LoginSign Up
← Back to all sparks
P

Psi4

DEVOPS
Velocity0.0

Open source suite of ab initio quantum chemistry programs

psicode.org ↗

Psi4 is closing the gap with ORCA on the methods that decide which code a lab installs

quantum-chemistrycoupled-clusterqcschemapython-packagingbuild-performancerelativistic-methods
◆Current state
Psi4 ships a major version roughly annually with a trail of conda-compatibility patch releases behind each one. The 1.11 cycle is the most competitively pointed in the window: DLPNO-CCSD and DLPNO-CCSD(T) become callable methods, with cutoffs deliberately tuned to match ORCA, and ZORA arrives for scalar-relativistic core Hamiltonians. Around the science, the project spends heavily on Python-ecosystem plumbing — QCSchema v2, Python 3.14 support, and the QCArchive dependency chain that most of its patch releases exist to unbreak.
◆Where it's heading
Two things are being built at once. Scientifically, the code is filling in the local-correlation and relativistic methods that users otherwise leave for commercial packages, plus external-potential and embedding machinery that makes Psi4 usable as a QM engine inside larger workflows. Structurally, it is betting on the QCArchive stack — qcelemental, qcengine, qcmanybody, optking, qcfractal — which delivers interoperability but also means a Python packaging change downstream can force a release, as 1.10.1 and 1.10.2 both did.
◆Prediction
The DLPNO work landed as energies only, so analytic gradients for DLPNO-CCSD are the natural next step. Expect at least one more 1.11.x patch driven by the QCFractal and pydantic constraints that the 1.11 notes flag as still unresolved for Python 3.14.

◆Recent moves

  1. 1mo ago

    Psi4 1.11 makes DLPNO-CCSD(T) callable and matches ORCA's cutoffs

    ⚡ SPARK

    The release where Psi4 stops ceding large-molecule coupled cluster: DLPNO-CCSD and DLPNO-CCSD(T) become first-class energy methods, with defaults explicitly aligned to ORCA. ZORA, MBIS across the full periodic table, and arbitrary one-electron external potentials extend the same pattern of covering what users currently leave the package to do elsewhere.

    View source ↗
  2. 1mo ago

    Psi4 1.10.2 unbreaks adcc with QCFractal 0.65

    A single-PR release to avoid a bad adcc and QCFractal v0.65 combination that failed snowflake jobs. Emblematic of the maintenance tax that comes with sitting on the QCArchive dependency chain.

    View source ↗
  3. 2mo ago

    Psi4 1.10.1 widens conda pins, fixes path-advisor solving

    Conda package allowances for OpenOrbitalOptimizer, QCElemental, QCEngine, QCManybody and pybind11, plus a psi4-path-advisor fix for optional dependencies that pull Psi4 back in as their own dependency. Packaging work, not science.

    View source ↗
  4. 11mo ago

    Psi4 1.10 adds MP2-F12, SAPT0-D4M and an LS-THC Python interface

    A 98-PR release that broadens the correlation and interaction-energy menu — conventional and DF MP2-F12/3C(FIX) energies, new SAPT variants, and a pybind interface for building LS-THC quantities to prototype tensor-hypercontraction methods. QCManyBody becoming a hard requirement marks the deepening commitment to the QCArchive stack.

    View source ↗
  5. 2y ago

    Psi4 1.9 adds unrestricted-LDA analytic Hessians and new SCF guesses

    A 79-PR release, note that the feed timestamp trails the actual 6 December 2023 release date by six months. Analytic Hessians for unrestricted DFT with LDA functionals plus the fitted SAP and MODHUCKEL guesses continue the incremental broadening of the SCF and property stack.

    View source ↗
  6. 2y ago

    Psi4 1.9.1 pins pytest 7 and prefers released libint

    Five PRs of conda pinning and small fixes, including a symbols list correction for high angular momentum and an import failure when the home directory is symlinked. The usual cleanup pass that follows a Psi4 major release.

    View source ↗