LFortran and Enzyme Enable Automatic Differentiation in Fortran Code

Researchers at Pasteur Labs have demonstrated automatic differentiation (autodiff) in Fortran by combining the LFortran compiler with the Enzyme autodiff framework. The integration allows Fortran code to be differentiated automatically, a capability long common in modern ML-oriented languages but historically absent in Fortran. This development is significant for scientific computing, where large legacy Fortran codebases are widely used in fields such as climate modeling and physics simulations. The work was published in a technical blog post in July 2026 as part of the Tesseract Core project documentation. By enabling differentiable Fortran, the effort could help bridge the gap between traditional scientific software and modern machine learning workflows.