Bad ordering of interpolated cross-section uncertainties?

[jshahinian]

Hi,

Thanks for putting together the HEPi-fast tool, it's super convenient! But I think I noticed a mismatch between the order of the outputs described in the README and what the tool actually spits out. The README says the ordering is like this:

id | Central value | error down | error up | error pdf down | error pdf up | error scale down | error scale up

while I think the scale uncertainties are actually reported before the pdf uncertainties, i.e.

id | Central value | error down | error up | error scale down | error scale up | error pdf down | error pdf up

At least that's what I glean from both the size of the uncertainties that are reported (pdf uncertainties should generally be larger than the scale uncertainties) and from the relevant print statement.

Should the order on the README be updated? As a side note, would it make sense to use the same ordering that is listed on the official SUSY XSection twikis, e.g. something like this?

id | Central value | error down | error up | error scale up | error pdf up | error scale down | error pdf down

[jshahinian]

In slightly worse news, we've realized that the up/down variations reported by the tool are also swapped with respect to the values on the twiki. Here's an example for Wino-like N2C1p production with a mass of 200 GeV:

$ hepi-fast 13600_wino_1000023_1000024_NNLL.json
> 200
0 200.0 1.2992818 -0.05042042126595758 0.05116085317111718 -0.012700500000000003 0.015379099999999868 -0.04879463270060147 0.04879463270060147

Now let's assume that the reported pdf and scale uncertainties are indeed swapped, as described above. Then I identify:

• Scale uncertainty down = -0.012700500000000003 pb --> roughly -1.0%
• Scale uncertainty up = 0.015379099999999868 pb --> roughly +1.2%

The corresponding twiki contains this point, but reports the opposite:

• Scale uncertainty down = -1.2%
• Scale uncertainty up = +1.0%

So there's a mismatch somewhere, but I'm not sure if it's coming from the interpolation tool or a mistake when these values were added to the twiki.

[APN-Pucky]

Quick note without having checked everything in detail (more will follow soon): https://arxiv.org/pdf/2304.11915 Table 6, second half at 200 says scale up is -1.0% and down 1.0% as in the twiki. Just for clarity the data is first produced in the paper (with resummino+hepi), then exported in to the json https://github.com/APN-Pucky/HEPi/blob/d44cc2ec64f53570a0c09afcd0d481829136748f/hepi/data/json/13600_wino_1000023_1000024_NNLL.json#L32 there the values are also still correct, then I uploaded to twiki. So the accidental mixture happens in the latest stage namely hepi-fast.

I think the readme/hepi-fast columns are swapped because I initially thought of using the same order as n(n)ll-fast. To be fixed!

[APN-Pucky]

I am not too much of a fan of the twiki ordering. I just kept it there as it was done there before. The fix I am going to merge will then order it as

ID | Central value | error up | error down | error scale up | error scale down | error pdf up | error pdf down

seems to align best with nnll-fast and one would naturally give first the upper error and then the lower.

[APN-Pucky]

Thank you for noticing and informing me about this issue! I have only used hepi-fast for the central values and quick plotting. Upcoming version 0.2.20 will include the update.

[jshahinian]

Thanks a ton for the quick fix and clear explanation, @APN-Pucky! It's totally fine that the ordering doesn't agree with the twiki (I'm not too fond of the twiki ordering either...), I trust users can handle that mapping themselves if they're ever doing these sorts of comparisons.