Quantum thermal state preparation for near-term quantum processors

This paper introduces a simple and efficient algorithm for preparing quantum thermal states on near-term processors by combining engineered bath resetting with modulated system-bath coupling, which achieves a fixed point approximating the Gibbs state with second-order accuracy in coupling strength and has been validated numerically for both the 2D Quantum Ising model and free-fermion systems.

The Gist

The Big Picture: Cooling a Quantum System

Imagine you have a chaotic, hyper-active room full of people (the quantum system) who are jumping around, shouting, and bumping into each other. This represents a quantum computer at a high temperature or high energy.

In the real world, if you want to calm these people down and get them to sit in specific, orderly seats (a thermal state or Gibbs state), you usually just open a window and let the heat escape into the cold outside air. The "outside air" is a massive heat bath (like the ocean or the atmosphere) that is so big it never gets warm; it just absorbs the heat forever.

The Problem:
Quantum computers are tiny. They don't have an ocean to dump their heat into. They only have a few extra "helper" qubits (ancilla qubits) to act as a mini-bath. If you just let the system talk to this small bath, the heat bounces back and forth like a ping-pong ball in a small room. The system never truly settles down; it keeps oscillating.

The Goal:
The authors want to build a recipe to force a quantum computer to settle into a calm, ordered state that looks exactly like it has been cooled by a giant ocean, even though it's only using a tiny, artificial "bath."

---

The Solution: The "Modulated Coupling" Protocol

The authors propose a clever three-step dance to achieve this cooling. Think of it as a DJ mixing a party to get everyone to slow down.

1. The Setup: The Tiny Bath

Instead of a giant ocean, they use a small group of "resettable" qubits (the DJ's turntable).

• The Trick: Every time the system gets too excited, the DJ hits a "Reset" button, instantly turning the helper qubits back to a calm, empty state (|0⟩). This prevents the heat from bouncing back.

2. The Dance: Modulated Coupling

This is the most important part. You can't just let the system and the bath talk to each other constantly. If they talk too loudly, they mess each other up.

• The Analogy: Imagine trying to teach a hyperactive dog to sit. If you shout "Sit!" constantly, the dog gets confused. But if you shout "Sit!" gently, then wait, then shout it again with a specific rhythm, the dog learns.
• The Science: The authors use a filter function. They turn the connection between the system and the bath on and off smoothly, like a dimmer switch. They tune this rhythm so that it only "resonates" with the specific energy levels the system needs to lose. It's like tuning a radio to a specific station to hear the music clearly while ignoring the static.

3. The Randomizer: Breaking the Loop

Sometimes, if the rhythm is too perfect, the system gets stuck in a loop (like a record skipping).

• The Analogy: Imagine a dancer who keeps doing the exact same move over and over. To fix this, the DJ suddenly changes the beat or spins the dancer randomly for a split second.
• The Science: They add a randomization step. They let the system evolve for a random amount of time before the next reset. This "shakes up" the system, destroying any weird quantum glitches (coherences) that might prevent it from settling into the right state.

---

Why This Matters: The "Near-Term" Advantage

Most previous theories said, "To cool a quantum computer perfectly, you need a super-advanced machine that can measure energy levels with infinite precision." That's like saying, "You can only bake a perfect cake if you have a kitchen that doesn't exist yet."

This paper says: "No, you can do it with the kitchen you have today."

• Simple Ingredients: Their method only requires basic operations: letting the system evolve, connecting it to a helper, and resetting the helper.
• Robustness: They proved mathematically that even if the connection isn't perfect, the errors are tiny (they scale with the square of the coupling strength).
• The Results: They tested this on a "Quantum Ising Model" (a classic physics problem about magnets). They showed that their method successfully cooled the system into the right state, even near the "phase transition" (the point where the material changes from a magnet to a non-magnet), which is usually the hardest part to simulate.

