Optimizer Benchmark for Burgers PINNs¶
This notebook compares optimizers on the continuous-time Burgers equation.
All runs share the same double-precision 4x20 tanh PINN, the same training points,
and the same Adam warm start.
Only the second-stage optimizer changes.
Main comparison¶
Final relative errors in this local benchmark are
Adam=5.11e-01,
L-BFGS=4.40e-01,
BFGS=3.07e-01,
and SSBFGS=2.66e-01.
