Initial State Memory in Finite Random Brickwork Circuits

✨

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 you have a giant, chaotic kitchen where you are cooking two different meals, Meal A and Meal B. You start with two very different ingredients (the Initial States).

Now, imagine a team of chefs (the Random Gates) who are constantly swapping ingredients between pots, stirring them, and mixing them up in a completely random way. This is like a Random Quantum Circuit.

The big question this paper asks is: If you take a small spoonful of soup from one specific pot (a Subsystem) after a long time, can you still tell which original meal it came from? Or has the soup become so mixed up that Meal A and Meal B taste exactly the same?

Here is the breakdown of their findings using simple analogies:

1. The "Halfway" Rule (The Magic Threshold)

The researchers discovered a magical tipping point based on how much of the kitchen you are looking at.

The Small Spoon (Less than half the system):
If you only look at a small part of the kitchen (less than 50% of the total pots), the chaos wins. The random mixing is so efficient that the "flavor" of the original ingredients gets completely washed away.
- The Result: After a while, the spoonful of soup from Meal A tastes identical to the spoonful from Meal B. The memory of the starting ingredients is lost. It's like trying to tell the difference between two different brands of flour after they've both been baked into identical loaves of bread.
The Big Bowl (More than half the system):
If you look at a large chunk of the kitchen (more than 50% of the total pots), the story changes. Even though the chefs are mixing everything, there is still so much "room" left over that the original ingredients leave a permanent fingerprint.
- The Result: You can still tell Meal A from Meal B, even after infinite time. The memory is retained. It's like if you kept 90% of the ingredients in a giant bowl; no matter how much you stir, you can still see the original seeds and spices because there wasn't enough "empty space" to hide them completely.

2. The "Universal Curve" (The Speed of Mixing)

The paper also looked at how fast this happens.

The Analogy: Imagine dropping a drop of ink into a glass of water.
The Finding: If you have a small glass (small subsystem), the ink spreads out and disappears quickly. If you have a massive ocean (large subsystem), the ink takes a long time to spread, but it never fully disappears if the ocean is big enough.
The Surprise: The researchers found that while the starting ingredients might be different, the speed at which the memory fades (or stays) follows a universal pattern. It doesn't matter if you started with chocolate or vanilla; the "mixing clock" ticks at the same rate for everyone once the system gets big enough.

3. The "Leaky Bucket" (Adding Dissipation)

What if the kitchen isn't perfectly sealed? What if there's a hole in the wall letting the soup leak out (this is called Dissipation or Noise)?

The Analogy: Imagine the kitchen has a leak. If the leak is big and constant, eventually, all the soup drains out, and you are left with an empty pot. No matter how big your bowl was, the memory is gone.
The Twist: However, if the leak is tiny and gets smaller over time (like a slow drip that stops), something interesting happens.
- If the leak is small enough, the system can still remember the original ingredients, even if you are looking at a huge part of the kitchen.
- The paper found a "critical point": a specific size of the leak where the system suddenly switches from "remembering everything" to "forgetting everything." It's like a dam holding back water; if the pressure (noise) gets too high, the dam breaks, and the memory floods away.

4. Mixed vs. Pure Ingredients

The paper also checked what happens if your starting ingredients weren't perfect "pure" meals but were already a bit messy (Mixed States).

The Finding: If you start with messy ingredients, it becomes harder to remember the original state. You need to look at an even larger portion of the kitchen (more than 50%) to find the memory. It's like trying to find a specific grain of sand in a pile of sand that's already been mixed with dirt; you need a bigger net to catch it.

Summary: Why Does This Matter?

In the world of quantum physics, we often worry that information gets scrambled and lost forever (like shuffling a deck of cards). This paper tells us:

Information isn't always lost: If you keep enough of the system, the original information is safe forever.
There is a limit: If you only look at a small piece, the information is gone.
Noise is the enemy: Even a little bit of "leakage" (noise) can destroy this memory, unless the noise is carefully controlled.

This helps scientists understand how to build better Quantum Computers. To store data (memory), you need to make sure your "bowl" is big enough and your "kitchen" is sealed tight enough so the random chaos doesn't wipe out your data!

1. Problem Statement

The paper investigates the conditions under which a finite quantum system, evolving under random unitary dynamics, retains or loses local information about its initial state.

Context: In closed quantum systems, unitary evolution preserves global information, but local information is typically "scrambled" or dispersed, leading to thermalization and the emergence of hydrodynamics.
Core Question: For a specific local subsystem $A$ of size $x$ within a total system of size $2L$ , does the reduced density matrix $\rho_A(t)$ eventually become indistinguishable from the reduced density matrix of a different initial state $\rho'_A(t)$ ?
Specific Focus: The authors analyze the transition between "memory loss" (where the subsystem forgets the initial state) and "memory retention" (where the subsystem retains distinguishable information) as a function of the subsystem size $x$ relative to the total system size $2L$ , and the effect of boundary dissipation.

2. Methodology

The authors employ a combination of random circuit theory, diagrammatic tensor network methods, and statistical mechanics mappings.

