Machine Learning Applications in Quantum Phase Transitions

Machine learning methods have emerged as powerful tools to characterise and predict quantum phase transitions. Rather than rely solely on conventional order parameters, these approaches extract patterns from high-dimensional data such as wavefunction snapshots or Monte Carlo configurations. Supervised models—including convolutional networks and graph-based encodings—classify phases and pinpoint critical points in models ranging from fermionic lattice systems to spin liquids. Unsupervised and semi-supervised techniques, such as variational autoencoders and clustering algorithms, reveal latent structures and novel phases by mapping many-body states to reduced representations. Neural-network quantum states—where variational parameters encode the many-body wavefunction—offer a complementary perspective, enabling efficient optimisation of ground-state energies and capturing complex entanglement across phase boundaries. These methods can dramatically reduce data and computational costs compared to traditional approaches, facilitating studies of larger systems and higher-dimensional models. Applications span topological transitions, quantum-spin liquids and disordered phases, with implications for materials design, quantum simulation and early detection of critical behaviour in emerging quantum technologies.

Research from Nature Portfolio

Recent studies have introduced classical machine-learning algorithms that predict ground-state properties of local Hamiltonians with unprecedented efficiency. By incorporating an inductive bias reflecting geometric locality, these models achieve logarithmic scaling in both data requirements and computational cost, enabling accurate predictions for systems of up to forty-five qubits using a modest training set. Another advance has developed a stochastic-reconfiguration optimiser tailored to deep neural quantum states with millions of parameters. This method resolves ground-state energies in frustrated spin models on square and triangular lattices, uncovering numerical evidence for gapless quantum-spin-liquid phases. Foundational work has also demonstrated that convolutional neural networks applied to Green’s-function data from auxiliary-field quantum Monte Carlo can overcome severe sign problems, correctly identifying phase transitions in interacting fermion systems and offering a robust framework for automated phase classification.

Research from all publishers

A novel network-theoretical framework represents wavefunction snapshots as graphs and analyses their algorithmic complexity, revealing scale-free structures that signal universal behaviour across critical points. This approach enables fully scalable cross-platform certification of quantum simulators, showing a clear decrease in complexity and an increase in correlation length upon crossing a phase transition. In continuous-variable systems, custom neural-network many-body states have been employed to simulate nonequilibrium dynamics of two-dimensional rotor models. By leveraging Hamiltonian Monte Carlo sampling, these methods capture quantities such as return probability and vorticity oscillations after a quantum quench, bridging the gap between experiment and theory for large lattices. Additionally, deep learning algorithms have been applied to classify eigenfunctions in disordered two-dimensional electron systems. Convolutional networks trained on raw wavefunction data successfully distinguish insulating, metallic and topological phases, overcoming randomness and establishing a versatile tool for disordered quantum matter.

Machine Learning Applications in Quantum Phase Transitions publication trend

The graph below shows the total number of articles in machine learning applications in quantum phase transitions across all publications each year (not limited to Nature Index journals).

Technical terms

Quantum phase transition: A transformation between distinct ground-state phases of a quantum system driven by quantum fluctuations at zero temperature.

Ground state: The lowest-energy eigenstate of a quantum Hamiltonian, determining equilibrium properties at zero temperature.

Neural-network quantum state: A variational ansatz in which the amplitudes of a many-body wavefunction are parameterised by a neural network.

Convolutional neural network: A deep-learning architecture that applies convolutional filters to extract spatial features, adapted here to recognise phase signatures in many-body data.

Kolmogorov complexity: A quantitative measure of the minimal description length of a dataset, used to assess structural changes in wavefunction snapshots.

Scale-free network: A graph whose node-degree distribution follows a power law, indicating hierarchical connectivity and universal features.

References

1. Improved machine learning algorithm for predicting ground state properties. Nature Communications (2024).
2. Empowering deep neural quantum states through efficient optimization. Nature Physics (2024).
3. Machine learning quantum phases of matter beyond the fermion sign problem. Scientific Reports (2017).
4. Wave-Function Network Description and Kolmogorov Complexity of Quantum Many-Body Systems. Physical Review X (2024).
5. Variational Quantum Dynamics of Two-Dimensional Rotor Models. PRX Quantum (2023).
6. Deep Learning the Quantum Phase Transitions in Random Two-Dimensional Electron Systems. Journal of the Physical Society of Japan (2016).