An AI model informed by calculations from a quantum computer can better predict the behaviour of a complex physical system over the long term than current best models that use only conventional computers, according to a new study led by UCL (University College London) researchers.
The findings, published in the journal Science Advances, could improve models predicting how liquids and gases move and interact (fluid dynamics), used in areas ranging from climate science to transport, medicine and energy generation.
The researchers say the improved performance is linked to a quantum device's ability to hold a large amount of information more efficiently. That is because instead of bits that are switched on or off, 1 or 0, like in a classical computer, the quantum computer's qubits can be 1, 0, or any state in between, and each qubit can affect any of the other qubits – meaning a few qubits can generate a vast number of possible states.
Senior author Professor Peter Coveney, based in UCL Chemistry and the Advanced Research Computing Centre at UCL, said: "To make predictions about complex systems, we can either run a full simulation, which might take weeks – often too long to be useful – or we can use an AI model which is quicker but more unreliable over longer time scales.
"Our quantum-informed AI model means we could provide more accurate predictions quickly. Making predictions about fluid flow and turbulence is a fundamental science challenge but it also has many applications. Our method can be used in climate forecasting, in modelling blood flow and the interaction of molecules, or to better design wind farms so they generate more energy."
Quantum computers have the potential to be far more powerful than even the fastest conventional computers but so far their practical usefulness has been limited.
To make predictions about a complex system, an AI model is trained on large amounts of simulation or observational data. In the quantum-informed method, this data is first fed to a quantum computer which learns the key statistical patterns of the data, or the invariant statistical properties, i.e. the patterns that stay the same over time. These quantum-learned patterns are then incorporated into the training of the AI model on a conventional supercomputer.
The quantum-informed method was about a fifth more accurate and remained stable over the long term in its predictions of how a complex, chaotic system would behave compared to the AI model that did not use the quantum-learned patterns, and it was much more efficient, requiring hundreds of times less memory.
This efficiency is a result of two quantum properties. One, entanglement, is where each qubit can affect any of the others, regardless of distance. The other, superposition, means a qubit can exist in different classical states simultaneously, until it is measured. These properties mean a quantum computer with only a few qubits still has massive computing power.
First author Maida Wang, based in the UCL Centre for Computational Science, said: "Our new method appears to demonstrate 'quantum advantage' in a practical way – that is, the quantum computer outperforms what is possible through classical computing alone. These findings could inspire the development of novel classical approaches that achieve even higher accuracy, though they would likely lack the remarkable data compression and parameter efficiency offered by our method. The next steps are to scale up the method using larger datasets and to apply it to real-world situations which typically involve even more complexity. In addition, a provable theoretical framework will be proposed."
The other first author, Xiao Xue, based at Advanced Research Computing at UCL, said: "In this work, we demonstrate for the first time that quantum computing can be meaningfully integrated with classical machine learning methods to tackle complex dynamical systems, including fluid mechanics. It is exciting to see this kind of 'quantum-informed' approach moving towards practical use."
The researchers suggested the quantum computer's ability to compactly capture the underlying physics of such complex systems may be because of the "quantum-like" chaos of these systems, with a movement in one part of the system affecting another part of the system far away (much like entanglement).
Their method, they said, avoids the limitations of current quantum computers – which are very noisy, prone to errors and interference, hence requiring too many measurements to be made – by only using a quantum device at just one stage of the process, rather than flitting data back and forth between classical and quantum systems.
The study used a 20-qubit IQM quantum computer linked to conventional supercomputing resources at the Leibniz Supercomputing Centre in Germany.
To achieve their quantum state, quantum computers are cooled to minus 273C (close to absolute zero, colder than anything in space).
The researchers received funding from UCL and the UK's Engineering and Physical Sciences Research Council (EPSRC) and support from IQM Quantum Computers and the Leibniz Supercomputing Centre in Munich.
