Asymptotic Replacement for Quantum Channel Products with Applications to Inhomogeneous Matrix Product States

This paper develops a trace-Dobrushin theory for quantum channel products to establish conditions for asymptotic replacement and memory loss in both deterministic and stationary random settings, subsequently applying these results to derive infinite-volume limits, boundary stability, and correlation bounds for inhomogeneous matrix product states.

The Gist

The Big Picture: Forgetting the Past

Imagine you have a machine that takes a piece of paper (a "quantum state") and shuffles the ink around on it. This machine is a Quantum Channel.

If you run the same machine over and over again, eventually, the original writing disappears. No matter what you wrote at the start, the machine produces the same blurry, static result. In physics, we call this "forgetting the initial state" or "loss of memory."

This paper is about what happens when you don't use just one machine, but a chain of different machines. Maybe the first one is a blender, the second is a printer, and the third is a photocopier. The environment might even change randomly every time you turn the crank.

The author asks: Does this chain of machines eventually forget what it started with? And if so, how fast?

The Core Concept: The "Trace-Dobrushin Coefficient"

To answer this, the author invents a specific measuring stick called the Trace-Dobrushin coefficient.

• The Analogy: Imagine you have two different starting messages, Message A and Message B. You run them through your chain of machines.
• The Measurement: The coefficient measures how different the two output messages are.
  • If the coefficient is 1, the machines are perfect mirrors; they keep the messages distinct forever.
  • If the coefficient is 0, the machines have completely erased the difference. Message A and Message B look exactly the same now.
  • If the coefficient is dropping toward 0, the machines are slowly "forgetting" the past.

The paper proves that if this number drops to zero, the entire chain of machines effectively becomes a single "replacement machine" that just outputs a specific, pre-determined result, regardless of what you fed it in.

Scenario 1: The Deterministic Chain (The Predictable Factory)

First, the author looks at a factory where the order of machines is fixed and known (e.g., Machine A, then B, then C, then A...).

• The Finding: Even if no single machine in the chain is good at erasing memory (maybe Machine A is a perfect mirror, and Machine B is also a perfect mirror), putting them together might create a "super-eraser."
• The Metaphor: Think of a lock. One key might not open it, and another key might not open it. But if you use Key A and then Key B in a specific sequence, the lock clicks open.
• The Result: The paper provides a way to calculate exactly how fast this "lock" opens. It shows that the chain eventually settles into a predictable pattern where the output depends only on the current time, not on the distant past.

Scenario 2: The Random Chain (The Chaotic Kitchen)

Next, the author looks at a kitchen where the chef changes every day, chosen randomly from a menu. Sometimes the chef is great at erasing memories; sometimes they are terrible.

• The Finding: Even in this chaos, if the "average" behavior of the chefs is to erase memories (mathematically, if a value called the Lyapunov exponent is negative), the system will still forget the past.
• The "Quenched" vs. "Annealed" Distinction:
  • Quenched (Frozen): This is looking at one specific random day. "If I pick this specific sequence of chefs, will I forget?" The paper says yes, almost always, and it happens exponentially fast.
  • Annealed (Averaged): This is looking at the average behavior across all possible days. The paper shows that if the chefs are independent (today's chef has no memory of yesterday's), the forgetting happens incredibly fast. If the chefs are somewhat correlated (today's chef is influenced by yesterday's), the forgetting is still fast, but the math is slightly more complex.

The Application: Matrix Product States (The Long String of Beads)

Finally, the author applies these findings to Matrix Product States (MPS).

• The Metaphor: Imagine a very long necklace made of beads. Each bead represents a particle in a quantum system. To understand the whole necklace, physicists often look at a "helper" system (an auxiliary space) that passes information from one bead to the next, like a relay race.
• The Connection: The "relay runners" in this race are exactly the quantum channels the author studied.
• The Result: By proving that the relay runners eventually forget the starting baton, the author proves that:
  1. Stability: The necklace has a stable, predictable shape at the ends (boundary stability), even if the beads in the middle are messy.
  2. Thermodynamic Limit: You can calculate the properties of an infinitely long necklace by just looking at a finite piece, provided the "relay" forgets the start fast enough.
  3. Correlations: If two beads are far apart, they don't "talk" to each other much. The further apart they are, the less they influence each other, and this influence drops off exponentially fast.

