Nuclear Fusion May Have Found a Decisive Ally: Quantum Computing

The availability of tritium represents one of the main obstacles to the commercial development of nuclear fusion. To address this issue, a group of researchers from Oak Ridge National Laboratory (ORNL), Cleveland Clinic, and IBM has developed a new method to study the behavior of molten salt FLiBe (lithium fluoride and bertrandium), considered one of the most promising candidates for fuel production inside future reactors.

To do this, and this is the important aspect, a hybrid supercomputing system, or as some prefer, "quantum-centric supercomputing," has been used, which combines quantum computers and traditional computing systems. The study, published as a preprint on arXiv, is described by the authors as the first known demonstration of using quantum computers to calculate the molecular configurations of this material and opens the door to more accurate simulations of the chemical processes occurring in the blankets of fusion reactors.

In a future tokamak, the neutrons released by the plasma during the fusion could bombard the surrounding molten salt blanket to generate tritium. The study uses quantum computers to model the interaction between tritium and a group of atoms present in the molten salt. FLiBe plays a central role in tokamak-type reactor designs. Besides cooling the system and shielding the magnets from neutron bombardment, the liquid salt has the task of producing tritium. When the neutrons generated by the fusion reaction hit the lithium-6 atoms present in the material, helium and new tritium are produced, theoretically allowing the reactor to sustain its fuel needs.

The problem is that tritium is extremely rare in nature. Current fission plants produce only a few pounds per year worldwide, while a single 1 GW fusion reactor could consume about a pound per day. Consequently, the ability to continuously produce and recover tritium within the reactor represents one of the greatest engineering challenges in the sector. Understanding the atomic behavior of tritium in FLiBe requires extremely precise simulations. If the isotope binds to the fluorine atoms forming tritium fluoride, it actually becomes more difficult to extract and can increase corrosive phenomena; if it remains in a gaseous state, it can be more easily recovered. However, predicting which of these scenarios prevails exceeds the capabilities of traditional simulation techniques based solely on classical computing.

For the first time, a quantum computer has simulated the properties of a real magnetic material. For this reason, the research group has adopted a hybrid approach where CPUs, GPUs, and quantum processors collaborate within the same workflow. Classical computers handle the less complex parts of the computation, while quantum ones tackle the problems related to the electronic structure of molecules, particularly suited to the quantum nature of the interactions between electrons.

The researchers analyzed nine different molecular configurations of FLiBe, each made up of clusters of 21 ions, calculating their energy both in the presence and absence of tritium. The results obtained were consistent with the best classical methods available for solving these molecular fragments, providing an initial validation of the approach. This work also extends techniques developed previously to simulate proteins containing over 12,600 atoms. In that case, a wave function-based embedding (EWF) methodology was employed, which breaks down molecules into smaller clusters: the less demanding parts are processed by traditional supercomputers, while those characterized by greater electronic entanglement are assigned to quantum processors using the Sample-based Quantum Diagonalization (SQD) technique.

According to Tom Beck, head of Science Engagement at ORNL's National Center for Computational Sciences, the progress has come much faster than expected. The ultimate goal of the project, part of the U.S. Department of Energy's Genesis Mission, is to build a platform where artificial intelligence, supercomputing, and quantum computing collaborate to accelerate the discovery of new materials intended for fusion reactors. In this vision, AI agents select the most promising candidates from decades of experimental data on molten salts, supercomputers run preliminary simulations of their atomic structure, and quantum computers intervene in areas where classical methods are not sufficiently accurate. The results are then reintegrated into the system to iteratively refine subsequent simulations.

IBM brings quantum computing into real chemistry: the "half-Moebius" molecule has been created and simulated. The work remains a proof of concept. Simulating the actual behavior of a blanket approximately one meter thick, made up of a practically limitless number of particles, is still well beyond the capabilities of current computing systems. The next step will be to progressively increase the size of the molecular clusters analyzed and the number of simulated configurations, while reducing the data transfer times between quantum computers and HPC infrastructures. The long-term goal is to provide engineers with tools capable of designing and virtually validating new materials even before their actual realization in the lab, reducing costs and experimental times.