Spanning trees, cycle-rooted spanning forests on discretizations of flat surfaces and analytic torsion

This paper establishes the asymptotic relationship between the determinants of graph Laplacians on discretized flat surfaces and zeta-regularized determinants to derive explicit formulas for the probabilities of cycle-rooted spanning forests inducing specific laminations and for the limits of associated topological observables.

The Gist

Imagine you have a giant, flat piece of fabric (a surface) that you want to study. Maybe it's a perfect rectangle, maybe it's a donut (a torus), or maybe it's a weird shape with corners and holes. In mathematics, we often want to know the "vibrations" of this fabric or how many ways we can connect all the points on it without creating loops.

This paper is like a sophisticated recipe for translating the behavior of a smooth, continuous fabric into the behavior of a pixelated, digital grid that approximates it. As we make the pixels smaller and smaller (zooming in infinitely), the digital version starts to look and act exactly like the smooth original.

Here is a breakdown of the paper's main ideas using simple analogies:

1. The Big Picture: From Pixels to Reality

Think of a high-resolution digital photo. If you zoom out, it looks like a smooth image. If you zoom in, you see individual square pixels.

• The Smooth Surface: The original, continuous fabric (mathematically called a "half-translation surface"). It has special points where the fabric is crumpled (conical singularities) or where the edges meet at sharp angles.
• The Discretization: The grid of pixels (graphs) we build to approximate that fabric.
• The Goal: The author wants to prove that if you count specific patterns on the pixel grid and then shrink the pixels to zero size, the result matches a very specific, complex number calculated from the smooth fabric.

2. The Two Main Characters: Trees and Forests

The paper focuses on two types of patterns you can draw on a grid of dots:

• Spanning Trees (The "Perfect Connection"): Imagine you have a city with many houses (dots). You want to build roads (lines) so that you can get from any house to any other house, but you never want to create a loop (a circle where you can drive around forever). A "Spanning Tree" is the most efficient way to connect everyone without any loops.

  • Why it matters: The number of ways you can build these trees is a measure of the graph's "complexity."

• Cycle-Rooted Spanning Forests (The "Looped Forests"): Now, imagine you allow loops, but with a rule: every separate cluster of houses must have exactly one loop in it. It's like a forest where every tree has a single ring of branches.

  • The Twist: The author adds a "monodromy" factor. Imagine walking around a loop and carrying a compass. If the compass spins a certain amount when you return to the start, that loop gets a special weight. This connects the geometry of the shape to the physics of the "compass" (a vector bundle).

3. The Magic Number: Analytic Torsion

In the smooth world, mathematicians have a tool called the Analytic Torsion. It's a bit like a "fingerprint" of the shape's geometry. It's calculated using an infinite product of all the possible vibration frequencies (eigenvalues) of the shape.

• The Problem: You can't multiply infinite numbers directly; it's messy. So, mathematicians use a trick called "Zeta-regularization" to make sense of it.
• The Discovery: The paper proves that the number of "Spanning Trees" and "Looped Forests" on the pixel grid, when you do some heavy-duty math to clean up the noise (renormalization), converges exactly to this Analytic Torsion fingerprint of the smooth shape.

4. The "Pillowcase" and the "Slit"

The shapes the author studies aren't just perfect squares. They are "Pillowcase covers."

• The Analogy: Imagine a pillowcase. If you sew the edges together, you get a flat shape. But sometimes, the fabric is folded or crumpled at specific points (conical singularities), creating angles that aren't 90 degrees.
• The Strategy: To solve the problem for a complex, crumpled shape, the author breaks it down into tiny, simple "model pieces" (like a perfect square, an L-shape, or a slit). They calculate the answer for these simple pieces and then stitch the results back together, accounting for how the corners and edges behave.

5. The "Probability" Application

One of the coolest applications is about probability.

