Torus one-point functions in critical loop models

Abstract: We show that in critical loop models, torus 1-point functions can be expressed in terms of sphere 4-point functions at a different central charge. Unlike in the Moore--Seiberg formalism, crossing symmetry on the sphere therefore implies modular covariance on the torus. We systematically compute torus 1-point functions in critical loop models, using a numerical bootstrap approach. We focus on the 1-point functions of the 6 simplest primary fields, which give rise to 10 solutions of modular covariance equations. Such 1-point functions are infinite linear combinations of conformal blocks. The coefficients are products of double Gamma functions, times polynomial functions of loop weights. For each solution, we determine the first 6 to 12 polynomials.

• The paper establishes a sphere–torus relation that maps selected torus one-point functions to sphere four-point functions at transformed central charge and nome, enabling modular-covariant data to be derived from crossing equations.
• It systematically computes polynomial factors for six primary fields, producing ten bootstrap solutions and validating them through arbitrary-precision numerical checks, exact degenerate cases, and transfer-matrix amplitudes.
• The results clarify how torus map symmetries restrict conformal spin, explain vanishing one-point functions, and extend the analytic solution of critical loop models while leaving generic-map ambiguities and the polynomial ansatz open.

Overview

This paper by Roux, Ribault, and Jacobsen (2604.24491) addresses a central open problem in the exact solution of two-dimensional critical loop models: the determination of torus one-point structure constants. Critical loop models — CFTs describing the continuum limits of non-intersecting loop ensembles such as the $O(n)$ and $Q$-state Potts models — were previously claimed to be solvable on the basis of 235 analytically determined sphere four-point structure constants. However, because sphere four-point structure constants do not factorize into three-point structure constants in these theories, knowledge of all three-point data is insufficient, and a complete solution requires systematic access to higher correlation functions.

The paper makes two principal contributions. First, it establishes a sphere–torus relation: torus one-point functions of primary fields $V_{(r_1,s_1)}$ with $r_1 \leq 3$ can be expressed as sums of sphere four-point functions of the type $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$, evaluated at a transformed central charge $\beta' = \beta/\sqrt{2}$ and nome $q' = \sqrt{q}$. Second, it systematically computes the polynomial factors $d_{(r,s)}$ appearing in torus one-point structure constants for the six simplest primary fields, yielding ten solutions of modular covariance equations, with 6 to 12 polynomials determined per solution via numerical bootstrap.

Combinatorial maps on the torus

A central structural insight is that correlation functions in loop models are labeled not only by their fields but also by combinatorial maps encoding leg connectivities. For a torus puncture carrying $2r_1$ legs, the legs form at most three topologically distinct loops, so maps are parametrized by triples $(m,n,p) \in \mathbb{N}^3$ with $Q$0, modulo cyclic permutations (or all permutations when an entry vanishes). The paper derives closed-form counting results:

$Q$1

with generating function $Q$2. Map symmetries constrain the second Kac index: every map has a $Q$3 symmetry forcing even conformal spin $Q$4, while special maps such as $Q$5 impose stronger constraints ($Q$6 for $Q$7). This explains why odd-spin fields like $Q$8, $Q$9, and $V_{(r_1,s_1)}$0 have vanishing torus one-point functions.

The sphere–torus relation

The key technical device is a dictionary between torus one-point data and sphere four-point data of the correlator $V_{(r_1,s_1)}$1, where $V_{(r_1,s_1)}$2 assigns weight zero to loops around it. The transformation preserves the modulus $V_{(r_1,s_1)}$3 while halving the nome ($V_{(r_1,s_1)}$4), rescaling $V_{(r_1,s_1)}$5, mapping external Kac indices $V_{(r_1,s_1)}$6, channel indices $V_{(r_1,s_1)}$7, and the contractible loop weight $V_{(r_1,s_1)}$8.

The authors verify this relation component by component:

• Combinatorial maps: each torus map lifts to a set $V_{(r_1,s_1)}$9 of one to six sphere maps; crucially, the three distinguishable cycles on the punctured sphere resolve the signature ambiguity that plagues direct torus characterization.
• Conformal blocks: using Zamolodchikov recursion residues, they show $r_1 \leq 3$0, leading to the full non-chiral block relation including logarithmic blocks.
• Structure constants: reference structure constants built from Barnes double Gamma functions satisfy explicit identities under the transformation, and residue relations $r_1 \leq 3$1 close consistently.

An important consequence concerns the Moore–Seiberg statement that modular covariance does not follow from crossing symmetry within a fixed CFT. The sphere–torus relation evades this because it relates CFTs at different central charges. The authors argue that, for Virasoro-symmetric theories with continuously varying central charge (loop models, Liouville theory), consistency on the sphere implies modular covariance on the torus, provided one includes a field of dimension $r_1 \leq 3$2. They illustrate this with A-series minimal models, where submodels that evade torus constraints are eliminated once the spectrum is accessed through the mapped four-point function. They are careful to state caveats: the argument requires the Virasoro algebra specifically, integer multiplicities on the torus have no sphere counterpart, and different central charges must be considered.

