UCLA & Caltech use quantum-enhanced AI on NVIDIA GPUs to steer molecules

Researchers at UCLA’s NarangLab, Caltech, and NVIDIA have developed a Fourier Neural Operator capable of designing control sequences for molecules in state spaces previously inaccessible to conventional methods. This work focuses on hydronium (H₃O⁺) molecules, whose sensitive inversion transitions offer a potential pathway to probe for physics beyond the Standard Model. “We need a model that has actually learned the underlying physics,” explains the UCLA group, and this new platform is being highlighted at IEEE Quantum Week in Toronto as part of NVIDIA’s launch of an open, programmable platform for fault-tolerant quantum computing applications.

Fourier Neural Operator Learns Hydronium Molecular Dynamics Achieving a target-state population of 0.98, the team demonstrated successful state initialization and control in hydronium molecules using pulse sequences designed by a Fourier Neural Operator, with a success rate of up to 86.2 percent. This advance bypasses the computational bottleneck of repeatedly simulating molecular dynamics during optimization, a challenge that limits conventional methods for complex molecules. Rather than relying on exhaustive numerical propagation, the researchers trained the Fourier Neural Operator to act as a surrogate simulator, predicting population trajectories with a single forward pass. A key innovation was a physics-informed embedding, where laser frequencies are encoded as detunings from relevant molecular transitions. This approach, detailed in “Inverse Design of Quantum Control Sequences with Fourier Neural Operators,” allows the model to focus on physically plausible parameters, improving both speed and accuracy. Because experiments measure a shared motional mode, the planner tracks molecular populations directly, a task made tractable by the FNO’s predictive capabilities. The resulting pulse sequences are then refined with gradient descent and validated using Monte Carlo rollouts, ensuring robustness. At cryogenic temperatures of 20 Kelvin, a trapped hydronium ion initially occupies hundreds of rotational and hyperfine levels, demanding precise quantum control to funnel it into a single, well-defined quantum state. Navigating this high-dimensional Hilbert space traditionally requires substantial computational resources, but the Fourier Neural Operator significantly reduces these demands. The collaborative effort between UCLA’s NarangLab, Caltech, and NVIDIA used NVIDIA’s GPU-accelerated computing platform to enable this hybrid quantum-classical workflow. NVIDIA provides quantum computing services and infrastructure through its CUDA Quantum platform and DGX Quantum systems, and reported a quantum calibration model working across six qubit modalities on July 28, 2026. The company’s NVQLink, launched in 2026, facilitates low-latency integration of quantum processors with GPU supercomputers, as demonstrated by recent integration with Quantinuum and QuEra Computing. The integration of Quantum Machines OPX with NVIDIA H100 GPUs within the DGX Quantum system further exemplifies this commitment to accelerating quantum computing research. Physics-Informed Embedding Accelerates Quantum Control Simulation This embedding encodes laser frequencies as detunings from actual molecular transitions, a technique that significantly improves the efficiency of the Fourier Neural Operator (FNO) used to predict molecular population trajectories. The FNO, trained to simulate the quantum dynamics of molecules, then identifies optimal control protocols within state spaces previously inaccessible to conventional optimization methods.

The team reports cutting sequence-generation time from around 10 hours to 10-20 minutes using this new method, a substantial reduction enabled by the speed of the learned surrogate model. NVIDIA’s role extends beyond providing computational infrastructure; the company’s expertise in GPU-accelerated computing is integral to the scalability of the FNO. According to NVIDIA records, the company has 10 families of patents and 4 publications related to quantum computing in the last twelve months, demonstrating a sustained investment in the field. The researchers emphasize the broader implications of this workflow, noting that operator learning provides a fast, differentiable surrogate model that enables efficient searching of large control spaces and refinement of identified solutions. The work was presented at IEEE Quantum Week in Toronto, signaling its growing recognition within the quantum computing community. FNO-SPMP Achieves 86.2% Success in Molecular State Preparation Achieving an 86.2% success rate, the Fourier Neural Operator-State Preparation with Molecular Protocols (FNO-SPMP) demonstrates a significant advancement in directing molecules to specific quantum states. This level of accuracy, reported by a collaboration between UCLA’s NarangLab, Caltech, and NVIDIA, surpasses previous methods for controlling complex molecular systems. The approach bypasses limitations of conventional optimization techniques when applied to expansive state spaces, a critical hurdle in precision molecular spectroscopy. The core innovation lies in the FNO’s ability to learn the underlying quantum dynamics of molecules, enabling it to design control protocols without relying on trial-and-error methods. This learned understanding allows the system to efficiently navigate the numerous rotational and hyperfine levels present in molecules like hydronium (H₃O⁺) at cryogenic temperatures. Hydronium’s sensitivity to fundamental constant variations positions it as a potential tool for probing physics beyond the Standard Model, making precise state control particularly valuable. The FNO-SPMP system tackles the challenge of preparing a trapped hydronium ion for experimentation. Starting from a distribution across hundreds of quantum levels at 20 Kelvin, the system uses learned protocols to funnel the molecule into a single, well-defined state. The process involves tracking population trajectories and identifying optimal laser frequencies to manipulate the molecule’s quantum state. NVIDIA’s commitment to quantum computing is evidenced by its partnerships with Quantinuum, QuEra Computing, and IonQ, among others. The recent integrations with Anyon Computing, as announced on September 14, 2026, further solidify NVIDIA’s position as a key player in the development of programmable, fault-tolerant quantum applications.

The team’s success with hydronium suggests the approach could be extended to other molecular ions and quantum platforms. The ability to accurately prepare molecular states is a critical step toward harnessing the full potential of quantum systems for scientific discovery. Source: https://naranglab.ucla.edu/teaching-a-neural-operator-to-steer-a-molecule/ More like thisQuantum Machine LearningResearchers Map Hamiltonian Control to Classifier Output FeaturesQuantum Machine LearningDesign Choice Limits Impossible Predictions to Less Than One Percent of CrystalsQuantum ApplicationsTrust Base partners with OQC to explore quantum finance workflowsQuantum SecurityStarkWare funds $20,000 challenge for Quantum-Safe BitcoinStay currentSee today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: Ivy Delaney Ivy Delaney has been working with neural networks and machine learning since the mid-nineties, back when a couple of hidden layers and a long afternoon of training counted as ambitious. She has watched the field go from academic curiosity to the thing quietly running underneath everything, and she brings that long view to quantum computing.

For Quantum Zeitgeist she covers the ground where the two fields meet. That means quantum machine learning and the variational algorithms it leans on, and it also means the less glamorous but more interesting story of classical machine learning already doing real work inside quantum machines, decoding error-correcting codes, calibrating noisy hardware and learning the error models that simulators depend on. She writes about the hardware those algorithms have to run on too, and about the post-quantum cryptography scramble that the same hardware has set off. Her stories typically start with the paper, whether that is peer-reviewed work, conference proceedings or an arXiv preprint, with the source linked so you can hold a claim up against the research it came from. She is unimpressed by benchmarks that will not say what they beat, and by demonstrations that only work in the press release.