# Tuesday

dealing with exp(-1000)

I noticed a problem in the dealing with upperlims/uncerts for very specific cases. In the case of matching to an EW=2.5 \pm 0.01 Ang. So I’m trying to match to that EW and sigma. However, there is at least one case where the only values simulated for this observation are 0.01

During analysis, I have to generate a Gaussian with \mu=2.5 and \sigma=0.01 and then plug in all the simulated values to generate a proper errorF. Since all simulated values are so far from the mean (and the Gaussian has such a small stdev), the output is value is zero. Normalizing an entire array of zero values will create a division by zero, and thus generate an array of ‘NaN’ – which is frustrating.
It seems this problem can be avoided by simply adding in a validation that circumvents the Gaussian normalization (pre-errorF) only when the entire Gaussian is zero. This will create an entirely zero output ‘steplike’ function, well…it will only be the yvalues of the output steplike function. The x-values (generated by the relative sizes of the clouds) will still be present.