Quantum Learning Replicates Complex Dynamics with Two New Circuit Designs

Antonio David Bastida Zamora and colleagues at Aix-Marseille University, France are enhancing the quantum lattice Boltzmann method, a technique for simulating complex fluid dynamics, by using quantum machine learning. Their research approximates a key, nonlinear component of the method’s collision operator with a trained variational quantum circuit. By constructing an operator capable of replicating the dynamics of the Bhatnagar-Gross-Krook approximation, the team presents two distinct circuit architectures, the R1 and R2 models, offering flexible options for both continuous, multi-step simulations and high-precision, single-step reconstructions of the nonlinear operator. This provides a sharp improvement to the efficiency and accuracy of fluid dynamics modelling.

Variational quantum circuits enable extended simulations of fluid dynamics

Multistep quantum lattice Boltzmann method (QLBM) simulations scaling to O(log2 N) are now possible, overcoming previous limitations in capturing nonlinear dynamics without constant quantum state measurement. Previously, quantum algorithms required frequent interruption of the simulation process to perform quantum state measurements, a procedure known as ‘state tomography’. This necessity severely restricted the ability to model fluid behaviour over extended periods, as each measurement collapses the quantum state and introduces decoherence. The new variational quantum circuit (VQC) bypasses this need by learning to predict fluid dynamics using the Bhatnagar-Gross-Krook approximation, effectively embedding the collision dynamics within the quantum circuit itself. This allows the simulation to proceed without repeated measurements, enabling longer and more complex simulations.

R1 and R2, two distinct circuit architectures, were developed to address varying simulation demands, with the former prioritising continuous evolution and the latter focusing on single-step precision. The R1 model is designed for simulations requiring numerous time steps, prioritising efficient propagation of the quantum state over extended periods. It achieves this through a shallower circuit depth, reducing the accumulation of errors inherent in quantum computations. Conversely, the R2 model prioritises accuracy in each individual time step, employing a deeper and more complex circuit to more faithfully represent the nonlinear collision operator. Current demonstrations remain limited to relatively small systems, and scaling to industrially relevant problem sizes with millions of lattice points still presents a substantial engineering challenge. This advance builds upon the foundations of the quantum lattice Boltzmann method, achieving multistep simulations previously limited to single iterations. The trained variational quantum circuit accurately mimics fluid behaviour, and the two circuit designs, R1 and R2, offer differing strengths. R1 excels at modelling continuous evolution over time, while R2 prioritises accuracy for individual simulation steps. Successful replication of nonlinear collision dynamics, key for modelling complex fluids, demonstrates the potential for scaling with the number of lattice sites as O(log2 N), opening possibilities for more efficient computational fluid dynamics. The logarithmic scaling is particularly significant, as it suggests that the computational cost of the simulation grows much more slowly with system size compared to classical methods, which typically scale polynomially.

Quantum machine learning tackles fluid simulation with initial simplifications

Fluid dynamics underpin countless processes, ranging from weather forecasting and climate modelling to designing more efficient aircraft and optimising industrial processes. Accurately modelling these flows remains a persistent computational challenge. Traditional computational fluid dynamics (CFD) methods often rely on discretising the fluid domain into a vast number of grid points, leading to extremely high computational costs, particularly for turbulent flows. Employing machine learning now offers a potential quantum shortcut, sidestepping the difficulties of simulating complex fluid behaviour on conventional computers. The approach relies on the Bhatnagar-Gross-Krook approximation, a simplification of the full Boltzmann equation, which raises an important question regarding how readily these quantum circuits will generalise to more realistic, and computationally demanding, fluid models lacking this convenient simplification. The Bhatnagar-Gross-Krook model introduces a relaxation time, τ, which governs the rate at which the fluid returns to equilibrium after a disturbance. While simplifying the calculations, this approximation may limit the accuracy of the simulation for certain fluid behaviours.

Despite the current reliance on the Bhatnagar-Gross-Krook approximation, a simplification of real-world fluid dynamics, this represents a major step forward. It provides a pathway for potentially accelerating simulations, and future work will focus on refining the models to handle more complex scenarios without this initial simplification. The ultimate goal is to develop quantum machine learning models capable of accurately simulating fluid dynamics without relying on such approximations, potentially unlocking new insights into complex fluid phenomena. A functioning quantum machine learning approach to fluid dynamics simulation has been established, bypassing limitations of prior algorithms. Multi-step simulations were achieved by training a programmable quantum computer to replicate the behaviour of fluids, without constant measurement of the quantum system. This offers flexibility through the distinct circuit designs of R1 and R2. The training process involves adjusting the parameters of the variational quantum circuit to minimise a cost function that quantifies the difference between the predicted fluid behaviour and the expected behaviour based on the Bhatnagar-Gross-Krook approximation. Further exploration of more efficient computational fluid dynamics is now possible due to the successful replication of nonlinear collision dynamics, a vital element in modelling complex fluids. The ability to accurately model these dynamics is crucial for simulating a wide range of phenomena, including turbulence, shock waves, and multiphase flows. The research opens avenues for investigating the potential of quantum computing to address longstanding challenges in fluid dynamics and related fields.

The researchers successfully demonstrated a quantum machine learning approach to simulate fluid dynamics, replicating nonlinear collision dynamics using a variational quantum circuit. This is significant because current fluid simulations often rely on approximations like the Bhatnagar-Gross-Krook model, which can limit accuracy. Two distinct circuit architectures, R1 and R2, were developed to offer flexibility in simulation approaches, with R1 enabling multi-step simulations and R2 focusing on single-step precision. The authors intend to refine these models to remove the need for the initial Bhatnagar-Gross-Krook simplification, potentially improving the accuracy of complex fluid simulations.