Summary of Claims

1. New Tool: The paper defines a precise way to measure how much a chain of quantum machines "remembers" its input.
2. Deterministic Rule: If this memory measure drops to zero, the chain becomes a "replacement machine" that outputs a specific state.
3. Random Rule: Even in a random environment, if the average memory loss is positive, the system forgets the past almost surely and converges to a unique, stable random state.
4. Speed: The paper gives formulas for how fast this forgetting happens (exponential decay) based on how "mixing" the environment is.
5. Application: These rules guarantee that certain quantum materials (described as Matrix Product States) have stable boundaries and that distant parts of the material stop influencing each other quickly.

What the paper does NOT claim:

• It does not claim to build a physical quantum computer.
• It does not claim to solve climate change or medical problems directly.
• It does not claim that every random system forgets the past; it specifically requires the mathematical condition of a "negative Lyapunov exponent" (which essentially means the system is, on average, chaotic enough to erase information).

Technical Summary

1. Problem Statement

The paper addresses the long-time behavior of products of quantum channels that are not necessarily identical (inhomogeneous) or are subject to random time-dependent noise. Specifically, it investigates:

• Memory Loss: Under what conditions does a product of quantum channels $\Phi_{t:s} = \Phi_{t-1} \circ \dots \circ \Phi_s$ "forget" its initial state?
• Asymptotic Replacement: Since inhomogeneous systems often lack a single stationary state, the output may converge to a moving state rather than a fixed density matrix. The paper seeks to characterize the convergence of channel products to a family of replacement channels (channels that erase the input and output a specific state).
• Matrix Product States (MPS): The theory is applied to inhomogeneous Matrix Product States in the left-canonical gauge, where the auxiliary transfer maps are quantum channels. The goal is to establish thermodynamic limits, boundary stability, and correlation decay for these states.

2. Methodology and Core Concepts

The Centered Trace-Dobrushin Coefficient

The central tool developed is the centered trace-Dobrushin coefficient, denoted $\kappa_{tr}(\Phi)$.

• Definition: For a positive trace-preserving map $\Phi$, it is defined as the operator norm of $\Phi$ restricted to the real, self-adjoint, trace-zero subspace $H_0$:
  $$\kappa_{tr}(\Phi) := \sup_{X \in H_0, X \neq 0} \frac{\|\Phi(X)\|_1}{\|X\|_1}$$
• Physical Interpretation: It quantifies the residual trace-norm memory of the input state.
  • $\kappa_{tr}(\Phi) = 0$ if and only if $\Phi$ is a replacement channel (outputs a fixed state regardless of input).
  • $2\kappa_{tr}(\Phi)$ equals the trace-norm diameter of the evolved state space $\Phi(S_d)$.
• Submultiplicativity: A key property is that for any two such maps $S, T$:
  $$\kappa_{tr}(S \circ T) \leq \kappa_{tr}(S)\kappa_{tr}(T)$$
  This allows the analysis of long products by bounding the coefficients of constituent blocks.

Comparison with Existing Criteria

The paper distinguishes $\kappa_{tr}$ from other contraction coefficients:

• Markov-Dobrushin / Doeblin: These provide sufficient upper bounds based on common lower bounds (minorization). $\kappa_{tr}$ is the exact coefficient. A channel can have $\kappa_{tr} < 1$ (strict contraction) even if it admits no common lower bound (e.g., amplitude damping channels).
• Hilbert-Birkhoff: Projective metrics often fail for channels converging to boundary states (non-full rank). $\kappa_{tr}$ remains effective for these boundary-attractor scenarios.
• Diamond Norm: Unconstrained diamond-norm variants are trivial (equal to 1) for state-memory loss because they allow perturbations in reference systems. $\kappa_{tr}$ focuses strictly on the system's trace-zero subspace.

3. Key Contributions and Results

A. Deterministic Inhomogeneous Products

The paper establishes a rigorous equivalence between the decay of the product coefficient and asymptotic replacement.

