Quantinuum Demonstrates Mathematically Proven Quantum Advantage

Claims of quantum advantage often come with a problem: though a classical computer is believed to be incapable of the task, definitive proof remains elusive. However, a team at Quantinuum recently conducted an experiment that eliminates this uncertainty. The performance gap between their quantum machine and the optimal classical approach is substantiated by mathematical proof, not merely by assumptions about the difficulty of the problem for conventional computers.

This research, peer-reviewed and published in Nature Communications (not a preprint or press release), was conducted by Quantinuum’s London-based team, which includes Harry Buhrman, who also holds positions at the University of Amsterdam and QuSoft. They devised a “complement sampling game” and executed it on trapped-ion hardware, utilizing up to 55 physical qubits.

The Game’s Design Favors Superposition

The experiment involves a referee and a player, both of which are roles within the experimental setup, not actual individuals.

In each round, the referee selects a set containing precisely half of all possible bit strings of a predetermined length. It then generates a quantum state, which is a uniform superposition of every string within that hidden half, and passes it to the player. Superposition implies that the state simultaneously encompasses all these strings. The player’s objective is to return any string from the other half (the complement). A correct answer earns one point, whereas a string from the original set results in a one-point deduction.

A classical player faces a significant disadvantage. Measuring the state only reveals one string from the hidden half. Any other submission is essentially a guess, as most of the information about the set remains inaccessible. As string lengths increase, the classical player’s probability of success over a random guess decreases exponentially.

A quantum player, however, employs a clever strategy. It applies Grover diffusion, an operation that reconfigures the state’s amplitudes, which determine the probability of each measurement outcome. For the specifically constructed sets in this game, this operation transforms a superposition of strings within the set into a superposition of strings outside it. An ideal quantum player would answer correctly every single round.

The sets are not arbitrary; they are built using the structure of the Bernstein-Vazirani problem, a classic example demonstrating quantum computers’ superior ability to find hidden information faster than classical ones. This structure enables the referee to prepare the states and verify answers without requiring impractically long calculations.

The Significance of a 137 Billion to One Advantage

For the largest problem, a 37-bit version, the theoretical disparity between the ideal quantum strategy and the best classical strategy exceeded 137 billion to one.

It’s crucial to interpret this number accurately. It does not mean the quantum computer completed a practical calculation 137 billion times faster than a laptop. Instead, it quantifies the score difference in this specific game. Its importance lies in the exponential growth of this separation with increasing problem size, a conclusion derived from proof.

This distinction sets it apart from random circuit sampling, the technique behind several prominent advantage experiments. Those setups involve a processor executing complex, mostly random operations and producing samples. Verifying the output can be extremely costly, sometimes necessitating a classical simulation of the entire circuit, and the advantage claim relies on the assumption that no clever classical shortcut exists. The complement sampling game, however, provides a classical referee with results that can be quickly verified, and the quantum-classical separation is not contingent on any hardness assumption.

Experimentation on a Racetrack of Ytterbium

The experiments were conducted on Quantinuum’s System Model H2 processors. These 56-qubit machines use charged ytterbium atoms as qubits and barium ions for cooling. The ions move within a racetrack-shaped trap, guided into zones where lasers perform operations. This architecture allows for flexible connections between qubits and enables in-circuit measurement and manipulation, which was essential for the most comprehensive version of the game.

The team tested bit-string lengths ranging from five to 37. The largest circuit employed 55 qubits and averaged approximately 228 native two-qubit gates. For shorter runs, up to 15 bits, the researchers used quantum teleportation to simulate a communication channel between the referee and player. Teleportation transfers a qubit’s state using shared entanglement and conventional classical messages; it does not involve moving matter or exceeding the speed of light.

The randomness for constructing the hidden sets came from Quantinuum’s Quantum Origin system, and combined a local source with a Bell test performed on a separate H1-1 machine. Across all tested sizes, the quantum scores consistently surpassed the best possible classical score, and the team rejected the classical-strategy hypothesis at a 1% significance threshold. The 37-bit run exhibited noticeable noise, which the researchers anticipated given its gate count.

What This Represents, and What It Doesn’t

The term “unconditional” applies to the mathematics, not the laboratory conditions. The primary demonstration assumed referee’s state preparation is trustworthy. Under this assumption, exceeding the classical limit indicates the player employed a nonclassical strategy. Without this trust, the test would possibly fail, as the referee and player could collude to achieve a perfect score.

Two further caveats are evident. Both the referee and player resided on the same H2 chip, which means no actual state transfer occurred between two independent quantum computers. Additionally, hardware noise imposes a ceiling, as errors accumulate with each two-qubit gate until the quantum signal becomes indistinguishable from the classical score. Standard error-mitigation techniques were ineffective here because they require multiple copies of an output, and this game uses a fresh random state in each round.

The authors are candid about these limitations. Minor improvements in gate fidelity might allow the game to scale to slightly larger sizes on current machines. True scaling will likely require fault-tolerant hardware, and the player’s operation, a variant of the multi-qubit Toffoli gate, would necessitate expensive error-corrected T gates.

In essence, this research presents a clear, verifiable, and mathematically-backed demonstration of a quantum machine performing a task that classical machines cannot. The aspiration of running it at a larger scale, between two separate computers, on non-existent error-corrected hardware, remains for the future.