Gravitational waves from the collision of binary neutron stars provide a unique opportunity to study the behaviour of supranuclear matter, the fundamental properties of gravity and the cosmic history of our Universe. However, given the complexity of Einstein's field equations, theoretical models that enable source–property inference suffer from systematic uncertainties due to simplifying assumptions. We develop a hypermodel approach to compare and measure the uncertainty of gravitational-wave approximants. Using state-of-the-art models, we apply this new technique to the binary neutron star observations GW170817 and GW190425 and to the sub-threshold candidate GW200311_103121. Our analysis reveals subtle systematic differences (with Bayesian odds of ~2) between waveform models. A frequency-dependence study suggests that this may be due to the treatment of the tidal sector. This new technique provides a proving ground for model development and a means to identify waveform systematics in future observing runs where detector improvements will increase the number and clarity of binary neutron star collisions we observe.

双中子星碰撞产生的引力波为研究超核物质的行为、引力的基本特性和宇宙的宇宙历史提供了一个独特的机会。然而，鉴于爱因斯坦场方程的复杂性，能够进行源属性推断的理论模型由于简化假设而受到系统不确定性的影响。我们开发了一种超模型方法来比较和测量引力波近似值的不确定性。使用最先进的模型，我们将这项新技术应用于双中子星观测 GW170817 和 GW190425 以及亚阈值候选 GW200311_103121。我们的分析揭示了波形模型之间的细微系统差异（贝叶斯几率约为 2）。一项频率依赖性研究表明，这可能是由于对潮汐部门的处理。这种新技术为模型开发提供了一个试验场，并提供了一种在未来观测运行中识别波形系统的方法，其中探测器的改进将增加我们观测到的双中子星碰撞的数量和清晰度。