Model: A 1D "brickwork" quantum circuit consisting of $2L$ qudits with local Hilbert space dimension $q$ . The time evolution operator $U$ is composed of layers of random unitary gates $U_{j, j+1/2}$ drawn independently from the Haar measure.
Metric: To quantify memory, they measure the normalized Frobenius distance between the reduced density matrices of two arbitrary initial pure states $|\Psi\rangle$ $∣Ψ ⟩$ and $|\Psi'\rangle$ $∣ Ψ^{'} ⟩$ :
$\Delta^2(t) = \frac{\|\rho_A(t) - \rho'_A(t)\|^2}{\|\rho_A(t)\|^2 + \|\rho'_A(t)\|^2}$
where $\rho_A(t) = \text{tr}_{\bar{A}}(|\Psi(t)\rangle\langle\Psi(t)|)$ $ρ_{A} (t) = tr_{\overset{ˉ}{A}} (∣Ψ (t)⟩ ⟨ Ψ (t) ∣)$ .
- $\Delta^2(t) = 0$ implies the states are indistinguishable (memory lost).
- $\Delta^2(t) > 0$ implies memory retention.
Averaging: Since the gates are random, they compute the annealed average $\langle \Delta^2(t) \rangle_a$ over the Haar ensemble. This involves calculating the average of the numerator and denominator separately:
$\langle \Delta^2(t) \rangle_a^2 = 1 - \frac{2 \langle \text{tr}[\rho_A(t)\rho'_A(t)] \rangle}{\langle \text{tr}[\rho_A(t)^2] \rangle + \langle \text{tr}[\rho'_A(t)^2] \rangle}$
Diagrammatic Technique: The authors map the calculation of the averaged traces to a 2D statistical mechanics model.
- They use a "folded" representation where the unitary gates become transfer matrices $W_2$ .
- The problem is mapped to a random walk of a domain wall between two types of states (denoted as "circle" and "square" states in the replicated space) propagating downwards through the circuit.
- Boundary conditions at the top (initial states) and bottom (tracing out the complement $\bar{A}$ ) determine the final overlap.

3. Key Contributions and Results

A. Unitary Dynamics: The "Half-System" Transition

The paper derives an exact expression for the annealed average distance at infinite time.

Subsystems smaller than half ( $x < L$ ):
- The distance $\langle \Delta^2(t) \rangle_a$ decays to zero exponentially fast.
- The timescale for this decay is proportional to the subsystem size $x$ .
- Conclusion: The subsystem completely forgets the initial state; the reduced state becomes maximally mixed (infinite temperature state).
Subsystems larger than half ( $x > L$ ):
- The distance does not vanish at infinite time. It approaches a non-zero constant determined by the overlap of the initial global states.
- Conclusion: The subsystem retains memory of the initial state forever. If the initial states are orthogonal, the reduced states can even become maximally distinct.
Universal Dynamics: For large scales, the time evolution of the distance follows a universal curve dependent on the ratio $r = x/2L$ and the entanglement velocity $v_e$ .

B. Effect of Mixed Initial States

The authors extend the analysis to mixed initial states with purity $q^{-2Ls}$ .

Result: Mixedness makes memory retention harder. The critical threshold for memory retention shifts from $x > L$ to a larger value:
$x > L \left( 1 + \frac{\min(s, s')}{2} \right)$
Interpretation: Mixed states can be viewed as purifications where the environment is larger, effectively reducing the "information capacity" of the subsystem relative to the total system.

C. Boundary Dissipation (Open Systems)

The authors introduce weak boundary dissipation (noise) at one edge of the circuit to model an open system.

Finite Dissipation: If the dissipation strength $p$ is constant ( $O(1)$ ), memory is eventually lost for all subsystem sizes, regardless of $x$ .
Scaling Dissipation: If the dissipation strength scales with time/depth $T$ $T$ as $p \sim (L/T)^\beta$ $p \sim (L / T)^{β}$ , a phase transition emerges.
- There exists a critical dissipation strength $a_c$ (or critical subsystem size $r_c$ ) separating a memory-preserving phase from a memory-loss phase.
- The critical condition is given by:
  $r_c = \frac{1}{2} + \frac{a(q^2 - 1)}{2q^2 \ln q}$
- As the Hilbert space dimension $q \to \infty$ , $r_c \to 1/2$ , recovering the unitary result.

4. Significance and Implications

Refinement of Scrambling Theory: The work challenges the simplistic view that random circuits always scramble information locally. It establishes that for finite systems, information is only lost if the observed subsystem is smaller than the "complement" (the rest of the system).
Connection to Page's Formula: The results are consistent with Page's formula for the average entropy of a subsystem, linking the loss of memory to the entanglement structure of random states.
Symmetry Restoration: The paper connects memory loss to the restoration of emergent symmetries. If the distance goes to zero, the subsystem has lost the symmetry-breaking information of the initial state.
Quantum Coding Transitions: The study of boundary dissipation connects to recent work on "quantum coding transitions," showing that even in chaotic systems, memory can be preserved if noise is sufficiently weak or decays appropriately.
Universality: The identification of universal scaling functions for the distance dynamics suggests that these phenomena are robust features of random unitary dynamics, independent of specific initial state details (provided they are not fine-tuned).

Summary

In essence, the paper proves that in finite random quantum circuits, local memory is retained if and only if the subsystem is larger than half the total system size (for pure states). This retention is robust against time evolution but fragile against boundary dissipation, which can induce a phase transition to a memory-less state. The authors provide exact analytical solutions for these transitions, bridging the gap between random matrix theory, quantum information, and non-equilibrium statistical mechanics.