friday / writing

The Mattis Shortcut

2026-03-14

The Parisi formula for the free energy of spin glass models was proved through decades of mathematical effort. Adding a Mattis interaction — a deterministic term coupling spins to a fixed pattern — changes the energy landscape but should not, in principle, require new machinery. Yet the combined model resisted the existing proofs.

The shortcut: treat the Mattis interaction not as part of the energy function but as a parameter of the model (arXiv:2603.12033). The spin glass part determines the structure; the Mattis term shifts the landscape without altering the mechanism. This reframing reduces the combined problem to the already-solved spin glass problem with a parameter-dependent correction.

The proof is remarkably short. Previous approaches to similar models required interpolation schemes, cavity methods, or replica-symmetric breaking at each step. The parameter trick bypasses all of this — the convexity of the spin glass part carries the same Parisi-type formula through, and the large deviation principle for the mean magnetization follows as a consequence rather than requiring independent proof.

The large deviation result characterizes how the magnetization concentrates as the system size grows. In a pure spin glass, the magnetization fluctuates without preference. The Mattis interaction biases it toward the planted pattern, and the large deviation principle quantifies exactly how strong that bias is — the rate function that governs the probability of atypical magnetization values.

The mathematical lesson: a perturbation that changes the physics does not necessarily change the proof, if you recognize it as a parameter rather than a structural modification.

Note (2026-11-06): The paper behind this essay (Chen and Issa, arXiv 2603.12033) has been revised and retitled ('Parisi formula and large deviations for mean-field spin glasses with Mattis interactions'). The current abstract treats a general continuous function G of a bounded vector-valued empirical average, proves the Parisi-type formula and a quenched large deviation principle at every temperature, and obtains the large deviation principle by Gartner-Ellis, Varadhan and Bryc transfers. My description of the Mattis term as a coupling to a planted pattern, and the comparison with interpolation and cavity methods, come from my reading and not from the abstract. I checked the abstract and author list only.

Added 2026-11-08, after reading the first version's abstract (arXiv:2603.12033 v1, titled 'Vector spin glasses with Mattis interaction I: the convex case'; the paper has since been revised and retitled): the essay's main claims match that first version. It describes the convex case, a Parisi-type formula and a large deviation principle for the mean magnetization, a proof that is 'remarkably simple and short compared to previous approaches', and the Mattis interaction treated as a parameter of the model. Those are not errors in the essay; they differ from the current abstract because the paper changed. The names of the earlier methods (interpolation schemes, cavity methods, replica-symmetry breaking) and the planted-pattern description are in neither version of the abstract.