Error: Missing file "MPCF122BB" when running FNO-CCSD(T)-F12

Dear MRCC user

Attached the output file. Ran the job on a single node parallelized with OpenMP in 32 cores.



Your kind assistance is really appreciated.

Compilation Details:
build.mrcc Intel -pOMP
Compiler: Intel 19.1.3.304
Blas/lapack: Intel MKL 2020.4
GCC: 13.3
Glibc: 2.28

Testing MRCC completed successfully with the following summary:
 %%%%%%%%%%%%          SUMMARY         %%%%%%%%%%%%%%%
 Warning! Warning! Warning! Warning! Warning! Warning!
 List of test jobs failed:
 Ethanol_ONIOM_ee_external_PBE_STO-3G_GFN2-xTB_water_sp_XTB
 Ethanol_ONIOM_ee_user_HF_STO-3G_GFN2-xTB_water_opt_XTB
 Ethanol_ONIOM_PBE_STO-3G_GFN-FF_grad_XTB
 Ethanol_ONIOM_PBE_STO-3G_GFN-FF_water_grad_XTB
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

A quic fix would be to change mem=50000MB to mem=90000MB or something larger so that the block version is not invoked.

Thanks a lot. I realized "mem" defines memory as per mpitask. After allocating a lot more memory, the error is solved.

I have a new error: "forrtl: severe (124): Invalid command supplied to EXECUTE_COMMAND_LINE", happening at the stage of "Fatal error in cp VARS_AO VARS". I am guessing that I ran out of local scratch disk space where I copy the files to run the job on the compute node after submitting the job. After moving to run job on a larger network file system, I completed the MP2F12 calculation, but I faced another error: "Fatal error in uccsd" with similar "forrtl: severe (124): Invalid command supplied to EXECUTE_COMMAND_LINE" after successfully completing calculation of (T) correction.

May I ask if this is the reason? If yes, how much disk space am I expecting to need? If not, your kind advice is really appreciated.

Dear limjiayang,

the issue with those lines is that their execution may require for a short time virtual memory identical to that allocated in MRCC via the mem keyword. The problem is that the swap size or the allowed overcommit limit of virtual memory seems to be very small in your system.
If you allocate x percent of physical memory for MRCC task(s), than you need 2*x percent available in physical mem+swap or allow up to this size overcommitting virtual memory.

You can:
- try to increase swap or vm.overcommit_ratio up to at least the size of your physical memory
- if this is not possible, use only half of your physical memory for (the total number of) MRCC tasks per node

I hope this helps,
Peter