⚛️ high-energy theory

Deep thermalization with and without quantum chaos

This paper establishes that while measuring a subsystem of a general quantum k-design typically degrades its randomness to at most a k/2-design, chaotic dynamics can reverse this loss, enabling the projected ensemble to achieve exact Haar-randomness even when the global state possesses only limited design properties.

Original authors: Soumik Ghosh, Arjun Mirani, Yihui Quek, Michelle Xu

Published 2026-10-06

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Soumik Ghosh, Arjun Mirani, Yihui Quek, Michelle Xu

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 quantum world, randomness is not just a lack of order; it is a specific, measurable resource. Physicists use a concept called a "design" to describe a collection of quantum states that are random enough to mimic the behavior of a perfectly random set, known as a Haar-random ensemble. Think of a design as a statistical shadow: if you look at the average behavior of the states in the collection, or even their more complex interactions, they look indistinguishable from true chaos. This is vital for quantum computing and cryptography, where generating genuine randomness is difficult, but generating a good approximation is often enough. The central mystery this research tackles is how much of this randomness can be concentrated into a small part of a system by simply measuring the rest. If you have a large, complex quantum system and you measure a portion of it, does the remaining unmeasured part become more random, less random, or stay the same?

The answer depends entirely on how the system was created. The researchers first established a strict, unavoidable limit for systems that are random in a generic way. They proved that if you start with a global system that has a certain level of randomness, measuring part of it will inevitably degrade the quality of the randomness in the remaining part. Specifically, the remaining system can never be more random than half the level of the original whole. If the original system was a "k-design," representing a certain depth of randomness, the leftover piece would, at best, only be a "k/2-design." This means that without special conditions, the act of measuring a quantum system actually throws away half of its statistical complexity. The researchers demonstrated this by constructing a specific, worst-case scenario where a highly random global state hides a subtle pattern that only reveals itself when looking at the remaining part with extreme precision, proving that the loss of randomness is real and unavoidable in general cases.

However, the story changes dramatically when the system is driven by chaotic dynamics, a specific type of evolution found in nature that scrambles information rapidly. The researchers found that if the global system evolves under the influence of a chaotic Hamiltonian—a mathematical description of energy that behaves like a random matrix—the rules of the game flip. In this chaotic setting, measuring a part of the system does not degrade the randomness; instead, it concentrates it. The remaining unmeasured part can become perfectly random, reaching a level of statistical purity that the original global system did not even possess. This phenomenon, known as deep thermalization, occurs at specific moments in time determined by the system's energy spectrum. At these moments, the leftover states become indistinguishable from a perfectly random set, effectively turning a modestly random global state into a maximally random local one.

This amplification of randomness is not just a mathematical curiosity limited to idealized models; the researchers showed it holds up even when the system is not perfectly chaotic. They proved that the effect persists as long as the system's energy levels have certain properties and its underlying structure is sufficiently complex, even if it falls short of being a perfect random matrix. Furthermore, this boost in randomness works even in finite systems, becoming more accurate as the measured portion grows larger. The work identifies global chaos as the specific mechanism that allows measurements to act as a lens, focusing scattered quantum randomness into a single, highly ordered point. It places tight limits on when this amplification is possible, showing that while generic randomness always loses ground, chaos can turn a simple measurement into a powerful tool for generating the highest quality of quantum randomness.

Technical Summary: Deep Thermalization with and without Quantum Chaos

Problem Statement
The paper addresses a fundamental question in quantum information and many-body physics: how much quantum randomness (specifically, the quality of a quantum state design) can be generated in a subsystem through measurement, given a global state evolved under specific dynamics? While "deep thermalization" describes the phenomenon where measuring part of a thermalizing system produces an ensemble on the remainder whose higher moments approximate Haar-random states (a "projected design"), the precise limits of this amplification were unclear. Specifically, the authors investigate whether the "design order" (the highest moment $k$ for which the ensemble matches Haar statistics) of the projected subsystem can exceed that of the global state, and if so, under what conditions.

