PyTC

JAX-based research framework for transcorrelated electronic-structure methods

Status: Active research and development at Yale University. The project was initiated and initially developed at the Max Planck Institute for Solid State Research.

Repo: nickirk/pytc

What is PyTC?

PyTC is a research framework for transcorrelated (TC) and explicitly transcorrelated (xTC) electronic-structure methods. It is built with JAX so that numerical kernels can use compilation, automatic differentiation, and accelerator backends while remaining accessible from Python.

The project connects transcorrelated Hamiltonian construction with correlated-electron solvers and modern low-rank numerical methods. Its development is guided by reproducible accuracy tests: compressed and accelerated calculations are checked against uncompressed references before performance conclusions are drawn.

The latest manuscript, Interpolative Separable Density-Fitting for Transcorrelated Hamiltonians, presents the ISDF-xTC method implemented in PyTC. It combines low-rank compression of grid-evaluated transcorrelated integrals with multi-GPU execution and automatic differentiation, and applies ISDF-xTC-CCSD to hydrogen chains through cc-pV5Z and benzene with up to 1200 orbitals (cc-pCV5Z).

Current research directions

  • Low-rank xTC Hamiltonians. Interpolative separable density fitting (ISDF) compresses the integral tensors that enter xTC calculations. Current validation covers hydrogen chains and large molecular calculations on benzene.
  • Neural quantum states and quantum embedding. PyTC provides a common numerical foundation for combining transcorrelated Hamiltonians with learned wavefunctions and embedding workflows for correlated materials.
  • Reliable scientific software. The code emphasizes explicit provenance, numerical diagnostics, CPU/GPU parity checks, and clearly separated reference and accelerated paths.

Why this matters

Transcorrelation moves difficult short-range correlation effects from the wavefunction representation into a similarity-transformed Hamiltonian. Low-rank factorization and accelerator-friendly implementations can then reduce the storage and computational cost of working with that Hamiltonian. PyTC is designed as a testbed for determining when these ideas deliver accurate, practical gains for molecules and periodic systems.

  • Interpolative Separable Density-Fitting for Transcorrelated Hamiltonians — arXiv:2607.17314 (2026)
  • Density Matrix Renormalization Group for Transcorrelated Hamiltonians: Ground and Excited States in Molecules — Journal of Chemical Theory and Computation (2023)
  • Towards Efficient and Accurate Ab Initio Solutions to Periodic Systems via Transcorrelation and Coupled Cluster Theory — Physical Review Research (2021)