Sphere-valued spins: decay, mass gaps, and continuum fields

Source and verification

This article explains family 215 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

Overview

Exponential decay, a spectral gap, a continuum quantum field, and an isolated particle are successively stronger kinds of information. Family 215 collects five papers on two-dimensional sphere-valued spin systems. Their conclusions must stay attached to the correct model and normalization.

🏷️ Preliminaries: spins and the meaning of mass

An $O(n)$ spin at a lattice vertex is a unit vector $s_x\in S^{n-1}$. On a finite graph its ferromagnetic Gibbs density, relative to independent uniform spins, is proportional to

$$\begin{array}{c}\exp \left(\sum _{\{x,y\}}{J}_{xy}{s}_x\cdot s_y\right),\\ 0\le J_{xy}\le \beta .\end{array}$$

The inverse temperature $\beta$ controls alignment. The correlation $⟨s_x\cdot s_y⟩$ measures how much information about one spin survives at another vertex.

An exponential estimate $|⟨s_x\cdot s_y⟩|\le A{e}^{-m|x-y|}$ gives a positive inverse correlation length in this observable. A full spectral gap is stronger: after reconstructing a transfer operator or Hamiltonian, every state orthogonal to the vacuum has energy at least some positive number. One observed spin correlation need not detect all low-energy states.

A particle pole is stronger still. The spin field’s spectral measure must have an isolated atom with positive weight. A gap alone allows the spectrum above it to be continuous, so it does not produce a particle automatically.

🏷️ Five distinct claims

The basic lattice companion claims exponential spin-correlation decay for every finite inverse temperature and every $n\ge 3$, uniformly over finite free-boundary subgraphs with bounded ferromagnetic couplings as above. This excludes the circle-valued $n=2$ model, treated separately in family 216. (OpenAI, 2026)

For the $O(4)$ model, another companion claims sharp two-sided bounds for the full lattice transfer gap, of order $\sqrt{\beta }{e}^{-\pi \beta }$ as $\beta \to \infty$, together with positivity at every finite temperature. (OpenAI, 2026b) The sharper companion claims the exact equivalent

$$m_{\mathrm{lat}}(\beta )\sim 32e^{\pi /4-1/2}\sqrt{\beta }{e}^{-\pi \beta }.$$

This is the full gap, with the original lattice transfer step retained, rather than a mass rescaled by an unspecified clock. (OpenAI, 2026a)

For $O(3)$, the continuum companion claims a canonical full-sequence scaling limit, normalized through susceptibility and second-moment correlation length. Its conclusions include the Osterwalder–Schrader conditions used to reconstruct a quantum theory, a unique vacuum, a positive full gap, and a nonzero connected four-point function. The last condition distinguishes the proposed field from a Gaussian field. (OpenAI, 2026d) A further companion claims an isolated positive atom at the lowest mass seen by the spin field. (OpenAI, 2026a)

These are not interchangeable statements about an arbitrary $O(n)$ model. In particular, an exact $O(4)$ mass constant does not furnish the $O(3)$ continuum construction.

🏷️ Two elementary diagnostics

First, an asymptotic mass gives a correlation-length scale. If the relevant length is $m_{\mathrm{lat}}^{-1}$, the displayed formula implies

$$\xi (\beta )\sim \frac{e^{1/2-\pi /4}}{32}{\beta }^{-1/2}{e}^{\pi \beta }.$$

The exponential dependence makes a continuum limit possible by shrinking lattice spacing as temperature decreases. It also makes normalization errors significant: multiplying the transfer time changes the quoted mass constant.

Second, consider a positive transfer operator with largest eigenvalue ${\lambda }_0$ and remaining eigenvalues ${\lambda }_j\ge 0$. Its long-cylinder partition sum is $Z_N={\sum }_j{\lambda }_j^N$. The ratio

$$\begin{array}{c}\frac{Z_{2N}}{Z_N^2}=\sum _j{w}_j^2,\\ w_j=\frac{{\lambda }_j^N}{Z_N},\end{array}$$

measures concentration of the normalized spectral weights. A value close to one forces almost all weight into one state. Unlike a single spin correlation, the partition sum counts every sector and every multiplicity. Uniform control as the transverse circumference grows is still needed to infer an infinite-volume gap.

🏷️ How the manuscripts propose to obtain that control

The finite-temperature decay proof uses rotations across annuli to compare twisted and untwisted partition functions. An embedded Ising-sign representation turns suitably pinned, separated annuli into repeated chances to destroy a long spin connection. The iteration yields exponential decay once its conditional estimates are uniform.

The full-gap work uses cylinder partition sums and the spectral concentration diagnostic above. Interchanging the two cylinder directions transfers estimates between long length and growing circumference. The exact-asymptotic paper adds a regulator calculation and a comparison that fixes the conversion back to the original lattice step. Controlling sectors invisible to the spin observable is part of this comparison.

