{"title":"旋转双黑洞合并的高阶波模式的人工智能预测","authors":"Victoria Tiki, Kiet Pham, Eliu Huerta","doi":"arxiv-2409.03833","DOIUrl":null,"url":null,"abstract":"We present a physics-inspired transformer model that predicts the non-linear\ndynamics of higher-order wave modes emitted by quasi-circular, spinning,\nnon-precessing binary black hole mergers. The model forecasts the waveform\nevolution from the pre-merger phase through the ringdown, starting with an\ninput time-series spanning $ t \\in [-5000\\textrm{M}, -100\\textrm{M}) $. The\nmerger event, defined as the peak amplitude of waveforms that include the $l =\n|m| = 2$ modes, occurs at $ t = 0\\textrm{M} $. The transformer then generates\npredictions over the time range $ t \\in [-100\\textrm{M}, 130\\textrm{M}] $. We\nproduced training, evaluation and test sets using the NRHybSur3dq8 model,\nconsidering a signal manifold defined by mass ratios $ q \\in [1, 8] $; spin\ncomponents $ s^z_{\\{1,2\\}} \\in [-0.8, 0.8] $; modes up to $l \\leq 4$, including\nthe $(5,5)$ mode but excluding the $(4,0)$ and $(4,1)$ modes; and inclination\nangles $\\theta \\in [0, \\pi]$. We trained the model on 14,440,761 waveforms,\ncompleting the training in 15 hours using 16 NVIDIA A100 GPUs in the Delta\nsupercomputer. We used 4 H100 GPUs in the DeltaAI supercomputer to compute,\nwithin 7 hours, the overlap between ground truth and predicted waveforms using\na test set of 840,000 waveforms, finding that the mean and median overlaps over\nthe test set are 0.996 and 0.997, respectively. Additionally, we conducted\ninterpretability studies to elucidate the waveform features utilized by our\ntransformer model to produce accurate predictions. The scientific software used\nfor this work is released with this manuscript.","PeriodicalId":501041,"journal":{"name":"arXiv - PHYS - General Relativity and Quantum Cosmology","volume":"14 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"AI forecasting of higher-order wave modes of spinning binary black hole mergers\",\"authors\":\"Victoria Tiki, Kiet Pham, Eliu Huerta\",\"doi\":\"arxiv-2409.03833\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"We present a physics-inspired transformer model that predicts the non-linear\\ndynamics of higher-order wave modes emitted by quasi-circular, spinning,\\nnon-precessing binary black hole mergers. The model forecasts the waveform\\nevolution from the pre-merger phase through the ringdown, starting with an\\ninput time-series spanning $ t \\\\in [-5000\\\\textrm{M}, -100\\\\textrm{M}) $. The\\nmerger event, defined as the peak amplitude of waveforms that include the $l =\\n|m| = 2$ modes, occurs at $ t = 0\\\\textrm{M} $. The transformer then generates\\npredictions over the time range $ t \\\\in [-100\\\\textrm{M}, 130\\\\textrm{M}] $. We\\nproduced training, evaluation and test sets using the NRHybSur3dq8 model,\\nconsidering a signal manifold defined by mass ratios $ q \\\\in [1, 8] $; spin\\ncomponents $ s^z_{\\\\{1,2\\\\}} \\\\in [-0.8, 0.8] $; modes up to $l \\\\leq 4$, including\\nthe $(5,5)$ mode but excluding the $(4,0)$ and $(4,1)$ modes; and inclination\\nangles $\\\\theta \\\\in [0, \\\\pi]$. We trained the model on 14,440,761 waveforms,\\ncompleting the training in 15 hours using 16 NVIDIA A100 GPUs in the Delta\\nsupercomputer. We used 4 H100 GPUs in the DeltaAI supercomputer to compute,\\nwithin 7 hours, the overlap between ground truth and predicted waveforms using\\na test set of 840,000 waveforms, finding that the mean and median overlaps over\\nthe test set are 0.996 and 0.997, respectively. Additionally, we conducted\\ninterpretability studies to elucidate the waveform features utilized by our\\ntransformer model to produce accurate predictions. The scientific software used\\nfor this work is released with this manuscript.