⚛️ high-energy theory

Do mixed states exhibit deep thermalisation?

This paper demonstrates that the conventional deep thermalisation framework fails for mixed initial states, necessitating a new paradigm where mixed-state deep thermalisation emerges dynamically in chaotic systems as a maximum-entropy ensemble on an augmented system that explicitly depends on the initial state's entropy.

Original authors: Alan Sherry, Sthitadhi Roy

Published 2026-07-21

📖 8 min read🧠 Deep dive

CC BY 4.0

Original authors: Alan Sherry, Sthitadhi Roy

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

Imagine the universe as a giant, chaotic dance floor where particles are the dancers. In physics, we've long known that if you let these dancers mix long enough, they eventually settle into a predictable, "thermal" rhythm, much like a cup of coffee cooling down to room temperature. This is called thermalisation. But recently, scientists discovered something even cooler: if you watch a specific group of dancers (a small subsystem) while keeping a record of every move made by the rest of the crowd (the bath), you don't just see a simple average. You see a rich, complex collection of possible dance routines. This is called deep thermalisation. It's like realizing that while the average temperature of the room is 70 degrees, the specific pattern of air currents in one corner is actually a masterpiece of randomness, exploring every possible way the air could move.

For a long time, physicists assumed this "masterpiece of randomness" only happened when the dancers started in a perfectly pure, crystal-clear state. But in the real world, nothing is perfect. Our quantum machines are noisy, and our starting states are often "mixed"—a bit fuzzy, like a photo taken with a shaky hand. The big question was: Does this deep thermalisation magic still happen if we start with a fuzzy, imperfect state? Or does the fuzziness ruin the whole show?

This paper tackles that exact question. The authors, Alan Sherry and Sthitadhi Roy, investigate what happens when we start with a "mixed" quantum state and let it evolve. They discover that the old rules of deep thermalisation fail dramatically if there is even the tiniest bit of fuzziness (mixedness) in the starting state. The beautiful, pure-state patterns we used to love simply don't appear in the same way. Instead, they find that mixed states create a new kind of deep thermalisation. It's not a total failure of the concept, but a transformation: the system settles into a different, more complex ensemble that is fundamentally distinct from the pure-state version and depends on the specific "fuzziness" of the start. They prove this mathematically for a special, solvable model and show through computer simulations that it happens in generic, chaotic systems too. Essentially, they've rewritten the rulebook for how imperfect quantum systems find their rhythm, showing that the phenomenon still exists but operates under a new paradigm.

The Story of the Fuzzy Dancers

Let's dive into the details with a story. Imagine you have a massive ballroom (the whole quantum system) filled with dancers. You are interested in a tiny VIP section, a small booth called Subsystem A. The rest of the ballroom is the Bath B.

In the old days of physics, we thought that if we let the dancers spin and mix for a long time, the VIP booth would just look like a blurry, average crowd. But modern experiments let us be nosy. We can watch the entire ballroom (Bath B) and record every single move they make. Based on what we see in the ballroom, we can guess what the VIP booth (A) is doing at that exact moment.

If we do this for a long time, we get a Projected Ensemble. Think of this as a photo album. Each photo in the album shows the VIP booth in a different state, depending on what the rest of the ballroom was doing.

Conventional Thermalisation is like looking at the average of all the photos. It tells you the general vibe (the density matrix).
Deep Thermalisation is like looking at the entire album. It asks: "Does this album contain every possible dance move the VIP booth could possibly do, distributed in the most random, maximum-entropy way possible?"

For a long time, scientists thought this "maximum entropy album" only appeared if the dancers started in a Pure State—a state of perfect clarity, like a laser beam. If the dancers started in a Mixed State—a state with some fuzziness or "noise," like a laser beam passing through fog—the old theory said the magic should still happen, just slightly tweaked.

The Plot Twist: Fuzziness Changes the Choreography

The authors of this paper decided to test this assumption. They asked: "What if we start with a mixed state? Does the deep thermalisation album still look like the perfect, random masterpiece defined for pure states?"

The answer, they found, is that the pure-state framework fails dramatically.

They showed that even a tiny amount of fuzziness in the starting state means the specific "pure-state" deep thermalisation (like the emergence of perfect quantum designs) does not occur. The resulting album of photos is no longer the perfect, random collection of pure dance moves. Instead, it's something else entirely. The "average purity" of the states in the album drops below 1, meaning the states in the album are themselves fuzzy. The old rules, which predicted a specific type of perfect randomness (called a "Haar ensemble" or "Scrooge ensemble"), simply do not apply to the states themselves.

