pysimpler

📖 Docs: https://alexlu997.github.io/pysimpler/

A small, readable numpy implementation of the SIMPLE / SIMPLER finite-volume pressure–velocity coupling (Patankar 1980) on a staggered grid — built as a study reference and a forkable skeleton for writing your own CFD solver.

• numpy only. TDMA (tri-diagonal solver) is hand-rolled; no scipy/PETSc required.
• SIMPLE and SIMPLER both implemented, side by side.
• Verified: pure-conduction manufactured solution to machine precision; lid-driven cavity matches the Ghia et al. (1982) benchmark.
• Textbook notation kept on purpose (aip/aim/ajp/ajm/ap/con, du/dv, diflow, solve/reset) so it maps cleanly onto any derivation and onto other languages.

Lid-driven cavity at Re=100 (SIMPLER): streamlines coloured by speed, and centreline velocity profiles overlaid on the Ghia et al. (1982) benchmark points. Reproduce with python docs/make_figure.py.

Install

pip install pysimpler-fvm       # from PyPI (import name stays: import pysimpler)
# or from a clone:
pip install -e .                # needs only numpy
pip install -e ".[test]"        # + pytest

The PyPI distribution is pysimpler-fvm (the bare name was taken), but the import name is unchanged — always import pysimpler.

Quickstart

import numpy as np
from pysimpler import Grid, SimpleField, simpler_iterate

g = Grid(L1=42, M1=42, XL=1.0, YL=1.0)     # 40x40 interior cells
s = SimpleField(g)
rho, mu = 1.0, 0.01                        # Re = rho*U*L/mu = 100

def lid(g, s):                             # BC hook (moving top lid)
    s.u[2, :] = 0; s.u[g.L1, :] = 0; s.v[:, 2] = 0; s.v[:, g.M1] = 0
    s.u[:, 1] = 0; s.u[:, g.M1] = 1.0; s.v[1, :] = 0; s.v[g.L1, :] = 0

for _ in range(400):
    smax, ssum = simpler_iterate(g, s, rho, mu, relax_u=0.8, relax_v=0.8, apply_bc=lid)
print("mass imbalance:", ssum)             # -> ~1e-15

Runnable demos in examples/: conduction.py (heat conduction, analytic check) and cavity.py (lid-driven cavity + a SIMPLE-vs-SIMPLER convergence comparison).

What's inside (pysimpler.py)

Verified results

pytest            # runs all four gates above

Extending it (the one idea)

Every physics — buoyancy, swirl, turbulence sources, variable properties — enters through just two things: the effective diffusion Γ and the linearised source S = S_C + S_P·φ (keep S_P ≤ 0). Add your term there, not as a new code path, and the same engine solves it. Non-Cartesian geometries add curvature source terms only.

Contributing

See CONTRIBUTING.md. In short: keep it readable, keep the pytest gates green, numpy-only in the core.

Publishing to PyPI (maintainers)

Releases publish to PyPI automatically via Trusted Publishing (OIDC — no token in the repo). One-time setup on pypi.org: add a pending publisher with PyPI Project Name pysimpler-fvm, owner alexlu997, repo pysimpler, workflow publish.yml, environment pypi. Then each GitHub Release (e.g. tag v0.1.0) triggers .github/workflows/publish.yml, which builds and uploads the wheel

• sdist. After the first successful publish: pip install pysimpler-fvm.

Provenance & license

Independent from-scratch re-implementation; structure follows the classic XJTU (Prof. 陶文铨 group) staggered-grid SIMPLER teaching code. The original teaching Fortran is not included (teaching-only). MIT licensed — see LICENSE and NOTICE. Please also credit Patankar (1980) in academic use.