Open michaeltryby opened 4 years ago
I have made progress on this bug. The 32 bit and 64 bit results are in agreement to approximately 12 decimal places. The results of stats update computations are different due to numerical noise beyond the 12 decimal place.
Furthermore this numerical noise is biasing the reported "Time of Max Occurrence" to be later than it otherwise would be. Filtering out numerical noise results in a shift towards earlier times of occurrence and brings the times being reported for 32 bit and 64 bit builds into agreement.
Under some conditions the time of peak flow is being converted differently causing an apparent shift in time (minutes column of a report table). This can occur when changing compiler versions or going from 32 bit to 64 bit compilation.