⚛️ high-energy theory

When the gravity path integral describes a statistical average

This paper establishes necessary and sufficient conditions to distinguish whether a gravitational path integral represents a true statistical average or merely a pseudo-average, demonstrating that the stronger condition for a statistical average imposes stricter constraints on wormhole amplitudes and significantly sharpens bounds on ensemble moments compared to the weaker condition required for an inner product interpretation.

Original authors: Gabriele Di Ubaldo, Luca V. Iliesiu, Henry W. Lin, Cynthia Yan

Published 2026-09-16

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Gabriele Di Ubaldo, Luca V. Iliesiu, Henry W. Lin, Cynthia Yan

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the quest to understand the universe, physicists often turn to a powerful mathematical tool called the path integral. Imagine trying to predict the behavior of a complex system by considering every possible way it could happen, from the most likely paths to the most bizarre, unlikely twists. In the realm of gravity, this tool sums up all possible shapes of spacetime to calculate probabilities. For decades, scientists have used this method to study black holes and the birth of the cosmos, treating the results as if they were describing a single, definite reality. However, a growing body of evidence suggests that the path integral might not be describing one specific universe at all, but rather an average over a vast collection of different universes, each with its own unique rules. This distinction is crucial: if the math describes a single reality, the universe is unique; if it describes an average, our reality might just be one member of a statistical crowd.

A team of researchers has now drawn a sharp line between these two possibilities. They have identified the specific mathematical conditions required to prove that the gravitational path integral is truly computing a statistical average, rather than just a fake or "pseudo" average that only looks like one on the surface. Their work reveals that while the standard rules of quantum mechanics allow for a consistent description of a single universe, they are not enough to guarantee that the math represents a genuine collection of many universes. By developing a new set of stricter tests, the authors show how to distinguish between a theory that describes a single, coherent reality and one that merely mimics the behavior of a statistical ensemble.

The researchers began by examining how the path integral calculates the "overlap" between different states of the universe. In a standard quantum theory, these overlaps act like distances or angles in a geometric space, and they must always be positive numbers to make physical sense. The team showed that if the path integral satisfies these basic positivity rules, it can always be interpreted as describing a valid quantum system with a well-defined space of possible states. This is the first, weaker condition. It ensures that the math doesn't break down, but it does not force the math to represent a collection of different universes. A theory can pass this test and still be describing a single, unique universe.

To find out if the math is actually describing a statistical average, the researchers looked for a second, much stronger condition. They asked whether the results of the path integral could be generated by averaging over a real, positive probability distribution of boundary conditions. In simpler terms, they asked if the numbers produced by the math could be explained by taking a weighted average of many different possible outcomes, where every outcome has a positive chance of occurring. They discovered that the first condition is necessary but not sufficient for this second condition. There are mathematical scenarios where the basic rules of quantum mechanics are satisfied, yet the results cannot be explained as a true statistical average. These scenarios produce what the authors call a "pseudo-average."

To demonstrate this difference, the team constructed a specific toy model of gravity. In this model, they combined a simple theory of gravity with a set of interacting matter fields. They carefully tuned the interactions so that the basic rules of quantum mechanics were obeyed, ensuring the theory was consistent. However, they also arranged the interactions so that the stronger statistical condition was violated. In this model, the path integral produced results that looked like they came from a statistical ensemble, but upon closer inspection, they failed the stricter test. The math was consistent, but it did not describe a genuine collection of universes. This proved that satisfying the basic quantum rules does not automatically mean the theory is an ensemble.

The researchers then extended this analysis to "open universes," which are universes with boundaries, like the space inside a black hole or the edge of our observable cosmos. Here, the constraints become even more powerful. In a statistical interpretation, the random variables representing the properties of these boundaries cannot take just any value; they must themselves form a consistent set of quantum states. This restricts the possible values these variables can take to a specific, limited domain. The team found that this restriction creates new, independent tests that the path integral must pass. They identified a specific mathematical inequality involving the properties of these boundaries that must hold true for a genuine statistical average. This inequality is not required by the basic quantum rules, meaning a theory could pass all the standard tests and still fail this new one.

The implications of these findings are profound for how we understand gravity and the nature of reality. The new, stricter tests provide a concrete way to check whether a theory of gravity is truly an ensemble or just a pseudo-average. If a theory fails these tests, it cannot be interpreted as a collection of different universes, no matter how much it might look like one. This distinction helps physicists classify different theories of gravity, separating those that describe a single, unique universe from those that describe a statistical mixture. It also suggests that the famous "wormholes" that connect different parts of spacetime in these calculations might not always represent a genuine statistical connection between different universes.

