[Wien] [WIEN2k] forrtl IO error in x_nmr_lapw for Heavy metal structures (TlF3, HgF2)

[Peter Blaha]

> 
> I recompiled lapw1 with LOMAX = 4 but then the scf cycle fails. 

As I said before, you also need to recompile lapw2, lapwso and nmr.

The real
> problem was the used cut off energy of -11 Ry in init_lapw. That 
> introduces also too much orbitals in x_nmr -mode in1 and a qtl-b errors 
> in the first two loops. After using the default value of -6 Ry for core 
> / valence separation, scf cycles converge without QTL-B errors and the 
> x_nmr initializations starts with less atomic orbitals and  the 
> calculations (x_nmr -p) no longer produces forrtl I/O errors.

Yes, such a low E-cut is dangerous and only necessary if you have 
extremely small spheres.

Note: Sometimes it is better to use -ecut 0.999 (or similar). This would 
not use the energy as core/valence separation, but the amount of charge 
inside the sphere for each orbital. In particular for 5d elements this 
is useful as it can put the 4f states into core, but lower energy 5p 
states remain valence.


> 
> 
> Thanks again for the help!
> 
> 
> Best regards,
> 
> Michael Fechtelkord
> 
> 
> Am 12.11.2023 um 23:28 schrieb Peter Blaha:
>> Once I've seen your in1 file, the solution is probably very simple:
>>
>> I did not know that you included the 4f states of Tl (near -8 Ry) as 
>> valence.
>> The nmr code constructs by default NMR-local orbitals up to 
>> "l-exception" + 1, i.e. up to l=4 when you have l=3 states listed in 
>> the regular case.in1
>>
>> While this is possible, it requires to recompile lapw1/2,nmr with a 
>> modified parameter  LOMAX =4 (param.inc in lapw1, in other codes in 
>> modules - do a search).
>>
>> This is necessary if you handle 4f elements or early 5d metals, 
>> however, I very much doubt that it is a good idea to include the 4f 
>> states for Tl (with RMT=2.5) as valence.  I would not use -ecut -11.
>> All it produces is noise as the 4f convergence can be quite 
>> problematic and SO effects might be of importance.
>>
>> Best regards
>> Peter Blaha
>>
>> Am 12.11.2023 um 22:12 schrieb Michael Fechtelkord:
>>> Lieber Herr Blaha,
>>>
>>> schon mal vorab herzlichen Dank für die schnelle Hilfe auch am 
>>> Wochenende. Anbei die gewünschten Daten und folgendermaßen bin ich 
>>> vorgegangen:
>>>
>>> im Verzeichnis TlF3
>>>
>>> 1) cif2struct TlF3.cif
>>>
>>> 2) Kontrolle und Nachbearbeitung mit struct generator in w2web
>>>
>>> 3) rmt gesetzt mit 0% Reduktion in w2web struct Generator (set 
>>> automatically RMT and continue editing)
>>>
>>> 4) Structfile abgeschlossen (save file and cleanup)
>>>
>>> Weiter im Terminalfenster:
>>>
>>> 5) init_lapw -b -rkmax 7 -numk 1000 -ecut -11 (endete mit ok)
>>>
>>> 6) run_lapw -p -ec 0.0001 -cc 0.0001 (konvergierte nach ca. 13 Zyklen)
>>>
>>> 7) save_lapw TlF3_pbe_rkmax_7_numk_1000_ecut_11_cc_0001
>>>
>>> 8) x kgen auf 10000 k points (habe es auch mit weniger probiert, 
>>> daran liegt es wohl nicht)
>>>
>>> 9)  x_nmr_lapw -mode in1
>>>
>>> 10)  x_nmr_lapw -p
>>>
>>> Ich hänge auch das cif File und das machines File der Vollständigkeit 
>>> halber an. NTMATMAX ist 40000, NUME 6000, OMP_NUM_THREADS 2
>>>
>>>
>>> Sollten Sie zusätzliche Daten benötigen, schreiben Sie mich kurz an.
>>>
>>>
>>> Viele Dank schon mal und einen guten Wochenstart
>>>
>>> wünscht
>>>
>>> Michael Fechtelkord
>>>
>>>
>>

-- 
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300
Email: peter.blaha at tuwien.ac.at    WIEN2k: http://www.wien2k.at
WWW:   http://www.imc.tuwien.ac.at
-------------------------------------------------------------------------