if mpirank == 0:
print 'Check gradient and Hessian against finite-difference'
InvPb.update_m(mtrueVm) #InvPb.update_m(1.0)
InvPb.Regul.update_Parameters({
'Vm': Vm,
'gamma': 0.0,
'beta': 0.0,
'm0': 1.0
})
InvPb.solvefwd_cost()
InvPb.solveadj_constructgrad()
nbcheck = 5
MedPert = np.zeros((nbcheck, InvPb.m.vector().local_size()))
for ii in range(nbcheck):
smoothperturb = dl.Expression('sin(n*pi*x[0])*sin(n*pi*x[1])',
n=ii + 1)
smoothperturb_fn = dl.interpolate(smoothperturb, Vm)
MedPert[ii, :] = smoothperturb_fn.vector().array()
checkgradfd_med(InvPb, MedPert, 1e-6, [1e-4, 1e-5, 1e-6], True, mpicomm)
print ''
checkhessfd_med(InvPb, MedPert, 1e-6, [1e-4, 1e-5, 1e-6], True, mpicomm)
sys.exit(0)
# Solve inverse problem
if mpirank == 0: print 'Solve inverse problem'
InvPb.inversion(1.0, mtrueVm, mpicomm, {'maxnbNewtiter': 200}, myplot=myplot)
#InvPb.update_m(mtrueVm)
#InvPb.update_m(1.0)
InvPb.update_m(dl.interpolate(dl.Expression("sin(x[0])*sin(x[1])"), Vm))
InvPb.regparam = 0.0
InvPb.solvefwd_cost()
InvPb.solveadj_constructgrad()
checkgradfd_med(InvPb, MedPert, 1e-6, [1e-4, 1e-5, 1e-6], True, mpicomm)
print ''
#checkhessfd_med(InvPb, MedPert, 1e-6, [1e-1, 1e-2, 1e-3, 1e-4, 1e-5], False, mpicomm)
checkhessfd_med(InvPb, MedPert, 1e-6, [1e-3, 1e-4, 1e-5, 1e-6, 1e-7, 1e-8],
False, mpicomm)
sys.exit(0)
#### Solve inverse problem
if mpirank == 0: print 'Solve inverse problem'
InvPb.inversion(minit,
mtrueVm,
mpicomm, {'maxnbNewtiter': 5000},
myplot=myplot)
InvPb.regparam = 0.0
InvPb.solvefwd_cost()
InvPb.solveadj_constructgrad()
checkgradfd_med(InvPb, MedPert, 1e-6, [1e-4, 1e-5, 1e-6], True, mpicomm)
print ''
checkhessfd_med(InvPb, MedPert, 1e-6, [1e-1, 1e-2, 1e-3, 1e-4, 1e-5], False,
mpicomm)