Publications
These span a few different method families, but they're all chasing the same thing: making PDE-based workflows cheaper and scalable.
Featured
Geometry that changes shape is one of the more annoying things to handle in reduced-order modelling, since it usually means remeshing every time. This combines unfitted methods with reduced basis ideas to avoid that.
2026 · Preprint
Sequential Bayesian filters and variational estimators behind one Julia package, with localisation, inflation, bias correction, and integration with ROM pipelines for efficiency. Tested on Lorenz-63, a Van der Pol oscillator with biased observations, Reynolds number discovery in a turbulent Navier-Stokes flow, and a heat equation running on a reduced-order surrogate instead of the full model.
2025
The software paper for GridapROMs.jl, the reduced-order modelling package I built and use daily. It spells out the interface design and performance choices behind it.
2025
Uses tensor-train structure to cut the cost of basis construction and operator approximation, while maintaining the typical robustness of a reduced-order model.
2024
Space-time interpolation for transient reduced-order models, with a focus on making residual and Jacobian evaluations more efficient.
2024
Compares space-time reduction strategies for unsteady Stokes flow on a haemodynamics-motivated benchmark. The established ST-LSPG approach turned out to have an expensive offline phase and lose accuracy over time; the Galerkin-projection method proposed here stayed cheap and accurate throughout.