Computational physics.
Verifiable results.

Implemented research prototype

Thermal memory in a synthetic Al–Zr slab

A coupled calculation follows heat conduction and integer-cluster kinetics across material regions. Processing and service share the evolving thermal and chemical state.

• Positive backward-Euler and independent BDF kinetic solvers.
• Command-line evaluation and bounded optimization of furnace programs.
• Conservation and detailed-balance checks, thermal-grid checks and selected size refinements.
• Structured input contracts, input hashes, diagnostics and execution journals.

Scientific AI evaluation

Modal sensitivity at repeated eigenvalues

Reference analysis checks AI-generated solutions to generalized eigenvalue problems where repeated modes invalidate ordinary individual-mode gradients.

• Projected perturbation matrices within repeated eigenspaces.
• Directional derivatives and first-order frequency predictions.
• Finite-difference comparisons and eigenspace basis-rotation checks.
• Clear explanations of mathematical errors and their consequences for optimization.

Reproducibility

Explicit commands and solver settings, structured inputs and outputs, source and input hashes, package installation checks, and records that distinguish successful calculations from failures or interrupted runs.

Recent thermal-memory execution is documented on Windows with PowerShell. Linux execution and container validation remain unverified for that prototype.

Connection to Earth Sciences

Heat transport, dynamical systems, numerical stability and verification are relevant methods for geophysical and environmental models.

Implemented Kepler applications currently focus on materials physics and mechanics. Direct processing of climate, satellite, hydrological or seismic observations is not claimed.