Skip to content

Latest commit

 

History

History
42 lines (34 loc) · 6.51 KB

psi4_guide.md

File metadata and controls

42 lines (34 loc) · 6.51 KB
Raw

Guide to using the Psi4 Workflow within Jobmanager

The Psi4 workflow is a utility that allows for one to do single-point DFT calculations using a number of different functionals in an automated way. First, a user converges a calculation using TeraChem, with B3LYP/LACVP* (or B3LYP/def2-SV(P)). Then, using the setup detailed below, the workflow does a one-step calculation in Psi4 to get a Psi4 wavefunction with B3LYP and the corresponding basis set. This is then used as the starting point to get a converged Psi4 wavefunction with B3LYP. Basis set projection can be carried out from LACVP* to def2-SV(P) to def2-TZVP, if desired. The B3LYP wavefunction is then used to initialize single-points using the other functionals, allowing DFT calculations on what should be the same electronic state on a number of different functionals. The single-points all have relatively small maximum number of iterations allowed, in order to ensure the structures have the same electronic state.

In order to run the Psi4 workflow, one needs to have their files set up in the following way:

  • A parent directory from which jobmanager will be run, and which contains the configuration files
  • A subdirectory for each structure, which each contain that structure's geometry (as a .xyz file), a .molden file of the structure, which was generated with TeraChem using B3LYP/LACVP*, and a .json file detailing some of the calculation parameters.

After all of these files are set up, one can simply run jobmanager in the parent directory (with the jobmanager conda environment active), which will then launch the Psi4 calculations specified in the configuration file. The specific inputs and outputs are detailed more below:

In the parent directory, one needs two files:

  • A file named configure, which contains the following: run_psi4: psi4_config.json, where psi4_config.json is the name of the .json file that one has specified the calculations they want to run.
  • A .json file that contains the following arguments (named psi4_config.json in this example folder):
    • functional: a list of functionals that you want to run on each structure. B3LYP will be run automatically, but can also be specified (although it has no effect). For functionals that support different Hartree-Fock exchange percentages, one can specify the degree of exchange by appending _hfx_xx to the end of the functional, where xx is the percentage of Hartree-Fock exchange one wants to use. Currently, only integer percentages of HFX are supported.
    • xyzfile: the name of the .xyz file in each subfolder. Each .xyz file must have this same name across all subdirectories that you want the Psi4 workflow to work on. This geometry should be the same as the one used for a TeraChem single point, or the final frame of a TeraChem geometry optimization (found in the /scr/ directory of a TeraChem optimization).
    • moldenfile: the name of the .molden file in each subfolder. Each .molden file must have this same name across all subdirectories that you want the Psi4 workflow to work on. Note that the Psi4 workflow is expecting a molden from a TeraChem calculation converged using B3LYP/LACVP*, which can be found in the /scr/ directory of a TeraChem calculation.
    • memory: The amount of memory that Psi4 will use for the calculations.
    • num_threads: The number of threads that Psi4 will use for the calculations.
    • basis: The basis set to use for the Psi4 calculations. lacvps will use LACVP*, and def2-tzvp will first project the LACVP* calculation to def2-SV(P), converge that, and then use that as an initial guess for a def2-TZVP calculation.
    • charge-spin-info: The name of the .json file in each subfolder that contains the charge, spin, and spin treatment for each structure. Note that this file has to have the same name across all subfolders.
    • hfx_rescue: If true, allows for the calculation to be converged at different HFX and then that used as an initial guess for the desired calculation. Should be left as false in order to keep the initial guess the same for all functionals.
    • wfnfile: Leave as b3lyp/wfn.180.npy. Gives the name of the Psi4 wavefunction file that all other calculations are initialized from. This should be converged in Psi4 using the specified basis set and B3LYP, and is generated automatically when the workflow is run.
    • bashrc: The user's .bashrc file, used to help initialize the conda environment.
    • conda_env: The path to the jobmanager conda environment.

In each subdirectory, one needs three files:

  • A .xyz file, matching the naming convention established in the parent directory's .json file. This should give the geometry of the structure corresponding to the wavefunction in the .molden file, so either the geometry used in a single-point or the final frame of a geometry optimization.
  • A .molden file, matching the naming convention established in the parent directory's .json file. This should be the converged molden for the structure, calculated using B3LYP/LACVP* (or B3LYP/def2-SV(P)), which is found in the /scr/ directory of a TeraChem calculation, by default.
  • A .json file, matching the naming convention established in the parent directory's .json file. This should contain three arguments:
    • charge: The overall charge of the structure.
    • spin: The spin of the structure.
    • ref: Either uks if one wants to run unrestricted Kohn-Sham for spin specialization, or rks if one wants to run restricted Kohn-Sham.

After these files are generated, one runs jobmanager in the parent directory (the one with the configure file), and it should perform the specified calculations on the structures in each subfolder.

The output structure is as follows (in each subfolder):

  • nohup.out files tell what actions were taken by jobmanager and which calculations had to be rerun.
  • nohup.err files show any error messages that appeared during the calculations.
  • For each functional, a folder will be created, which contains the file output.dat, containing the output from the Psi4 calculation using that functional.
  • For B3LYP specifically, a Psi4 wavefunction file wfn.180.npy will be generated as well, which is used as an initial guess for the other functionals. The 1step wavefunctions are generated in the process of converting the TeraChem molden to a Psi4 wavefunction, and can be mostly ignored.

The given example folder (example_psi4_run) should run if the jobmanager command is run from within that folder, and gives a sense for the desired file structure and components needed.