Quantum computing has garnered a lot of attention, and with good reason. Futuristic computers aim to mimic what happens in nature on a microscopic scale to better understand its quantum world and develop new materials, including medicines, eco-friendly chemicals and more. has the potential to accelerate discovery. However, experts say existing quantum computers are still a decade away. What should researchers do in the meantime?
A new Caltech-led study in Science describes how machine learning tools running on classical computers can be used to predict quantum systems and help researchers solve some of the toughest problems in physics and chemistry. Although this concept has already been proven experimentally, this new report is the first to demonstrate mathematically that this method works.
"Quantum computers are ideal for many problems in physics and materials science," said lead author Xin-Yuan (Robert) Huang, John Preskill Professor of Theoretical Physics, Richard P. Feynman and Allen WC Davis graduate student. Lenabel Davies, Executive President of the Institute for Quantum Science and Technology (IQIM). “But we haven't gotten that far yet, and we were surprised to learn that classic machine learning methods can still be used for a while. Ultimately, this work is about showing what humans can learn about the physical world.”
At the microscopic level, the physical world is becoming a very complex place governed by the laws of quantum physics. In this world, particles can be in overlapping states or in two states at the same time. And overlapping situations can lead to entanglement, a phenomenon where particles connect or come together without touching. It is very difficult to describe mathematically these strange situations and relationships that are characteristic of both natural and artificial materials.
"Predicting the low-energy state of a material is very difficult," Huang said. "There are many atoms, and they pile up and get entangled. You can't write equations to describe them all."
This new research is the first mathematical proof that classical machine learning can be used to bridge the gap between us and the quantum world. Machine learning is a computer program that mimics the human brain to learn from data.
"We are classical beings living in a quantum world," says Preskill. "Our brains and computers are classical, which limits our ability to interact with and understand quantum reality."
While previous research has shown that machine learning programs are capable of solving some quantum problems, these techniques typically work in ways that make it difficult for researchers to study how machines find solutions.
“Typically, with machine learning, you don’t know how machines solve problems. It's a black box," Huang said. "But now, with our digital simulations, we know what's going on inside the box." Huang and his colleagues worked with Caltech's AWS Quantum Computing Center to run extensive numerical simulations and validate their theoretical results.
The new research will help scientists better understand and classify the complex and exotic phases of quantum matter.
"The concern is that if a person creates a new quantum state in the lab, they won't be able to understand it," Preskill explained. "But now we can get classic data that makes sense to explain what's going on. Classic machines don't answer us like oracles, they lead us to a deeper understanding.”
Co-author Victor W. Albert, a physicist at the National Institute of Standards and Technology (NIST) and a former DuBridge Award-winning postdoctoral fellow at Caltech, agrees. "What excites me most about this work is that we are now closer to a tool that will help us understand the fundamental stages of a quantum state without requiring us to know much about the previous state."
According to scientists, future quantum learning tools will eventually surpass classical methods. In a related study published in Science on June 10, 2022, Huang, Preskill and their colleagues report using Google's Sycamore processor, the first quantum computer, to show that quantum machine learning outperforms classical approaches.
"We're still in the early stages of this field," Huang said. "But we know that quantum machine learning will ultimately be the most effective."