What is published
Ripple tank solves the two-dimensional wave equation with finite differences and absorbing edges. Orbits integrates Newtonian gravity for several bodies with a symplectic leapfrog scheme and reports energy drift, so you can see how step size affects accuracy. Wavepacket evolves a one-dimensional quantum state with the split-step Fourier method and measures how much of it tunnels through a barrier.
Questions being worked on
Learned models that respect physics
When should conservation laws, symmetries and boundary conditions be built into a learned surrogate rather than hoped for? The useful test is generalisation across scales and geometries, not accuracy inside the training range.
Uncertainty in inverse problems
Inferring hidden parameters from incomplete observations means separating measurement noise, model error, numerical error and the uncertainty a learned component adds.
What a quantum algorithm actually improves
A claim needs the task, input model, error tolerance, hardware assumptions and the best classical comparison. A small state-vector simulator is listed as a concept under study.
How results are recorded
Each computational result keeps the system and regime, governing equations and approximations, the numerical method and its convergence, the comparison baseline, and a label saying whether it is theory, classical simulation, hardware execution or an independent reproduction. A simulation is reported as a simulation.
Revised 25 October 2026. Questions and corrections: [email protected]