Can you use clustering results to constrain physics models, or mainly just flag potential anomalies for follow-up?

Clustering results can do both—flag anomalies *and* directly constrain physics models—depending on how you use them.

Anomaly Flagging (the straightforward application)

Clustering excels at model-independent anomaly detection. In particle physics, unsupervised clustering methods can be trained on background data to spot deviations without knowing what signal you're looking for [AG-2024.03-1202]. The Cluster Scanning method illustrates this: k-means clustering separates potentially anomalous event clusters in low-level observable space, then checks whether a resonant signal emerges in the anomalous region [AG-2024.02-1209]. This is powerful because you don't need to assume a specific signal model upfront—you're letting the data speak first.

Direct Model Constraints (the powerful but less obvious path)

But clustering patterns themselves can constrain physics models quantitatively. In cosmology, the spatial clustering of galaxies measured from large surveys directly constrains galaxy formation and dark matter physics. Researchers compared two competing models—a halo occupation distribution (HOD) model and a semi-analytical model (SAM)—by fitting both to clustering measurements of massive galaxies [AG-2025.05-291]. The clustering data revealed that simpler HOD models could recover the velocity field predictions of more complex SAMs with 1–3% accuracy, effectively constraining which galaxy-halo connection physics fits the data.

The key insight: clustering encodes information about the underlying physics. When you measure how objects cluster in space, you're measuring the effects of gravity, matter distribution, and the processes that created those structures. Fitting a physics model to clustering observables is just as constraining as fitting to any other measurement—sometimes more so, because clustering integrates information across scales and environments.

When to use which approach

Use clustering for anomaly flagging when you want sensitivity to unexpected signals and don't have a prior model [AG-2025.11-1368].

Use clustering to constrain models when you have candidate theories and want to test which one(s) survive the data [AG-2025.05-291].

In practice, you often do both: cluster the data to spot something interesting, then fit physics models to the anomalous region to interpret what you've found.

None of the provided preprints directly address frontier propulsion physics or relativity constraints from clustering, so this answer draws on particle and astro applications only.