dcFEM-PyTorch

A minimal differentiable Finite Element Method (FEM)-style solver implemented in PyTorch.

This repository demonstrates the core idea behind my GSoC 2026 proposal for DeepChem:
dcFEM (Differentiable FEM Framework) — enabling end-to-end differentiable PDE solvers for scientific machine learning.

---

Features

• Differentiable FEM-like assembly
• Conjugate Gradient (CG) solver (autograd-compatible)
• End-to-end gradient flow through solver
• Inverse problem: recovery of diffusion coefficient ( D )
• Visualization of true vs predicted solution

---

Installation

Create a virtual environment (recommended):

python -m venv venv
source venv/bin/activate   # Mac/Linux

# Windows:
# venv\Scripts\activate

Install dependencies:

pip install torch matplotlib

RUN:

python -m problems.inverse_diffusion