The Takeaway

Think of this paper as a new recipe for baking a cake in a microwave instead of a professional oven.

• Old way: You needed a perfect, massive oven (infinite bath) to get the cake right.
• New way: The authors found a way to use a small, imperfect microwave (near-term quantum processor) by carefully timing the heating pulses (modulated coupling) and shaking the pan (randomization) to get a perfect cake (thermal state).

This is a huge step forward because it means we don't have to wait for "future" quantum computers to study complex chemistry, materials science, or optimization problems at finite temperatures. We can start doing it now with the hardware we have.

Technical Summary

1. Problem Statement

Preparing quantum thermal states (Gibbs states, $\hat{\sigma}_\beta \propto e^{-\beta \hat{H}_S}$) is a fundamental challenge for quantum simulation, with applications in physics, chemistry, and optimization. While real-world systems naturally thermalize by dissipating heat into macroscopic environments, simulating this process on quantum computers is difficult due to two main obstacles:

1. Finite Bath Recurrences: Quantum simulators have limited qubit counts ($N_B \ll N_S$). Coupling a system to a finite bath leads to Poincaré recurrences rather than irreversible decay to a steady state.
2. Energy-Time Uncertainty: Accurately resolving energy levels to satisfy detailed balance typically requires simulation times scaling inversely with the energy level spacing, which can be exponentially large in system size. This renders algorithms relying on Quantum Phase Estimation (QPE) inefficient for near-term devices.

Existing theoretical solutions (e.g., exact Lindblad samplers) often require complex, high-level operations or resources beyond the reach of current hardware. Conversely, simple cooling protocols often fail to produce accurate thermal states at finite temperatures or near phase transitions.

2. Methodology: Modulated Coupling Protocol

The authors propose a simple, efficient algorithm designed for near-term digital and analog quantum processors. The protocol generates a quantum channel $\mathcal{E}$ that drives the system toward a fixed point approximating the Gibbs state. The core mechanism involves engineered dissipation using a small, resettable bath of auxiliary qubits.

The protocol consists of repeated "reset cycles" comprising four stages (see Fig. 1 in the paper):

1. Initialization: A bath of $N_B$ auxiliary qubits is initialized in the state $|0\rangle^{\otimes N_B}$.
2. Joint Unitary Evolution: The system and bath evolve under a time-dependent unitary $\hat{U}(\tau)$ for a duration $T$. The interaction is modulated by a filter function $f(\tau)$:
   $$\hat{V}(\tau) = \theta f(\tau) \sum_{\mu} \hat{A}_{S,\mu} \otimes \hat{Y}_{B,\mu}$$
   where $\theta$ is the coupling strength, $\hat{A}_{S,\mu}$ are local system operators, and $f(\tau)$ is a Gaussian filter.
3. Bath Reset: The auxiliary qubits are reset to $|0\rangle$, effectively removing entropy from the system.
4. Randomization (Key Innovation): A random unitary evolution $\hat{R}$ is applied to the system for a random duration. This step acts as a dephasing map in the energy eigenbasis, suppressing unwanted coherences and preventing resonant divergences.

Key Parameters:

• Filter Width ($a$): Set by $a = \sqrt{4h/\beta}$, where $h$ is the bath energy scale and $\beta$ is the target inverse temperature. This ensures the protocol approximately satisfies Quantum Detailed Balance (QDB).
• Reset Time ($T$): Must be large enough ($T \gg 1/a$) to capture the filter function's tails but scales only as $O(\sqrt{\beta})$.

3. Key Contributions and Theoretical Analysis

A. Approximate Detailed Balance and Error Scaling

The authors prove that the steady state $\hat{\sigma}$ of their protocol approximates the target Gibbs state $\hat{\sigma}_\beta$ with an error scaling as:
$$\|\hat{\sigma} - \hat{\sigma}_\beta\|_1 \sim O(\theta^2)$$
This result holds in the weak-coupling limit ($\theta \ll 1$).

