Tuesday 11 March 2025
Researchers have made a significant breakthrough in developing a new approach to preparing thermal states for quantum systems, paving the way for more accurate simulations of complex quantum phenomena.
Thermal states are critical in understanding quantum systems, as they represent the equilibrium state that these systems reach at high temperatures. However, achieving thermal states for large-scale quantum systems is an extremely challenging task due to the exponential growth in the complexity of the parameterized quantum circuits required to prepare them.
To overcome this challenge, scientists have turned to reinforcement learning (RL), a type of artificial intelligence that enables machines to learn from their environment and make decisions. In this case, RL has been used to optimize the preparation of thermal states for a specific class of quantum systems known as Sachdev-Ye-Kitaev (SYK) models.
The SYK model is a simplified variant of the Sachdev-Ye model, which was introduced in a series of seminal talks by Kitaev. It consists of N Majorana fermions in 0+1 dimensions with random couplings between q fermions, drawn from a Gaussian distribution with zero mean and variance proportional to J2/Nq−1.
The RL framework used in this study employs an agent that interacts with its environment to learn an optimal policy for preparing thermal states. The agent’s behavior is governed by a stochastic policy, which defines the probability of choosing each possible action given the current state of the system.
In this case, the actions correspond to the application of specific quantum gates to the system, while the states represent the configuration of the quantum circuit at each step. The agent learns to optimize its policy through a process of trial and error, receiving rewards or penalties based on the quality of the thermal state achieved.
The results demonstrate that the RL approach is able to prepare thermal states with high accuracy for SYK models with up to 12 Majorana fermions, outperforming traditional methods such as first-order Trotterization. The method also scales well with increasing system size, making it a promising tool for simulating complex quantum phenomena.
The researchers used a 3D convolutional neural network (CNN) to process the tensor-based encoding of the quantum circuits, which captures the structure and arrangement of the gates in the circuit. This approach allows the agent to learn more complex features and relationships between the gates, leading to better performance.
The study also highlights the importance of careful postprocessing to achieve an optimal balance between free energy, Hamiltonian expectation value, and entropy of the thermal state.
Cite this article: “Breakthrough in Quantum Simulation: Reinforcement Learning Prepares Thermal States with High Accuracy”, The Science Archive, 2025.
Quantum Systems, Thermal States, Sachdev-Ye-Kitaev Model, Reinforcement Learning, Artificial Intelligence, Quantum Circuits, Majorana Fermions, 3D Convolutional Neural Network, Postprocessing, Quantum Simulation.
