Bootstrapping Tensor Integrals

This paper introduces a bootstrapping methodology combining Dyson-Schwinger equations and positivity constraints to efficiently approximate moments of large-$N$ random $U(N)^D$ invariant tensor models, successfully validating the approach on quartic and hexic rank-three models and proposing new explicit formulae for the quartic case.

The Gist

The Big Picture: Building a Universe from Lego

Imagine you are trying to understand the rules of a massive, complex universe. In physics, we often use matrices (grids of numbers) to describe the universe in 2 dimensions (like a flat sheet of paper). But our real universe has 3 dimensions (and maybe more). To describe these higher dimensions, physicists use tensors.

Think of a matrix as a flat sheet of paper. A tensor is like a 3D block of Lego bricks. It's much more complex because the "bricks" can connect in many more directions.

The problem? Calculating the behavior of these 3D Lego blocks is incredibly hard. It's like trying to predict the weather by simulating every single air molecule. The math gets so messy that for decades, scientists could only solve these problems for very simple, specific cases.

The New Tool: "Bootstrapping with Positivity"

The authors of this paper introduce a clever new way to solve these problems, which they call Bootstrapping with Positivity.

Here is the analogy:
Imagine you are trying to guess the weight of a mysterious, sealed box. You can't open it. However, you know two things:

1. The Rules (Dyson-Schwinger Equations): You know the physics inside the box must follow specific balance laws (like a scale that must stay level).
2. The Constraints (Positivity): You know that the weight of any part of the box must be a positive number. You can't have "negative weight."

Instead of trying to solve the whole box at once, you start guessing.

• "If the weight is 5, does it break the balance laws?"
• "If the weight is 5, is it positive?"
• "If the weight is 10, does it break the rules?"

By testing thousands of guesses and throwing out the ones that break the rules or become negative, you slowly narrow down the possibilities until you find the only answer that fits all the constraints. This is "bootstrapping"—pulling yourself up by your own bootstraps to find the answer without needing the full, impossible formula.

What Did They Do?

The team applied this "guess-and-check" method to three specific types of 3D Lego universes (called Tensor Models):

1. The Quartic Model: A universe built with 4-way connections.
2. The Hexic Cyclic Model: A universe built with 6-way connections in a circle.
3. The Hexic Pillow Model: A universe built with 6-way connections in a different shape.

The Results:

• It Works: Their method found the answers very quickly.
• It's Accurate: For the models where scientists already knew the answer (from old, difficult math), the new bootstrapping method found the exact same numbers.
• It's Fast: It converged (found the answer) much faster in some regions than others, but it worked reliably.

The Big Discovery: A New Conjecture

While testing these models, the authors noticed a surprising pattern. They realized that for the "Quartic Model," the complexity of the shape didn't matter as much as they thought.

The Analogy:
Imagine you have a bag of different Lego structures. Some are simple towers, some are complex castles.

• Old Thinking: To know the weight of the castle, you need to count every single brick and every specific connection.
• New Conjecture: The authors suspect that for this specific type of universe, the weight depends only on the total number of bricks, not on how they are arranged. Whether it's a tower or a castle, if they have the same number of bricks, they behave the same way.

They used their computer code (called feyntensor) to check this and found it held true. They are now proposing a new "Universal Formula" for these models based on this idea.

Why Does This Matter?

1. Solving the Unsolvables: This method gives physicists a way to study complex 3D (and higher) universes that were previously too hard to calculate.
2. Understanding Gravity: These tensor models are believed to be the mathematical building blocks for Quantum Gravity (how gravity works at the smallest scales). By understanding these models, we get closer to understanding how space and time are formed.
3. A New Toolkit: Just as this method helped solve problems with 2D matrices, it can now be applied to 3D, 4D, and even higher-dimensional problems. It opens the door to studying "universes" of any complexity.

Summary in One Sentence

The authors invented a smart "guess-and-check" method that uses the laws of physics and the rule that "numbers must be positive" to solve incredibly complex 3D math problems, revealing that the shape of these mathematical universes might be simpler than we thought.

Technical Summary

1. Problem Statement

The paper addresses the challenge of computing the moments (expectation values of unitary invariants) of random $U(N)^D$-invariant tensor models in the large $N$ limit.

• Context: While random matrix theory is well-understood, random tensor models (generalizations to rank $D \geq 3$) are significantly more complex. Analytic solutions for tensor integrals are scarce, and computing eigenvalues/eigenvectors is NP-hard.
• Specific Gap: Existing methods often rely on specific solvable cases (e.g., melonic dominance) or intermediate field theories. There is a lack of a general, non-perturbative numerical framework to bound and approximate moments for arbitrary tensor models, particularly those involving non-melonic or higher-genus graphs.
• Goal: To adapt the "bootstrapping with positivity" methodology—previously successful in random matrix models—to rank-3 tensor models to approximate moments, identify critical points, and test conjectures about the factorization of invariants.

2. Methodology: Bootstrapping with Positivity

The authors propose a numerical scheme combining Dyson-Schwinger Equations (DSE) with positivity constraints derived from the inner product structure of the tensor space.

