SEAMM VASP Plug-in

A SEAMM plug-in for VASP

• Free software: BSD-3-Clause

• Documentation: https://molssi-seamm.github.io/vasp_step/index.html

• Code: https://github.com/molssi-seamm/vasp_step

History

2026.10.8 – Bugfix: whole cells converge reliably again

• A whole cell run as a model-chemistry task now uses ALGO = All with VASP’s legacy line search (ISEARCH = 0), replacing ALGO = Normal (Davidson) from 2026.10.7. Davidson diverged in a 32-molecule FEC cell with r2SCAN, and with gentler density mixing it diverged in a 32-molecule EC cell. ALGO = All with the new line search (ISEARCH = 1) aborts on the FEC cell (“EDWAV: internal error, the gradient is not orthogonal”). With ISEARCH = 0 both cells converge, to the same energies the other settings give where they work.

• Fragments are unchanged (ALGO = All, ISEARCH = 1). Setting options ["electronic method"] chooses another minimizer, with the line search the step uses elsewhere (ISEARCH = 1).

2026.10.7 – The step declares what its cost model is made of

• energy.TIMING_SPEC – valence electrons and the grid volume (the cell volume scaled by (ENCUT/500 eV)^1.5, now a descriptor of the record) as size variables, the functional as the method class, the task, ionic steps as the unit, k-points as a multiplier – is passed when a run is recorded (seamm-exec 2026.10.6.1 writes it beside the records), so the cost model is fitted from the step’s own description rather than a table in seamm-exec.

• Removed the unused Docker option from the vasp.ini template.

• A whole cell run as a model-chemistry task now uses ALGO = Normal (Davidson) rather than ALGO = All: VASP’s conjugate-gradient minimizer aborts reproducibly on some cells (“EDWAV: internal error, the gradient is not orthogonal”, e.g. a 32-molecule FEC cell). The converged energy, forces and stress are the same. Fragments keep ALGO = All; options["electronic method"] chooses another.

• Requires Python 3.12 and seamm-exec 2026.10.6.1.

2026.10.6 – Timing records that a cost model can be fitted to

• Each VASP run – the Energy step’s, and a model-chemistry task when analyze_task is given its task – appends a record to ~/.seamm.d/timing/vasp.csv through seamm_exec.timing: the machine class, ranks, wall time and outcome, and the descriptors of the calculation – the variables of the cost model (valence electrons, cell volume, ENCUT, k-points), ALGO, EDIFF, PREC, ISPIN, IBRION/NSW, the functional and potentials, the atoms, and from the OUTCAR the electronic and ionic steps, ranks and VASP’s own times. This replaces the step’s own CSV, which carried the POSCAR, INCAR and KPOINTS text and grew without bound (#18). See seamm_exec’s campaign of 2026-10-05.

• Requires seamm-exec 2026.10.6.

2026.10.3.1 – Realistic time estimates for VASP tasks

• The estimated time of a VASP calculation, which decides how calculations are bundled into batch jobs, is now fitted to about 400,000 VASP runs on ARC’s TinkerCliffs and checked against the MBE prototype’s 64,656: it was up to five times too low for large cells.

• A periodic cell’s calculation gets a time limit of three times its estimate, at least an hour.

2026.10.3 – VASP as a model chemistry, for the MBE step

• VASP can be used as a model chemistry, VASP:DFT@<functional>/<potentials>@<ENCUT> (e.g. VASP:DFT@r2SCAN-D4/PAW-hard@1200), so steps that evaluate many structures – first the MBE step’s periodic low level – run VASP as separate calculations, locally or bundled on a cluster, and reuse finished ones on a rerun.

• A molecule can be placed in a box registered on a parent cell’s FFT grid (whole grid steps, the same atom-to-grid offsets as in the cell), so that many-body increments of plane-wave calculations cancel the “egg-box” error; the cell’s grid is set explicitly.

• -D4 functionals add the D4 dispersion with the dftd4 program after VASP (set dftd4 in vasp.ini), since many VASP builds lack D4: periodic for a cell, for the isolated molecule for a fragment.