Methodology
The authors employ a combination of rigorous mathematical construction, probabilistic analysis, and random matrix theory:

Worst-Case Construction: To establish upper bounds on design preservation without assuming specific dynamics, the authors construct explicit counterexamples. They build a global $(2k-1)$ -design where the projected ensemble fails to be even an approximate $k$ -design. This involves creating a global state with a "hidden sign preference" in its projected distribution that is orthogonal to all polynomials of degree less than $k$ , making it invisible to moments below $2k$ in the global view but detectable in the projected view.
Probabilistic Lower Bounds: To establish what is generically achievable, they analyze typical states drawn from global $2k$ -designs, using trace-norm approximations and frame potential calculations to show that a projected $k$ -design is achievable with high probability.
Random Matrix Theory (GUE): To explore the role of chaos, the authors model global dynamics using the Gaussian Unitary Ensemble (GUE). They utilize the replica trick and Weingarten calculus to compute the frame potential of the projected ensemble in the thermodynamic limit.
Robustness Analysis: They extend the GUE results to weaker assumptions, analyzing systems where the Hamiltonian has specific spectral properties (vanishing first trace moment) and eigenbases drawn from approximate unitary 4-designs rather than full Haar randomness.

Key Contributions and Results

Generic Limit: The $2k \to k$ Bound
The paper establishes a tight, optimal bound for generic global designs. Without assumptions on the dynamics generating the state, a global $2k$ -design can guarantee at best a projected $k$ -design.
- Upper Bound: The authors prove that for any fixed $k$ , there exists a global $(2k-1)$ -design whose projected ensemble is not an approximate $k$ -design. The $k$ -th moment of the projected ensemble remains at a constant trace-norm distance from the Haar moment.
- Lower Bound: Conversely, they show that a typical state from a sufficiently accurate global $2k$ -design yields an approximate projected $k$ -design with high probability.
- Conclusion: In the absence of specific dynamical structure, projection can preserve at most half the design order of the global state. This resolves an open question regarding the optimality of the $2k \to k$ guarantee.
Chaos-Induced Amplification: The GUE Case
The authors demonstrate that global quantum chaos can reverse the generic degradation of design quality.
- Infinite Boost: When the global state is generated by evolving under a GUE Hamiltonian for specific discrete times (defined by the zeros of the first normalized trace moment, $J_1(2t)/t = 0$ ), the projected ensemble becomes an exact Haar-random design in the thermodynamic limit.
- Mechanism: This occurs even if the global unitary evolution is only an $O(1)$ -design (constant order) at these times. The chaotic dynamics effectively "concentrate" randomness onto the unmeasured subsystem, boosting a seemingly low-order global design into an infinite-order local design.
- Spectral Condition: The phenomenon relies on the vanishing of the first trace moment of the GUE evolution, suggesting that controlling only the first spectral moment is sufficient to reproduce Haar moments of arbitrarily high order in the projected ensemble.
Robustness and Physical Realizability
Recognizing that GUE Hamiltonians have exponential circuit complexity and are unphysical, the authors prove the robustness of this amplification phenomenon under weaker, more physical assumptions:
- Spectral Relaxation: The full GUE eigenvalue distribution is unnecessary; it suffices that the normalized first trace moment vanishes and concentrates.
- Eigenbasis Relaxation: The requirement for Haar-random eigenvectors can be replaced by the condition that the eigenbasis is drawn from a sufficiently accurate approximate unitary 4-design.
- Finite-Size Effects: The phenomenon persists in finite systems. For a measured subsystem of size $N_B$ and unmeasured size $N_A$ , if $N_B > N_A^{k+3}$ , the projected ensemble forms an approximate $k$ -design with high probability at the specific chaotic times.

Significance
The paper identifies global chaotic dynamics as a specific mechanism that allows measurements to "concentrate" quantum randomness, enabling the generation of high-order designs on a subsystem that far exceed the design order of the global state. This contrasts sharply with the generic case, where projection inevitably degrades design quality by a factor of two.

The authors position these results as a sharpening of the conjecture that deep thermalization and design formation are universal features of quantum chaotic systems. By distinguishing between the behavior of generic designs (which degrade) and chaotic designs (which amplify), the work places tight limits on when such amplification is possible and suggests a bridge between the Eigenstate Thermalization Hypothesis (ETH) and the generation of quantum state designs. The findings provide a theoretical foundation for understanding how simple constant-time chaotic evolution followed by measurement can serve as a resource for generating high-quality pseudorandomness.

Technical Summary: Deep Thermalization with and without Quantum Chaos

More like this