Roadmap

JaQMC’s near-term direction is to make advanced capabilities fit naturally into the same modular workflow model that already powers current molecule, solid, and quantum Hall workflows. Rather than growing a long list of isolated features, the focus is on turning useful legacy and system-specific ideas into reusable, first-class framework components.

In practice, that means investing in estimator infrastructure, Hamiltonian support, and training objectives that broaden what can be expressed without forcing users onto special-case code paths.

This page is informational: it highlights the areas JaQMC is currently evolving toward. It is not a public task list, contribution roadmap, or commitment to a specific timeline or implementation order. The directions below reflect current project emphasis and may be useful context for contributors.

Legacy Feature Migration

We will gradually integrate features from the previous version of JaQMC into the current framework.

• Sparse forward Laplacian support Improve the current forward Laplacian handling to support baking sparsity information directly into the network definition. Ref: Forward Laplacian.

• Diffusion Monte Carlo Ref: NNDMC.

• Pseudo Hamiltonian support Bring pseudopotential-related machinery into the current framework as first-class, configurable support rather than legacy add-ons. Ref: NNQMC-PH.

• Spin- and overlap-aware training penalties Modernize useful legacy penalty mechanisms so they integrate cleanly with today’s training stack. Ref: Spin-plus Penalty.

Long-term Roadmap

We aim to continuously improve JaQMC, with efforts focused on, but not limited to, the following areas:

• Neural Network Ansatz

• Optimization Methods

• Systems and Hamiltonians

• Reusable Estimators

• Other Cutting-edge Techiniques