A. Theoretical Framework

1. Tensor Models: The study focuses on $U(N)^D$ invariant tensors $T$ of rank $D=3$. Invariants are represented by bipartite, $D$-colored graphs (vertices represent tensor indices, edges represent contractions).
2. Dyson-Schwinger Equations (DSE): Derived via integration by parts on the partition function. These equations relate the expectation value of an invariant to a sum of expectation values of "glued" invariants (where vertices are removed and half-edges reconnected). In the large $N$ limit, these provide a recursive system of linear equations for the moments.
3. Positivity Constraints: Based on the inequality $\langle P \cdot \bar{P} \rangle \geq 0$ for any polynomial $P$ of the tensor. By choosing $P$ as a linear combination of derivatives of invariants, the authors construct a moment matrix $M$ where entries are expectation values of products of invariants. The condition that $M$ must be positive semi-definite (all eigenvalues $\geq 0$) imposes non-linear inequality constraints on the moments.

B. The Algorithm

1. Truncation: The infinite system of DSEs and the infinite-dimensional moment matrix are truncated to a finite size (cutoff on the number of vertices/edges and matrix dimension).
2. Optimization Problem: The problem is formulated as a linear optimization problem with non-linear inequality constraints.
   • Objective: Minimize a linear combination of moments (parameterized by an angle $\theta$) to probe the boundaries of the allowed solution space.
   • Constraints: The truncated DSEs (linear equalities) and the positive semi-definiteness of the truncated moment matrix (non-linear inequalities).
3. Solver: The authors use the fmincon package in Matlab with an interior-point algorithm to find the feasible region of moments for given coupling constants $g$.

3. Key Contributions

A. Application to Rank-3 Tensor Models

The authors successfully applied the bootstrap method to three specific rank-3 models:

1. Quartic Model: Interaction involving four tensors (melonic bubbles).
2. Hexic Cyclic Bubble Model: Interaction involving six tensors in a cyclic arrangement.
3. Hexic Pillow Model: Interaction involving six tensors in a "pillow" configuration (non-cyclic).

B. Conjecture on Moment Factorization (The "Strong Conjecture")

A major theoretical contribution is a new conjecture regarding the large $N$ limit of the quartic rank-3 tensor model:

• Hypothesis: The leading order expectation value of any connected 3-colored graph invariant depends only on the number of vertices (specifically, the number of melonic insertions), and not on the specific coloring or topology (genus) of the graph, provided the graph is melonic.
• Implication: This implies a massive simplification of the DSE system, where all moments of the same order become equal.
• Verification: The authors support this using:
  • Feyntensor: Explicit double-series computations (in $g$ and $1/N$) using the feyntensor package.
  • Bootstrap Results: The conjectured analytic solutions fit perfectly within the bootstrapped bounds.
  • Analytic Recovery: Under this conjecture, the authors derive explicit closed-form formulae for all moments (e.g., $m_4, m_6, m_8$) that match known solutions for the second and fourth moments and extend them to higher orders.

C. Non-Factorization and Genus Dependence

The paper investigates the factorization of multi-trace invariants.

• Finding: While melonic graphs (planar) often factorize into powers of the two-point function, non-planar (higher genus) graphs do not necessarily factorize.
• Evidence: Using series expansions, the authors show that for a specific genus-1 hexic interaction, the leading order of the correlator does not match the expected power of the two-point function, confirming that genus plays a decisive role in non-factorization, distinct from the melonic property.

4. Results

• Convergence: The bootstrap method converges rapidly for coupling constants $g$ away from critical points. Convergence slows significantly near critical points ($g_c$), a behavior also observed in matrix models.
• Validation:
  • For the Quartic Model, the bootstrapped solution space for the moment $m_4$ tightly bounds the known analytic "melonic" and "instanton" solutions.
  • For the Hexic Models, the bootstrapped results for $m_2$ and $m_6$ (or $m_{6,d}$) agree excellently with known analytic solutions derived from intermediate field theory and topological recursion.
• Critical Points: The method successfully identifies the critical points of the models (e.g., $g_c = -1/12$ for the quartic model, $g_c = -4\alpha/243$ for hexic models) where the solution space collapses or becomes singular.
• Free Energy: Using the conjectured relations, the authors derive explicit integral formulae for the free energy of the hexic models in the large $N$ limit.

5. Significance and Future Directions

• Generalizability: The work demonstrates that bootstrapping with positivity is a viable, general tool for tensor models of any rank, not just matrices. It opens the door to studying models where analytic solutions are currently unknown.
• Critical Exponents: The method provides a numerical route to estimating critical exponents and identifying phase transitions in tensor models, which are relevant for understanding higher-dimensional quantum gravity and the continuum limit of discrete spacetimes.
• Theoretical Insight: The conjecture regarding the dependence of moments on vertex count rather than graph topology simplifies the landscape of rank-3 quartic models, suggesting a hidden universality class similar to vector models.
• Extensions: The authors note the method can be adapted to $O(N)$ and $Sp(N)$ invariant tensor models (relevant for SYK-like models) and models with broken symmetries (Laplacian terms).

In summary, this paper establishes a robust numerical framework for solving tensor integrals, validates it against known analytic results, and proposes a powerful new conjecture that simplifies the understanding of the large $N$ limit in rank-3 tensor models.