Bugfix for r_power_n_segmented branch

[krisvanneste]

Claudio,

I fixed the bug which I introduced in the r_power_n_segmented branch.
In order to catch the case where hinge_distances is None, I had written:

hinge_distances or []

which implies a comparison, which is not supported if hinge_distances is an array.

Now that this is fixed, I would like to ask your advice concerning the current design:
Currently, hinge_distances (corresponding to geom_spread_n_distances configuration parameter) only contains the distances separating the different segments, with the smallest allowed distance hard-coded to 1.
Do you agree with this or would you prefer to include the smallest distance in the list?

[krisvanneste]

Is it possible I have no longer write access to this branch after a pull reqest?
I wanted to push a commit to remove the brackets from the geom_spread_n_exponents and geom_spread_n_distances example in the comments in config_files/configspec.conf, but access is denied.

[claudiodsf]

Is it possible I have no longer write access to this branch after a pull reqest? I wanted to push a commit to remove the brackets from the geom_spread_n_exponents and geom_spread_n_distances example in the comments in config_files/configspec.conf, but access is denied.

That's bizarre. You should retain the write access. Did you try with a force push?

[krisvanneste]

Sorry, again my mistake. I was pushing to the wrong remote...

[claudiodsf]

Nice ! I'll look into details later today

[claudiodsf]

Now that this is fixed, I would like to ask your advice concerning the current design:
Currently, hinge_distances (corresponding to geom_spread_n_distances configuration parameter) only contains the distances separating the different segments, with the smallest allowed distance hard-coded to 1.
Do you agree with this or would you prefer to include the smallest distance in the list?

Why not allow the user to choose the smallest one? I would do something like this (where I also clarified that those are hypocentral distances):

# Exponents and hypocentral distances (in km) separating segments with different exponents
# for the "r_power_n_segmented" geometrical spreading function.
# Note that the smallest distance is generally set to 1, so that at small distances
# the attenuation is the same as "r_power_n".
# Example:
#   geom_spread_n_exponents = 1., 0., 0.5
#   geom_spread_n_distances = 1., 70, 130
geom_spread_n_exponents = float_list(min=0, default=None)
geom_spread_n_distances = float_list(min=0, default=None)

Of course, the check on the length of geom_spread_n_exponents and geom_spread_n_distances should be modified: they must have the same length now.

What do you think?

[krisvanneste]

I can agree with this implementation, but actually the purpose of the smallest distance is to have no further amplification at smaller distances. This also requires a modification in the geom_spread_r_power_n_segmented function. We need to replace

hinge_distances = np.hstack([[Rref], hinge_distances])

with

Rref = hinge_distances[0]

[claudiodsf]

Ok. I just quickly scanned the Boore (2003) paper, so feel free to propose a better wording for the inline help in the config file.

I'll let you do the required commits. Let me know when it's ready to merge 😉

[krisvanneste]

OK

[claudiodsf]

Maybe write "1 km" ?

[krisvanneste]

OK, will do.

[krisvanneste]

Done.
The behaviour below the smallest distance is not explicitly mentioned in the paper, but I simplified it from the Fortran code of SMSIM/EXSIM.

[claudiodsf]

Great.

Just a curiosity: setting the smallest distance to 1 km isn't also a way to have a "classical" $1 / R^n$ attenuation law in the first segment? (as you were showing in the figure in the previous PR)

[krisvanneste]

Not per se.
Classical $1 / R^n$ attenuation is achieved by specifying 1 segment, i.e. 1 exponent and 1 distance.
However, the 1st distance also limits the spreading correction to the distances above it (because $1 / R^n$ attenuation for distances < 1 is actually amplification). So, for distances below the 1st distance, there will be a difference between r_power_n and r_power_n_segmented, even if the exponents are the same.

[claudiodsf]

Ok, clear, thanks.

P.S. You can write LaTeX equations by putting them between $ ... $

[claudiodsf]

It's good to go for me. Ok to merge it?

[krisvanneste]

Yes

[krisvanneste]

Switching back to main

[claudiodsf]

Switching back to main

I'm also rebasing v2, I'll let you knwo when it's done

[claudiodsf]

v2 is rebased and force-pushed. Could you check that this branch also works correctly for you?

[krisvanneste]

v2 is rebased and force-pushed. Could you check that this branch also works correctly for you?

OK. I haven't been working on v2 for a while, but I will try. I guess I will have to force-pull?

[claudiodsf]

I guess I will have to force-pull?

Yes, and there's absolutely no hurry to test v2 😉