jump to navigation

Lubos the experimentalist November 20, 2008

Posted by dorigo in internet, personal, physics, science.

Let us get one thing straight at the beginning of this post, which centers on Lubos Motl’s arrogant pretense of being able to understand details of experimental collider physics better than the experimentalists who did the measurement.

The heart of the matter is the effective cross section of “ghost events” collected by the recent CDF study. The paper says in the abstract that the integrated luminosity used in the study is 2100 inverse picobarns: a number which basically says how many proton-antiproton collisions were used to collect the data where the “ghost events” are found.

When, then, one sizes up at 153,000 the “ghost sample”, the effective cross section of these events is easy to compute: it is just \sigma = N/L, number events divided by integrated luminosity. This is what Matthew Strassler did, finding \sigma=75 pb.

This unfortunately is wrong, because the integrated luminosity of the data CDF used to extract the 153,000 events is not 2100 inverse picobarns, but a smaller sample, amounting to 742 inverse picobarns.

The above fact, however, is slightly concealed in the CDF publication, and you only realize it if you read on page 28 (after all discussions of event counts have finished -the mention of the size of the 153000 event sample is 20 pages above): “For this study, we use a larger statistics data sample[5]“, where footnote [5] explains why the full dataset is now used, 2100/pb, rather than the smaller one of 742/pb used that far: data taken after the first 742/pb were collected is prescaled, which makes it less straightforward to extract cross section measurements from it.

[Motl was deceived because he dismissed too quickly the possibility that the data amounted to 742/pb, reasoning that 153000/742 is 200 picobarns, a number much higher than that given by the authors themselves elsewhere. He made another mistake here: he did not understand that the authors do not care to estimate the cross section of ghost events, which are 50% polluted by backgrounds, but concentrate on the half of it which is indeed of unknown origin: so they get something like 100 picobarns. On this he may be excused: indeed, experimentalists are much more used to talk of “cross sections” for mixed samples, while it would be cleaner to leave that name for physical processes, and only speak of rates when more processes mix together. This time, though, the roles got inverted…]

So, in a nutshell: Strassler, and then Motl, stepped on a point where the CDF publication is not too clear, and err. Not a tragedy, of course. I used a bit of irony when I pointed it out while commenting Strassler’s paper the other day. My point is that theorists should be very careful when interpreting experimental papers. Such numerical details should be checked with the experimentalists who did the measurement.

Now, Motl is not known for his ability to admit his own mistakes. So, what happened yesterday was that after I pointed out he was wrong, and tried in good faith to explain what he had failed to realize, he went postal. His strategy, when cornered, is to counterattack, polluting the waters. But in so doing, he digs his own grave, for now not only his original mistake gets amplified, but he ends up adding a lot of detail about his ignorance of basic facts of collider physics.

The whole thread is available here, but let me nit-pick a few pearls for the lazier of you.

“The 742/pb figure you refer to is just a way to measure a particular smaller subset of the (total) 2100/pb collisions – namely those collisions which are relevant for the correlated b-bar cross section measurements. These special 742/pb collisions are those that are acquired by the dimuon trigger.”

Here Lubos shows his confusion. ALL of the 2100/pb of data the CDF paper is based on are collected by a dimuon trigger. He regards 742/pb as a figure only relevant to an earlier measurement, which is used to compute the sample composition of the data also in the “ghost” analysis.

“The calculation you suggest would lead to three times higher cross section than the real one – something like 200 pb? You’re just completely deluded […]”

He insists: 200 pb is too high (why, one could ask. We are talking of a mixed sample, but in any case there is no reason why 200 pb is outrageous and 75 is fine, for a unknown new physics signal. Just as an example, the cross section of W events in CDF is 20,000 pb, that of Z events is 6,000 pb. But of course, the ghost sample is very large and unaccounted for: it does look very vweird that a signal in the 100pb range has escaped detection in the past).

At this point, my christian soul forces me to warn Lubos, and I say:

“If I tell you the luminosity with which those 153000 events have been collected is 742/pb, you should pause before insisting in your mirror climbing, lest you end up injuring yourself badly. It is NOT 2100/pb, ok ?”

Now, anybody would pause for a moment, when confronted with an author of a paper stating facts very clearly. Does this raise a doubt in Lubos’ beautiful mind ? No. He insists, showing he is basically ignorant of collider physics:

“The number 742/pb only tells us that in about 1/3 of the proton-antiproton collisions, a bottom-antibottom pair was believed to be created at the very beginning, as evidenced by the dimuon trigger. One could use this number, 742/pb, if it were true that all these events recorded with the dimuon trigger had the bottom-antibottom pair at the beginning of the event, and if all the cross sections were somewhat re-expressed as cross sections for bottom-antibottom collisions. Well, except that the bottom-antibottom are not really colliding here. They are just produced in 1/3 of the collisions.”

Here he really does it. He does not appear to know the first thing about a proton-antiproton collision: bottom-antibottom pairs are produced about a thousandth of the times at the Tevatron. Lubos must ignore that the total proton-antiproton cross section is about 80 millibarns, while the b \bar b cross section is more than three orders of magnitude smaller. Not “a third”, Lubos. Please stop it. Stop it… Noooo, he continues:

“But with some handwaving, one could perhaps re-express the cross sections as cross sections based on fluxes of bottom quarks and antiquarks, and the corresponding integrated luminosity would only be 742/pb because the fluxes would refer to fluxes of bottom quarks and antiquarks.”

Some handwaving ? This is not handwaving, this is crazy! Fluxes of bottom quarks! Anyway, the thread ends there. Instead than continuing in my blog, where he cannot control matters well, he brings it to his own, where he has an illuminating post out today, in which he pretends to explain where I was wrong. Poor Lubos. I sincerely feel sorry for his blunder, and did try to stop him even with a private, conciliatory message. To no avail.

Now, my only concern at this point is that he is spreading false data on the CDF analysis. I will have him contacted by somebody in CDF to stop this.


1. mfrasca - November 20, 2008

I am somewhat shocked. It seems you put Lubos completely off the track. Having to choose between unleash an attack against a person or trust his competence, he took the wrong turn, the former, worsening the consequences. An harsh lesson indeed!


2. goffredo - November 20, 2008

Some people think scientists and physicists in particular are some special breed of humans.

3. Gordon - November 20, 2008

Has anyone had a response from Strassler to your criticisms?
One would think either he would admit his error or correct yours.

4. dorigo - November 20, 2008

Gordon, do not fall in Lubos’ mistake.
I do not of course claim to be infallible, but I signed the paper, I know internal information about it, and I said publically here that I even contacted the main author, Paolo Giromini, to triple-check. Strassler’s figure of 75 pb comes out of the blue, but he does not say how he computes it. Lubos, instead, really explains it in detail, and he is doubly wrong.

If you still have doubts, I am sorry. I gave you all the inputs, plus some… The bottom of page 28 of the CDF paper is however rather clear: the luminosity of the data used for all studies until that page is 742/pb.


5. Gordon - November 20, 2008

OK I’ll look at it.

Sorry comments are closed for this entry

%d bloggers like this: