SCESS icon

Active2023 – now

SCESS

Schrödinger's Cat's Electronic System Solver, an ab initio quantum chemistry package with a GPU-rendered viewer.

  • C++20
  • Quantum chemistry
  • Eigen
  • Libint2
  • Libxc
  • LAPACK
  • Dahlia
  • Dear ImGui
  • ImPlot
Water's lowest unoccupied orbital (RHF/STO-3G), drawn as an isosurface in the viewer, with the orbital energies, results table and SCF convergence monitor.
13 Electronic structure methods
27k Lines of C++
100 Commits
3 Programs (library, CLI, viewer)

Hartree–Fock, DFT, coupled cluster, configuration interaction and quantum Monte Carlo methods, with explicitly correlated F12 variants, geometry optimisation and an interactive viewer for exploring the results.

SCESS implements a range of ab initio electronic structure methods from first principles, along with a graphical viewer for exploring the solutions. It is split into three programs:

  • QuantumSCESS, a shared library holding the quantum chemistry code
  • SCESS, a command-line driver for single-point calculations
  • ViewerSCESS, a graphical front end built on Dahlia: Metal by default on macOS, OpenGL 4.1 elsewhere, and Vulkan 1.3 too

Methods

Self-consistent field

  • Restricted and unrestricted Hartree–Fock, both with DIIS
  • Kohn–Sham DFT (functionals from Libxc)

Correlation

  • MP2 and MP2-F12, with fixed and optimised amplitudes
  • Full and truncated CI, with Davidson diagonalisation and string-driven σ-vectors for full CI
  • Closed-shell CCSD (with DIIS) and CCSD(T), with triples over the unique virtual triples
  • CCSD-F12b (3C(FIX) ansatz with the CABS approximation)
  • Coupled-cluster and full-CI quantum Monte Carlo (CCQMC, FCIQMC), and the hybrid CCSD-QMC[MP2-F12]

The F12 methods add explicitly correlated terms that depend directly on the inter-electron distance \(r_{12}\), which recovers much of the basis set limit without very large basis sets.

Geometry optimisation

Geometries are optimised with GDIIS (Császár–Pulay, with the Farkas–Schlegel safeguards) in Cartesian coordinates. Optimisation starts from Lindh's model Hessian, applies BFGS updates, and falls back to rational function optimisation (RFO) steps within a trust radius:

  • analytic RHF and UHF gradients, from Libint2 first-derivative integrals
  • central finite-difference gradients for every other deterministic method

The viewer

The viewer opens with its panels docked around the 3D view:

  • Molecule. Load an XYZ file and see the formula, electron count, nuclear repulsion and basis size.
  • Calculation. Choose the method, reference, basis set (searchable) and CABS. Problems are flagged before you run, for example RHF with an odd electron count.
  • Results. Every calculation in the session, with energy differences in hartree, kcal/mol, kJ/mol, eV or cm⁻¹, and CSV export.
  • Orbitals. Molecular orbital isosurfaces from the selected result.
  • Live Monitor. SCF, CC and CI convergence, plus QMC energies and walker populations as the calculation runs.

Performance

Every method runs on all hardware threads. The QMC methods split each cycle's spawning attempts into chunks, each with its own random stream seeded from the cycle and chunk index. For a given seed, a result is therefore bit-for-bit reproducible on any number of threads.