Multi-mode inversion yields models that do not predict the data about evodcinv

[keurfonluu]

Hi @ariellellouch,

Totally my fault.
disba and evodcinv are using different default values for the parameter dc (the increment for the root finding algorithm), 0.005 and 0.001 respectively.
The default value in disba is too high for multimodal dispersion curves, resulting in what's called modal "jumps" (i.e., wrong association of root and mode).

Setting the same value for both disba and evodcinv will give you an output consistent with what has been calculated in evodcinv:

model.configure(
    optimizer="cpso",  # Evolutionary algorithm
    misfit="rmse",  # Misfit function type
    optimizer_args={
        "popsize": 200,  # Population size
        "maxiter": 200,  # Number of iterations
        "workers": 6,  # Number of cores
        "seed": 0,
    },
    dc=0.001,
)

and

pd = PhaseDispersion(*final_model.T, dc=0.001)

[ariellellouch]

Super, thanks a lot!
Not sure if you want to change the defaults or put a note of it in the manual (these picks are from field data we just acquired, not some theoretical case), but that's a great explanation :)

[keurfonluu]

Yes, I should. I am just not sure which one to use. 0.005 is usually enough for the fundamental mode and is the default value in CPS. But 0.001 is more accurate (but slower). Anyway, if you have other questions and requests regarding evodcinv, feel free to ask!

[ariellellouch]

One idea would be a simple heuristic based on the mode number, maybe? From my experience, your defaults are OK for fundamental, mode 1, and mode 2. So, for example, even something like dc = max(0.001, 0.005-0.001*mode number) could be a starting point...