Currently, acoustic FWI (MODE=1, PHYSICS=2) does only converge, when using PCG optimization (GRAD_METHOD=1). The l-bfgs optimization implementation (GRAD_METHOD=2) does not converge.
I suspect there is a bug when using an even number of parameter classes, because the l-bfgs implementation works in GERMAINE for the mono-parameter Vp inversion and in DENISE with 3-parameter Vp, Vs, Density inversion.
Currently, acoustic FWI (MODE=1, PHYSICS=2) does only converge, when using PCG optimization (GRAD_METHOD=1). The l-bfgs optimization implementation (GRAD_METHOD=2) does not converge.
I suspect there is a bug when using an even number of parameter classes, because the l-bfgs implementation works in GERMAINE for the mono-parameter Vp inversion and in DENISE with 3-parameter Vp, Vs, Density inversion.