Neural Force Regression on Gravitational Wave Data: Time-Averaged Effective Potential Extraction and Multi-Event Comparison Across BBH and BNS Mergers

We apply the gauge-free force regression framework (Paper 6) to real LIGO gravitational wave strain data from three events: GW150914 (36+29 Msun BBH), GW151226 (14+8 Msun BBH), and GW170817 (1.4+1.4 Msun BNS). For each event a neural network is trained to fit observed strain accelerations, then decomposed into conservative and dissipative components by velocity averaging, yielding the effective potential V_eff(h) without gauge ambiguity. All three events show systematic negative deviations from the simple-harmonic GR baseline, identified as a consequence of the time-varying frequency structure of GW chirp signals rather than new physics. The key physical finding is a 2.22× ratio between BNS and BBH effective coefficients, reflecting the chirp-mass dependence of frequency evolution. This is Paper 8 of the Neural Lagrangian Series. LIGO data accessed via GWOSC.