• Imagine you pick a "Looped Forest" at random from all possible options on your pixel grid.
• The paper calculates the probability that the loops in your random choice form a specific pattern (a "lamination") on the surface.
• The Result: As the pixels get smaller, this probability settles down to a precise value determined by the geometry of the surface and the "compass" (the vector bundle). It's like saying, "If you randomly weave a net on a donut, the chance that the net forms a specific knot is determined by the donut's shape."

Summary: Why is this important?

This paper is a bridge between the discrete world (computers, grids, counting) and the continuous world (smooth geometry, physics, calculus).

It tells us that the chaotic, messy counting of patterns on a grid isn't just random noise. If you look at it through the right mathematical lens (the "renormalized logarithm"), it reveals a deep, hidden order that perfectly matches the smooth geometry of the universe it approximates. It confirms that the "fingerprint" of a shape (Analytic Torsion) can be found by simply counting trees and forests on a grid, provided you know how to clean up the math.

In a nutshell: The author showed that if you count the ways to connect dots on a grid without (or with) loops, and you zoom in infinitely, you aren't just counting dots—you are actually calculating the fundamental geometric "soul" of the shape itself.

Technical Summary

1. Problem Statement

The paper addresses the asymptotic behavior of the number of spanning trees and the weighted sum of cycle-rooted spanning forests (CRSFs) on a sequence of graph discretizations ($\Psi_n$) approximating a continuous surface ($\Psi$).

Specifically, the author investigates the relationship between the discrete combinatorial quantities and the analytic torsion (the zeta-regularized determinant of the Laplacian) on the continuous limit surface. The setting involves:

• Surfaces: Half-translation surfaces ($\Psi$) with piecewise geodesic boundaries, conical singularities, and corners. These surfaces admit a flat metric and can be tiled by Euclidean squares (pillowcase covers).
• Bundles: A flat unitary vector bundle $(F, h_F, \nabla_F)$ over the surface.
• Discretization: A family of graphs $\Psi_n$ constructed by taking the centers of $n^2$-subdivisions of the square tiling. The edges connect nearest neighbors, and the bundle is discretized via parallel transport.
• Goal: To establish an asymptotic expansion for the logarithm of the determinant of the discrete twisted Laplacian ($\det' \Delta_{\Psi_n}^{F_n}$) as the mesh size $n \to \infty$, and to relate this limit to the analytic torsion of the continuous Laplacian ($\det' \Delta_{\Psi}^F$).

2. Methodology

The proof strategy combines techniques from spectral geometry, probability theory, and discrete analysis.

A. Spectral Convergence and Zeta Functions

The core of the proof relies on comparing the zeta functions of the discrete and continuous Laplacians.

• Discrete Zeta Function: Defined as $\zeta_{\Psi_n}^{F_n}(s) = \sum_{\lambda \neq 0} \lambda^{-s}$, where $\lambda$ are eigenvalues of the rescaled discrete Laplacian $n^2 \Delta_{\Psi_n}^{F_n}$.
• Continuous Zeta Function: Defined via the meromorphic continuation of the trace of the heat kernel for the Friedrichs extension of the continuous Laplacian.
• Convergence Strategy: The author proves that while the discrete zeta functions do not converge pointwise to the continuous one everywhere (due to poles), they converge after subtracting "local" contributions arising from the boundaries and singularities. This is achieved using a parametrix construction (inspired by Müller's work on the Ray-Singer conjecture), decomposing the surface into local model patches (squares, cones, and angles).

B. Uniform Bounds and Weyl's Law

To handle the infinite sums in the zeta functions, the paper establishes:

• Uniform Weak Weyl's Law: A lower bound on the eigenvalues of the discrete Laplacian ($\lambda_i \geq C i$) that holds uniformly in $n$.
• Uniform Bounds on Powers: Estimates on the operator norms of powers of the discrete Laplacian and their differences from local models, ensuring the convergence of the renormalized traces.