• Mechanism: The protocol induces a Lindblad equation in the interaction picture. While the "Lamb shift" (coherent term) in their protocol does not perfectly match the ideal Hamiltonian required for exact QDB (as defined in prior works like Chen et al.), the deviation is small.
• Role of Randomization: The randomization step is crucial. Without it, the perturbative expansion of coherences diverges at resonant frequencies ($\omega_{ab} = m\pi/T$). The randomization effectively "washes out" these resonances, ensuring the error remains bounded and perturbative.

B. Perturbative vs. Mixing Time Bounds

Unlike previous works that bound errors by the mixing time ($t_{mix}$) multiplied by the coupling error (which could be large if $t_{mix}$ is large), this paper argues that the relevant timescale is the dephasing time of off-diagonal errors.

• Errors in populations (diagonal elements) relax slowly ($\sim 1/\theta^2$).
• Errors in coherences (off-diagonal elements) relax rapidly due to dephasing in the Schrödinger picture.
• Consequently, the total error scales as $O(\theta^2)$, independent of potentially exponential mixing times associated with symmetry breaking, provided the randomization step is active.

C. Renormalized Hamiltonian View

The authors show that the steady state can be interpreted as the exact Gibbs state of a renormalized Hamiltonian $\hat{H}'_S = \hat{H}_S + \theta^2 \hat{C}$. This provides a complementary physical intuition: the bath induces a small level shift, but the state remains thermal with respect to this shifted Hamiltonian.

4. Numerical Results

The authors validated the protocol using two primary models:

A. 2D Quantum Ising Model

• Setup: Simulated on lattices up to $4 \times 4$ spins using Google's qsim simulator.
• Performance: The protocol successfully prepared thermal states across the finite-temperature phase diagram, including near the critical point ($J_c \approx 0.33$).
• Observables:
  • Energy: Matched exact thermal values with high precision.
  • Heat Capacity: Correctly reproduced the peak signature of the phase transition.
  • Correlations: Mutual information and magnetization matched exact diagonalization results.
• Scaling: Errors in energy and trace distance decreased quadratically with the coupling $\theta$, confirming the perturbative prediction. The randomized protocol significantly reduced errors compared to the unrandomized version, particularly near resonances.

B. Free-Fermion Chain (1D Ising)

• Setup: Simulated for large system sizes (up to hundreds of sites) using Gaussian state techniques.
• Scaling:
  • Trace distance error scales as $\|\hat{\sigma} - \hat{\sigma}_\beta\|_1 \leq O(\theta^2 \sqrt{N_S})$.
  • Local observable errors (e.g., energy density) scale as $O(\theta^2)$, independent of system size.
• Significance: This demonstrates that the protocol is robust for large systems in the weak-coupling limit, provided the mixing time is not exponentially large (which is true for free fermions).

5. Significance and Impact

• Near-Term Feasibility: The algorithm requires only local operations, a small number of auxiliary qubits ($N_B \propto N_S$ or even constant), and standard reset capabilities available on current superconducting and trapped-ion processors. It avoids the need for deep QPE circuits.
• Critical State Preparation: Unlike many ground-state preparation methods that struggle near phase transitions, this protocol effectively prepares thermal states near quantum critical points, as evidenced by the correct heat capacity peaks in the Ising model.
• Theoretical Insight: The work clarifies the relationship between detailed balance violations and steady-state errors in the Schrödinger picture, showing that dephasing mechanisms can suppress errors even when the mixing time is long.
• Future Directions: The paper suggests immediate applications in simulating 2D spin models (XY, Heisenberg) and quantum chemistry problems. It also highlights the need to study the stability of these protocols against device noise, which couples directly to populations.

In summary, this paper provides a practical, theoretically grounded "recipe" for preparing thermal states on current quantum hardware, bridging the gap between theoretical Lindblad engineering and experimental realization.