\",\"PeriodicalId\":501041,\"journal\":{\"name\":\"arXiv - PHYS - General Relativity and Quantum Cosmology\",\"volume\":\"14 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-05\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - PHYS - General Relativity and Quantum Cosmology\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.03833\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - PHYS - General Relativity and Quantum Cosmology","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.03833","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
摘要
我们提出了一个物理学启发的变压器模型,该模型预测了准环形、旋转、非预处理双黑洞合并所发射的高阶波模式的非线性动力学。该模型预测了从合并前阶段到环减阶段的波形演变,从输入时间序列开始,跨度为 $ t \ in [-5000\textrm{M}, -100\textrm{M}) $。 合并事件定义为包括 $l =|m| = 2$ 模式的波形的峰值振幅,发生在 $ t = 0\textrm{M} 。然后,变换器在 [-100\textrm{M}, 130\textrm{M}] $ 的时间范围内生成预测。我们使用 NRHybSur3dq8 模型制作了训练集、评估集和测试集,考虑了由质量比 $ q (在 [1, 8] $ 之间)、自旋分量 $ s^z_{\{1, 2\}} 定义的信号流形。\in [-0.8, 0.8] $; modes up to $l \leq 4$, including the $(5,5)$ mode but excluding the $(4,0)$ and $(4,1)$ modes; and inclinationangles $\theta \ in [0, \pi]$.我们在14,440,761个波形上训练了模型,使用Deltas超级计算机中的16个英伟达A100 GPU在15个小时内完成了训练。我们使用 DeltaAI 超级计算机中的 4 个 H100 GPU,在 7 个小时内利用 840,000 个波形的测试集计算了地面实况与预测波形之间的重叠度,发现测试集重叠度的平均值和中位数分别为 0.996 和 0.997。此外,我们还进行了可解释性研究,以阐明我们的变压器模型利用哪些波形特征进行准确预测。这项工作所使用的科学软件随本稿一起发布。
AI forecasting of higher-order wave modes of spinning binary black hole mergers
We present a physics-inspired transformer model that predicts the non-linear
dynamics of higher-order wave modes emitted by quasi-circular, spinning,
non-precessing binary black hole mergers. The model forecasts the waveform
evolution from the pre-merger phase through the ringdown, starting with an
input time-series spanning $ t \in [-5000\textrm{M}, -100\textrm{M}) $. The
merger event, defined as the peak amplitude of waveforms that include the $l =
|m| = 2$ modes, occurs at $ t = 0\textrm{M} $. The transformer then generates
predictions over the time range $ t \in [-100\textrm{M}, 130\textrm{M}] $. We
produced training, evaluation and test sets using the NRHybSur3dq8 model,
considering a signal manifold defined by mass ratios $ q \in [1, 8] $; spin
components $ s^z_{\{1,2\}} \in [-0.8, 0.8] $; modes up to $l \leq 4$, including
the $(5,5)$ mode but excluding the $(4,0)$ and $(4,1)$ modes; and inclination
angles $\theta \in [0, \pi]$. We trained the model on 14,440,761 waveforms,
completing the training in 15 hours using 16 NVIDIA A100 GPUs in the Delta
supercomputer. We used 4 H100 GPUs in the DeltaAI supercomputer to compute,
within 7 hours, the overlap between ground truth and predicted waveforms using
a test set of 840,000 waveforms, finding that the mean and median overlaps over
the test set are 0.996 and 0.997, respectively. Additionally, we conducted
interpretability studies to elucidate the waveform features utilized by our
transformer model to produce accurate predictions. The scientific software used
for this work is released with this manuscript.