• PAW-hard uses the hard potentials (H, B, C, N, O, F, P, S, Cl); an ENCUT below the potentials’ ENMAX, and small cells without k-points, are refused.

• Without the graphical interface, elements without a chosen potential now get the set’s default instead of an error. The inputs of the VASP step itself are unchanged.

2026.9.29 – Bugfix: a vasp.ini template to edit, and an installer that writes it

• The step expected to create ~/SEAMM/vasp.ini from a template, but the template was never included, so running VASP without the file failed. The template is now included, with comments explaining each option for a local build or environment modules.

• Added vasp-step-installer, which the SEAMM Manager runs when the step is installed, to write the template to ~/SEAMM/vasp.ini for you to edit.

• If vasp.ini gives no command line, vasp_std (and vasp_gam and vasp_ncl) are looked for on the PATH; otherwise the error says what to set. Gamma-point calculations use code when there is no Gamma-only build, and non-collinear calculations without noncollinear_code say that it is needed.

• The documentation describes installing with the SEAMM Manager and configuring vasp.ini.

2026.9.27 – The VASP potentials can belong to the installation

• The PAW potentials were always taken from ~/SEAMM/Parameters/VASP. They are now taken from the Parameters/VASP directory of the SEAMM installation in use, falling back to ~/SEAMM/Parameters/VASP if that installation has none. A second installation such as ~/SEAMM_DEV therefore works without copying the potentials, but can have its own. Requires seamm-util 2026.9.27.1.

2026.7.28: Initial wavefunction can reference another job

• Initial wavefunction can now reference another job’s WAVECAR, via job://<job number>/<name> (SEAMM’s Node.file_path gained read-only cross-job references) – useful for seeding from a wavefunction computed in a different job.

2026.3.1: Internal: switching from deprecated library pkg_resources to importlib

2026.2.5: Corrected r2SCAN and added functionality

• Corrected an error with r2SCAN which resulted in running rSCAN instead!

• Added atomic and elemental reference energies for PBE, PBE-D3BJ, r2SCAN, and r2SCAN-D3BJ with a 700 eV cutoff. Have all atoms through Pu and elements through Ar except for S which is still running.

• Added changes in the cell parameters, density and volume to the available results, as well as the elapsed time for the VASP calculation and the number of processors used and the energy per atom.

• Add the standard system and configuration handling to create new configurations and systems to store the results of calculations.

• Added the ability to start from a previous WAVECAR file, be default that from the previous step of the VASP calculation.

• Switched the default to using the new line search (ISEARCH=1) in optimizations.

• Corrected the geometry convergence parameter to be the maximum force on an atom or cell coordinate, not the square.

2026.2.1: Added -D3BJ functionals and reference energies for r2SCAN-D3BJ

• Added the natively supported functionals with D3(BJ) dispersion corrections, PBE-D3BJ, PBEsol-D3BJ, RPBE-D3BJ, revPBE-D3BJ, M06-l-D3BJ, TPSS-D3BJ, and SCAN-D3BJ. Also added r2SCAN-D3BJ using the D3 parameters from ORCA.

• Added reference energies for r2SCAN-D3BJ with a 700 ev cutoff for elements He-Ca except P and S.

2026.1.28: Bugfix: protected against missing data calculating energies of formation

• Fixed issues with missing data when trying to calculate the energies of formation, etc. Now the code notes the error and continues safely.

2026.1.1: Added more reference energies for r2SCAN

• Added reference energies for r2SCAN for H-S and Ar

2025.11.26: Added cohesive and formation energies, and saving gradients

• Added an option to save the gradients in the configuration.

• Calculate the cohesive energy and energy of formation if tabulated atom and element energies are available for the DFT functional and planewave cutoff.

2025.11.2: Small improvements and better output.

• Added missing precision control

• Added control over using HDF5 files

• Improved the output, particularly for optimization

• Standardized pressure and stress units to GPa

• Changed some of the defaults to make standard calculations work better.

2025.10.31: Added optimization and improved the output.

2025.10.22: First working version

• Handles single point energies reasonably well.

2025.10.12: Plug-in created using the SEAMM plug-in cookiecutter.