Total cross sections in Sherpa: How do they compare to LO/NLO and, in particular, to Pythia

[Frank Krauss]

(email by Lenzi) I can confirm that when comparing virtuality ordered pythia shower and sherpa+up to 1 jet they look just the same (which is not the case for alpgen btw). What is still different however is the total cross section, that is lower for sherpa. I know that ME corrections in pythia cannot alter the cross section, so in pythia the ME corrected cross section is the same as the cross section for the qqbar-->e+e- process; but I'm a bit confused about the fact the in sherpa I get a different number. Can you remind me where this differece comes from?

(Follow up question) Is it correct that if I raise Q_cut up to, say s^0.5/2, I shoud get sigma_i ~ 0 for i > 0 and sigma_0 ~ sigma_pythia? About the formal accuracy, during the school in Debrecen Leif Lonnblad said that in CKKW for hadron collision "the total cross section becomes beyond LO (although not NLO)". I did not understand this statement very well. Is this the source of the difference (less that 10%) that I observe when comparing Pythia with Sherpa + up to 1 jet, Q_cut=20GeV?

[Frank Krauss]

the answer is that Pythia takes the LO plain cross section - in contrast we compose our total cross section of the various entries, such that

sigma_tot = sigma_0 * weight_0 + sigma_1 * weight_1 ...

where sigma_i is the cross section for Z+i jets (with kt-jet measure as specified) and weight_i is the average Sudakov rejection weight on this configuration. In principle, both have the same formal accuracy, but they will probably always differ by up to 20-30 %. Depending on jet definitions, process etc., Sherpa's cross section then may be larger or smaller than the ordinary LO cross section.

yes, you're right, raising Q_cut lowers sigma_i and sigma_0 becomes the LO xsec, ~sigma(pythia) - there may be residual differences due to slightly different scale choices in the factorisation scale ... . But this is minor, conceptually you're right.

Concerning Leif's remark, well, let me put it that way: I know that for, e.g., inclusive Z production the NLO-LO K-factor is about 1.3, give or take. We found that in some setup our xsec is smaller than he LO one, so we move away from NLO ... . I do not take this as a sign for our xsec being "improved LO", closer to NLO. So, in my opinion, it's probably best to think of the xsec's as having LO accuracy.

From a more technical side, let me argue on orders: Typically CKKW is good for logs (as is the parton shower), but typically, NLO xsecs are dominated by finite terms (since most of the time they're quite inclusive - there are no logs then). CKKW has no well-defined way to get hold of these finite terms, and that's why I do not believe to produce anyting like an NLO xsec.

On the other hand, shapes are typically dominated by logs (erspecially jet pt's etc.) - therefore, CKKW can be expected to perform reasonably well there.

[Stefan Hoeche]

Milestone old deleted