History#

2026.11.9 – Bugfix: the Thermochemistry results name the method that ran

  • With ‘use existing parameters’ the header of the Thermochemistry results named the step’s own, unused method (e.g. Kohn-Sham DFT) instead of the method of the calculation it analyzed, and correlated methods showed ‘R-None’. It now names the method that ran.

  • The developer guide’s ALMO-EDA campaign records the validation and cost of the Psi4 backend of the Interaction EDA step.

2026.11.8.1 – Bugfix: optimizations did not store the optimized structure

  • Since version 2025.3.9 an Optimization looked for Psi4’s optimized structure in the wrong directory, did not find it, and silently kept the starting coordinates – while still naming the configuration as optimized. The optimized structure is now written for each Optimization and stored, aligned with the starting structure, and a missing structure is an error rather than silently ignored. Later steps in the same Psi4 step (e.g. Thermochemistry) were not affected: Psi4 itself used the optimized geometry.

  • The results table again gives the number of optimization steps and whether the optimization converged when cclib cannot read Psi4’s output (as for the current conda-forge build of Psi4 1.11).

2026.11.8 – Bugfix: wrong correlated gradients in optimizations; derivatives and CC convergence options

  • Bugfix: Psi4’s optimizer keeps the previous step’s scratch files, and a correlated gradient computed with them present can be wrong – for the all-electron CCSD(T) gradient of CH4 near its minimum, ten times too large, so a tight optimization wandered uphill and never converged. Optimizations with correlated methods now clean the scratch files before each gradient. The previous step’s orbitals are still used as the starting guess, and Hessians, whose displaced gradients Psi4 computes afresh, were not affected.

  • New “Derivatives” in Optimization and Thermochemistry: “analytic where available” (as before) or “finite differences of energies”, which uses only energies even where Psi4 has analytic gradients – a check on them, or a way around them.

  • New “CC amplitude convergence” for the correlated methods: Psi4’s r_convergence, the convergence of the amplitude and lambda equations. “default” leaves Psi4’s own.

  • Bugfix: with “use existing parameters”, Thermochemistry described the step’s own unused settings (e.g. “B3LYP”) rather than the previous step’s method, which is what runs. It now names the previous step’s method, integrals and frozen core.

  • Bugfix: a Thermochemistry step with its own parameters ignored its “Correlation integrals” setting.

2026.11.7.2 – Density-fitted and Cholesky correlation integrals, with analytic DF gradients

  • New “Correlation integrals” setting for MP2, MP2.5, MP3, LCCD, CCD, CCSD, CCSD(T) and the orbital-optimized methods: Psi4’s default for the method, conventional, density fitting (with an optional fitting basis), or Cholesky decomposition.

  • Density-fitted CCSD(T) and CCSD have analytic gradients with a frozen core, via Psi4’s DFOCC module, so frozen-core coupled-cluster optimizations and Hessians no longer need finite differences of energies. The step uses DFOCC for the energy too, so energy and gradient come from the same code.

  • A model whose integrals differ from the method’s default is named with a prefix – DF-CCSD(T)/cc-pVTZ, CD-…, CONV-MP2 – so its results are kept apart from the others.

  • The rules for analytic gradients now cover the integrals: conventional MP2, MP2.5, MP3 and LCCD gradients are all-electron only, and Cholesky-decomposed integrals have none; the step then uses energies and says so.

2026.11.7.1 – Bugfix: Psi4 runs on node-local scratch under a queueing system

  • Under a queueing system such as SLURM, Psi4 now runs in a temporary directory on the node’s local disk ($TMPDIR), with its scratch files there too, and the output and results are brought back to the step’s directory. Before, it ran in the job’s directory, often on a shared filesystem: on a busy day the many small writes Psi4 makes (its output, timer and scratch files) stalled it, so that a CCSD(T) gradient that takes 20 s ran for many minutes. Interactive runs are unchanged.

  • The developer guide has the design of a Psi4 backend for the Interaction EDA step (second-generation ALMO-EDA on CPUs).

2026.11.7 – Bugfix: analytic CCSD(T) gradients, and no silent finite differences

  • CCSD(T) optimizations and thermochemistry used finite differences of energies, not Psi4’s analytic gradients: Psi4 uses them only when told to use its ccenergy module, which the step did not do. All-electron CCSD(T) now uses the analytic gradients, which for a Hessian is many times faster.

  • Psi4’s analytic CCSD(T), CCSD, EOM-CCSD and LCCD gradients are all-electron only. With the core frozen the step now asks Psi4 for finite differences of energies explicitly and warns in the output, rather than Psi4 switching silently (or, for CCSD, stopping with an error).

  • An energy calculation computes the gradient as well only when it is analytic, so a CCSD(T) energy with a frozen core no longer costs a finite-difference gradient.

  • CCD is marked as having no analytic gradient. The user guide explains which methods have analytic gradients.

  • A Psi4 output that cclib cannot read, e.g. from a development build of Psi4, no longer stops the step; the results come from Psi4’s own data as before.