The paper rules out the idea that mixed states just "slow down" or "slightly distort" the pure-state result. It's not a matter of degree; it's a matter of kind. The deep thermalisation for mixed states is fundamentally distinct from that of pure states.

The New Paradigm: The "Shadow" Ensemble

So, if the old magic is gone, what is the new magic? The authors propose a new way to understand it.

Imagine the mixed state (the fuzzy starting point) is actually a shadow of a larger, perfect reality. To understand the fuzzy state, you have to imagine it as part of a bigger system that includes some "auxiliary" dancers (let's call them X) that we can't see. If you take a perfect, pure state of this bigger system (A + B + X) and let it dance, it creates a perfect album. But because we can't see the auxiliary dancers (X), we have to "trace them out" (ignore them) to see what happens to our VIP booth (A).

The authors show that the new "Mixed-State Deep Thermal Ensemble" is exactly this: it's the result of taking a perfect, maximum-entropy album from a bigger system (including the hidden X dancers) and then ignoring the X part.

The Catch: Unlike the pure-state version, where you only needed to know the local state of the VIP booth to predict the album, the mixed-state version requires you to know the entire spectrum of the initial fuzzy state. It's like needing to know the entire history of the fog to predict how the shadow will look.

They call this a "mixed-state deep thermal ensemble." It's a new paradigm where the maximum entropy principle is applied to a larger, augmented system, and the result is a shadow cast onto our local subsystem.

The Proof: From Simulations to Exact Math

How sure are they about this? The authors didn't just guess; they proved it.

The General Case (Simulations): They ran computer simulations on a "Kicked Ising Chain," a model of a quantum system that is chaotic and hard to solve. They started with mixed states and let the system evolve. They measured the distance between the resulting "album" and the theoretical "mixed-state deep thermal ensemble" they predicted. The results showed that as time went on, the distance shrank to zero. The system did converge to their new prediction. This suggests that this new behavior is robust and happens in generic, chaotic systems.
The Exact Case (Math): To be absolutely certain, they looked at a special version of the system called the Self-Dual Kicked Ising (SDKI) chain. This system has a special mathematical symmetry that makes it "exactly solvable."
- They proved mathematically that for a specific class of mixed initial states, the system reaches this new deep thermal ensemble at a finite time (specifically, when time $t$ is greater than or equal to the size of the subsystem plus the size of the mixed region, $t \ge |A| + |S|$ ).
- They calculated the exact "moments" (statistical properties) of the ensemble and showed they matched their new formula perfectly.

Why This Matters

This discovery is a big deal because it changes how we think about quantum chaos and information.

Real-World Relevance: In real quantum computers, we can't create perfect pure states. There's always noise (mixedness). This paper tells us that we shouldn't expect the "perfect" deep thermalisation we see in textbooks for pure states. Instead, we should expect this new, shadow-based thermalisation.
New Tools: The authors suggest that we can actually measure this new behavior. They mention that the probabilities of these probabilities (a fancy way of saying how likely certain outcomes are) follow a specific distribution called the Erlang distribution. This is different from the distributions seen in pure states. This gives experimentalists a new way to check if their quantum systems are behaving "deeply thermal" even when they are noisy.

In short, the paper tells us that the universe is more flexible than we thought. Even when we start with a messy, imperfect state, the system still finds a way to explore all possibilities, but it does so through a new, richer mechanism that involves hidden layers of reality. The "fuzziness" doesn't break the dance; it just changes the choreography to a fundamentally distinct style.

Technical Summary: Deep Thermalisation of Mixed States

Problem Statement
The paper addresses the fate of "deep thermalisation" when the initial state of a quantum system is mixed (possessing finite entropy) rather than pure. Deep thermalisation is a phenomenon where, under unitary dynamics, the projected ensemble (PE)—the ensemble of local subsystem states conditioned on measurements of the complement—approaches a universal maximum-entropy ensemble (such as the Haar ensemble or Scrooge ensemble). While this framework is well-established for pure initial states, the authors investigate whether it holds for mixed states, a scenario relevant to experimental platforms where state preparation and measurement (SPAM) errors introduce finite entropy. The central question is whether the standard definitions of deep thermalisation (e.g., emergence of quantum designs) survive even infinitesimal amounts of initial mixedness.

