Researchers at Oak Ridge National Laboratory have cut the quantum gate count needed to simulate fluid flow by more than 95 percent — from 2 million gates to 91,000 — using an algorithm they call LuGo, developed and validated on the Oak Ridge Leadership Computing Facility's Frontier supercomputer.

The work, led by ORNL postdoctoral researcher Chao Lu and published in Future Generation Computer Systems 178:108270 (2026), tackles one of the most stubborn bottlenecks in quantum modeling of fluid dynamics: quantum phase estimation, the step that converts equations into quantum form. That conversion consumed more than 90 percent of the computational effort in the team's earlier attempts. The innovation received a 2026 R&D 100 Award from R&D World, one of a record 22 such awards for ORNL this year.

"We wanted to find a smarter approach to dealing with a major computational bottleneck in quantum modeling of fluid dynamics," Lu said. "Now we're interested to see what kind of acceleration it can enable for other applications."

The study builds on a previous ORNL effort to solve the Hele-Shaw flow equation — the flow of liquids and gases between two flat, parallel plates set extremely close together — using the Harrow-Hassidim-Lloyd (HHL) algorithm, a quantum method for solving linear equations. Modeling such flows matters well beyond the laboratory. Unsteady flow of air and liquids over machinery parts generates turbulence that drags down performance in fields from aerodynamics to oil refining. The problem is scale: a recent 3D simulation of oceanographic turbulence on ORNL's Summit supercomputer required modeling trillions of grid points.

Classical simulations approximate flow with simplified equation sets that can miss fine detail; direct simulation captures more physics but devours computing time. Quantum computing promises a way out because qubits can encode combinations of values where classical bits hold only one. Realizing that advantage on today's noisy intermediate-scale quantum hardware is another matter. Qubits degrade quickly, error rates run high, and the industry has not settled on a standard error-mitigation protocol or even a canonical qubit medium — neutral atoms, trapped ions and superconductors all remain in play.

The gate-count problem sat at the center of this fragility.

"We needed 2 million gates to perform the necessary calculations," said Muralikrishnan Gopalakrishnan Meena, an ORNL computational scientist and co-author. "These gates are like switches between functions or like intersections on a busy highway. Just as more intersections mean more traffic, more gates mean more potential for error, or noise. That's especially true in quantum computing, because the volatile nature of qubits causes them to degrade quickly and introduces a high degree of noise already."

To develop the algorithm, the team obtained allocations on Frontier — the OLCF's 1.4-exaflop flagship, capable of up to 1.4 quintillion calculations per second and currently ranked No. 2 on the TOP500 list with a peak of 2 exaflops — and on Perlmutter, the National Energy Research Scientific Computing Center's 113-petaflop machine. "For these kinds of calculations, we needed machines with lightning speeds," Gopalakrishnan Meena said. The team validated LuGo by running classical simulations of the quantum circuits, an exercise that required just one of Frontier's nearly 10,000 nodes.

LuGo's trick is architectural rather than exotic: it performs more preprocessing on the classical side before encoding data as quantum circuits, delaying the expensive quantum conversion. "With LuGo, we reduced the computational effort tremendously and observed better overall performance," Lu said.

The team then obtained time on commercial quantum hardware through the OLCF's Quantum Computing User Program, part of the DOE's Quantum User Expansion for Science and Technology Initiative: Quantinuum's H-1, which relies on trapped ions; IBM's Marrakesh and Sherbrooke; and IQM's Garnet and Sirius — the IBM and IQM machines built on superconducting qubits. Those systems served to evaluate the algorithm's quantum capabilities.

"What LuGo does is extend the HHL solution's capability to new levels of detail in much less time," said Kalyan Gottiparthi, an ORNL computational scientist and co-author. "As quantum computing grows as a field and as we move toward a fault-tolerant generation of quantum computers, we expect we'll find more of these kinds of approaches that allow us to leverage established classical solutions in new ways adapted for a quantum advantage."

The research was supported by the DOE Office of Science's Advanced Scientific Computing Research program. The team presented its results at the 2025 IEEE International Conference on Quantum Computing and Engineering, and the findings point toward applications in microfluidics, groundwater flow and porous media flow as the field works toward fault-tolerant quantum computers with lower error rates. As hardware vendors and national labs hunt for hybrid classical-quantum workflows that reduce gate counts today, LuGo offers evidence that preprocessing on machines like Frontier can meaningfully shrink what quantum processors must do tomorrow.