2026.10.24 – The optimized structure keeps its name

  • The structures that optimizations write are now named ‘{current name} optimized with {model}’ by default, keeping the name the structure had, such as a title read from an SDF file: ‘DMC_cc optimized with B3LYP/def2-SVP’. Before, every structure was named ‘optimized with <model>’, so the titles of several molecules were lost. Flowcharts already saved keep their setting, and the previous choice remains in the list. Needs seamm 2026.10.24.

2026.10.18 – Internal: read the output files from disk

  • seamm-exec 2026.10.18 no longer reads the files a calculation returns into memory, since they can be gigabytes. The Psi4 step now reads the ones it needs from the step’s directory, which works with earlier versions of seamm-exec too.

2026.10.11 – The step declares what its cost model is made of

  • psi4.TIMING_SPEC – basis functions, electrons and atoms as the size variables, the method as the class, the task, and SCF runs as the unit – is passed with each timing record (seamm-exec 2026.10.6.1 and later write it beside the records), so the cost model is fitted from the step’s own description.

  • Removed the Docker image (and its recipe and release job) and the Docker settings in the default psi4.ini, which were no longer used or maintained.

  • Requires Python 3.12 or later.

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

  • Each Psi4 run appends a record to ~/.seamm.d/timing/psi4.csv through seamm_exec.timing: the machine class, processes, wall time and outcome, and the descriptors of the calculation: the method and basis, the structure, and from the output the basis functions, electrons, SCF iterations and Psi4’s own wall time. This replaces the step’s own CSV (SMILES, formula and the whole parameter dictionary as JSON), which grew without bound. See seamm_exec’s campaign of 2026-10-05.

  • Requires seamm-exec 2026.10.6.

2026.10.1 – Settings that depend on each other, and three bug fixes

  • The dialogs show only the settings that apply with the current choices, and SEAMM’s flowchart tools use the same rules: the functional only for DFT; the frozen core only for correlated methods; a dispersion correction only from the functional’s own list; the damping, level-shift and second-order SCF controls only when switched on.

  • BSSE no longer offers the 17 Energy settings it does not use (Psi4’s own SCF settings are used, and no plots are made), and Thermochemistry’s plots apply only when it uses its own settings.

  • Bugfix: Thermochemistry right after Initialization tried to use the previous step’s method; it now uses its own settings.

  • Bugfix: a functional given by its short name (e.g. a variable set to ‘b3lyp’) raised KeyError.

  • Bugfix: keeping the dispersion choice valid in the dialog raised a TypeError.

  • Documented in the user guide. Needs seamm 2026.10.1.

  • Internal: CI now installs the package’s declared dependencies with uv rather than a conda test environment; seamm-exec, which it uses, is now declared.

2026.8.7 – Bugfix: catch a mis-assigned per-fragment BSSE charge before running Psi4

  • The BSSE (counterpoise) sub-step now passes the cluster’s atomic numbers into seamm_bsse.validate_fragments, which checks that every fragment has an even electron count at its assigned charge. Previously, a per-fragment charge given to the wrong fragment (e.g. an ion’s charge assigned to its neutral partner instead of the ion) could still sum correctly to the complex’s own charge and pass validation, only to fail deep inside Psi4 once that fragment ran as an impossible odd-electron singlet. This is now caught up front with a clear message naming the offending fragment.

  • Leaving fragment charges empty no longer always means all-neutral: if the structure carries per-atom formal charges (e.g. an ion marked with an SDF/MOL M  CHG record, such as the Na+ in a Na+..H2O complex), each fragment now defaults to the sum of its own atoms’ formal charge. This covers monatomic ions (Na+, Cl-) and polyatomic ions (NH4+, BF4-) alike without the caller needing to spell out fragment charges at all – removing the main way to trigger the mis-assignment above in the first place. An explicit fragment charges still overrides it.

2026.8.4 – Added a BSSE (counterpoise) sub-step

  • New BSSE sub-step: computes the counterpoise-corrected (Boys-Bernardi) energy and gradient of an N-fragment complex, with an independent charge per fragment – needed for ions, e.g. a Na+/Cl- pair or a Na+/Cl-/H2O cluster. A thin wrapper around Psi4’s own native N-fragment counterpoise driver (bsse_type=['cp', 'nocp']); requires the new seamm-bsse package for the fragment-definition/validation layer shared with orca_step’s own BSSE sub-step. Validated against real Psi4 and cross-checked against orca_step’s independently-implemented BSSE sub-step on the water dimer, charged ion/water pairs, and a three-fragment Na+/Cl-/H2O cluster – agreement to a few hundredths of a kcal/mol throughout.

2026.3.1 – Internal: switching from deprecated library pkg_resources to importlib

2025.3.9 – Optimization and thermochemistry enhancements

  • Expanded the output to include gap, HOMO, LUMO, RMSD between initial and final structure for optimizations, dipole moment, etc.

  • Added control for updating the structure after optimization.

