Scientists spent years arguing that only a quantum computer could decode nitrogenase, the enzyme that makes most of life on Earth's nitrogen available — a classical computer at Caltech solved the key problem first

In January 2026, a team of quantum chemists at the California Institute of Technology posted a preprint to arXiv titled “Classical solution of the FeMo-cofactor model to chemical accuracy and its implications.” The paper’s core claim is contained in its title. A problem that had spent nearly a decade serving as an informal proof-of-concept benchmark for quantum computing, held up as an example of what only a fault-tolerant quantum machine would eventually be able to solve, had been solved without one.

The problem was the ground-state electronic structure of FeMo-co, the active site of the enzyme nitrogenase. The team was led by Garnet Chan, a quantum chemist who had been studying nitrogenase for over two decades and who had, for most of that period, disagreed quietly but persistently with the consensus that the problem required quantum hardware. Kevin Hartnett’s reporting in Quanta Magazine, published 29 May 2026, is the basis for much of what follows.

Why nitrogenase matters

Nitrogen makes up roughly 80 per cent of Earth’s atmosphere, but atmospheric nitrogen exists as the diatomic molecule N2, whose triple bond is chemically inert. Life cannot use it directly. For most of Earth’s history, the only natural mechanism for breaking that bond was high-energy events: lightning strikes, certain geological processes, the slow work of rare atmospheric chemistry. Organisms, in the words of Daniel Suess, a chemist at MIT who studies nitrogenase, were “literally waiting for lightning to strike.”

Roughly two to three billion years ago, nitrogenase evolved in early prokaryotes and changed that situation completely. The enzyme catalyses nitrogen fixation, breaking the N2 triple bond and converting inert atmospheric nitrogen into ammonia that living things can incorporate into organic matter. Every protein, every nucleic acid, every organism that has existed since depends on nitrogen fixed either by nitrogenase or, since 1909, by the industrial Haber-Bosch process that Fritz Haber and Carl Bosch developed and that now feeds roughly half the world’s population.

Understanding how nitrogenase accomplishes this, mechanically and chemically, is both a foundational question in biochemistry and a practical one: a cleaner, more energy-efficient alternative to Haber-Bosch, which requires high pressure and temperatures and accounts for 1–2% of global industrial energy use, has been a goal in chemistry for decades. The enzyme manages the same reaction inside an ordinary soil bacterium.

What makes it computationally hard

The difficulty lies in FeMo-co, nitrogenase’s active site: a cluster of seven iron atoms and one molybdenum atom, each iron carrying four or five unpaired electrons. The electrons cannot be treated independently; their quantum mechanical behaviour is entangled, each one’s state dependent on the others. This is what chemists call the electron correlation problem, and FeMo-co is considered one of its most extreme instances in biology.

The standard entry point for characterising a system like this is to determine its ground state, the lowest-energy electronic configuration from which the reaction proceeds. But FeMo-co’s electrons produce more than 78,000 plausible configurations, and the ground state is a quantum superposition of all of them, with different weighted contributions. In principle, the Schrödinger equation governs the full picture. In practice, solving it exactly for a system of this complexity is not feasible with either quantum or classical methods. Both require starting with an educated guess about which configurations dominate, then building from there.

The quantum computing case for nitrogenase, made most prominently in a 2017 paper in PNAS by Markus Reiher, Matthias Troyer, and colleagues at ETH Zurich. A quantum computer, unlike a classical one, can in principle represent a quantum state directly and evolve it forward until it naturally finds the lowest-energy configuration. A classical computer must instead proceed by ruling out large numbers of configurations, showing they don’t contribute materially to the ground state, and progressively accounting for those that do. The worry was that for FeMo-co, this classical pruning process would become intractable.

Chan had doubted that conclusion since the quantum computing community first latched onto nitrogenase, describing his position as akin to “trying to resist the ocean tide.” His argument, developed in work including a 2023 paper in Nature Communications co-authored with colleagues, was that the quantum advantage claim had never been rigorously established for this specific problem, and that classical techniques were maturing faster than critics assumed.

What Chan’s team actually did

The result reported in January 2026 uses a 76-orbital model of FeMo-co’s resting state, the same model used in most prior quantum resource estimates. The team applied two complementary classical approaches.

The first involved incrementally adjusting the correlated behaviour of small numbers of electrons and demonstrating that extending the same treatment to larger groups produced no meaningful change in the energy estimate. This gave the team a principled criterion for which configurations could be safely ignored and which could not.

The second was the approach Chan had spent the preceding two decades refining: a method that breaks the quantum state into components and allows only a limited amount of quantum information to flow between them, then demonstrates that extending the limit further doesn’t alter the result. Chan described the key insight as “realising that the description could be achieved by ‘simpler’ methods and pushing these methods extremely hard.”

Both methods produced the same ground-state energy for FeMo-co, and both matched experimental observations. The agreement between independent approaches, and with experiment, is the basis for the team’s confidence in the result.

The preprint is not yet peer-reviewed, and peer review of results at this computational frontier can surface significant complications. The claim should be read as a finding that warrants serious attention, not as a settled result.

What this does not resolve

The FeMo-co ground state is the resting position of the enzyme before the reaction begins. It is not a description of how the reaction proceeds. To model the full nitrogen fixation mechanism, researchers would need to calculate energies for an entire sequence of intermediate chemical states, tracking how the system transitions through them. That is considerably harder, and Chan’s team is nowhere near that stage. Suess put it plainly: “We’re not even close to achieving the holy grail of this. We’ve still just described the resting state.”

The result also does not settle the broader debate about quantum computing’s prospects in chemistry. James Whitfield, a quantum computing theorist at Dartmouth College, argues that computing a single ground-state energy value was never where quantum machines were expected to show a decisive advantage. Their likely edge, he says, would appear in the harder problem of modelling how chemical systems evolve over time, where the classical pruning methods Chan uses become progressively more expensive. “If we pick any optimisation problem and you put 20 years into it, you can figure out that one system,” Whitfield told Quanta Magazine. “But whether that solution is transferable? Questions like that won’t be answered by solving one instance of one molecular system.”

Chan does not dispute that quantum computers will eventually play a role in quantum chemistry; he has said he would gladly use one if it were available. His narrower point is about the damage done by a claim that became entrenched before it was properly established. In an email to Quanta Magazine, he wrote: “Science is self-correcting, but quite often, the corrections do not receive the same attention as the initial claim, because the field has moved on to other claims.”

What to watch next

Chan’s stated aim now is to extend the classical methods his team developed to model the full catalytic cycle of nitrogenase, not just its resting state. Whether classical approaches remain tractable as the problem grows more complex is the open question his result raises, not answers.

Progress in fault-tolerant qubit coherence times will eventually make it possible to test the competing predictions about where the classical-quantum threshold actually falls in molecular simulation. That experiment has not happened yet. For the moment, Chan’s result establishes that at least one landmark of that debate was placed in the wrong colum.