the scf cycles, typically during the ﬁrst cycle, but

sometimes later as well. They are recorded the lapw2( n).error ﬁles as:

’ l2main ’ −QTL−B.GT. 1 5 . , Ghostbands , check s c f f i l e s

Ghostbands are arise in a number of situations, each of which requires a diﬀerent solution.

Some of these are described below:

•Muﬃn Tins badly chosen: These appear when the muﬃn tins are too diﬀerent, or

if you ignore everything that setrmt does. Solution: Diagnose this by running the

calculation with the setrmt values (might need to remove the core hole) or make the

muﬃn tins more reasonable (might need to allow core leakage).

•Local Orbitals need better initial guess: Sometimes the starting points for local

orbitals are not close enough to the converged values resulting in divergence. Solu-

tions: Figure out which orbitals are causing the issue by looking in case.scf2( n). The

last lines of these ﬁles should have a line like this:

:WARN : QTL−B v al ue eq . 9 9 .9 5 i n Band o f energy −0.12674 ATOM=

2 L= 0

This message also tells us to look in case.in1( st) and that the problem is with atom 2

and the L=0 orbital. In case.in1( st), every independent atom is listed, eg:

To handle ghostbands, either delete the relevant local orbital line (lines that match

the atom and orbital number with energy guesses that are not 0.30) and reduce the

number of orbitals for that atom correspondingly. A more reﬁned option is to adjust

the initial guesses to another value, and rerunning run lapw.

9.3 NN in Optimization

Crashes the ﬁrst scf cycle almost immediately, due to overlapping muﬃn tins resulting from

a decreased cell size. Solution: decrease all muﬃn tin sizes before running “x optimize”.

9.4 NR and NT

Before considering a spectra as ﬁnal, it is important to verify that it is converged with respect

to the Q mesh as well. The Q mesh deﬁnes which changes in momentum are collectable by

the detector. More information on exactly how the mesh is determined can be found in the

Telnes section of the userguide. To test convergence, increase these values from NR=5, and

NT=2 to larger values, (eg. 7,3), rerun telnes( qtl and telnes3) and make sure that the ﬁne

structure does not change.

9.5 GMax Value less than Gmin

Occurs in dstart, ﬁx is to bump up the Gmax value in case.in2 from 12.00 to 14.00 or 16.00.

9.6 Ordering

If a calculation run apparently unchanged, a possible reason is that all the edits were over-

written by init lapw. The order of a calculation should be:

•Make edits to case.struct ﬁle, ie. Muﬃn Tin Radius

•run init lapw

•Edit .in ﬁles. eg. setting RKMax, dealing with ghostbands, inserting core hole, in-

creasing k points...

•run lapw

•Increase Kpoints, run case speciﬁc calculations, and/or loop back to and run forwards

from there (ie increasing k points/RKMax does not require rerunning init lapw, but

changing muﬃn tin radii does).

Guide

Navigation menu

Versions of this User Manual:

Views

Navigation