The authors also showed that these new constraints are not just theoretical curiosities; they can be used to sharpen our understanding of the universe. By applying these stricter tests to known theories, physicists can place much tighter limits on the possible values of physical quantities. For example, in the study of random matrices, which are often used to model complex systems, adding these statistical constraints significantly narrows down the range of possible outcomes. This means that if we can verify that a theory of gravity satisfies these stronger conditions, we gain a much more precise picture of what that theory allows and forbids.

Ultimately, this work provides a new diagnostic tool for the foundations of quantum gravity. It clarifies that the path integral is a versatile tool that can describe both single realities and statistical ensembles, but the two are not the same. The basic rules of quantum mechanics ensure the math is consistent, but only the stricter statistical conditions guarantee that the math represents a genuine average over many possibilities. By identifying exactly where these two interpretations diverge, the researchers have given the scientific community a way to tell the difference. This allows for a more rigorous testing of gravitational theories, ensuring that when we claim to be describing a statistical ensemble of universes, we are not just seeing a mirage. The path forward is now clearer: to truly understand the statistical nature of gravity, we must look beyond the basic rules and subject our theories to these more demanding, yet essential, tests.

Technical Summary: When the Gravity Path Integral Describes a Statistical Average

Problem Statement

The gravitational path integral (GPI) is a central tool in modern theoretical physics, used to describe black hole thermodynamics, the Page curve, and cosmological transition amplitudes. However, a fundamental ambiguity remains regarding the mathematical interpretation of the GPI's output. Two primary interpretations are widely discussed:

Hilbert Space Interpretation: The GPI computes inner products between states in a Hilbert space of open or closed universes.
Statistical Ensemble Interpretation: The GPI computes an ensemble average (or coarse-graining) of boundary observables over a positive probability measure.

While recent developments in black hole physics often rely on the statistical interpretation (treating multi-boundary amplitudes as moments of an ensemble), it is unclear whether the conditions required for a valid Hilbert space structure are sufficient to guarantee a valid statistical ensemble. This paper addresses the necessary and sufficient conditions for the GPI to admit each interpretation, specifically investigating whether the Hilbert space interpretation implies the statistical one.

Methodology

The authors employ a rigorous mathematical framework based on the moment problem and positivity cones to distinguish between the two interpretations.

Hilbert Space Positivity (SOS):
- For the GPI to define a Hilbert space, the norm of any linear combination of states must be non-negative. This translates to the requirement that all Gram matrices of states are positive semi-definite (PSD).
- In terms of the GPI linear functional $L_G$ acting on polynomials of boundary observables, this condition is equivalent to $L_G[P] \geq 0$ for all Sum-of-Squares (SOS) polynomials. An SOS polynomial $P$ can be written as $P = \sum |Q_i|^2$ .
- The authors denote the cone of such test polynomials as $\mathcal{C}^{\text{Hilbert}}$ .
Statistical Positivity (Non-SOS):
- For the GPI to represent a statistical average over a positive measure $d\mu(\alpha)$ , the functional must be non-negative on all polynomials that are pointwise non-negative on the support of the measure, not just SOS polynomials.
- The authors utilize the fact that for more than two variables, there exist polynomials that are pointwise non-negative but are not sums of squares (non-SOS).
- The condition for a statistical ensemble is $L_G[P] \geq 0$ for all $P \in \text{Pos}(\mathcal{K})$ , where $\mathcal{K}$ is the support of the measure. This defines a strictly larger cone, $\mathcal{C}^{\text{stat}}$ , than the Hilbert space cone.
Open vs. Closed Universes:
- Closed Universes: The support of the statistical measure is unconstrained by internal structure, making the distinction between SOS and non-SOS polynomials the primary diagnostic.
- Open Universes: The random variables (boundary partition functions) in each ensemble member must themselves define valid inner products. This restricts the support $\mathcal{K}$ to a non-trivial domain (e.g., completely monotonic functions for thermal partition functions). The authors analyze constraints arising from "doubly non-negative" matrices versus "completely positive" matrices to distinguish the interpretations in this context.
Toy Models and Bootstrap Analysis:
- The authors construct explicit gravitational toy models (coupling topological gravity to interacting matter) to demonstrate cases where Hilbert space positivity holds but statistical positivity fails.
- They perform a bootstrap study on multivariate probability distributions and multi-matrix integrals to quantify how much tighter the bounds on moments become when imposing statistical positivity versus Hilbert space positivity.

