PDL unfortunately cannot help with this, since it (currently) only works with C types, in this case probably a double-precision floating point number. However, the techniques discussed elsewhere in this thread to avoid underflow (especially scaling) will work great to overcome that.