Structure constants and bootstrap results

Torus one-point functions decompose as $r_1 \leq 3$3, with the conjectural ansatz

$r_1 \leq 3$4

where $r_1 \leq 3$5 involves double Gamma functions, $r_1 \leq 3$6 is a known polynomial in the loop weight $r_1 \leq 3$7, and $r_1 \leq 3$8 is conjectured to be polynomial in $r_1 \leq 3$9, the channel weight $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$0, and the external weight $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$1. Numerically determined degree bounds include $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$2, matching $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$3.

The ten solutions computed cover the fields $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$4, $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$5, $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$6, $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$7, $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$8, and $\langle V'_0 V'_1 V'_0 V'_0\rangle^{\text{sphere}}$9 across their admissible maps. Each solution exhibits shift equations $\beta' = \beta/\sqrt{2}$0 (with $\beta' = \beta/\sqrt{2}$1 for maps of type $\beta' = \beta/\sqrt{2}$2), and parity $\beta' = \beta/\sqrt{2}$3 holds in most cases — notably it fails for $\beta' = \beta/\sqrt{2}$4 at $\beta' = \beta/\sqrt{2}$5 and $\beta' = \beta/\sqrt{2}$6, an explicitly reported exception to the otherwise general parity pattern. Two solutions (maps $\beta' = \beta/\sqrt{2}$7 for $\beta' = \beta/\sqrt{2}$8 and $\beta' = \beta/\sqrt{2}$9) required the sphere–torus relation rather than direct torus computation. All results were validated computationally: the Julia package BootstrapVirasoro.jl parses the LaTeX output and checks modular covariance numerically at arbitrary precision.

Exact special cases

Several degenerate-field one-point functions are obtained in closed form:

• Partition function: identifying $q' = \sqrt{q}$0 with the $q' = \sqrt{q}$1 model partition function $q' = \sqrt{q}$2 in the presence of a combinatorial defect, the paper verifies the prediction $q' = \sqrt{q}$3 against its numerics, where $q' = \sqrt{q}$4 are Chebyshev-type polynomials.
• Energy operator of the $q' = \sqrt{q}$5 model $q' = \sqrt{q}$6: all structure constants are analytic, reference constants reduce to ordinary Gamma functions, and the full series expansion is written explicitly.
• Potts energy operator $q' = \sqrt{q}$7: fusion rules leave a single conformal block, which reduces to $q' = \sqrt{q}$8 — a strikingly simple result whose existence was noted previously but whose closed form is new here.

For orientable-loop models (Potts, $q' = \sqrt{q}$9), bicolourability of maps selects combinations $d_{(r,s)}$0 that automatically annihilate half-integer-$d_{(r,s)}$1 contributions, consistent with the absence of such fields in those spectra. The paper also predicts vanishing of several Potts one-point functions ($d_{(r,s)}$2, $d_{(r,s)}$3, $d_{(r,s)}$4, $d_{(r,s)}$5), though for the non-diagonal fields it concedes that no lattice interpretation of these vanishings is available.

Lattice cross-checks

Using transfer matrices in the periodic dilute Temperley–Lieb algebra, Jacobsen's lattice computation of $d_{(r,s)}$6 on $d_{(r,s)}$7 tori confirms that ratios of transfer-matrix amplitudes between two values of $d_{(r,s)}$8 equal the corresponding ratios of CFT structure constants $d_{(r,s)}$9, independently of the module index $2r_1$0 and of system size $2r_1$1. Notably, these amplitude ratios hold even away from criticality, suggesting they are algebraic rather than conformal data. The $2r_1$2 computation provides an independent check of all polynomials for the map $2r_1$3.

Limitations and open questions

The paper is candid about boundaries of its results. The polynomial ansatz for $2r_1$4 remains a conjecture, supported numerically rather than proven. For generic maps with $2r_1$5 all distinct and nonzero (first occurring at $2r_1$6), bootstrap equations cannot distinguish $2r_1$7 from $2r_1$8 even on the sphere; only the two-dimensional solution space is accessible. The sphere–torus relation is established only for the Virasoro algebra, leaving larger chiral algebras untreated. The fermionic solutions suggested by mapping individual (rather than summed) sphere solutions to the torus — with spins in $2r_1$9 and potentially arbitrary complex spins — remain speculative, and the algebraic justification of the lattice amplitude-ratio observations is deferred to future work.

Conclusion

This work extends the analytic bootstrap program for critical loop models from the sphere to the torus, delivering the first systematic set of torus one-point structure constants together with a structural sphere–torus correspondence that resolves the combinatorial ambiguities inherent to genus-one geometries. The demonstration that crossing symmetry at modified central charge encodes modular covariance challenges the standard Moore–Seiberg picture for families of Virasoro-symmetric CFTs, and the agreement with both exact degenerate-field formulas and finite-size lattice data substantiates the polynomial structure constant ansatz beyond the sphere.