pysisphus implements Chain Of States (COS) methods like Nudged Elastic Band (NEB) and Simple Zero Temperature String (SZTS) to converge minimum energy paths and provide initial guesses for transition states. In addition pysisyphus provides serveral Intrinsic Reaction Coordinate algorithms. The required gradients and/or hessians are calculated by calling external quantum chemistry codes. By default everything is done in cartesian coordinates but an internal coordinates implementation is in progress.
This software is still work in progress and shouldn't be used for production runs. Use at your own risk.
pysisyphus provides three hooks that can be called from the shell (command line). The available commandas can be queried with the -h or --help arguments:
# Run optimizations of single molecules or chain of states (NEB, string method)
pysisrun
# Visualize the results of chain of states calculations
pysisplot
# Manipulate a .trj file or multiple .xyz files
pysistrj
Usually called with a YAML-formatted file containing all input for the optimizations. The --restart is quite experimental for now.
usage: pysisrun [-h] [--clean] [--fclean] [--restart] [yaml]
positional arguments:
yaml Start pysisyphus with input from a YAML file.
optional arguments:
-h, --help show this help message and exit
--clean Ask for confirmation before cleaning.
--fclean Force cleaning without prior confirmation.
--restart Continue a previously crashed/aborted/... pysisphus run.
cos:
type: neb # optimize a NEB
parallel: 4 # Run 4 calculations in parallel
opt:
type: cg # Use conjugate gradient
max_cycles: 50 # Optimize for at most 50 cycles
align: True # Remove translation and rotation with partial procrustes
climb: True # Use climbing image NEB to get a better guess for the TS
interpol:
idpp: True # Interpolate using the image dependent pair potential (recommended)
between: 15 # Interpolate 15 images for a total of (1 + 15 + 1 = 17) images
calc:
type: xtb # Use xTB calculator
charge: 0
mult: 1
# Interpolate between these two geometries. One can supply also more than two geometries,
# but then the between value in the interpol section has to be lowered. Supplying a third
# geoemtry here would lead to a chain of states containing 33 images (1 + 15 + 1 + 15 + 1).
xyz: [15_tpss_tzvp_full_scan_acn_013_xtb.xyz, 15_tpss_tzvp_full_scan_acn_024_xtb.xyz]
usage: pysisplot [-h] [--until UNTIL]
(--saras | --tddft | --params PARAMS | --cosgrad | --energies)
optional arguments:
-h, --help show this help message and exit
--until UNTIL Only show until cycle [until].
--saras Plot OpenMolcas state average potential energy surfaces
over the course of the NEB.
--tddft Plot ORCA TDDFT potential energy surfaces over the course
of the NEB.
--params PARAMS Follow internal coordinates over the course of the NEB. All
atom indices have to be 0-based. Use two indices for a
bond, three indices for an angle and four indices for a
dihedral. The indices for different coordinates have to be
separated by ','.
--cosgrad Plot image gradients along the path.
--energies Plot energies.
Usually called wit multiple .xyz files or a single .trj file.
usage: pysistrj [-h] [--idpp] [--xyz XYZ [XYZ ...]]
(--between BETWEEN | --align ALIGN [ALIGN ...] | --split SPLIT | --reverse REVERSE | --cleantrj CLEANTRJ | --spline SPLINE)
optional arguments:
-h, --help show this help message and exit
--idpp Use Image Dependent Pair Potential instead of simple
linear interpolation.
--xyz XYZ [XYZ ...]
--between BETWEEN Interpolate additional images.
--align ALIGN [ALIGN ...]
Align geometries onto the first geometry read from
multiple .xyz or one .trj file.
--split SPLIT Split a supplied .trj file in multiple .xyz files.
--reverse REVERSE Reverse a .trj file.
--cleantrj CLEANTRJ Clean a .trj file.
--spline SPLINE Evenly redistribute geometries along a splined path.
- Initial guess
- Linear interpolation
- Image Dependent Pair Potential (https://doi.org/10.1063/1.4878664)
- Removal of translation and rotation (http://pubs.acs.org/doi/abs/10.1021/acs.jctc.7b00360)
- Either the first and/or the last image of the path can be fixed
- Improved tangent (https://doi.org/10.1063/1.1323224)
- 1 Climbing Image (https://doi.org/10.1063/1.1329672)
- 2 Climbing Image (https://doi.org/10.1063/1.4905209)
- Different optimizers (see below)
- Steepest Descent Optimizer only (https://doi.org/10.1063/1.2720838)
- Splined images
- Equal spacing
- Energy dependent spacing, leading to higher resoultion at high energy regions
- Can be used for single molecules or COS-paths
- Currently implemented
- Using only first derivatives
- Steepest Descent (SD)
- Conjugate Gradient (CG)
- QuickMin (QM)
- FIRE
- Using second derivatives (Hessian)
- BFGS
- Rational Function Optimization (RFO)
- Wrapper for serveral SciPy optimizer that don't require an explicit hessian (Experimental and not recommended, as the step size can't be controlled easily)
- Using only first derivatives
- Very experimental
- Mass-weighted or non-mass-weighted coordinates
- Initial displacement using the transition vector
- Can be called in parallel or in serial mode
- Wavefunctions from &rasscf and gradients from &alaska
- Reuse of orbitals from previous iterations
- State average calculations to converge excited state MEPs
- Fast semi-emperical method for ground state calculations (http://pubs.acs.org/doi/abs/10.1021/acs.jctc.7b00118)
- All methods that have gradients
- Recognizes TDDFT optimizations
- Density functional methods
- Recognizes TDDFT optimizations
- (Probably) All methods that support Force run
- Recognizes TDDFT optimizations
- All expressions that can be handled by sympy
- Automatic differentation to give gradients and hessian
- Easy visualization of optimization progress with COS methods (AnimPlot class)
All coordinate input read from .xyz or .trj files is expected to be in Ångström. Internally atomic units (Hartree, Bohr, etc.) are used throughout. All coordinate output written is given in Ångström again.
- Get everything to work very smoothly;)
tmpdir: export TMPDIR=[tmpdir]