The $O(3)$ construction retains marked sources through exact blocking. Their normalizations must converge along the full sequence, while positivity and locality survive the limit. The particle companion then needs spectral separation and a positive residue, not just positivity of the reconstructed Hamiltonian. These additional steps explain why the five results are presented as a family rather than one undifferentiated “mass gap” theorem.

🏷️ Key steps and proof outlines

From an annular twist to exponential decay

For three-component spins, write the first coordinate as an amplitude times a sign. Conditional on amplitudes and the other coordinates, the signs have a ferromagnetic Ising representation. Its open-bond clusters transmit the sign correlation.

A crucial comparison says that pinning selected spins to $s_{\ast }=(1,0,0)$ increases increasing bond events. To see the mechanism, first condition their amplitudes to be at least $1-\epsilon$ and their signs positive. The induced amplitude-bond law is tilted by

$$T_{\epsilon }=1_{\{r_x\ge 1-\epsilon \text{ on the pin set}\}}{2}^{-k_B},$$

where $k_B$ counts open clusters meeting that set. Opening a bond can only decrease $k_B$, so this tilt is increasing. The source’s joint association inequality proves the comparison; continuity of finite-volume weights then permits $\epsilon \downarrow 0$ and exact pinning.

Pin the inner and outer layers of an annulus. Let $Z(0)$ use the same pinned sign on both, and $Z(\pi )$ the opposite inner sign. The cluster expansion gives the exact identity

$${\mathbb{P}}_{\mathrm{pinned}}(\text{annulus crossing})=1-\frac{Z(\pi )}{Z(0)}.$$

Changing the inner sign removes exactly the bond configurations connecting the two layers. The annular rotation estimate makes this probability at most $p_N\to 0$ as the annulus radius $N$ grows. Pinning the boundary layers of disjoint annuli separates their interiors, so simultaneous crossings cost at most $p_N$ to the number of annuli.

Choose one fixed $N$ with $4p_N^{1/25}<1$. For $d=\Vert y{\Vert }_{\infty }>4N$, a connection visits a simple coarse path of at least $d/N-1$ boxes. Among its boxes, one of the $25$ residue classes modulo $5N$ contains at least a twenty-fifth of them; their outer annuli are disjoint. There are at most $4^k$ coarse paths of length $k$, hence

$$\mathbb{P}(0\leftrightarrow y)\le p_N^{1/25}\sum _{k\ge \lceil d/N-1\rceil }(4p_N^{1/25}{)}^k.$$

This is a geometric series with exponential decay in $d$. Bounded distances are absorbed by increasing the prefactor. The conditional reduction to three active components extends the estimate to $n\ge 3$. The constants depend on the temperature but are uniform over the allowed finite graphs and edge strengths. (OpenAI, 2026c)

Why the finite coupling shift determines the exact mass constant

The sharp $O(4)$ bounds retain the factor $\sqrt{\beta }$ as well as $e^{-\pi \beta }$. Controlling only $\log m/\beta \to -\pi$ would allow an extra factor $e^{-\sqrt{\beta }}$, which would destroy a positive mass in the prescribed physical units. The full transfer estimates therefore control a bounded multiplicative error, not only the leading exponential rate. (OpenAI, 2026d)

The exact companion compares the lattice coupling $\beta$ with a Brownian-regulator coupling $b$. Matching helicity responses in the two directions gives

$$\beta -b=c+o(1),c=\frac{1}{4}-\frac{1+\log 2}{2\pi }.$$

The constant comes from differences of propagator sums. Each separate sum diverges near zero momentum, so the proof pairs the sums before passing to integrals. In the common region, their difference is bounded near the origin; the remaining Brownian momentum tails are summable. One of the finite terms reduces to

$$I(z)={\int }_0^{\pi /2}\frac{1-(1+z{\sin }^{2}y{)}^{-1/2}}{\sin y}dy.$$

Differentiation under the integral gives $I^{\prime }(z)=1/[2(1+z)]$, and $I(0)=0$, producing the $\log 2$ term at $z=1$.

