Skip to content

Latest commit

 

History

History
182 lines (119 loc) · 5.17 KB

prog_interface.rst

File metadata and controls

182 lines (119 loc) · 5.17 KB
Raw

overview of external interfaces

customization

Most system calls to programs can be modified at run-time using $HOME/xoptrc. The program's default values are set in default.f90.

All system calls (besides for conical intersection optimizations) are handled by getgrad.f90, if further modification are needed.

setting custom system calls via $HOME/.xoptrc

call_orca = orca_scipt_or_binary
call_gaus = run-g09
call_psi4 =psi4
call_gei=mygrad.sh
call_mpi=mpiexec -bynode
call_mopac=mopac16
scratch=/local_scratch/
maxiter=500  # max iterations
maxd=0.35    # max displacement bohr
gconv=1e-3   # gradient norm threshold
econv=1e-7   # energy change threshold
maxgrad=1e-3 # max. gradient threshold

numerical gradients

Arbitray numerical gradients can be computed via a script/executable named mygrad.sh located in the working directory. It does not need to be bash script. It will be excuted by a system call inside xopt. It needs to do 3 things:

- Have the program to be executed use ``xopt.xyz`` as input
- run the program
- provide the energy output as a single number inside ``xopt.energy.tmp``

An example for Turbomole would look like:

#!/bin/bash
#1. make use of xopt.xyz
babel -ixyz xopt.xyz -otmol coord 
#2. execute
dscf > dscf.out
#3. write energy into file
grep "|  total energy" dscf.out | awk '{print $5}' > xopt.energy.tmp

xopt will read the xopt.energy.tmp value for a 2-point (central) finite difference computation of the gradient with a step size of 0.005 bohr.

general external interface

The general external interface (GEI) option allows the user to call arbitrary programs and feed the gradient into xopt. Is uses similar technology as the above numerical gradient.

Use any script that transforms a xopt.xyz file into an energy and gradient file with the following format in a.u.:

energy
grad1(x,atom 1), grad1(y,atom 1),grad1(z,atom 1)
.
.
grad1(x,atom n), grad1(y,atom n),grad1(z,atom n)

The default name of the script is mygrad.sh and the default gradient file is xopt.grad. The script name can be changed either in the .xoptrc options or with -gei_command <script name>.

ORCA

Requirement:
  • input file: orca.in
  • orca40 startup script to run orca (provided)

It is necessary that the following input lines are included

! ENGRAD
*xyzfile <chrg> <mult> xopt.xyz

Turbomole

Prepare a standard input using define. Supported are HF/DFT calculations using ridft and dscf, and calculations using the ricc2 module (eg. for MP2.). A 'coord' file is automatically written and updated. If $rij is found in the control file, ridft will be called instead of dscf.

Requirement:
  • ridft/dscf/ricc2 in $PATH

Gaussian09

Requirement:
  • input file: g.in
  • run-g09 startup script for Gaussian (provided)

Mopac201X

Requirement:
  • binary name: mopac2016
  • input directives using a SETUP file

SETUP must contain at least:

1SCF GRAD XYZ aux(42,PRECISION=9,MOS=-99999,COMP)

use of PRECISE is recommended.

gamess

experimental

nwchem

experimental

PSI4

needs to be adapted to changes in psi4 v1.2

Amber (sander)

requires modified sander

conical intersection optimization

Penalty function-based (no non-diabatic coupling) CI optimizer following Levine/Martinez DOI: 10.1021/jp0761618. We can do Gaussian, Turbomole(dscf,ricc2), Orca and (modified) Amber and numerical gradients.

You need to make 2 directories named stateJ.xopt` andstateI.xopt. Prepare the input for each state inside the directories. It should work for Turbomole(dscf/ricc2), ORCA, G09, AMBER and the GEI custom interface. For the GEI each subdirectory must contain it's own script that writes thexopt.gradfile. We assume state I < state J, e.g. J=I+1. Thexoptoutput will print something like:gap[eV]: 0.024 penalty: 13.1 E(low): -546.9431436 E(high): -546.9422645 root flip: F, where E(low) denotes the lower state (eg. groundstate) calculated as state I in stateI.xopt and E(high) as the higher state calculated as state J inside stateI.xopt. If at any stage during the optimization E(low)>E(high) (I>J) root flip will be set true (=T) and E(low/higher) will be interchange, e.g. E(low) will be the energy obtained instateJ.xopt``. This is checked for each optimization step individually and is not tracked through previous steps.

The strategy to increase the penaly σ is as follows:


σ = σ + (2ΔEij/α) ,

where ΔEij is the energy gap between state I and J, $alpha$ the smoothing factor (see paper). $sigma$ is increased when ΔEij is larger than 1e-3 and the penalty function change ΔFij smaller than 5e-5.

Note 1: for gaussian groundstate SA-CASSCF calculations add IOp(10/97=100) and IOp(5/97=100). It switches the CI vectors internally so one gets the proper groundstate gradient (e.g. for CASSCF(2,2,nroot=2,stateaverage)). Not sure about nroot > 2.