C. Local Model Analysis (Fourier Analysis)

The asymptotic behavior of the "local" terms (contributions from squares, corners, and conical points) is computed explicitly using discrete Fourier analysis on square grids.

• The spectrum of the discrete Laplacian on a square grid is known explicitly.
• The author derives Szegő-type asymptotic expansions for the trace of the logarithm of the Laplacian on these local models, identifying terms proportional to $n^2 \log n$, $n^2$, $n \log n$, and $\log n$.

D. Probabilistic Interpretation

The paper utilizes the Matrix-Tree Theorem and its generalization by Forman and Kenyon, which relates the determinant of the Laplacian to the sum of CRSFs weighted by the monodromy of the connection. This bridge allows the translation of analytic results into probabilistic limits regarding loop measures.

3. Key Contributions and Results

A. Asymptotic Expansion of the Determinant (Theorem 1.2)

The main result provides a precise asymptotic formula for the renormalized logarithm of the discrete determinant:
$$\log_{\text{ren}}(\det' \Delta_{\Psi_n}^{F_n}) \to \log(\det' \Delta_{\Psi}^F) - \log(2) \cdot \frac{\text{rk}(F)}{16} \cdot \# \text{Ang}_{=\pi/2}(\Psi) + \text{Universal Constants}$$

• Renormalization: The logarithm is renormalized by subtracting terms proportional to the area ($n^2$), perimeter ($n$), and a logarithmic term involving the number of non-convex corners.
• Universality: The error term depends only on the topological type of the surface (specifically the sets of conical angles and boundary angles) and universal constants (like Catalan's constant $G$).
• Significance: This generalizes previous results by Duplantier-David (for rectangles) and Kenyon (for simply connected domains) to arbitrary half-translation surfaces with vector bundles.

B. Conformal Invariance and Topological Observables

• Corollary 1.4 & Theorem 1.6: The paper proves that the ratio of determinants for two surfaces with identical geometric invariants (area, perimeter, angle sets, and dimension of flat sections) converges to the ratio of their analytic torsions. This implies a form of conformal invariance for the probability measures associated with CRSFs.
• Theorem 1.8: Establishes that the expected value of the product of monodromy traces over cycles in a CRSF converges to a quantity proportional to $\sqrt{\det' \Delta_{\Psi}^F}$. This provides a geometric interpretation for a functional $G(F)$ previously defined by Kassel-Kenyon.

C. Limit of Lamination Probabilities (Theorem 1.10)

The paper gives an explicit formula for the limit probability that a uniformly sampled CRSF with non-contractible loops induces a specific lamination (a collection of disjoint non-contractible loops).

• The limit is expressed as a sum over laminations involving the analytic torsion and coefficients derived from the Gram-Schmidt orthonormalization of trace functions on the representation variety of the fundamental group.
• This resolves an open problem regarding the explicit geometric meaning of the limit probability in the work of Kassel-Kenyon.

4. Significance

• Unification of Discrete and Continuous: The paper rigorously bridges the gap between combinatorial graph theory (spanning trees, CRSFs) and spectral geometry (analytic torsion, zeta functions) on surfaces with singularities.
• Resolution of Open Problems: It answers Open Problems 2 and 4 from Kenyon's 2006 survey, specifically regarding the asymptotics of spanning trees on multiply connected domains and the relation to analytic torsion.
• Geometric Interpretation: It provides the first explicit geometric formula for the limit of probabilities related to non-contractible CRSF loops, linking them directly to the analytic torsion of the surface.
• Methodological Advancement: The use of a partition of unity to reduce global problems to local model cases (squares, cones, angles) combined with uniform spectral bounds offers a robust framework for studying asymptotics on singular surfaces.

In summary, Finski's work demonstrates that the combinatorial complexity of discretized surfaces, weighted by vector bundle monodromy, converges to the analytic torsion of the underlying flat surface, with precise corrections determined by the local geometry of corners and singularities.