Researchers from Oak Ridge National Laboratory, IBM, and Cleveland Clinic have shown that a workflow integrating quantum computers with high-performance computing and artificial intelligence can simulate chemical reactions in molten salt used in future fusion reactor designs. The study was published as a preprint on arXiv on June 29, while IBM Quantum published details of it on July 6, 2026.
The study focuses on a fundamental problem in fusion reactors: supplying tritium, a rare radioactive isotope of hydrogen required for fusion reactions with deuterium. There are no significant natural sources of tritium on Earth, while the entire world produces only a few pounds of it annually. According to the article, a one-gigawatt fusion reactor could consume about one pound of tritium per day, meaning that any future plant would need to produce its own fuel during operation.
The Role of Molten Salt in a Fusion Reactor
One design proposes surrounding the hot plasma in a tokamak reactor with a thick blanket of molten lithium salt. When fusion-generated neutrons collide with lithium-6 atoms, the atoms split to produce helium and tritium. Beryllium helps multiply the neutrons, while the fluorine-and-lithium mixture remains liquid and stable at reactor temperatures.
The blanket’s role is not limited to producing fuel; it is also intended to protect the reactor’s magnets from neutrons, cool the plasma-facing wall, and transfer heat to operate a turbine. However, neutron bombardment changes the salt’s chemistry, and the way tritium bonds with its components affects how easily it can be extracted. If it bonds with fluorine, tritium fluoride forms, which is corrosive and difficult to remove, whereas if it remains a free gas, it may leave the salt on its own.
Why Does the Simulation Need Quantum Computing?
Researchers typically use density functional theory to simulate the arrangement of electrons on conventional computers. This method remains fast and useful in many cases, but Tom Beck’s team at Oak Ridge National Laboratory found in previous work that it could miscalculate the salt’s free energy by as much as 10%. This level of accuracy is insufficient to predict whether tritium will become tritium fluoride or remain a gas.
To address this, the team used a method called wavefunction-based embedding, in which the problem is divided into smaller clusters. Conventional computers handle the simpler parts, while quantum computing uses a sampling-based quantum eigensolver method for the clusters with more complex interatomic entanglement; the results are then recombined on conventional computers.
The researchers extracted nine configurations from FLiBe salt, with each configuration containing a small cluster of 21 ions. Its energy was calculated with and without tritium, and the results were compared with advanced classical methods for calculating the chemical components. The quantum-computing calculations agreed with those methods, providing an initial proof that the approach is viable.
A Workflow Combining Artificial Intelligence and Quantum Computing
The researchers want to develop a three-stage computational loop:
- Artificial intelligence agents propose candidate salts and screen them using a database at Oak Ridge National Laboratory containing 70 years of molten-salt research. The screening includes estimating the tritium production ratio and the salt’s ability to remain liquid and transfer heat.
- High-performance computers simulate the best candidates atom by atom using density functional theory, with the help of artificial intelligence models trained to reproduce physics in order to accelerate computationally expensive processes.
- Quantum computing performs the high-accuracy chemical calculations that density functional theory cannot resolve, particularly determining how tritium bonds with the salt.
Next Steps
The researchers acknowledge that the result remains limited. Fully solving the free-energy problem requires studying a moving salt blanket one meter thick and containing an enormous number of particles, a scale beyond the capabilities of computational chemistry for the foreseeable future. The team also plans to increase the cluster size beyond 21 ions and run hundreds of configurations instead of nine.
Oak Ridge National Laboratory researchers hope these tools will enable fusion engineers to design and evaluate molten salts computationally before mixing and heating them in the laboratory. The same methodology may later extend to other chemical problems, but the current study does not yet demonstrate that it can be used to design a practical fusion reactor or produce tritium on an industrial scale.