Noise-Assisted Variational Quantum Thermalization

Jonathan Foldager, Arthur Pesah, Lars Kai Hansen

TL;DR

The Noise-Assisted Variational Quantum Thermalizer (NAVQT) turns a bug into a feature by using hardware noise as a resource to prepare thermal quantum states. By controlling depolarizing noise levels and using a clever closed-form entropy approximation, the algorithm prepares Gibbs states on NISQ devices without the need for the extra "ancilla" qubits typically required for state purification.

What Problem Does This Paper Address?

Preparing thermal (Gibbs) states—mathematical descriptions of quantum systems at a specific temperature—is a foundational task for quantum chemistry, many-body physics, and quantum machine learning. However, doing this on near-term "Noisy Intermediate-Scale Quantum" (NISQ) hardware is notoriously difficult for three reasons:

  1. The Qubit Tax: Most methods use "purification," which requires doubling the number of qubits (half for the system, half to act as an environment) to represent a mixed state.
  2. The Entropy Hurdle: To find a thermal state, you must minimize Free Energy ($F = E - TS$). While energy ($E$) is easy to measure, the Von Neumann entropy ($S$) is not an observable and is computationally expensive to estimate.
  3. The Noise Paradox: Usually, researchers fight against hardware noise. Here, the challenge is how to integrate unavoidable noise into a variational framework rather than just trying to mitigate it.

Methods

The authors introduce NAVQT, a variational algorithm that treats noise as the source of entropy.

  • The Ansatz: Instead of a purely unitary circuit, they use a sequence of unitary layers (inspired by QAOA) where each layer is followed by a depolarizing channel. Crucially, they assume the ability to control the noise level ($\lambda$)—a technique already used in error mitigation (like Zero-Noise Extrapolation).
  • Entropy Approximation: To bypass the difficulty of measuring entropy, the authors derived a closed-form analytical expression. They did this by approximating a circuit where all noise is "pushed" to the beginning. This allows them to calculate the gradient of the entropy with respect to the noise parameter $\lambda$ directly, without performing full state tomography.
  • Optimization: The algorithm performs gradient descent on two fronts simultaneously: the unitary gate parameters ($\theta$) to minimize energy and the noise level ($\lambda$) to reach the target entropy.

Datasets (Physical Models)

The method was evaluated via simulation on three standard many-body Hamiltonians with system sizes ranging from $N=3$ to $7$ qubits:

  • Classical Ising Chain: Both uniform and random coefficients.
  • Transverse-Field Ising (TFI) Chain: Adds non-diagonal terms to the Ising model.
  • 1D Heisenberg Model: A more complex model of magnetic systems, tested with both uniform and random coefficients.

Key Insights and Ideas

  • Noise as a Resource: In a thermal state, entropy is mandatory. By using the circuit's own depolarizing noise to generate that entropy, the authors eliminate the need for ancilla qubits, making the algorithm much more "hardware-efficient."
  • The "U-Shaped" Difficulty Curve: The algorithm performs best at the extremes. At very high temperatures (high noise, maximally mixed state) and very low temperatures (zero noise, ground state), fidelities are near-perfect. The "hard" part is the intermediate temperature regime where the balance between quantum correlation and thermal noise is most delicate.
  • Heuristic Validity: Even though the entropy formula is based on a simplified "noise-at-the-start" circuit, the authors showed it serves as an excellent heuristic for the actual noisy circuit, following nearly identical optimization trajectories.

Key Results

  • High Fidelity: For uniform Ising and TFI chains, the algorithm consistently achieved fidelities above 0.9 across all temperatures and system sizes.
  • Learning the Physics: The optimizer successfully "learned" the relationship between noise and temperature, automatically tuning $\lambda \to 1$ for high temperatures and $\lambda \to 0$ for low temperatures.
  • Complexity Limits: The Heisenberg model with random coefficients proved the most challenging, with fidelities occasionally dipping below 0.7. The authors suggest this is likely due to the "expressibility" of the chosen circuit (the ansatz) rather than a failure of the optimization logic itself.

Limitations and Future Work

  • Ansatz Expressibility: The current QAOA-style ansatz struggled with the complex Heisenberg model. Future work could explore more flexible or hardware-efficient architectures.
  • Simplified Noise Model: The study used a single shared noise parameter $\lambda$ for the whole circuit. Allowing per-qubit or per-layer noise control could significantly increase the state-space the algorithm can reach.
  • Barren Plateaus: As with many variational algorithms, the authors warn that "noise-induced barren plateaus" (where gradients vanish) could hinder scalability as system sizes grow.
  • Hardware Implementation: While simulated here, the next step is testing this on real NISQ hardware where noise isn't just a parameter in a simulation but a physical reality to be harnessed.

Summary updated

Abstract

Preparing thermal states on a quantum computer can have a variety of applications, from simulating many-body quantum systems to training machine learning models. Variational circuits have been proposed for this task on near-term quantum computers, but several challenges remain, such as finding a scalable cost-function, avoiding the need of purification, and mitigating noise effects. We propose a new algorithm for thermal state preparation that tackles those three challenges by exploiting the noise of quantum circuits. We consider a variational architecture containing a depolarizing channel after each unitary layer, with the ability to directly control the level of noise. We derive a closed-form approximation for the free-energy of such circuit and use it as a cost function for our variational algorithm. By evaluating our method on a variety of Hamiltonians and system sizes, we find several systems for which the thermal state can be approximated with a high fidelity. However, we also show that the ability for our algorithm to learn the thermal state strongly depends on the temperature: while a high fidelity can be obtained for high and low temperatures, we identify a specific range for which the problem becomes more challenging. We hope that this first study on noise-assisted thermal state preparation will inspire future research on exploiting noise in variational algorithms.
BibTeX
@inproceedings{foldager2022noise,
  author = {Jonathan Foldager and Arthur Pesah and Lars Kai Hansen},
  title = {{Noise-Assisted Variational Quantum Thermalization}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-07296-z},
}
Figure