Segmentation fault when using the ExternalField block in 3D

[mlyto]

Dear developers,

I'm running tests with the ExternalField block for various field configurations.
While this works well in 1D, I experienced a segmentation fault in 3D tests. For example, when I use analytical solutions for a 3D Gaussian beam to set the laser in this way. The same field solution installed at the boundary of the simulation box works correctly.

This is what I see in the slurm file

Initial fields setup

     Applying external fields at time t = 0

[nia1230:54934:0:54934] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x231)
[nia1230:54960:0:54960] Caught signal 11 (Segmentation fault: address not mapped to object at address 0xa1)
[nia0151:253841:0:253841] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x241)
[nia0151:253826:0:253826] Caught signal 11 (Segmentation fault: address not mapped to object at address 0xa1)
[nia0151:253844:0:253844] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x231)
[nia1230:54954:0:54954] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x231)
[nia0151:253840:0:253840] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x231)
[nia1230:54961:0:54961] Caught signal 11 (Segmentation fault: address not mapped to object at address 0x31)

[beck-llr]

Hi
Did you make sure your simulation starts if you comment the ExternalField block ?

[mlyto]

Hi,
Yes, if I comment the ExternalField block, the simulation starts and I see in the slurm file

Initial fields setup

     Applying external fields at time t = 0
     Applying prescribed fields at time t = 0
     Applying antennas at time t = 0

[beck-llr]

Ok, can you share how you define your external fields please ?

[mlyto]

I use different configurations, here is the simplest one - just with the 3D paraxial Gaussian beam in vacuum. I'm not sure if I also need to set profiles for components equal to 0, but after many crashed I also added these functions (didn't help)

---

def myExternal_Ex(x,y,z):
    efield_x = myEx(c_double((- t0) * T_r), c_double((x - Lsim[0]/2.) * L_r), c_double((y - Lsim[1]/2.) * L_r), c_double((z - Lsim[2]/2.) * L_r),
                    c_double(lambda0), c_double(w0), c_double(fwhm * T_r))
    return efield_x * E0 / E_r

def myExternal_Ey(x,y,z):
    return 0

def myExternal_Ez(x,y,z):
    return 0

def myExternal_Bx(x,y,z):
    return 0

def myExternal_By(x,y,z):
    bfield_y = myBy(c_double((0.5*dt - t0) * T_r), c_double((x - Lsim[0]/2.) * L_r), c_double((y - Lsim[1]/2.) * L_r), c_double((z - Lsim[2]/2.) * L_r),
                    c_double(lambda0), c_double(w0), c_double(fwhm * T_r))
    return bfield_y * B0 / B_r

def myExternal_Bz(x,y,z):
    return 0

ExternalField(
    field = "Ex",
    profile = myExternal_Ex
)

ExternalField(
    field = "Ey",
    profile = myExternal_Ey
)

ExternalField(
    field = "Ez",
    profile = myExternal_Ez
)

ExternalField(
    field = "Bx",
    profile = myExternal_Bx
)

ExternalField(
    field = "By",
    profile = myExternal_By
)

ExternalField(
    field = "Bz",
    profile = myExternal_Bz
)

P.S: edited by beck-llr just to make the code a bit more readable.

[beck-llr]

You don't have to set other components to 0.

Can you reproduce the issue if you set efield_x=0 and bfield_y=0 instead of myEx and myBy which are not defined in the piece of code you shared ?

[mlyto]

Yes, I just tried it - it crashes even with efield_x=0 and bfield_y=0

[beck-llr]

Ok that looks bad indeed :-)
Thanks for reporting we'll try to have a look asap.
Are you using the latest version of the code ? If not try it out. If yes, try to roll back to 4.8 if you're in a hurry.

[mlyto]

I run my codes on a supercomputer. This version I uploaded a year ago.

[mlyto]

Update: In version 5.0 I have the same problem with the ExternalField block (previous version was 4.7)
Also, now when I run the job, there is an info.shelf file in the directory. This was not the case in the previous version, what is this file for?

[mccoys]

Please try to use external fields on a smaller case. This is too check whether it is a memory issue.

The new file serves as a storage for simulation information. Happi uses it to obtain information about the configuration

[mlyto]

It works in 1D. What do you mean by "smaller case"? Smaller simulation box size? This one isn't big 24x24x32 (in lambda)

If possible, one more question wrt info.shelf file. Does it mean that updated happi doesn't work with the old simulation results since it requires info.shelf?

[mccoys]

By smaller case i mean something occupying les memory. Like less particles or lower resolution.

The new happi falls back to the previous approach if it can't find the file

[mlyto]

Thank you for your explanation about happi.

As to ExternalField block, the test case I reported was related to vacuum, so no particles at all, and the number of cells 384x384x512, which is pretty low wrt my simulation for plasma with Laser or Prescribed field blocks.

[mccoys]

Can you provide a case that crashes?

[mlyto]

vac_ext_zeros.txt
It's a Python file, I just can't attach it as a .py

If I comment all ExternalField blocks it works (with all zero results), however if all block are uncommented I get
Caught signal 11 (Segmentation fault: address not mapped to object at address 0x361) ...

[beck-llr]

Thanks for sharing. I was able to reproduce your problem. There seems to be something off with the E fields (it works for the B fields). We'll look into it and let you know asap.

[mlyto]

Thank you.

[beck-llr]

I've pushed a fix in our local repository. We'll try to make it available to you tomorrow. Thanks again for reporting and doing all the tests !

[mlyto]

Thank you again!

[beck-llr]

The last commit should fix your issue. Please confirm and close the issue if it does.

[mlyto]

Thank you for fixing this issue. Now it works.