Large deviations and conditioned monitored quantum systems: a tensor network approach

This paper introduces a tensor network framework that enables the application of large deviation theory to large monitored quantum many-body systems, successfully identifying first-order dynamical phase transitions and characterizing the coexistence of distinct dynamical phases through conditioned quantum states.

The Gist

Imagine you are watching a massive, chaotic dance party where thousands of dancers (quantum particles) are moving around a room. In a normal party, everyone eventually settles into a predictable rhythm, like a slow, steady sway. But in this specific quantum party, something strange happens: the dancers get stuck in weird patterns, some moving frantically while others stand completely still for long periods. This is what physicists call "glassy dynamics"—a state where things are frozen in a chaotic mess, like honey that won't flow.

The problem is, this party is happening at the quantum level. It's invisible, unpredictable, and the number of possible ways the dancers could move is so huge that it's impossible for a computer to calculate them all. It's like trying to count every possible shuffle of a deck of cards while the deck is being shuffled by a hurricane.

The Big Idea: A New Way to Watch the Party

The authors of this paper have built a new "super-microscope" using a mathematical tool called Tensor Networks. Think of this tool not as a camera, but as a clever way of organizing a massive library of stories.

Instead of trying to read every single book (every single possible history of the dance) one by one, which would take forever, this method groups similar stories together. It allows the researchers to look at the entire library at once and ask: "What are the most common stories? Are there any rare, weird stories that happen only once in a million years?"

The "Biased" Lens

To understand the glassy behavior, the researchers used a trick called Large Deviation Theory. Imagine you are a film director watching the dance party. Usually, you just record what happens naturally. But to find the weird patterns, you put on a special pair of glasses (a "bias") that makes you care more about specific types of movements.

• If you put on "Active Glasses," you only pay attention to the dancers who are jumping and spinning wildly.
• If you put on "Inactive Glasses," you only care about the dancers who are standing still.

By adjusting these glasses, the researchers could force the computer to focus on the rare, extreme moments of the party. They discovered that under certain conditions, the party splits into two distinct modes: a "hyper-active" mode and a "frozen" mode.

The Discovery: A Phase Transition

The most exciting finding is that the system doesn't just slowly change from active to inactive. Instead, it undergoes a First-Order Phase Transition.

Think of water. You can have liquid water or ice. There is a specific temperature where they coexist. You can have a glass with ice cubes floating in water. The paper found that in this quantum dance party, there is a "temperature" (actually a specific interaction strength between the particles) where the system is stuck in a state of coexistence.

Some parts of the quantum system are dancing wildly (active), while other parts are frozen in place (inactive), and they can switch back and forth. This is the hallmark of a "glass"—a material that is stuck between being a liquid and a solid.

Why This Matters

Before this paper, scientists could only simulate small parties or look at individual dancers. They couldn't see the big picture of how the whole room behaves when it gets "glassy."

This new method is like having a super-powerful telescope that can see the entire galaxy of possibilities. It allows scientists to:

1. Map the landscape: They created a "map" (a phase diagram) showing exactly when the system becomes glassy.
2. See the microscopic details: They didn't just see the statistics; they could actually look at the quantum state of the particles during these rare events to understand why they get stuck.

The Analogy of the "Conditioned Ensemble"

Finally, the paper talks about "conditioned ensembles." Imagine you are a detective looking at a crime scene. Usually, you look at the average evidence. But sometimes, you want to know: "If the suspect had done X, what would the rest of the scene look like?"

This method allows the researchers to reconstruct the quantum state of the system as if a specific rare event had happened. It's like rewinding the tape and asking, "Show me the version of the party where everyone was frozen for 10 minutes straight." By doing this, they proved that the "glassy" behavior isn't just a statistical fluke; it's a real, physical property of the quantum system itself.

In Summary

This paper is a breakthrough because it gave physicists a new set of tools to solve a problem that was previously too big to handle. They used a clever mathematical shortcut (Tensor Networks) to explore the "what-ifs" of a quantum system, discovering that these systems can get stuck in a chaotic, glassy state where active and frozen regions coexist. This helps us understand not just quantum computers, but also the fundamental nature of disorder and how things get stuck in the universe.

Technical Summary

1. Problem Statement

The paper addresses the challenge of analyzing dynamical phase transitions and glassy behavior in monitored quantum many-body systems.

• Context: In classical kinetically constrained models (KCMs), large deviation (LD) theory has successfully identified the coexistence of "active" and "inactive" dynamical phases, a hallmark of glassy dynamics. This is achieved by analyzing the statistics of space-time trajectories.
• The Gap: While similar phase coexistence has been suggested in monitored quantum systems, a systematic LD analysis has been impossible for large systems.
  • Reason: The trajectory space dimension grows exponentially with system size ($L$) and time ($T$).
  • Technical Barrier: In classical KCMs, the "tilted generator" (used to bias trajectory ensembles) can be mapped to a Hermitian quantum Hamiltonian, allowing the use of standard tensor network (TN) ground-state methods. In monitored quantum systems, the tilted operator is a non-Hermitian quantum channel acting in Liouville space. No general mapping to a local Hermitian Hamiltonian exists, rendering standard variational TN methods inapplicable.
• Goal: Develop a scalable framework to compute LD statistics (specifically the scaled cumulant generating function, SCGF) and access conditioned quantum states for large monitored quantum systems.

2. Methodology: Tensor Network Framework for Trajectory Ensembles

The authors propose a novel Tensor Network (TN) approach that treats the quantum channel and trajectory ensembles directly, without mapping to a Hermitian Hamiltonian.

