Re: Bug in LS calculations

[Alexander Mitrushchenkov]

Dear Nayla,

Thanks for sending me your full output. From that I can see that you used 2010.1 version. Unfortunately we do not support any more such old molpro version. Though, if I have time (and will be able to compile that..) I will investigate it further. Just to note that you ask for 5GW=40GB memory per CPU, so for 24 CPUs it becomes quite large memory. You job is very small, it perfectly runs with 1GW and I guess much smaller memory, using molpro 2015 or later.

So if you can update to 2015 or later, you must not have any problems. Otherwise, please wait, I can not give exact delays when I can look for this.

Best wishes,

Alexander

Le 13/05/2021 à 15:31, Nayla el kork a écrit :

Dear Alexander,

Thank you for your email, and very sorry for the late reply, I got confused with the MOLPRO mail list. I have attached to this e-mail all the output file, I have tried changing the memory many times. When I used a memory of 1400 000 000 the program stopped at the Singlet type calculation. However, when I used a memory of  3 000 000 000 and 5 000 000 000 the program continued to the triplet section and stopped at the spin-orbit calculations. I had also tried higher values such as  9 000 000 000, however the calculations did not start on the HPC in this case. I truly hope you can help out.

Sincerely,

Nayla

On Mon, May 10, 2021 at 1:44 PM Alexander Mitrushchenkov <Alexander.Mi...@univ-eiffel.fr> wrote:

Dear Nayla,

I tried to run your job with 2015 version of Molpro, I have no problem
with that. Could you please send me the full failing output? Please note
that you asked for 16GW=128GB of memory per CPU in your job. If your
machine is limited in memory, it may lead to high swapping and very slow
run. Actually, your job is very small so you can reduce the memory to
1GW or smaller.


Best wishes,

Alexander

--
Dr. Alexander Mitrushchenkov, IGR
Laboratoire de Modélisation et Simulation Multi Echelle
UMR 8208 CNRS
Univ Gustave Eiffel
5 Bd Descartes
77454 Marne la Vallée, Cedex 2, France

Phone:    +33(0)160957316
Fax:      +33(0)160957320
e-mail:   Alexander.Mi...@univ-eiffel.fr

-- 
Dr. Alexander Mitrushchenkov, IGR
Laboratoire de Modélisation et Simulation Multi Echelle
UMR 8208 CNRS
Univ Gustave Eiffel
5 Bd Descartes
77454 Marne la Vallée, Cedex 2, France

Phone:    +33(0)160957316
Fax:      +33(0)160957320
e-mail:   Alexander.Mi...@univ-eiffel.fr

[Alexander Mitrushchenkov]

Dear Nayla,

Sorry, I was wrong in my previous mail. I was not running your job but rather the small one  cited in your original mail. The job you were writing about, is indeed quite lengthy. So I am running it now to see how it is with current molpro and 2015 version. Please note however that it does not change the fact that we hardly support the bugs in the 2010 version. I'll let you know as soon as I have results.

Best wishes,

Alexander

[Nayla el kork]

Dear Alexander, 

Thank you for your kind help. I will be waiting for your updates. I have been using the old license indeed, but my institution bought the molpro 2020 version this year. I will look into installing it shortly. Thank you again.

Sincerely, 

Nayla

[Alexander Mitrushchenkov]

Dear Nayla,

This is good if you can update the molpro. I can confirm that your input, with minor modifications (below) work smoothly with current molpro. I just commented out full MRCI calculations, as they were not used in properties; but I can confirm that these full MRCI calculations (clearly slow) works with 1GW of memory without problems.

The things to fix in the input:

  * Normally you should not use file 1 for storing your data, like orbitals or CI records. This file is for internal molpro records, like integrals. In your case running SO integral evaluation ecrases all records starting from 1300.1 so all your saved MRCI records are not available, which leads to error. Please use files 2 or 3. I guess it is mentioned in manual but perhaps not very clear. Also, in parallel case, the file 1 is a distributed file so not suitable for storing. Changing all your MRCI records to file 2 makes the calculations running without problem.

  * As you have many SO calculations, it is advantages to calculate SO integrals once by putting "lsint" before first HLSMAT/TRANLS calculation. This will avoid recalculating of integrals. For some reason when you change the SO component, old integrals are ecrased (unless you used 'lsint'). This is probably a bug and I'll take care of that.

Best,

Alexander