Skip to content
  • Danny Hindson's avatar
    Improve sampling of scatter pts in absorption correction simulation · e2af417f
    Danny Hindson authored
    The sampling is now performed according to the volume of each part of
    the sample\environment that intersects the beam profile
    Following changes have been made:
    a) the function IObject::generatePointInObject has been modified so that it returns
    false rather than raising an exception if it fails to generate a point inside
    the object that is also in the active region. This is more efficient when it is
    being called with a maxAttempts value of 1 in an attempt to fairly sample the
    scatter points among the sample + environment components. This involved a change
    in IObject and various child classes (MeshObject, CSGObject, MeshObject2D, Container)
    b) the code that calls IObject::generatePointInObject and cycles through the
    various parts of the environment\sample has been moved from SampleEnvironment into
    MCInteractionVolume so that the sample can be included. There is a new function
    MCInteractionVolume::generatePoint that randomly generates a scatter point across the
    sample and environment components. Possible the SampleEnvironment class could be retired
    entirely and just replaced by a vector of IObject items attached to the sample.
    c) change CSGObject::generatePointInObject to stop calling the fallback method when the
    maxAttempts parameter equals 1. The fallback method always returns a point if the object's
    bounding box is inside the active area which doesn't produce the required sampling across
    the env components. This change has modified the random number sequences used in various
    tests - including the "sample only" tests
    d) added some logging to the simulation to show where the scatter points
    occurred. These show that for Pearl around 6% of the scatter points are
    in the sample which is less than the 50% assumption previously in the code
    e) Several changes to the unit tests (MCInteractionVolumeTest.h, MonteCarloAbsorptionTest.h, The updated sampling means that the absorption corrections
    are slightly different than before for cases with a sample + environment.
    For DirectILLSelfShieldingTest, an extra parameter has been added to underlying algorithm
    (DirectILLSelfShielding) so that this test can continue to use 300 events per point while
    the ILLDirectGeometryReductionTest can use 5000 events per point
    f) some changes to system tests (ILLDirectGeometryReductionTest, complexShapeAbsorptions)
    The calculation used in ILLDirectGeometryReductionTest wasn't converged (changing the seed
    gave a ~25% change in the output) so have increased the number of events per point from 300
    to 5000. I didn't increase the number of events further because I didn't want to make the
    runtime of the system test (esp in debug mode) too large