For the first time, a quantum computer has calculated the molecular configurations of FLiBe, the molten salt composed of fluorine, lithium, and beryllium that is the leading candidate for tritium breeding inside fusion reactors. The result comes from a joint team at IBM, Oak Ridge National Laboratory, and Cleveland Clinic, with the findings published on arXiv on July 6, 2026.
The team calculated nine molecular configurations of FLiBe clusters, with and without tritium, using quantum-centric supercomputing techniques that combine quantum processors with classical computers. Results were benchmarked against classical reference methods, with the two approaches aligning to a mean absolute error of 0.3 kilocalories per mole. This hybrid approach enabled the determination of the material’s electronic structure and the measurement of how tightly each configuration binds tritium at the molecular level—revealing properties that classical methods alone struggle to capture with comparable precision.
Tritium is an isotope of hydrogen that is extremely rare in nature, yet essential for powering nearly all fusion reactors currently in development. Future plants will need to breed their own supply by passing neutrons emitted from the plasma through a surrounding blanket of molten salt. FLiBe is among the most studied materials for this purpose, but optimizing its composition is far from straightforward: the salt operates under intense neutron irradiation, extreme temperatures, and powerful magnetic fields, with its chemical structure constantly evolving. Until now, research into these properties has relied on costly experiments or classical numerical approximations that can lose accuracy precisely where quantum-level detail matters most.
The work is part of the U.S. Department of Energy’s Genesis Mission, which aims to integrate supercomputing, artificial intelligence, and quantum computing across all 17 national laboratories to tackle strategic scientific challenges. IBM is contributing as an industry partner, bringing its quantum architectures to work alongside CPUs, GPUs, and QPUs. The same computational approach had previously been applied to the simulation of proteins comprising more than 12,000 atoms in collaboration with Cleveland Clinic—its extension from biology to fusion materials chemistry underscores the broad applicability of this methodology.
The researchers are candid about current limitations: the full problem requires modeling a molten salt blanket roughly one meter thick, containing a number of particles on the order of a trillion trillion. That scale will remain beyond the reach of any computational approach for a very long time. But the direction is clear: progressively scaling up the number of atoms that can be simulated, sharpening the ability to predict FLiBe behavior under real reactor conditions. Every step in that direction reduces reliance on physical experiments—which are slow and expensive—and brings fusion closer to a genuinely viable fuel cycle.