The regulator calculation, together with the full-sector trace comparison and removal of the extra circle contribution, gives

$$m_{\mathrm{lat}}(\beta )=32\sqrt{2}\sqrt{b}{e}^{-\pi b}(1+o(1))+o\left(\sqrt{\beta }{e}^{-\pi \beta }\right).$$

Substituting $b=\beta -c+o(1)$ makes its normalized limit

$$32\sqrt{2}{e}^{\pi c}=32e^{\pi /4-1/2}.$$

Thus an order-one coupling shift changes the leading mass constant. The full-sector comparison is also needed to rule out a lower state invisible to a selected spin correlation. (OpenAI, 2026b)

Two scale relations eliminate subsequence dependence

For the $O(3)$ continuum construction, exact blocking initially gives normalized limiting correlation hierarchies $G_s$ depending continuously on an offset $s$. Let $\ell (s)>0$ be the corresponding unnormalized second-moment length. Comparing two descriptions of the same microscopic law gives

$$\begin{array}{rlrl}{G}_{s-\gamma }& =G_s,& \ell (s-\gamma )& =L^{-1}\ell (s),\\ G_{s+d}& =(O_{+}^{-1}{)}_{\ast }{G}_s=(O_{-}^{-1}{)}_{\ast }{G}_s,& \ell (s+d)& =5^{-1}\ell (s).\end{array}$$

Here $L$ is a power of two, and $O_{\pm }$ are opposite rotations with cosine $3/5$ and sine $4/5$. Equality of the two pushforwards gives invariance under twice that rotation angle. Its multiples are dense in the rotation group, so continuity on test functions gives full rotation invariance. The second relation then becomes $G_{s+d}=G_s$.

Extend the relations from their overlapping offset intervals using the $\gamma$-periodicity. If $d/\gamma =p/q$ were rational, iterating the length relations would imply $5^{-q}=L^p$, impossible for a power of two. Thus $\gamma$ and $d$ generate a dense additive subgroup. Every continuous test evaluation of $G_s$ has both periods and must be constant. This removes the offset dependence and supplies full-sequence uniqueness after canonical normalization. The comparison uses both the marked correlation functions and the positive length; neither relation alone proves uniqueness. (OpenAI, 2026e)

Counting low-energy states forces a particle atom

For the particle companion, let $\rho$ be the positive spectral measure of the spin field, and let $m_{\ast }>0$ be the least mass in its support. The key additional input is a cylinder state-count bound

$${\mathcal{N}}_w(3m_{\ast }/2)\le Cw$$

in the cutoff limit, where $w$ is circumference and ${\mathcal{N}}_w(u)$ counts all transfer states below energy $u$, including multiplicities. Exchanging cylinder time and circumference, approximating an energy cutoff by positive functions of the transfer operator, and dyadic iteration produce this linear bound.

Suppose there were $J$ distinct support masses in $[m_{\ast },5m_{\ast }/4]$. Fix a small momentum interval $[-p_0,p_0]$ so that every corresponding energy

$$E(p,\mu )=\sqrt{p^2+{\mu }^2}$$

lies below $3m_{\ast }/2$. Choose disjoint neighborhoods of the $J$ energy graphs. Positivity of the spectral measure gives each neighborhood positive spin weight uniformly over that compact momentum interval. Moment convergence to finite-cylinder spectral measures then forces at least $J$ distinct energy levels at each allowed momentum, once the cutoffs are sufficiently fine.

There are at least $c_{0}w$ allowed momenta in the interval, so $J{c}_{0}w\le Cw$. The constants $c_0,C$ do not depend on the chosen masses or on $J$. Hence the low-mass support is finite. Its smallest point is isolated, and a positive locally finite measure assigns an isolated support point a strictly positive finite mass. This proves the positive residue and spectral separation claimed for the spin field. It does not follow merely from exponential decay. (OpenAI, 2026a)

See Also

• Family 216 treats circle-valued spins, where a massless phase and a BKT transition replace finite-temperature exponential decay.
• Ising preliminaries introduces Gibbs weights and the distinction between finite-volume and infinite-volume states.

🏷️ Manuscripts and verification scope

• The canonical massive continuum limit of the two-dimensional O(3) model — pinned manuscript and companion source.
• An Isolated Particle Pole for the Two-Dimensional O(3) Spin Field — pinned manuscript and companion source.
• Exact mass asymptotics for the two-dimensional O(4) lattice model — pinned manuscript and companion source.
• Sharp mass bounds for the two-dimensional O(4) model — pinned manuscript and companion source.
• Exponential decay in two-dimensional classical O(n) models — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. The documented selection covers finite-free-volume exponential spin-correlation decay for at least three spin components. It does not cover the full O(4) spectral gap, its exact asymptotic, the O(3) continuum construction, or the isolated particle pole. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026a. An Isolated Particle Pole for the Two-Dimensional O(3) Spin Field.

🐻  OpenAI 2026b. Exact mass asymptotics for the two-dimensional O(4) lattice model.

🐻  OpenAI 2026c. Exponential decay in two-dimensional classical O(n) models.

🐻  OpenAI 2026d. Sharp mass bounds for the two-dimensional O(4) model.

🐻  OpenAI 2026e. The canonical massive continuum limit of the two-dimensional O(3) model.