[Rivet] Validation & tuning phone meeting, Monday 29th Sept, 4pm (UK)

[Andy Buckley]

Lars Sonnenschein wrote:
> On Fri, 10 Oct 2008, Andy Buckley wrote:

>> The bin width factors should be canceled out in the division, but it's
>> possible that LWH's implementation of the histogram division gets this
>> wrong and includes the width factor when creating the DataPointSet.
>> That's worth checking (and fixing if it's the case).
> 
> I checked already the division code. It looks fine.

Me too.

> Now, how could the histogram which has been produced by the division of 
> two other histograms possibly know that it should be treated differently 
> from the other histograms when converted to DataPointSet?

The problem is that the division produces a Histogram rather than a DPS.
I didn't realise that that was the case. So yes, it gets bin values
wrong by a factor of the bin width in the ratio plot.

> The histo would need e.g. a variable which could be set by the 
> division function and the when converting to datapoint sets it could be 
> checked if this variable is set, but it would need to be a publically 
> accessible variable like the histo axis() and bin information variables.

This isn't the proper solution, but I could tolerate it provided we get
rid of LWH soon. But wouldn't an easier way be to make the DPS
explicitly in your analysis code and then rescale it by the bin widths
(= x-errors in DPS language)?

Andy

-- 
Dr Andy Buckley
Institute for Particle Physics Phenomenology
Durham University
0191 3343798 | 0191 3732613 | www.insectnation.org