⚡ GPU-accelerated 🔁 Fully differentiable 📐 Spectral methods 🧠 ML-ready 🧩 Voxel-based simulations 🔍 Inverse material identification
FFTjax is a next-generation FFT-based spectral solver framework inspired by classical FFT homogenization approaches such as FFTMAD — reimagined in JAX for differentiable, GPU-accelerated scientific computing.
At its core, FFTjax implements variational FFT solvers for periodic unit cells, enabling efficient solutions of mechanical, thermal, and multi-physics boundary value problems using spectral methods.
Built for modern computational mechanics, FFTjax bridges:
- 🏗 Variational FFT homogenization
- ⚙️ JIT-compiled, hardware-accelerated execution
- 🔁 End-to-end automatic differentiation
- 🎯 Inverse material calibration
- 📊 Bayesian optimization & uncertainty quantification
- First doc implementation