pplotn
commented
3 years ago In my experience, it seems that using SWS_TAPER_GRAD_HOR=1 is very important. Also, FWI is very instable, because there a lot of simulation parameters, which can blow up the simulation. So I think I was using not good simulation and preconditioning parameters.
Hello, I have a question, regarding the work of FWI. In PSV mode I get error(Simulation is unstable !). Why do I get this, if CFL condition is satisfied and I have small enough dt? Model on this iteration also looks reasonable. If you need more data from me, please ask.
There is a log file below.
Calculate residuals
iteste = 3 itests = 3 step1 = 1 step2 = 0 eps_scale = 3.375000e-02 countstep = 2 stepmax= 6 scalefac = 2.000000e+00 MYID = 0 L2t[1] = 1.148769e-16 L2t[2] = 1.112919e-16 L2t[3] = 1.081774e-16 MYID = 0 pimaxr = 4.228599e+03 gradmaxr = 3.681690e+05 MYID = 0 umaxr = 2.412584e+03 gradmaxr_u = 1.170411e+05 MYID = 0 rhomaxr = 2.555284e+03 gradmaxr_rho = 1.396093e+05
**Message from matcopy (written by PE 0): Copy material properties at inner boundaries ... finished (real time: 0.00 s).
***** Starting simulation (test-forward model) no. 3 for shot 1 of 51 (rel. step length 0.03375000) Message from function wavelet written by PE 0 1 source positions located in subdomain of PE 0 have been assigned with a source signal.
Time step: 2776; pvy: -nan Message from PE 0 R U N - T I M E E R R O R: Simulation is unstable ! ...now exiting to system. Message from PE 4 R U N - T I M E E R R O R: Simulation is unstable ! ...now exiting to system.
I tried to make my issue reproducible with data itself. Please download the archive and try to reproduce the issue. https://drive.google.com/file/d/1fhIx2rlPzHXRfql-gPEr6HV9l9a3LY63/view?usp=sharing