Quantum and AI to Accelerate Scientific Discovery
Chi Chen (Microsoft)
AI has rapidly accelerated scientific discovery by enabling fast prediction and optimization across chemistry and materials science. Quantum computing, meanwhile, promises unprecedented accuracy in modeling complex systems. Together, they offer a powerful synergy: AI delivers speed and scalability, while quantum provides physics-grounded accuracy. This combination could dramatically shorten discovery cycles and unlock new classes of materials and molecules.
Reducing runtimes of ground state calculations on quantum computers
Raffaele Santagati (Boehringer Ingelheim)
Quantum chemistry simulations are among the most promising applications of fault-tolerant quantum computers. However, while recent algorithmic advancements, such as qubitization and improved Hamiltonian representations, like tensor hypercontraction, have significantly reduced resource requirements, achieving practical runtimes for industrially relevant systems remains challenging. We improve upon these advancements and combine these improvements with a novel active volume (AV) compilation technique that optimizes resource utilization by reducing the overhead associated with idling logical qubits. Estimating runtimes reductions of our approach and applying it to the challenging cytochrome P450 system, a key enzyme in drug metabolism, we demonstrate the potential of our strategy in bringing quantum computing closer to practical applications in pharmaceutical research and other industries.
Classical Baselines for Flagship Chemical Targets of Quantum Computing
Matthew Otten (University of Wisconsin – Madison)
Quantum computing papers often point to flagship catalytic systems, including FeMoco, cytochrome P450, iron–sulfur clusters, and Ru‑based catalysts, as compelling applications, yet rigorous classical baselines are uneven. We present a systematic study of active‑space Hamiltonians for these targets using semistochastic heat‑bath configuration interaction (SHCI). For each system we perform careful convergence with respect to the variational selection threshold and perturbative correction, and we report state‑of‑the‑art classical reference data, including the ground-state energies, estimated Hartree-Fock wavefunction overlaps, and reproducible performance metrics (determinant counts, wall‑clock time, etc). We provide analysis of the estimated absolute error and find that some of these systems are probably converged to sub-mHa accuracy. The picture that emerges is a sharper boundary for quantum advantage in chemical simulation: several widely cited instances appear classically tractable at the stated accuracy goals, while others remain credible candidates for early fault‑tolerant quantum algorithms. By establishing rigorous and transparent baselines, our work provides concrete milestones for algorithm validation, resource estimation, and principled target selection in chemical applications of quantum computing.
Quantum-Classical Auxiliary Field Quantum Monte Carlo with Matchgate Shadows on Trapped Ion Quantum Computer
Evgeny Epifanovsky (IonQ)
Quantum-classical auxiliary-field quantum Monte Carlo (QC-AFQMC) with quantum tomography matchgate shadows is a promising near-term quantum algorithm with the potential to deliver practical value in quantum chemistry and materials science. However, prior studies have raised significant concerns about its viability, citing the prohibitive cost of classical post-processing. In this work, we demonstrate that, through algorithmic innovations and highly optimized implementations leveraging state-of-the-art NVIDIA GPUs, we achieve several orders of magnitude speedup in the post-processing step compared to previous approaches. We further showcase an end-to-end workflow modeling a transition metal catalyzed synthetic reaction, executed on IonQ’s Forte quantum computer. This 24-qubit simulation represents the largest QC-AFQMC matchgate shadow experiment performed on quantum hardware to date. Combined with advanced error mitigation techniques, our results achieve accuracy competitive with leading classical methods. These advances mark a substantial step toward the practical application of quantum computing in real-world quantum chemistry simulations.
Harnessing Intrinsic Noise for Efficient Quantum Simulation of Open Quantum Systems
Yu Zhang (Los Alamos National Laboratory)
Simulating open quantum systems on quantum computers is often achieved by embedding non-unitary dynamics into a larger unitary framework, which requires many ancillary qubits. We propose an alternative approach that directly exploits the intrinsic noise of qubits as a computational resource. By preserving and harnessing specific noise channels, we emulate non-unitary processes without explicit unitary encodings, thereby reducing qubit overhead. I will present a noise-assisted quantum algorithm that models open system evolution with a few qubits, demonstrating that carefully tailored noise can enable practical quantum simulation on near-term hardware.
Finding Quantum Advantage for Quantum Chemistry
Birgitta Whaley (UC Berkeley)
he potential of quantum algorithms to enable significant computational speed up for molecular electronic structure calculations has generated significant interest in the quantum computing community. Â While worst case evaluation of molecular ground states is hard even for quantum computers, empirical studies have suggested that generic chemical systems show no exponential quantum advantage over standard classical algorithms. However, for the important class of strongly correlated electronic systems, the situation is different. I shall describe quantum algorithms for systems with both strong and weak correlations that provide both algorithmic and practical advantages over the best-known classical algorithms for these systems, enabling efficient solutions for both ground and excited states of molecular systems undergoing bond breaking and catalysis, and for systems possessing multiple unpaired electrons or multivalent metal atoms. These approaches are based on non-orthogonal methods, employing linear combinations of either exponentially correlated UCC ansatz states (the non-orthogonal quantum eigensolver, NOQE) or matrix product states (the tensor network quantum eigensolver, TNQE), with the latter allowing calculations linear in the system size. I shall also describe recent work proving the hardness of classical simulation of the corresponding quantum circuits, and showing that exponential quantum speedups are possible in quantum chemistry with linear depth, for important molecular systems in which both strong and weak electronic correlations are important.
Quantum Assisted Ghost Gutzwiller Ansatz
Kristine Rezai (IQM)
The ghost Gutzwiller ansatz (gGut) embedding technique was shown to achieve comparable accuracy to the gold standard dynamical mean-field theory method in simulating real material properties, yet at a much lower computational cost. Despite that, gGut is limited by the algorithmic bottleneck of computing the density matrix of the underlying effective embedding model, a quantity which must be converged within a self-consistent embedding loop. We develop a hybrid quantum-classical gGut technique which computes the ground state properties of embedding Hamiltonians with the help of a quantum computer, using the sample-based quantum-selected configuration interaction (QSCI) algorithm. We study the applicability of SCI-based methods to the evaluation of the density of states for single-band Anderson impurity models within gGut and find that such ground states of interest become sufficiently sparse in the CI basis as the number of ghost orbitals is increased. Further, we investigate the performance of QSCI using local unitary cluster Jastrow (LUCJ) variational quantum states in combination with a circuit cutting technique, prepared on IQM’s quantum hardware for system sizes of up to 11 ghost orbitals, equivalent to 24 qubits. We report converged gGut calculations which correctly capture the metal-to-insulator phase transition in the Fermi-Hubbard model on the Bethe lattice by using quantum samples to build an SCI basis with as little as 1% of the total CI basis states.
Leveraging Quantum Computers and Machine Learning to Simulate Biomolecular Processes
Norm Tubman (NASA)
Reactive molecular dynamics simulations using highly accurate force field potentials offer valuable insights into complex biomolecular processes. Ab initio molecular dynamics, powered by quantum-computed potential energies, holds the promise of delivering virtually exact dynamics for modeling such systems. However, the high computational cost and data limitations associated with quantum hardware remain significant barriers. Challenges such as noisy qubit environments, the computational expense of gradient evaluations, and the large number of qubits required for simulating realistic systems hinder practical implementation. In this work, we demonstrate that recent advances in machine learning can help overcome some of these limitations. By integrating transfer learning with machine-learned potential energy surface construction, we propose a novel framework that paves the way for more feasible and accurate molecular dynamics simulations on quantum hardware.