A. The Physical Model

• System: A 1D chain of $L$ qubits interacting with a set of ancilla qubits via a collision model.
• Dynamics:
  1. The system and ancillas evolve unitarily under a collision Hamiltonian $H_{CM}$ for time $\Delta t$.
  2. Ancillas are measured in the computational basis and reset.
  3. This generates a stochastic sequence of measurement outcomes (a quantum trajectory).
• Hamiltonian: The system Hamiltonian $H_S$ includes a transverse field ($\Omega$) and nearest-neighbor interactions ($V$). In the limit of strong interactions, it resembles the PXP model (relevant for Rydberg atom simulators).

B. Large Deviation Formalism

• Observable: Dynamical Activity ($A$), defined as the total number of "1" outcomes in the measurement record over time $T$.
• Biasing: A biased ensemble is introduced with probability weights $e^{-sA} \pi(\eta)$, where $s$ is the counting field.
• Tilted Channel: The dynamics of the biased ensemble is governed by a tilted quantum channel $\mathcal{E}_s$:
  $$\mathcal{E}_s(\rho) = \sum_k e^{-s A(k)} K_k \rho K_k^\dagger$$
  where $K_k$ are Kraus operators and $A(k)$ is the activity of outcome $k$.
• SCGF: The scaled cumulant generating function $\theta(s)$ is derived from the dominant eigenvalue $\Lambda(s)$ of $\mathcal{E}_s$: $\theta(s) = \lim_{T \to \infty} \frac{1}{LT} \ln \Lambda(s)$. Non-analyticities in $\theta(s)$ signal dynamical phase transitions.

C. Tensor Network Implementation

The core innovation is representing the non-Hermitian channel $\mathcal{E}_s$ and its eigenvectors using Matrix Product Operators (MPOs) and Matrix Product States (MPS).

1. MPO Representation: The quantum channel is represented as an MPO. The unitary evolution is decomposed via a Trotter expansion into local gates, which are converted into MPOs.
2. Bias Insertion: The bias operator $e^{-s \hat{A}}$ is inserted into the "ancilla legs" of the MPO transfer matrix.
3. Power Method: To find the dominant eigenvalue $\Lambda(s)$ and eigenvector $|\rho_s\rangle$:
   • An initial state (MPS) is iteratively acted upon by the tilted MPO channel.
   • After each application, the bond dimension is truncated (via SVD) to a maximum value $D_{max}$ to maintain efficiency.
   • The process converges to the dominant eigenpair.
4. Conditioned States: The method naturally provides access to the conditioned ensemble. By sampling specific measurement outcomes and projecting the ancilla legs, the authors can reconstruct the quantum state of the system conditioned on specific trajectory histories.

3. Key Contributions

1. First Scalable LD Analysis for Monitored Quantum Systems: The authors overcome the non-Hermitian barrier by applying TN methods directly to the quantum channel in Liouville space, enabling the study of systems with $L \sim 60$ qubits.
2. Access to Conditioned Quantum States: Unlike previous methods that only provided trajectory statistics, this framework allows the microscopic characterization of the quantum state conditioned on rare dynamical events (e.g., low-activity trajectories).
3. Exact Derivative Calculation: The authors derive a method to compute the mean activity $a(s)$ directly from the eigenvectors using a Hellmann-Feynman-like relation, avoiding noisy numerical differentiation of $\theta(s)$.

4. Results

The framework was applied to a driven-dissipative quantum system inspired by Rydberg atom simulators.

• Dynamical Phase Transitions:
  • For weak interactions ($V/\Omega = 2$), the system exhibits smooth behavior with a single dynamical phase.
  • For strong interactions ($V/\Omega \approx 5.875$), the SCGF $\theta(s)$ develops a non-analyticity near $s=0$.
  • The activity $a(s)$ shows a sharp crossover, and the rate function $\phi(a)$ becomes flat (Maxwell construction), indicating the coexistence of active and inactive phases.
• Finite-Size Scaling:
  • The crossing point of the SCGF branches shifts toward $s=0$ as system size $L$ increases.
  • Extrapolation to the thermodynamic limit ($L \to \infty$) confirms a first-order dynamical phase transition exactly at $s=0$ for specific interaction strengths ($V/\Omega \approx 4.3, 5.875, 6.8$).
• Glassy Signatures: The coexistence of active (fluctuating) and inactive (frozen) trajectories in the thermodynamic limit is identified as a signature of glassy dynamics.
• Microscopic Characterization:
  • By analyzing string correlators in the conditioned ensemble, the authors showed that "inactive" trajectories correspond to quantum states where sites remain excited for long periods (frozen dynamics).
  • This confirms that the phase transition is not merely a statistical artifact of measurement records but is imprinted in the quantum dynamics itself.

5. Significance

• Bridging Classical and Quantum Glassiness: The work establishes a rigorous link between classical large deviation theory (used for glasses) and monitored quantum dynamics, proving that quantum systems can exhibit similar glassy phase coexistence.
• New Tool for Quantum Simulation: The method provides a powerful tool for analyzing rare events and non-equilibrium behavior in quantum devices, which is crucial for understanding measurement-induced phase transitions and learnability.
• Generalizability: The framework is not limited to activity bias; it can be extended to other observables, space-time correlations, and models with long-range interactions (naturally realized in Rydberg platforms).
• Microscopic Insight: By providing access to conditioned states, the method moves beyond "black box" statistics to offer a microscopic understanding of why dynamical phases coexist, revealing the underlying quantum mechanisms of glassiness.

In summary, this paper introduces a robust tensor network framework that solves the long-standing problem of applying large deviation theory to large-scale monitored quantum systems, revealing a rich landscape of dynamical phase transitions and glassy behavior.