Quantum proposals improve a fixed-size molecular calculation

Calculating a molecule’s electronic energy means accounting for many possible arrangements of its electrons. Keeping every arrangement quickly becomes impractical. Keeping only a useful selection makes the calculation smaller—but creates another problem: how do you find the arrangements worth retaining?

In a preprint submitted to arXiv on November 9, 2026, Kentaro Yamamoto and colleagues propose quantum residual exploration, or QRE. A quantum circuit supplies candidates, while classical calculations decide whether they deserve space in the final calculation. The paper combines noiseless computer simulations with a demonstration on Quantinuum’s Reimei trapped-ion quantum computer. Its findings should be treated as preprint results, not established general performance guarantees.

Finding configurations that local searches overlook

The classical starting point is selected configuration interaction. Here, an electronic configuration is represented by a Slater determinant: a mathematical description specifying which electron states are occupied while respecting the rules for identical electrons. The calculation combines selected determinants to approximate the molecule’s lowest-energy state.

Classical selection methods often judge a new determinant by how strongly it connects to the current approximation. But an important determinant may connect only through intermediate arrangements. Repeated classical searches can eventually reach it; a limited search may not.

QRE is designed to provide complementary suggestions along these multistep routes. It does not require the quantum processor to prepare an accurate version of the molecule’s final electronic state.

The researchers first build a classical baseline using heat-bath configuration interaction. A smaller classical calculation supplies a reference state for the quantum circuit. Ten ordered operations move amplitude between electronic configurations. Halfway through, a measurement filters out configurations that retain a specified occupation pattern on selected orbitals. The remaining operations explore the accepted component further, before final measurements produce candidate configurations.

These quantum suggestions receive no reserved places. They compete with classical candidates under the same energy-based ranking. The final energy is then calculated classically in the selected space.

Testing the suggestions on stretched hydrogen

The benchmark is an equally spaced chain of twelve hydrogen atoms, represented using a small orbital basis called STO-3G. Within the chosen electron and spin sector, the complete calculation contains 853,776 determinants. That makes an exact reference calculation possible for assessing the approximations.

Every main comparison ends with 20,000 determinants. This tests whether quantum suggestions improve the use of a fixed final space—not whether they reduce total computing time.

In noiseless simulations, the team examined seven atom separations from 1.2 to 1.8 angstroms. Stretching the chain makes a description involving multiple electronic configurations increasingly important.

The benefit of QRE grew with stretching. Using the researchers’ collective ranking method, called CEN2, quantum proposals reduced the energy error relative to classical-only proposals by 0.81 millihartree at the shortest separation and 40.51 millihartree at the longest. A millihartree is one-thousandth of the atomic energy unit called a hartree. Even after the larger improvement, a substantial error remained.

CEN2 accounts for indirect connections through other candidates, rather than only direct connections to the baseline. Quantum proposals also helped with the simpler direct ranking, but CEN2 provided a further improvement. Adding a size-matched random candidate pool produced almost no benefit in the reported control, suggesting that candidate quality mattered more than simply enlarging the pool.

A smaller benefit on hardware

The Reimei experiment used the most stretched chain. Its circuit required 31 qubits and 939 two-qubit gates. Of 10,000 attempted measurements, 1,072 passed both the intermediate filter and checks on electron numbers. Those accepted outcomes included repeats, so they were not 1,072 distinct discoveries.

After classical selection, the measured proposals lowered the calculated energy relative to the classical-only comparison at the same final size. The gain was modest and smaller than in the ideal simulations. Device noise cannot explain that gap by itself: the hardware also used an approximate reference preparation and a limited measurement sample.

What remains unproven

This is not a demonstration of quantum computational advantage. Total computational costs were not matched, and substantial classical work remains necessary, including candidate scoring, storage and the final energy calculation.

The evidence concerns one hydrogen-chain model in a small basis. It does not establish performance for larger molecules, chemical accuracy, or the independent benefit of the intermediate filter. The practical question is whether future circuits can generate enough useful, distinct candidates to justify their preparation and measurement costs.