Machine learning speeds up quantum computer calculations

Published on August 11, 2026, a new study details how researchers at the Indian Institute of Technology Bombay are accelerating quantum computer calculations. The work, conducted in collaboration with the Technical University of Denmark, combines perturbation-theoretic measures and generative machine learning to improve computational speed. Researchers detail their approach in Quantum Science and Technology, focusing specifically on applications for quantum computing in chemistry. The code used to generate the study’s data is publicly available on GitHub.

Researchers at the Indian Institute of Technology Bombay and the Technical University of Denmark have developed PIGen-SQD, a workflow designed to accelerate quantum computations for complex chemical simulations. This approach combines generative machine learning with perturbation-theoretic measures to improve the efficiency of quantum-centric supercomputing, a technique aiming to solve problems currently intractable for classical computers. The team’s work, published on August 11, 2026, focuses on accurately reconstructing fermionic states, essential for modeling electron behavior in molecules, despite the noise present in current quantum hardware.

PIGen-SQD addresses a critical bottleneck in quantum chemistry: the need to sample relevant configurations from an enormous Hilbert space. Quantum-centric supercomputing leverages quantum computers to identify these key configurations, then uses classical processors to perform the computationally intensive task of Hamiltonian diagonalization. However, errors in quantum measurements necessitate robust methods for recovering accurate configurations; the new workflow introduces physics-informed pruning based on perturbative measures.

These measures, used in conjunction with samples from quantum hardware, provide substantial overlap with the target state, effectively guiding the machine learning models to focus on the most important sectors of the Hilbert space. The system’s generative machine learning component then stochastically explores this reduced space, identifying additional crucial configurations in a self-consistent manner. Numerical experiments conducted on IBM Heron R2 and R3 quantum processors, utilizing up to 58 qubits, demonstrate the effectiveness of this approach.

The researchers report that PIGen-SQD produces compact, high-fidelity subspaces, significantly reducing the cost of diagonalization while maintaining chemical accuracy even under strong electronic correlations. This reduction in computational burden is a key step toward practical quantum simulations of complex molecules.

By intelligently combining classical insights with the power of machine learning and quantum computation, the researchers have created a system capable of tackling problems previously beyond reach. This work highlights the growing trend of integrating diverse computational techniques to overcome the challenges of realizing practical quantum utility in chemistry and materials science.