论文标题
将有效的一体精度和降低的阶次级速度与机器学习相结合
Combining effective-one-body accuracy and reduced-order-quadrature speed for binary neutron star merger parameter estimation with machine learning
论文作者
论文摘要
储层计算是预测湍流的有力工具，其简单的架构具有处理大型系统的计算效率。然而，其实现通常需要完整的状态向量测量和系统非线性知识。我们使用非线性投影函数将系统测量扩展到高维空间，然后将其输入到储层中以获得预测。我们展示了这种储层计算网络在时空混沌系统上的应用，该系统模拟了湍流的若干特征。我们表明，使用径向基函数作为非线性投影器，即使只有部分观测并且不知道控制方程，也能稳健地捕捉复杂的系统非线性。最后，我们表明，当测量稀疏、不完整且带有噪声，甚至控制方程变得不准确时，我们的网络仍然可以产生相当准确的预测，从而为实际湍流系统的无模型预测铺平了道路。
We present mlgw-bns, a gravitational waveform surrogate that allows for a significant improvement in the generation speed of frequency-domain waveforms for binary neutron star mergers, at a negligible cost in accuracy. This improvement is achieved by training a machine-learning model on a dataset of waveforms generated with an accurate but comparatively costlier approximant: the state-of-the-art effective-one-body model TEOBResumSPA. When coupled to a reduced-order scheme, mlgw-bns can accelerate waveform generation up to a factor of ~35, outperforming all other approximants of similar accuracy. By analyzing GW170817 in realistic parameter estimation settings with our scheme, we showcase an overall speedup against TEOBResumSPA greater than an order of magnitude. Our methodology will bear a significant impact on the scientific program of next generation detectors by allowing routine usage of accurate effective-one-body models.