Methodology
The authors employ a combination of analytical derivations and numerical simulations:

Reference Ensemble Definition: They define a reference ensemble, $E_{\text{ref}}[\rho_0]$ , generated by applying a Haar-random unitary ( $U_{\text{Haar}}$ ) to an initial mixed state $\rho_0$ and performing projective measurements on the bath $B$ . This serves as the benchmark for "maximal scrambling."
Moment Analysis: The properties of the PE are characterized by its moments, $\rho^{(k)}_{\text{EPE}}$ . The authors analytically compute these moments for $E_{\text{ref}}[\rho_0]$ in the thermodynamic limit ( $|B| \to \infty$ ).
Purification and Augmentation: To characterize the structure of $E_{\text{ref}}[\rho_0]$ , the authors utilize a purification of $\rho_0$ on an extended system $X \cup A \cup B$ . They analyze the ensemble obtained by tracing out auxiliary degrees of freedom ( $X$ ) from a maximum-entropy ensemble defined on the augmented system.
Model Systems:
- Generic Chaotic Dynamics: Numerical simulations are performed on a non-integrable, periodically kicked (Floquet) Ising chain to demonstrate the emergence of the reference ensemble at late times.
- Exactly Solvable Model: Analytical proofs are derived for the self-dual kicked Ising (SDKI) chain, a dual-unitary circuit, specifically for a class of mixed initial states where the mixedness is confined to a subsystem.

Key Contributions and Results

Fragility of Pure-State Deep Thermalisation:
The authors demonstrate that the standard deep thermalisation framework fails dramatically for mixed initial states. Even for infinitesimal initial mixedness, the reference ensemble $E_{\text{ref}}[\rho_0]$ does not converge to the Haar ensemble or any ensemble of pure states.
- Specifically, the average purity of the ensemble is strictly less than 1 for any finite entropy in $\rho_0$ .
- The trace distance between the moments of $E_{\text{ref}}[\rho_0]$ and the Haar ensemble grows monotonically with the second Rényi entropy of the initial state.
- Consequently, the emergence of exact quantum designs is "fine-tuned" to pure initial states; any finite entropy leads to a finite departure from a design.
New Paradigm: Mixed-State Deep Thermalisation:
The paper proposes a new definition for deep thermalisation in the presence of mixedness. The reference ensemble $E_{\text{ref}}[\rho_0]$ is shown to be equivalent to an ensemble obtained by:
- Defining a maximum-entropy ensemble (specifically a Scrooge ensemble) on an augmented system $X \cup A$ , where $X$ represents auxiliary degrees of freedom purifying $\rho_0$ .
- Tracing out the auxiliary degrees of freedom $X$ from the states in this ensemble.
- Crucially, this construction depends on the full spectrum of the initial density matrix $\rho_0$ , unlike the pure-state case where the reduced density matrix of the subsystem often suffices.
Dynamical Emergence:
- Numerical Evidence: For generic locally interacting chaotic systems (the kicked Ising chain), numerical results show that the moments of the PE induced by time evolution converge to those of $E_{\text{ref}}[\rho_0]$ as $t \to \infty$ .
- Exact Finite-Time Emergence: For the self-dual kicked Ising chain with specific mixed initial states, the authors analytically prove that the PE moments exactly match the mixed-state deep thermal ensemble at a finite time $t_{\text{ref}} = |A| + |S|$ , where $|A|$ is the subsystem size and $|S|$ is the size of the mixed subsystem.

Significance and Claims
The paper claims to provide fundamental insights into how maximum entropy principles manifest in the unitary dynamics of states with finite entropy.

Conceptual Shift: It establishes that deep thermalisation for mixed states is fundamentally distinct from that of pure states, requiring a reformulation of the maximum entropy principle that accounts for the purification of the initial state on an augmented system.
Experimental Relevance: The authors note that while conventional thermalisation (looking only at the reduced density matrix) cannot distinguish between decoherence and unitary scrambling (both yielding featureless states), deep thermalisation measures (probing higher moments) can.
Operational Probes: The paper highlights that signatures of mixed-state deep thermalisation are experimentally accessible via observables based on "probabilities of probabilities" (PoPs) and fidelity estimators. Preliminary results suggest these follow universal Erlang distributions, distinct from the Porter-Thomas statistics of pure-state designs.

The work concludes that while the pure-state paradigm is fragile, a robust, well-defined "mixed-state deep thermal ensemble" exists and emerges dynamically in chaotic systems, offering a new lens for understanding non-equilibrium quantum dynamics in realistic, noisy settings.