Argonne National Laboratory is using its exascale supercomputers and artificial intelligence to simulate how carbon behaves under extreme heat and pressure, with the goal of designing entirely new nanodiamond and post-diamond materials. The work, announced in mid-March 2026, pairs billion-atom molecular dynamics simulations with AI models trained to predict carbon phase transitions, a combination that could accelerate the discovery of materials stronger than steel and lighter than aluminum. The effort raises a sharp question for materials science: can computational predictions at this scale actually translate into real-world synthesis, or does a gap remain between simulated promise and laboratory proof?
Why Carbon Demands Exascale Computing
Carbon is one of nature’s most versatile elements, capable of forming structures ranging from soft graphite to the hardest known natural material, diamond. That versatility also makes carbon difficult to model. Predicting how trillions of atoms rearrange under pressures found deep inside planets or during explosive detonations requires computing power that simply did not exist a few years ago. The Carbon at Extremes initiative, funded through the INCITE program at Argonne’s Leadership Computing Facility, is running quantum-accurate, billion-atom molecular dynamics simulations on both the Frontier and Aurora exascale systems to map these transformations.
The specific target that makes this project unusual is BC8, a post-diamond phase of carbon predicted to be even harder than diamond itself. Diamond exhibits extreme metastability, meaning it resists transformation into other carbon phases under most conditions. Simulations described in a preprint on the diamond-to-BC8 transition have identified a narrow pressure-temperature acceptance window where the transformation can occur and proposed a double-shock pathway that could guide experimental synthesis. Without exascale resources, scanning that narrow window across billions of atoms would be computationally impossible, and subtle nucleation events would remain hidden in smaller-scale models.
The AI Model Behind Billion-Atom Accuracy
Raw computing power alone does not solve the problem. Traditional force fields used in molecular dynamics are fast but approximate, while full quantum mechanical calculations are accurate but too slow for large systems. The bridge between these two limits is Allegro-FM, an E(3)-equivariant foundation model designed specifically for atomistic simulations at exascale. The model preserves the symmetry properties of physical systems, which means it can generalize across different carbon configurations without retraining from scratch for each new arrangement.
Allegro-FM scales to multi-billion-atom systems on Argonne’s Aurora, according to the preprint describing the model. That scaling matters because carbon phase transitions are collective phenomena. A handful of atoms switching from one arrangement to another tells researchers almost nothing about whether a bulk material will actually form. Billion-atom simulations capture the grain boundaries, defects, and shock-wave propagation that determine whether a new phase is stable enough to exist outside a computer. By combining quantum-level accuracy with classical-scale system sizes, the model allows researchers to explore complex transformation pathways that would otherwise be inaccessible.
Experimental Roots in Detonation Nanodiamonds
The computational work does not exist in a vacuum. It builds on decades of experimental research into detonation nanodiamonds, tiny diamond particles that form inside the fireball of a high-explosive blast in less than a microsecond. Researchers at Argonne and collaborating institutions have used time-resolved small-angle X-ray scattering during detonation events, combined with follow-up SAXS and TEM analysis, to characterize how these nanodiamonds nucleate and grow. That experimental data provides the ground truth against which AI-generated predictions can be tested, anchoring the simulations in measured particle sizes, growth rates, and phase fractions.
The tension between simulation and experiment is real. Computational models can now predict that a double-shock compression pathway should produce BC8 carbon, but no laboratory has yet confirmed that synthesis. The predicted pressure-temperature window is narrow enough that even small experimental uncertainties in shock timing, sample purity, or diagnostic calibration could prevent researchers from hitting it. This gap between computational design and physical validation is the central challenge the field faces, and it is one that exascale AI alone cannot close. Instead, the simulations are beginning to function as detailed roadmaps for experimental campaigns, indicating not just which conditions to try, but which diagnostics will be most sensitive to the fleeting formation of new phases.
Aurora’s Role in Accelerating Discovery
Argonne’s Aurora supercomputer, an exascale system opened to the research community for simulation, AI, and data analysis, is the hardware backbone of this effort. Its architecture is designed to handle the mixed workloads that modern materials science demands: running molecular dynamics on one set of nodes while training machine learning models on another, then feeding results back into the next round of simulations. High-bandwidth interconnects and fast storage are crucial, because each large-scale trajectory can generate petabytes of data that must be distilled into usable physical insight.
The carbon project is not the only materials discovery effort benefiting from this infrastructure. Argonne scientists have separately used AI workflows on Aurora to screen vast chemical spaces for carbon capture materials, demonstrating that the same exascale-plus-AI approach can be applied across very different materials challenges. The common thread is scale: problems that once required years of trial-and-error laboratory work can now be narrowed to a manageable set of candidates in weeks of computation. For carbon under extremes, that means moving from a nearly infinite landscape of possible pressures, temperatures, and shock profiles to a handful of promising synthesis routes.
In parallel, Argonne researchers have emphasized that these workflows depend on a tight feedback loop between simulation and data-driven models. As described in an Argonne feature on designing carbon materials, AI tools are trained not just on experimental measurements but also on the massive simulation datasets produced on exascale systems. That co-training allows models like Allegro-FM to improve as more trajectories are generated, effectively turning each new supercomputer run into an investment in future predictive capability.
Real-World Stakes for Quantum Tech and Medicine
The practical payoff, if these materials can be synthesized, extends well beyond the satisfaction of confirming a theoretical prediction. BC8 and related post-diamond phases are expected to combine extreme hardness with unusual electronic properties, potentially enabling new kinds of quantum devices that operate under high pressure or in harsh radiation environments. Nanodiamonds already play a role in quantum sensing, where defects in the crystal lattice act as sensitive probes of magnetic and electric fields. More controllable phases and architectures could yield sensors that are smaller, more robust, and better matched to integration in semiconductor platforms.
Medical applications are another major driver. Detonation nanodiamonds have been explored as drug delivery vehicles and imaging agents because their surfaces can be functionalized with biomolecules while the carbon core remains mechanically and chemically stable. If exascale-guided simulations can point to synthesis routes that produce tighter size distributions, cleaner surfaces, or novel core structures, they could improve the safety and effectiveness of nanodiamond-based therapeutics. In that sense, the extreme physics of shock compression may eventually translate into gentler, more targeted treatments inside the human body.
There are also implications for planetary science and national security. Understanding how carbon transforms under multi-megabar pressures informs models of carbon-rich exoplanets and the deep interiors of icy giants, where unusual carbon phases may influence heat transport and magnetic field generation. At the same time, better predictive control over detonation-driven synthesis has obvious relevance for stockpile stewardship and explosive safety, where the goal is to understand and, where possible, engineer the microscopic products of high-energy events.
From Demonstration to Discipline
The broader significance of Argonne’s work lies in how it showcases the role of exascale computing and AI in modern science. As an Oak Ridge summary of the nanodiamond effort notes, carbon’s complexity makes it an ideal proving ground for methods that fuse large-scale simulation with machine learning. If those methods can successfully predict and guide the synthesis of BC8 or other exotic phases, they will strengthen the case for applying similar workflows to metals, ceramics, and energy materials.
Still, the central question remains unresolved: can billion-atom simulations and sophisticated AI truly close the loop from design to demonstration? Argonne’s researchers are betting that the answer is yes, but they are also clear that progress will depend on sustained collaboration with experimentalists, continued investment in exascale infrastructure, and careful validation at every step. The next few years of high-pressure experiments will test not just specific predictions about carbon, but the broader proposition that computation can take a leading role in discovering and engineering the materials that future technologies will require.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
