• Steve Russell's avatar
    Fixed two bugs in the 1D field classes, FM1DDynamic, FM1DDynamic_fast, · 058f85d9
    Steve Russell authored
    FM1DElectroStatic, FM1DElectroStatic_fast, FM1DMagnetoStatic and FM1DMagnetoStatic_fast.
    
    The first bug had to do with field normalization. Although the 1D field profile
    derivatives were normalized properly, the actual field was not. Instead of
    looping through all of the field values, the loop was over the Fourier 
    accuracy parameter. This typically had the effect that the on axis field was not
    normalized at all. For input fields that were already normalized to 1, this
    had no effect. However, if the input field was not then the field expansion
    was incorrect.
    
    The second had to do with the Fourier expansion used to calculate the 
    derivatives of the on axis field. The code reflects the input
    field in order to ensure that it is periodic. For best accuracy, this 
    reflection should be about the first point of the input data (z = 0). However,
    what was actually implemented was a reflection about -delta / 2, where delta
    is the 1D grid spacing. This caused an error that was small, and became
    smaller with decreasing delta, but significant. Using the new implementation,
    the reconstruction of the field file for the Envelope-Tracker-Phase-1
    (FINSS-RGUN.dat) is about two orders of magnitude more accurate than when
    using the old Fourier expansion implementation. Even though this error was
    small, there was a noticeable difference in the beam envelope when running
    the Envelope-Tracker-Phase-1 and ExternalFieldTest regression tests. The
    reference files for both of the tests were updated using the new field
    classes.
    
    A quick glance shows that the Fourier expansion bug may also be present in the
    1D Astra field classes, but I need to look more carefully to be sure.
    
    Finally, all six classes were cleaned up to conform to the new coding style.
    058f85d9
FM1DDynamic.cpp 10.4 KB