The Google Quantum AI team has published results in the journal Science from a large-scale experiment demonstrating contextuality on the superconducting Willow quantum processor—a fundamental property of quantum systems where the result of one measurement depends on what other compatible measurements are performed simultaneously. The authors suggest that contextuality may be one of the physical resources providing quantum algorithms with a computational advantage over classical ones. The experiment encompassed tasks on lattices of up to 105 qubits and included a series of tests—from two-qubit games to multi-party GHZ schemes—each of which confirmed a deviation from classical limits.

Contextuality: A property that cannot be "pre-explained"

In classical physics, it is assumed that measurement results are determined by the system's state at the moment of measurement, regardless of what other compatible quantities are being measured in parallel. Quantum mechanics allows for a violation of this principle: the result may depend on the "context," or the set of simultaneously performed measurements. Theoretically, this property, rather than just superposition or entanglement, is viewed as a possible source of quantum computational advantage. Until recently, contextuality remained a largely theoretical construct, verified only on small systems. The Willow experiment marks the first time this has been tested on an industrial-scale processor with dozens of qubits.

The Magic Square Game: The classical limit and its transcendence

To verify contextuality, researchers used the so-called magic square game. The task is defined as follows: fill a 3×3 table with +1 or −1 values such that the product of numbers in each row equals +1, and in each column equals −1. Mathematically, such a table is impossible: row conditions require an even total number of negative values, while column conditions require an odd number. In the game version, two participants are tasked with filling a row and a column respectively without communicating, and they win if their answers match in the intersecting cell. Any pre-prepared (classical) strategy can win at most 8 out of 9 times, or 88.89% of games. A quantum system using entanglement allows for exceeding this limit. In the experiment, participants had two pairs of entangled Bell qubits. Google conducted all 9 variations of the game 100,000 times and obtained a result of 98.30% ± 0.01%. The processor lost approximately one game out of 60, whereas any classical strategy is bound to lose at least one out of nine.

Bell-Kochen-Specker Inequality: Quantifying contextuality

After demonstrating superiority over the classical limit, the team measured the degree of contextuality itself using the Bell-Kochen-Specker inequality. This inequality checks if measurement results can be explained by pre-determined values of physical quantities (the so-called non-contextual model). For classical systems, the corresponding index χ cannot exceed 4, while quantum mechanics allows a value of up to 6. Google obtained χ = 5.590 ± 0.008—a result significantly higher than the classical boundary and only 7% below the absolute quantum limit. A key feature of this test is that it does not depend on the initial state of the qubits: the value obtained characterizes the processor itself as a physical system, not just a specifically prepared input dataset.

Scaling: 71-qubit GHZ game and 45-qubit entanglement

A separate set of experiments was dedicated to checking whether quantum advantage persists as the number of qubits increases. In the GHZ game, based on a specific multi-qubit entanglement state, the experiment involved up to 71 participants, each corresponding to a qubit. Quantum mechanics allows for a win in every game, while the best classical strategies yield 75% for three participants, with the result asymptotically approaching 50% as the number of participants grows. Due to hardware noise, Google did not reach a perfect result, but the quantum system remained above the classical boundary at all tested lattice sizes. Furthermore, during these experiments, the team confirmed genuine quantum entanglement between 45 qubits—an important result for assessing the processor's scalability.

Linear function search algorithm and practical equipment verification

Another experiment tested a hidden linear function search algorithm, for which the possibility of quantum advantage using constant-depth circuits—where the number of sequential computational steps does not grow with problem size—was theoretically proven in 2018. Google implemented this algorithm on real hardware at a large scale for the first time: the processor solved tasks on lattices of up to 105 qubits, showing a success probability of over 50%. Beyond their fundamental importance, contextuality tasks have found practical applications in equipment verification. They require the coordinated operation of many qubits and help identify subtle systematic errors that may remain unnoticed during standard tests using random quantum circuits. The physical limits (χ = 6 for contextuality measurement and 100% for winning probability) act as a benchmark. This approach may become a standard tool for assessing processor accuracy when executing complex algorithms, including molecular modeling for chemistry and materials science.