Publications
Our research
Preprints and software are archived on Zenodo with permanent DOIs. Each link below resolves to the archived record.
Thermal and two-phase cooling
Trustworthy machine learning for boiling heat transfer and critical heat flux.
- Preprint
Leakage-Safe Evaluation for Multimodal Boiling Machine Learning: A Domain-Shift Baseline on BoilingBench
A conservative, leakage-aware baseline evaluation on the public BoilingBench multimodal release, showing how random-window splits can overstate how well boiling models transfer to new conditions.
- Preprint
CHF-Watch: Leave-One-Surface-Out Evaluation of Multimodal Boiling Heat-Flux Learning, Uncertainty Calibration, and Research Alarms
Heat flux learning from multimodal boiling signals evaluated by holding out entire surfaces, with uncertainty calibration and research alarms for out-of-distribution conditions.
- Software
CHF-Watch
The archived code release behind the CHF-Watch preprint: a leakage-safe, leave-one-surface-out evaluation package for multimodal boiling and critical heat flux machine learning on the public NED3-007 dataset (acoustic emission, hydrophone and thermal channels).
Computational mathematics
We also publish in computational number theory, applying the same standards that guide our thermal work: precise statements, computations anyone can rerun, and a clear line between what is proved, what is certified by computer, and what remains conditional. These preprints study unit signatures and class groups of totally real number fields, with PARI/GP verification scripts and their outputs included in each source archive.
- Preprint
Signature Deficiency and Subfield Unit Indices in Totally Real Biquadratic Extensions
Relates the Kuroda unit index of a totally real biquadratic extension to the real signatures of its subfields, with certified examples and a counterexample showing where the bound fails.
- Preprint
Primitive Unit-Signature Relation Modules in Degree-Eight Fields
Studies signature deficiency in degree-eight fields through a module invariant, supported by a reproducible PARI/GP-assisted census. Computational results are marked as unconditionally certified or conditional on GRH where applicable.