Insider Brief
- In an essay, researchers from IonQ and Microsoft suggest that quantum computers could help generate highly accurate data that can train artificial intelligence models to simulate chemical systems more efficiently.
- The approach would combine the precision of quantum simulations with the speed of AI models running on classical computers, potentially accelerating materials and drug discovery.
- The researchers say quantum-generated training data could improve predictions of electron behavior in molecules, helping scientists design catalysts, batteries and other materials while large-scale fault-tolerant quantum computers are still under development.
- Photo by geralt on Pixabay
Two researchers, now from rival quantum computing companies, have teamed up to propose that quantum technology’s most practical near-term role may be training artificial intelligence (AI) systems to simulate chemistry, a strategy they propose could accelerate the discovery of new materials and drugs long before large-scale quantum machines arrive.
In an essay published in IEEE Spectrum, Chi Chen, of IonQ, and Matthias Troyer, of Microsoft, report on a hybrid approach that combines quantum computing with artificial intelligence to overcome a longstanding bottleneck in chemistry simulations. Their proposal suggests quantum computers could generate highly accurate data about the behavior of electrons in molecules, which would then be used to train AI models capable of making fast predictions on conventional computers.
The idea highlights a rare alignment between competitors in the emerging quantum industry. IonQ and Microsoft are developing different quantum computing systems and business strategies, yet the researchers write that both fields — quantum computing and artificial intelligence — will likely advance faster together than separately.
At the center of the proposal is a concept familiar to computational chemists called “Jacob’s Ladder,” a metaphor introduced in 2001 by physicist John P. Perdew to describe how scientists simulate the behavior of electrons in materials.
Understanding the Electron Problem
Jacob’s Ladder represents a hierarchy of computational methods where each rung on the ladder represents a different level of approximation and computational effort.
At the lowest levels, scientists simplify the problem dramatically. Atoms may be treated as simple particles connected by springlike forces, allowing simulations to track millions of atoms over long time periods. These calculations are relatively fast but only provide rough descriptions of atomic behavior.
Higher levels of the ladder incorporate more detailed physics. Methods such as density functional theory — often called DFT — attempt to describe the quantum behavior of electrons more accurately by approximating how they interact with each other. DFT is widely used in chemistry and materials science but still relies on mathematical shortcuts that limit accuracy in certain cases.
At the top of the ladder are the most precise calculations, which attempt to model the interactions among electrons exactly. These calculations quickly become computationally overwhelming. The number of possible electron arrangements grows exponentially as molecules get larger, creating what researchers often call an “exponential wall” in computing complexity.
According to the IEEE Spectrum essay, classical computers struggle to cross that barrier for many chemically important systems, particularly those involving strongly interacting electrons such as catalysts or materials with unusual electronic properties.
Quantum computers could help overcome this limitation because their basic units of information, called qubits, can theoretically exist in multiple states at once. This property allows them, in principle, to represent many possible electron configurations simultaneously, making them well suited for simulating quantum systems such as molecules.
Yet practical quantum computers remain limited. Current devices contain relatively small numbers of qubits and suffer from errors that accumulate during calculations. Fully fault-tolerant machines capable of handling large chemical simulations may require millions of physical qubits and remain years away.
Combining Quantum Accuracy With AI Speed
Chen and Troyer suggest that quantum computers do not need to run every chemical simulation themselves to be useful. Instead, they could be used to generate small amounts of extremely accurate data about electron behavior, data that would be prohibitively expensive to compute using classical methods.
That data could then train machine learning models running on classical computers. Once trained, the models could predict chemical properties with far greater speed than traditional simulations.
The approach mirrors a growing trend in science and engineering where AI models act as surrogates for complex simulations. Instead of solving difficult equations every time, a trained model can estimate results almost instantly.
The researchers describe how this hybrid approach could effectively bend Perdew’s “Jacob’s Ladder.” Rather than climbing step by step toward more computationally demanding methods, scientists could use quantum-generated data to train AI systems that deliver high levels of accuracy without the same computing cost.
The strategy could allow scientists to reach higher levels of predictive accuracy in chemical modeling while still running calculations on ordinary computers.
Testing the Concept in Materials Discovery
The essay points to recent work by Microsoft and researchers at Pacific Northwest National Laboratory as an example of how AI can already accelerate materials research.
In that project, AI models evaluated more than 32 million candidate materials for potential use in battery electrolytes. Traditional computational methods might have required decades to analyze such a large set of materials.
According to the researchers, the AI screening process narrowed the pool to about 500,000 stable materials and eventually to roughly 800 promising candidates within less than a week. High-performance computing systems then performed more detailed simulations on the most promising options.
Researchers at Pacific Northwest National Laboratory later synthesized and tested one of the candidate materials, a solid-state electrolyte that used sodium and significantly reduced the amount of lithium required compared with conventional lithium-ion batteries.
The example illustrates how machine learning can dramatically speed up the early stages of materials discovery by narrowing the list of candidates that must be tested experimentally.
Quantum computing could extend that approach by improving the accuracy of the AI models themselves. According to the researchers, AI systems trained on quantum simulation data could capture complex electron interactions that classical methods struggle to represent.
Accurate modeling of electron behavior is critical for predicting chemical reactions and designing new materials. Small errors in energy calculations can lead scientists to the wrong conclusions about how molecules behave.
This is particularly important when studying reaction pathways — the series of steps that occur when chemicals transform into new substances. Researchers often describe these pathways as a landscape of hills and valleys, where the height of each hill represents an energy barrier that a reaction must overcome.
Even small inaccuracies in calculating those barriers can change predictions about which products will form or how quickly reactions occur.
Improved modeling tools could therefore influence a wide range of industries, including pharmaceuticals, energy storage and environmental chemistry. More accurate simulations could help scientists identify better catalysts, design more efficient batteries or discover chemical reactions capable of breaking down persistent pollutants.
The researchers write that hybrid quantum-AI systems could significantly improve the reliability of these predictions by combining quantum-level accuracy with the speed of machine learning.
Remaining Technical Hurdles
Despite the potential benefits, several technical challenges remain.
Quantum computers must become significantly larger and more reliable before they can perform the most demanding chemistry simulations. According to the researchers, useful simulations beyond the reach of classical computers may require hundreds or thousands of high-quality qubits with extremely low error rates.
Achieving that reliability will likely require fault-tolerant quantum computing, in which quantum information is protected through error-correcting codes. Each logical qubit may require hundreds of physical qubits, meaning full-scale machines could require millions of qubits.
Even with those challenges, the researchers suggest that the hybrid approach could begin to deliver benefits sooner. AI models already trained on classical simulation data may only need to be fine-tuned with smaller amounts of quantum-generated data to improve their accuracy.
That incremental strategy could allow scientists to integrate quantum computing into scientific workflows gradually as the hardware improves.
If successful, the combination of quantum computing and artificial intelligence could transform how scientists study matter at the atomic scale, making high-precision chemical modeling accessible on everyday computers rather than specialized supercomputers.
