Researchers at UCLA’s NarangLab and Caltech, working with NVIDIA, have developed an artificial intelligence model capable of designing control sequences for quantum applications in scenarios exceeding the limits of conventional optimization. The team’s Fourier Neural Operator learns the quantum dynamics of molecules, enabling the identification of control protocols for complex systems like hydronium (H₃O⁺), chosen for its potential to probe physics beyond the Standard Model. This approach replaces computationally expensive simulations with a surrogate model running up to ~10⁷× faster on NVIDIA’s CUDA-Q Dynamics platform.
Fourier Neural Operator Learns Hydronium Molecular Dynamics
The ability to reliably prepare a single quantum state in the hydronium ion (H₃O⁺) has moved closer to reality, thanks to a novel application of artificial intelligence developed through a collaboration between UCLA, Caltech, and NVIDIA. Researchers have demonstrated a Fourier Neural Operator capable of designing control protocols for this polyatomic molecule, achieving a target-state population of 0.98 with a success rate of up to 86.2%, a significant improvement over existing methods.
This advance addresses a critical bottleneck in precision spectroscopy, where initial state preparation for complex molecules like hydronium has historically been the most challenging step. The team’s approach bypasses the need for computationally intensive simulations within the optimization loop by training the Fourier Neural Operator to act as a surrogate model for the molecule’s quantum dynamics.
Unlike traditional optimal-control methods that rely on repeated numerical propagation of dynamics, a process that scales poorly with system complexity, the FNO predicts the full population trajectory of the molecule given a laser pulse in a single forward pass. This efficiency stems from two key design choices: a physics-informed embedding that encodes laser frequencies as detunings from relevant molecular transitions, and the model’s ability to track molecular populations directly, using experimental measurement of a shared motional mode.
“We show that a Fourier Neural Operator can learn the quantum dynamics of a real molecule well enough to design control protocols for it, in a state space far too large for conventional optimization,” the researchers state in their published work. Hydronium was specifically chosen for this research due to the potential for its inversion transitions to reveal physics beyond the Standard Model, making precise control of its quantum state particularly valuable.
At 20 Kelvin, a trapped hydronium ion begins in a smeared distribution across hundreds of rotational and hyperfine levels, requiring a carefully orchestrated sequence of laser pulses to funnel it into a single, well-defined state. The difficulty lies in the high-dimensional Hilbert space that describes the molecule’s possible states, with each candidate pulse sequence demanding a full numerical simulation to evaluate its effectiveness. By replacing the simulator with the learned surrogate model, the team significantly reduced sequence-generation time from around 10 hours to 10-20 minutes.
This speedup is not merely a matter of convenience; it unlocks the possibility of exploring a much larger control space, leading to more effective and robust control protocols. Compared to a reinforcement-learning baseline in the same discrete control space, the FNO-SPMP method, the team’s implementation, roughly doubled the success rate, used about half as many pulses, and dramatically reduced computation time.
The underlying principle, according to the researchers, is that operator learning provides a fast and differentiable surrogate model, enabling both efficient search over the control space and refinement of the resulting sequences. NVIDIA’s involvement extends beyond providing computational infrastructure; the work is being highlighted at IEEE Quantum Week in Toronto as part of the company’s announcement of an open, programmable platform for designing and testing fault-tolerant quantum computing applications. NVIDIA’s commitment to quantum computing is evidenced by its CUDA Quantum platform and DGX Quantum systems, providing both services and infrastructure for the field.
Physics-Informed Embedding Accelerates Quantum Control Simulation
The resulting system, dubbed the FNO stochastic pulse-measurement planner (FNO-SPMP), significantly accelerates the process of finding effective control sequences. This speedup isn’t achieved at the expense of accuracy; the surrogate model maintains sufficient fidelity to enable effective control. NVIDIA played a key role in this advancement, not only by providing computational infrastructure but also by highlighting the work as part of its broader push into programmable, fault-tolerant quantum computing applications.
This collaboration extends beyond infrastructure, with NVIDIA actively promoting an open ecosystem for quantum software development. Recent developments, such as the NVQLink architecture launched in 2026, demonstrate NVIDIA’s commitment to integrating quantum processors with its GPU-accelerated computing platform, enabling hybrid quantum-classical workflows for simulation, optimization, and machine learning. The researchers believe the workflow, combining physics-informed operator learning with powerful computational resources, is broadly applicable to other molecular ions and quantum platforms where state spaces exceed the capabilities of traditional simulators.
This approach addresses a critical limitation of large language models, which often lack a fundamental understanding of the physical world. “AI models that learn the physics directly, rather than the language about it, can do things in the physical world that language models cannot,” the team explains.
The research, detailed in a paper on arXiv, acknowledges contributions from Anastasia Pipi, Valentin Duruisseaux, Emily Been, Xuecheng Tao, Taylor Patti, Anima Anandkumar, and the NVIDIA team responsible for CUDA-Q. This combination of advanced modeling techniques and robust computational infrastructure promises to enable new advances in quantum control and precision spectroscopy.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
