tpt-chemistry

A pure-Rust, AI-native computational chemistry and molecular simulation library: molecular dynamics (MD), quantum chemistry primitives (Hartree–Fock with Gaussian basis sets), reaction kinetics, and crystallography — without C/C++ FFI or external solver dependencies.

No LAMMPS. No Gaussian. No Quantum ESPRESSO. SCF iteration, McMurchie– Davidson electron-repulsion integrals, Ewald summation, Particle Mesh Ewald (on an in-house radix-2 FFT), and neighbor lists are all implemented from scratch, leaning on the sibling tpt-math crates for typed units and numeric utilities.

Part of the TPT science stack. Full design rationale lives in spec.txt; the build progress in todo.md.

Examples

Runnable end-to-end demos live in the umbrella crate (cargo run -p tpt-chemistry --features <feat> --example <name>):

Example	Features	Shows
h2_hf	quantum	RHF/STO-3G bond-length scan, dissociation
water_hf	quantum	SCF energies, orbital spectrum, Molden output
lj_argon_nve	md	NVT equilibration → NVE production, energy drift
nacl_pme	md	Direct Ewald vs Particle Mesh Ewald convergence
lotka_volterra_ssa	kinetics	Gillespie SSA vs mass-action ODE
michaelis_menten	kinetics	Enzyme kinetics, both regimes
cpp_calcium_phosphate	aqueous	Calcium-phosphate onset pH vs temperature, with and without a chelator
silicon_xrd	crystal	Diamond-structure powder pattern as CSV
nacl_xrd	crystal	Rock-salt extinctions and peak families
cif_to_xrd	io, crystal	CIF text → structure → diffraction pipeline

Building

This workspace depends on the sibling repository tpt-math via path dependencies (../tpt-math/...), so a fresh clone needs both checkouts side by side:

git clone https://github.com/tpt-solutions/tpt-math.git
git clone <this repo> && cd tpt-chemistry
cargo build && cargo test

CI performs the same sibling checkout in every job (see .github/workflows/ci.yml).

Design principles

• Zero external solver dependencies — everything from scratch in safe, pure Rust.
• Strict licensing — workspace is MIT OR Apache-2.0, and the dependency graph is held to a stricter bar: no Apache-2.0-only crates (deny.toml enforces it in CI).
• Type-state & dimensional safety — phantom types distinguish Cartesian from internal (Z-matrix) coordinates; physical quantities are wrapped with compile-time units (tpt-math-units).
• Native formal verification — Kani bounded model checking for critical loops, proptest for statistical verification of physical laws (energy conservation, mass conservation, phase-space volume preservation).
• no_std + alloc core — tpt-chem-core builds for bare-metal targets.

Crates

Crate	Purpose
tpt-chem-core	Molecular graphs, force fields, physical constants, unit-safe types
tpt-chem-md	Integrators, thermostats/barostat, cell/neighbor lists, Ewald + PME (in-house FFT), virial/pressure, NVE/NVT/NPT run loops, energy minimisation
tpt-chem-quantum	Gaussian basis sets, McMurchie–Davidson ERIs, SCF / Hartree–Fock (charged species), Molden orbital output
tpt-chem-kinetics	Mass-action ODEs, Gillespie SSA, Arrhenius/Eyring rates
tpt-chem-aqueous	Activity coefficients (Debye–Hückel, Davies, Pitzer, eNRTL), Ksp/saturation indices, chelator-controlled calcium phosphate precipitation onset
tpt-chem-crystal	Bravais lattices, space groups, reciprocal space, XRD simulation
tpt-chem-io	XYZ, PDB, MOL2, CIF readers/writers, trajectory streaming
tpt-chem-verify	proptest strategies, conservation-law checks, Kani harnesses
tpt-chemistry	Feature-gated umbrella crate re-exporting all of the above

Synergies (spec.txt §5)

• tpt-materials — micro-to-macro: MD/DFT results feed homogenized properties (elastic constants, thermal conductivity).
• tpt-math — dense linear algebra and ODE machinery.
• tpt-dsp (future) — could replace the in-house 3D FFT behind PME; currently tpt-math-signal-fft — FFT utilities (PME ships its own in-house radix-2 transform) vibrational density of states.
• tpt-fem / tpt-physics — QM/MM coupling of the quantum region into continuum FEM/CFD environments.
• AI-native agents (tpt-eve, tpt-anima) — forces and energy surfaces expose autodiff hooks for training machine-learned interatomic potentials and differentiable molecular design.

Quick start

just check      # fmt + clippy + test + cargo deny
just no-std     # verify the core builds for thumbv6m-none-eabi
just kani       # bounded model checking (Linux/WSL only)

use tpt_chem_core::molecule::{BondOrder, Molecule};

let mut mol = Molecule::new("H2");
let h1 = mol.add_atom::<2>([0.0, 0.0, 0.0].into());
let h2 = mol.add_atom::<2>([0.0, 0.0, 0.74].into());
mol.add_bond(h1, h2, BondOrder::Single).unwrap();
assert_eq!(mol.mass(), 2.0 * 1.008);

Hartree–Fock for H₂ at R = 1.4 Bohr (STO-3G, lit. −1.11676 Eₕ):

use tpt_chem_core::molecule::Molecule;
use tpt_chem_core::units::BOHR_ANGSTROM;
use tpt_chem_core::vec3::Vec3;
use tpt_chem_quantum::hf::rhf_energy;

let mut h2 = Molecule::new("H2");
let r = 1.4 * BOHR_ANGSTROM;
h2.add_atom::<1>(Vec3::new(0.0, 0.0, -r / 2.0));
h2.add_atom::<1>(Vec3::new(0.0, 0.0, r / 2.0));
assert!((rhf_energy(&h2).unwrap() - (-1.11675931)).abs() < 5e-4);

License

Dual-licensed under MIT or Apache-2.0, at your option.