Tag
This paper identifies a failure mode in physics-informed neural networks where derivative fidelity is compromised during function-value training and provides strengthened benchmark evidence.