iOpenShell » Q-Chem specific questions » ADC(2) Calculations

ADC(2) Calculations

Page: 1

Author Post
Member
Registered: Feb 2017
Posts: 2
Recently I was benchmarking ADC(2) calculations for excited states in medium sized CT complexes.
I have a few questions regarding the performance of these calculations in Q-Chem.
First it is a "undocumented" (I only found it in the examples section of the manual) THREADS variable.
As far as I can see it affects number of threads spawned for RI-MP2 stage and further calculations.
I am wondering why the number of threads in this case is controlled by the separate rem-variable, but not the standard "-nt" command line option.
Anyways, the performance of ADC calcs itself seems to be far from optimal. With "htop" utility I can see that at the stage of ADC-calcs there are spawned a number of threads as many as were requested by THREADS variable, but all the process are loaded by ~10%.
In my case of the 48 processes launched, load average on the major part of the calcs was about 2.5-3, and calculations was way longer than I expected (comparing to the time of such calculations in TM).

Finally, I was trying to run spin opposite scaling calculations (sos-adc(2)), but they failed with following message on the stage of RI-MP2:


"--------------------------------------------------------------------------
SOS-MP(2) calculation ... FAILED!
[ERROR] [libctx::root_storage::get(const std::string&)] Data key "/canmo/sos_mp2/energy" doesn't exist.

Q-Chem fatal error occurred in module /scratch/qcdevops/jenkins/workspace/build_qchem_linux_distrib/tags/qc512/qchem/adcman/adcman_main.C, line 46:

[libctx::root_storage::get(const std::string&)] Data key "/canmo/sos_mp2/energy" doesn't exist.
"
Administrator
Registered: Oct 2017
Posts: 22
Dear spinor,

Could you provide an example of the input, giving performance problems? Which version of Q-Chem do you use? I didn't run a lot of ADC calculations, but I run many CC/EOM-CC calculations, which are sensitive to the amount of given memory. If the specified amount of memory in the input is too small, the calculation will have to use the disk. This can easily spoil performance and parallelism. As for the number of threads, the "-nt" option should be the only legit way to execute Q-Chem with several threads. I run a small job with and without "THREADS" rem keyword, and did not notice any influence. I guess it is a historical artifact and it should not be used.

Page: 1

iOpenShell » Q-Chem specific questions » ADC(2) Calculations