  • Added enthalpy of formation to the thermochemistry output for known basis sets.

  • Updated the installation to work with recent changes.

2024.12.7 – Minor update due to changes in molsystem

  • The molsystem was updated to improve the handling of properties. This required changes in this module to match the new molsystem.

2024.10.15 – Bugfix: error if used in a loop and previous directories deleted.

  • The code crashed if called with a loop in the flowchart, and the last directory of a previous loop iteration was deleted before running the next iteration.

2024.10.5 – Enhancements and bug fixes for thermochemistry

  • Improved GUI for thermochemistry so that it automatically recognizes whether it is after e.g. an optimization and configures appropriately.

  • Fixed and issue with transferring the multipole moments to the JSON file

2024.7.30 – Enhanced results and fixed problem with psi4.ini

  • Added to the results by using cclib to parse the output.

  • If ~/SEAMM/psi4.ini did not exist the version that was automatically created was not complete, causing calculations to fail.

2024.5.23.3 – Added auxiliary codes that help Psi4

  • Psi4 uses codes like DFTD4 and DFTD4 for parts of e.g. dispersion calculations. This release adds them to the Conda install SEAMM uses for Psi4.

2024.5.23.2 – Bugfix: incorrect name for gradients

  • There was a typo in the name for the gradients, such that they could not be output to Results.json.

  • The units for the energy and gradients in Results.json were incorrect.

2024.5.23.1 – Internal fix for creating Docker image.

2024.5.23 – Added standard energy and gradients to results

  • Added ‘energy’ and ‘gradients’ to optional results to support e.g. Energy Scan

  • Fixed crashing bug in description of the Energy substep.

2024.3.17 – Updated the installer

  • Updated the installer to use the new version of the SEAMM installer.

  • Finalizes installing either with Conda or Docker

2024.3.4 – Allowing short names for method and DFT functionals

  • Added short names for the methods (Hamiltonians) and DFT functionals.

  • Catch errors in Psi4 calculating properties for e.g. CISD(T) method

2024.2.29 – Completed support for containers

  • Fixed issues with running amd64 containers on arm64 systems.

2024.1.11 – Changes to allow running in containers.

  • Moved to the new executor and ensured it still runs directly.

  • Fixed bugs in printing the summary output.

2023.8.23 – Fix for installation of Psi4

  • Psi4 is now available on CondaForge, so install from there if requested.

  • Psi4 crashes if asked to optimize and atom, so change to doing a single point energy is optimization is requested for and atom.

2023.8.22 – Enhancement of orbital plots

  • Added structure to the orbital plots

  • Fixed a bug if the default number of cores was not ‘available’

2023.6.4 – Enhancements

  • Added thermochemistry substep to compute the vibrational and other corrections for thermochemistry.

  • Added ability to create .cube files for plotting the density and orbitals.

2023.2.21 – Added control over SCF convergence

  • Able to set convergence criteria.

  • Options for damping. level shifting and second-order SCF

  • Set default for number of steps for optimization to 6*nAtoms to make it more explicit.

2023.2.17 – Checking more thoroughly for errors in Psi4

  • Check and throw errors for various issues when running Psi4, including whenever it does not complete successfully according to the output.

2023.2.16.1 – OptKing will not die!!!

  • Yet another part needed to be removed.

2023.2.16 – Thoroughly removing calls to OptKing

  • OptKing is causing compatibility issues, and is not currently being used in the Python, so am temporarily removing it.

2023.1.23 (23 January 2023)#

  • Corrected bugs, including if no dispersion term requested

  • Added preliminary version of accelerated optimization using MOPAC for Hessians, etc. May not fully work at the moment.

  • Updated documentation theme and structure to match new style.

2022.11.18 – Added total charge and multiplicity to Psi4 input

2021.2.11 (11 February 2021)#

  • Updated the README file to give a better description.

  • Updated the short description in setup.py to work with the new installer.

  • Added keywords for better searchability.

2021.2.4 (4 February 2021)#

  • Updated for compatibility with the new system classes in MolSystem 2021.2.2 release.

2020.12.5 (5 December 2020)#

  • Internal: switching CI from TravisCI to GitHub Actions, and in the process moving documentation from ReadTheDocs to GitHub Pages where it is consolidated with the main SEAMM documentation.

2020.11.2 (2 November 2020)#

  • Updated to be compatible with the new command-line argument handling.

2020.9.28 (28 September 2020)#

  • Improved handling of results

    • Selection depends on both calculation and method.

    • Arrays can be stored in variables (but not tables!).

    • Multipoles gathered as arrays rather than a large number of scalars.

2020.8.1 (1 August 2020)#

  • Fixed minor bugs.

2020.7.3 (24 July 2020)#

  • Added control options for number of threads and amount of memory to use.

2020.7.2 (23 July 2020)#

  • First pass at specifying number of threads and amount of memory.

2020.7.1 (23 July 2020)#

  • Added substep for optimization.

2020.7.0 (21 July 2020)#

  • First release on PyPI of initial working version.