High-temperature low-scaling GW calculation

[bprobinson102]

VASP team,

I am attempting to run a single-shot low-scaling GW calculation at an electronic temperature of Te = 10 eV. However, these calculation are failing to complete due to errors that stem from the choice of NOMEGA.

1. NOMEGA = 12 leads to the following warning in the standard output

Code: Select all

WARNING: Greens function poles not found, increasing NOMEGA might help

2. NOMEGA = 16 leads to the following error, I understand that this error may be due to the convergence criterion for the Remez-Algorithm being too tight.

Code: Select all

 -----------------------------------------------------------------------------
|                     _     ____    _    _    _____     _                     |
|                    | |   |  _ \  | |  | |  / ____|   | |                    |
|                    | |   | |_) | | |  | | | |  __    | |                    |
|                    |_|   |  _ <  | |  | | | | |_ |   |_|                    |
|                     _    | |_) | | |__| | | |__| |    _                     |
|                    (_)   |____/   \____/   \_____|   (_)                    |
|                                                                             |
|     internal error in: minimax.F  at line: 4008                             |
|                                                                             |
|     internal ERROR, DETERMINE_ALTERNANT was not able to find alternant      |
|     31 8192 10                                                              |
|                                                                             |
|     If you are not a developer, you should not encounter this problem.      |
|     Please submit a bug report.                                             |
|                                                                             |
 -----------------------------------------------------------------------------

Could you please provide any suggestions on how to resolve this issue so that I can successfully run this calculation at this temperature and potentially at higher temperatures?

The files are attached and let me know if you need anything else.
Thank you for your time.

Brian

[ahampel]

Dear Brian,

thank you for your question on the official VASP forum.

I checked your input files and you did set SIGMA=10 eV to set the temperature using LFINITE_TEMPERATURE=TRUE . This is a very high temperature for GW. as this would correspond to 116040 K . This could explain why VASP does not find any poles. Can you try to run the calculation for SIGMA=0.1 (1100K) or maybe SIGMA=0.05 (550 K) . The temperature is set in absolute values of energy not inverse eV (beta), so 10 is really really high and all features would be smeared out by temperature. See also : wiki/index.php/LFINITE_TEMPERATURE .

Can you try this, and let me know if the error persists than I will have a closer look at your calculation. Thank you.

Best regards,
Alex H.

[bprobinson102]

Alex,

Thank you for your response. I have ran several calculations with temperatures ranging from 0.10 to 7.0 eV without issue. I am interested in the results for even higher temperatures, such as 10 eV. Do you think there is anything I can do to achieve this? Or is it past the limits of what is currently feasible. Thanks.

Brian

[ahampel]

Hi Brian,

I would question that this works in any finite temperature GW code. I understand that you probably do not want to hear an answer that questions your motifs, but let me try to explain why this does not work at a certain temperature. VASP uses at finite temperatures Matsubara Green functions for. Those "live" at a finite temperature. You might know all this, but let me graphically explain why this becomes a problem at very high temperatures. Let's assume you have a non-interacting semicircular density of states with bandwidth of 4 eV:

dos.png

This could represent a typical system that you want to run GW on. Now let's take a look at G(iw_n)=[ iwn - mu - epsilon]^-1 at different temperatures (in the end the Green functions stores the non interacting density of states as series of poles/excitations):

imG.png

I marked each line with the temperature value used in VASP and the value in Kelvin. All physical data of such Green function is contained in the first part of the function, which you can see as a sharp peak reaching 0. This is because for iwn-> 0 Im G(iwn) has the value of the DOS. All other values are far away from the real frequency axis but encode all physical data. However, when your system temperature gets larger than the bandwidth of your input then you basically smear out all possible excitations. Hence, if you set SIGMA larger then the spectral width of your problem there is no way that you can represent any meaningful data in it. This is exactly what the code tells you. Even if you could run the calculation it would not tell you anything about the system I would say.

Let me know if this makes sense to you. I understand that you would like to scan the temperature dependence of your system, but at some value SIGMA this will stop working. Is there any specific physical reason why you must reach that high temperatures in this system?

Best,
Alex H.

[bprobinson102]

Alex,

Thank you for that very helpful explanation!

We are interested in high temperatures to understand temperature effects on e-e scattering for these extreme conditions.

Brian

[ahampel]

Hi Brian,

okay thanks for clarifying. I fear at a certain point you will have problems with the finite temperature GW. Also please note that this is just the electronic temperature in GW. It does not reflect the full temperature of the system. You can try to increase the precision and number of frequencies, but if that does not help at some point there is no way to make it work (just due to theory reasons in Matsubara Green functions).

Best,
Alex H.