Skip to content

mechdsl-core

Write your mechanics in LaTeX. Get a tested finite-element solver back.

mechdsl-core is a LaTeX-to-code compiler for computational solid mechanics. You describe a boundary-value problem — geometry dimension, element type, formulation, constitutive model, boundary conditions — using % mechanics directives embedded in an ordinary LaTeX document. The compiler parses that, derives the kinematics and stress/tangent symbolically, and emits deterministic Taichi solver code.

The same .tex file renders normally through pdflatex (the directives are LaTeX comments) and is executable input to the compiler. Your paper's source can be your simulation's source.

It is published on PyPI under the MIT license:

pip install mechdsl-core              # emit code — Taichi-free, small install
pip install "mechdsl-core[verify]"    # the full engine — run and verify solves

Why a colleague should care

Without MechDSL With MechDSL
Hand-code element routines, stress kernels, tangents Write the energy / model in LaTeX; the compiler derives stress S = ∂Ψ/∂E and tangent C = ∂²Ψ/∂E² for you
Constitutive math drifts from the paper it came from The model in the paper is the model that runs
Voigt ordering, sign, and index bugs cost days Conventions are enforced in one place and verified against reference kernels
Every new material = a new error-prone kernel New hyperelastic models are derived, not hand-written

The core principle: derive models from LaTeX; don't hand-code what the compiler should generate.

A 60-second taste

from mechdsl import compile_latex

source = r"""
% mechanics dim 3
% mechanics cell hex8
% mechanics formulation total_lagrangian
% mechanics material svk --E 200e3 --nu 0.3
% mechanics boundary fix  --type dirichlet --value 0 --components 0 1 2
% mechanics boundary load --type neumann --traction "0 0 -1000"
"""

bundle = compile_latex(source)
print(bundle.element_ir_summary)   # what got localised (element, quadrature, einsum specs)
print(bundle.content_hash())       # deterministic — same input, same hash, every time

That call runs the full pipeline: parse the directives → build the Mechanics IR → localise to an Element IR → plan the tensor contractions → emit Taichi. See Getting started to run it yourself.

What it supports today

  • Elements: Hex8 (MVP-stable), plus Hex8-R, Hex20, Tet4, Tet10 (experimental)
  • Formulations: Total Lagrangian (MVP-stable), Updated Lagrangian, convected curvilinear coordinates
  • Elasticity: St. Venant–Kirchhoff, Neo-Hookean
  • Hyperelasticity: Mooney-Rivlin, Ogden, HGO (fiber-reinforced/anisotropic)
  • Plasticity: J2 with power-law isotropic, linear kinematic, and mixed hardening (return-mapping, transpiled via algo2code)
  • Viscoplasticity & damage: Perzyna, Johnson-Cook, Lemaitre
  • Backends: Taichi (MVP-stable, GPU-capable); MFEM and MOOSE (experimental)

See the full constitutive model catalog and the support-tier policy.

Where to go next