Frequencies calculation acceleration

[Andrew]

  Hello, dear NWChem forum users,

I am currently trying to obtain normal vibrational modes for some compounds at the PBE0-D3(BJ)/def2-tzvp level, using the following DFT settings:
 dft
 direct
 disp vdw 4
 grid nodisk xfine
 xc pbe0
 convergence fast energy 1e-08
 end
The hardware used is as follows: AMD Ryzen 9 5950X 16-Core processor and memory Kingston KF3200C16D4/32GX 32x4 Gb with MPI parallelization. The calculation took 82 minutes of wall-time, an input is attached. When the  charge density fitting basis set were used, the SCF procedure coiuldn't reach the convergence.

Is there a way to accelrate the calculation without loosing much in accuracy?

Thank you for your attention!

[Edoardo Aprà]

The input file looks good to me, I would simply suggest to use a spherical basis set by changing the basis set input lines to
basis  spherical
 * library def2-tzvp

 end

I would focus on checking if you have a NWChem installation meant to give the best performances, instead.

How was NWChem installed? What were the environment variables used?

My  recommendation is to use containers, as discussed in earlier threads (https://groups.google.com/g/nwchem-forum/c/82RJbQB60Tc)

[Andrew]

Dear Prof Apra,
Thank you for your reply! I was using the standard version of NWChem, installed with sudo apt-get install command. Following your advise, I have used the program in a container with MYNPROC=16 variable set, and observed tremendous acceleration: thrice as less of wall-time with no accuracy loss! Spherical basis set did not work so well though, leading to the appearance of the imaginary modes, but it works fine with the Cartesian ones.

 Anyway, now it is optimization procedure that determines the overall calculation time. But if I understand correctly, there is no way to make it faster without reducing the gradient and coordinates.