• Theorem 1 (Fixed-Start Replacement): For a sequence starting at time $s$, the condition $\lim_{t \to \infty} \kappa_{tr}(\Phi_{t:s}) = 0$ is equivalent to the product converging in the induced trace-norm to a family of replacement channels $R_{t:s}(X) = \text{Tr}(X)\eta_{t:s}$.
• Theorem 2 (Pullback Replacement): In a two-sided setting ($I=\mathbb{Z}$), if $\lim_{s \to -\infty} \kappa_{tr}(\Phi_{t:s}) = 0$ for fixed $t$, there exists a unique pullback boundary state $(\rho_t)_{t \in \mathbb{Z}}$ satisfying the consistency relation $\rho_{t+1} = \Phi_t(\rho_t)$. The product converges to the replacement channel centered at this unique moving state.
• Rate Estimates: The paper provides quantitative bounds (exponential, polynomial, or block-based) for the decay of $\kappa_{tr}$ based on "good blocks" (sub-intervals where contraction occurs).

B. Stationary Random Products (Cocycles)

The theory is extended to random environments driven by an invertible, probability-preserving system $(\Omega, \mathcal{F}, \text{pr}, \theta)$.

• Trace-Dobrushin Lyapunov Exponent: Defined as $\lambda_{tr}(\omega) = \lim_{n \to \infty} \frac{1}{n} \log \kappa_n(\omega)$.
• Theorem 3 (Quenched Convergence): If $\lambda_{tr} < 0$ almost surely, there exists a unique dynamically stationary random state $\rho_\omega$ (satisfying $\Phi_\omega(\rho_\omega) = \rho_{\theta\omega}$). The random products converge quenched (almost surely) to the replacement channel centered at $\rho_\omega$ with exponential rates governed by $\lambda_{tr}$.
• Annealed Estimates (Theorem 4): Under additional stochastic decorrelation assumptions (e.g., $\varrho$-mixing or independence of the channel sequence), the paper derives annealed (averaged) convergence rates:
  • $\varrho$-mixing: Yields super-polynomial decay ($O(n^{-p})$).
  • Independence: Yields exponential decay ($O(e^{-\gamma n})$).

C. Applications to Inhomogeneous Matrix Product States (MPS)

The channel results are transferred to inhomogeneous MPS in the left-canonical gauge, where auxiliary transfer maps are CPTP.

• Thermodynamic Limit (Theorem 5): Under the condition that right-tail auxiliary products forget memory ($\kappa_{tr}(\Theta_{q,n}) \to 0$), there exists a unique infinite-volume state $\phi_\infty$. The finite-volume normalized expectations converge to $\phi_\infty$.
• Correlation Decay (Theorem 6 & 8): Spatial correlations between observables separated by a gap $L$ decay according to the trace-Dobrushin coefficient of the transfer product across that gap.
  • Deterministic: $|\phi_\infty(AB) - \phi_\infty(A)\phi_\infty(B)| \leq C \|A\|\|B\| \kappa_{tr}(\Theta_{gap})$.
  • Random: The random infinite-volume MPS exhibits quenched exponential clustering if $\lambda_{tr} < 0$.
• Significance: This provides a unified framework for MPS limits that does not require translation invariance or eventual strict positivity of the transfer maps, relying instead on the intrinsic contraction of the trace-norm.

4. Significance and Impact

1. Unified Framework: The paper unifies the study of deterministic inhomogeneous systems and random stationary systems under a single "trace-Dobrushin" theory, bridging the gap between operator theory and statistical mechanics.
2. Beyond Strict Positivity: Unlike previous works on random MPS that often rely on "eventual strict positivity" (which implies the channel maps everything to the interior of the cone), this theory works for channels converging to boundary states (e.g., pure states or dephased states), which are common in physical models like amplitude damping.
3. Quantitative Precision: By using the exact product coefficient rather than loose upper bounds, the paper provides sharp, quantitative estimates for convergence rates and correlation decay, applicable to both deterministic and random settings.
4. Robustness: The results hold for general CPTP maps without requiring unitality or specific spectral gaps, making them applicable to a broader class of open quantum systems and tensor network states.

In summary, Pathirana develops a robust "product-level" theory of memory loss for quantum channels, proving that the decay of the centered trace-Dobrushin coefficient is the necessary and sufficient condition for asymptotic replacement. This theory is then successfully applied to establish the existence, uniqueness, and correlation properties of thermodynamic limits for inhomogeneous and random Matrix Product States.