在大型强子对撞机中，质子每年约有九千万次碰撞会产生顶夸克与反顶夸克这对最重的基本粒子，它们在约十的负二十四次方秒内衰变但保持量子纠缠。其衰变产物保留自旋资讯，使实验得以将两种自旋视为代表 0 与 1 的量子位并重建纠缠。ATLAS 实验在 2023 年首次量测顶夸克对的纠缠，此时距离对撞机启用已超过十七年。

这些量测显示高能物理与量子资讯理论正在汇流，被研究者形容为新兴与如淘金热般的领域。2025 年春天，CMS 实验定量分析顶夸克对的魔性，魔性指的是经典电脑难以高效率模拟的纠缠量子态。由于可扩展量子演算法需要稳定供应此类魔态作为计算燃料，对撞机资料如今提供统计量极大、自然产生的样本来检验量子优势。

过去人们以为更强的纠缠必然带来指数级加速，但 1990 年代的 Gottesman-Knill 定理证明，高度纠缠的稳定子态仍可在经典硬体上高效率模拟，因此不具量子优势。2014 年的研究指出真正关键资源是情境性：在魔态中，测量结果本质上依赖同时测量的其他物理量。这种强烈情境性使其统计行为无法用非情境模型描述，也使魔态成为最难以用经典方法仿真的量子状态。

At the Large Hadron Collider, proton collisions generate top–antitop pairs roughly ninety million times per year, and these heaviest known quarks survive only about a trillionth of a trillionth of a second before decaying while remaining quantum entangled. Their decay products preserve spin information, letting experiments infer entanglement between two effective qubits whose spin-up and spin-down correspond to 0 and 1. In 2023, the ATLAS detector reported the first measurements of top–antitop entanglement, more than seventeen years after the collider began operations.

These measurements signal a new convergence between high energy physics and quantum information theory, described by researchers as an emergent field and a gold rush. This spring, the CMS experiment quantified the magic of top–antitop pairs, where magic denotes entangled qubit states that classical computers simulate only with great difficulty. Because scalable quantum algorithms require a continual supply of such magic states as computational fuel, collider data now provide statistically rich, naturally occurring examples for benchmarking quantum advantage.

Earlier expectations that more entanglement automatically yields exponential speedups were overturned in the 1990s by the Gottesman–Knill theorem, which showed that highly entangled stabilizer states can be efficiently simulated on classical hardware and thus offer no quantum advantage. Work in 2014 identified contextuality as the missing resource: in magic states, measurement outcomes intrinsically depend on which other observables are probed. This strong contextuality makes their statistics incompatible with noncontextual models and secures their status as the hardest states to emulate classically.