Variational quantum algorithms for climate and weather applications
Meng
Wang
University of British Columbia
Talk
Climate and weather prediction workflows increasingly depend on large-scale optimization and variational procedures, including data assimilation, observing-system design, sensor placement, ensemble selection, resource allocation, and the calibration of reduced-order or surrogate models. Variational Quantum Algorithms (VQAs), such as the Quantum Approximate Optimization Algorithm (QAOA) and the Variational Quantum Eigensolver (VQE), offer a near-term model for exploring whether quantum processors can support these workloads. In a VQA, a parameterized quantum circuit estimates an objective function while a classical optimizer updates the circuit parameters, a structure that resembles familiar variational methods in the geosciences.
However, practical VQAs face two major barriers before they can contribute to climate and weather applications. First, hardware noise can distort the objective landscape seen by the optimizer, especially as larger problems require more qubits and deeper circuits. Second, VQAs impose a heavy execution burden: a single optimization may require hundreds or thousands of circuit evaluations, often with multiple restarts, on shared quantum devices with uneven queue times and error rates.
This talk presents two complementary approaches for making VQAs more robust and efficient. Red-QAOA reduces the cost of QAOA parameter search by constructing a smaller distilled graph whose optimization landscape approximates that of the original problem. Parameters are trained on this lower-noise surrogate and then transferred back to the full instance for refinement, preserving the original problem while improving search stability. Qoncord addresses the systems side by scheduling different VQA stages across devices with different fidelity and queue characteristics. Early exploratory iterations and restart screening run on lower-load devices, while final exploitation is reserved for higher-fidelity hardware.
Together, Red-QAOA and Qoncord address algorithmic and execution bottlenecks that limit near-term VQAs. By improving noise resilience, time-to-solution, and hardware utilization, these techniques move VQAs closer to becoming useful components in future climate and weather computing workflows.
However, practical VQAs face two major barriers before they can contribute to climate and weather applications. First, hardware noise can distort the objective landscape seen by the optimizer, especially as larger problems require more qubits and deeper circuits. Second, VQAs impose a heavy execution burden: a single optimization may require hundreds or thousands of circuit evaluations, often with multiple restarts, on shared quantum devices with uneven queue times and error rates.
This talk presents two complementary approaches for making VQAs more robust and efficient. Red-QAOA reduces the cost of QAOA parameter search by constructing a smaller distilled graph whose optimization landscape approximates that of the original problem. Parameters are trained on this lower-noise surrogate and then transferred back to the full instance for refinement, preserving the original problem while improving search stability. Qoncord addresses the systems side by scheduling different VQA stages across devices with different fidelity and queue characteristics. Early exploratory iterations and restart screening run on lower-load devices, while final exploitation is reserved for higher-fidelity hardware.
Together, Red-QAOA and Qoncord address algorithmic and execution bottlenecks that limit near-term VQAs. By improving noise resilience, time-to-solution, and hardware utilization, these techniques move VQAs closer to becoming useful components in future climate and weather computing workflows.