How does surrogate accuracy degrade when detector dimensions change versus when sensor spacing patterns are altered?

Question

Antigravity · Accepted Answer

This is an excellent technical question, but the preprints provided don't directly address surrogate model degradation under geometric parameter changes versus spatial configuration changes. Let me explain what's close and what's missing.

What the archive contains (partially relevant)

Two preprints develop neural surrogates for detector design, but focus on *training* rather than *failure modes*:

Surrogate robustness in detector design: Papers [AG-2025.11-530] and [AG-2025.03-1551] both train deep learning surrogates to approximate expensive physics simulators—one for gravitational wave detectors, one for calorimeter optimization. However, neither systematically compares how surrogate *accuracy* degrades when you vary detector dimensions versus sensor spacing. Paper [AG-2025.11-530] mentions that "small changes in physical parameters can change the output by orders of magnitudes," which suggests surrogates may be sensitive to geometry—but the paper doesn't isolate whether dimensional changes versus spacing pattern changes degrade accuracy differently.

Sensor robustness more broadly: Paper [AG-2024.05-1982] addresses robustness in quantum sensing by decorrelating target signals from nuisance parameters using machine learning. This is thematically related (ML + sensor design + noise filtering) but concerns parameter correlation, not surrogate accuracy degradation modes.

Spatial measurement precision: Papers [AG-2024.06-2230] and [AG-2024.07-2168] study optimal measurements in spatial superresolution and find that measurement tradeoffs exist—suggesting that different spatial configurations have different information-theoretic costs. But these don't address surrogate model accuracy specifically.

What's missing

The archive doesn't contain a systematic study comparing:

How neural surrogate accuracy changes with *absolute detector size* (e.g., larger calorimeter layers)
versus changes in *relative spacing* (e.g., sensor array pitch or interferometer arm separation ratios)
or which type of geometric change is a larger source of surrogate error.

This would require an ablation study on a surrogate model across a controlled parameter sweep—work that isn't present here.