Hi all
I’m trying to obtain an initial_source.h5 file for a fixed source simulation. I have been using a bit of a hack to get this to work in version 0.12.2 but it appears to no longer work on the develop branch.
I’m just wondering if anyone knows what has changed to stop this working.
import xml.etree.ElementTree as ET
import numpy as np
import openmc
def create_initial_particles(source, number_of_particles=2000):
"""Accepts an openmc source and creates an initial_source.h5 that can be
used to find initial xyz, direction and energy of the partice source
"""
# no real materials are needed for finding the source
mats = openmc.Materials([])
# just a minimal geometry
outer_surface = openmc.Sphere(r=100, boundary_type="vacuum")
cell = openmc.Cell(region=-outer_surface)
universe = openmc.Universe(cells=[cell])
geom = openmc.Geometry(universe)
# Instantiate a Settings object
sett = openmc.Settings()
sett.run_mode = (
"eigenvalue" # this will fail but it will write the initial_source.h5 file first
# "fixed source" # write_initial_source is not supported by this run mode
)
sett.particles = number_of_particles
sett.batches = 1
sett.inactive = 0
sett.write_initial_source = True
sett.source = source
model = openmc.model.Model(geom, mats, sett)
model.export_to_xml()
# this just adds write_initial_source == True to the settings.xml
tree = ET.parse("settings.xml")
root = tree.getroot()
elem = ET.SubElement(root, "write_initial_source")
elem.text = "true"
tree.write("settings.xml")
# This will crash but it should write the initial_source.h5
openmc.run()
my_source = openmc.Source()
my_source.space = openmc.stats.Point((0, 0, 0))
my_source.angle = openmc.stats.Isotropic()
my_source.energy = openmc.stats.Discrete([14e6], [1])
output from openmc 0.12.2 with initial source .h5 file produced
output from openmc 0.13 dev branch with initial source .h5 not file produced
Have you got OpenMC built in debug mode? You can attach GDB and probe from there, guessing in the case it bugs out before writing the file. Is there not a ‘less broken’ way you can create this file?
@Shimwell I tried this on the develop branch and I was able to produce an initial_source.h5 file without running into a segfault. Looking at the traceback you have there, it looks like something went wrong with parallel HDF5. Your parallel HDF5 installation is linked against MPICH, but the fact that libmpi.so.40 shows up in the backtrace makes me suspect you compiled with OpenMPI.
That aside, I love your ingenuity in figuring out how to get an initial source file, but we (OpenMC developers) ought to make it easier! At the very least, it looks like write_initial_source is not connected to the Python API which should be fixed.
Hello and sorry for resurrecting an old post! One of my students needs to extract the sampled source locations from a fixed-source calculation. Has a permanent solution been implemented for this? It didn’t seem like we could get the write_initial_source flag to work.
Model.sample_external_source() is probably what you are looking for!
Thank you, Paul! I’ve done some additional digging and want to make sure I’m understanding the process of sampling fixed source particles correctly.
I am generating a fixed source using Model.sample_external_source() and write it to a SourceFile that I then export to HDF5 and use to run a fixed source transport calculation.
If I match the total number of particles (batches times particles per batch) to be equal to the number of source particles in the HDF5 file, am I guaranteed to hit all of the particles, or does OpenMC sample from that list of particles?
The reason I’m asking this is because we have a need to normalize our tallies (which are using a CellBornFilter and a CellFilter) from the fixed-source calculation by the region the inducing source particles are coming from. As such, I would need to count the source particles that are coming from each specific region, and it needs to be exactly consistent with the particles that are run in the calculation.
I have a student that is implementing this directly into OpenMC, but for now we wanted to make a few analyses by obtaining the equivalent behavior via pre- and post-processing. Does any obvious approach jump to mind?