Key Contributions and Results

1. Necessary and Sufficient Conditions

The paper establishes that:

Hilbert Space Positivity requires the GPI to be positive on the cone of SOS polynomials ( $\mathcal{C}^{\text{Hilbert}}$ ).
Statistical Positivity requires the GPI to be positive on the cone of all pointwise non-negative polynomials on the support $\mathcal{K}$ ( $\mathcal{C}^{\text{stat}}$ ).
Result: $\mathcal{C}^{\text{Hilbert}} \subsetneq \mathcal{C}^{\text{stat}}$ . The conditions for (i) are necessary but not sufficient for (ii).

2. Concrete Diagnostics (Closed Universes)

The authors identify specific constraints on wormhole amplitudes that are invisible to Gram matrix positivity but required for a statistical interpretation.

Motzkin Polynomial: For two boundary conditions, the Motzkin polynomial $P_M = x_1^4 x_2^2 + x_1^2 x_2^4 + 1 - 3x_1^2 x_2^2 \geq 0$ is non-SOS. Statistical positivity requires the GPI to satisfy a specific inequality involving 6-boundary wormhole amplitudes (Eq. 17), which is not implied by the positivity of any Gram matrix.
Choi-Lam Polynomial: For four boundary conditions, the Choi-Lam polynomial provides a constraint on 4-boundary amplitudes (Eq. 19) that distinguishes a "pseudo-average" from a true statistical average.

3. Concrete Diagnostics (Open Universes)

In the open universe case, the distinction is captured by matrix positivity classes:

Hilbert Space: Requires the moment matrix $M$ to be doubly non-negative (PSD and entry-wise non-negative).
Statistical Ensemble: Requires $M$ to be completely positive (a sum of outer products of non-negative vectors).
Result: The set of completely positive matrices is a strict subset of doubly non-negative matrices for dimensions $n \geq 5$ . The authors use the Horn matrix (Eq. 54) to construct a constraint (Eq. 55) involving at most two boundaries that is satisfied by Hilbert space positivity but violated by a valid statistical ensemble.

4. Gravitational Toy Models

The authors construct a model coupling the Marolf-Maxfield topological gravity to interacting matter:

The matter sector is chosen to satisfy reflection positivity (SOS) but violate non-SOS positivity (e.g., $\ell(P_{CL}) < 0$ ).
Finding: When coupled to gravity, the full theory retains a well-defined Hilbert space (all Gram matrices are PSD) but fails the statistical positivity test. The GPI computes a pseudo-average (a pseudo-distribution) rather than a true statistical ensemble.
They further show that even if the direct violation is suppressed, one can construct higher-boundary amplitudes using sourceless boundaries to isolate the violation, proving that the failure of statistical positivity persists.

5. Bootstrap Sharpening

In Appendix C, the authors apply these constraints to bootstrap studies of multivariate distributions and multi-matrix integrals.

Finding: Imposing statistical positivity (non-SOS constraints) significantly sharpens the bounds on the moments of the ensemble compared to imposing only Hilbert space (SOS) constraints.

Significance and Claims

The paper claims to provide a definitive diagnostic for determining whether a gravitational path integral computes a genuine statistical average or merely a pseudo-average.

Classification of EFTs: The derived inequalities allow for the classification of low-energy effective field theories based on which interpretation of the GPI they admit. The authors note that while previous work (e.g., axion-wormhole analysis) utilized Hilbert space positivity, determining if these theories admit the stronger statistical interpretation requires computing specific combinations of multi-boundary amplitudes identified in this work.
Algebraic Interpretation: The failure of statistical positivity in the presence of Hilbert space positivity implies that the boundary creation operators $\hat{Z}[J]$ commute algebraically but do not strongly commute. Consequently, they do not admit a joint spectral representation, preventing the construction of a classical ensemble of $\alpha$ -states for the full theory with sources.
General Applicability: The logic extends beyond gravity to any statistical theory where observables are moments of a distribution, including matrix models and critical points of statistical systems (unitary and non-unitary).

The authors conclude that these positivity conditions impose nontrivial constraints on wormhole amplitudes, thereby restricting which gravitational effective field theories can be embedded in consistent theories of quantum gravity that admit a statistical interpretation.