This Week’s Rumor

A couple months ago CDF made the New York Times by releasing results claiming to see a resonance in the invariant mass spectrum of two jets produced together with a W. Last week they released a new analysis with twice as much data, claiming the signal was still there, now at a statistical significance of nearly 5 sigma.

I recently wrote about this here, explaining the reasons for being skeptical, despite the high statistical significance. One very good reason for being a skeptic is that CDF’s Tommaso Dorigo doesn’t believe this is real, going so far as to put his money where his mouth is, offering a $100 wager to back up his arguments. The crucial question in everyone’s mind has been whether D0, CDF’s competition and sister detector at Fermilab, would see the same thing in its data. If there’s really something there, D0 should see it. This Friday there will be a Wine and Cheese talk at Fermilab, where the D0 results will be unveiled, and you can watch this as a live video stream here. But, as one might expect, now that the D0 result is ready to be revealed, people do things like leave print jobs on printers, etc., causing well-sourced rumors to spread. Blogs such as this one seem to be a place where such information tends to end up, so I can report a rumor (based on excellent sources) that Tommaso is right. D0 will report on Friday that there’s nothing there, that they find no evidence for a dijet resonance in the region from 110-170 GeV. They reject the CDF hypothesis of a resonance with a cross section of 4 pb at a significance level of over 4 sigma. In other news, the LHC is running very well, with the official goal of this week being to reach an integrated luminosity of 1 fb-1, something that had been the official goal for the entire year (although, unofficially, 2-3 fb-1 is more like it). Right now, they’re around .8 fb-1. This sort of luminosity should finally start to allow in coming months results that either rule out a Higgs in the region it is expected or see first indications if it is there. At the KITP, this week marks the start of a program on The First Year of the LHC. Unfortunately for theorists, the only result of data from the first year of the LHC has been to shoot down some of their favorite models, ruling out for instance a large amount of the parameter space where supersymmetry was expected to be found, making the most popular theoretical idea of the last thirty years significantly less popular. The first talk held at the KITP program was this afternoon, and it dealt not with the LHC data, but with the supposed CDF resonance (it appears that news of the D0 result hadn’t yet made it to Santa Barbara). Update: The KITP talk is now available here. Update: The D0 PRL submission that has been circulating privately for the past few days is supposed to be available at 9am Friday Fermilab time here. See here for other material to be released publicly today. Update: Now that the D0 results are officially out, as usual your best bet for informed explanation is to check out what Jester and Tommaso have to say. This entry was posted in Experimental HEP News. Bookmark the permalink. 34 Responses to This Week’s Rumor 1. Anonymous says: As always, the devil will be in the details. When D0 concludes no resonance, one should ask compared to what. An interesting D0 thesis on W+2 jets is http://www.slac.stanford.edu/spires/find/hep/www?r=fermilab-thesis-2010-48. Same luminosity as in D0 talk on friday, different cuts. Starting at pg 101, a procedure called reweighting the theory prediction is explained. Of particular interest is fig 44 (the dijet invariant mass before and after reweighting). The crucial thing to look for during Friday’s D0 talk is how does the dijet mass distribution compares to background estimates (theory prediction+detector simulation) before reweighting. Only then do we compare apples to apples. Everything after that, including reweighing, is interpretation and will be open to discussion….. 2. anon. says: The first talk held at the KITP program was this afternoon, and it dealt not with the LHC data, but with the supposed CDF resonance (it appears that news of the D0 result hadn’t yet made it to Santa Barbara). Everyone expects D0 will say they see nothing. This doesn’t mean we shouldn’t be talking about the CDF resonance. It’s raised a lot of questions — how do we know when to believe signals that come from jet physics? How can you successfully make a case that a peak is real if it’s a wiggle on a steeply falling background? What can we do to convince ourselves something is new physics, if even high statistical significance isn’t a good guide? (See also: pentaquarks.) Even if D0 rules it out so decisively that everyone is convinced, we’ll still have the puzzle of what went wrong in the CDF analysis, and how to keep it from going wrong again. None of the potential mismodeling explanations seem entirely convincing. The LHC is doing amazing work slicing away parameter space using missing ET or leptons, but if new physics shows up in very jetty channels, the same issues raised by the CDF analysis are going to be important. Whether or not the CDF bump is real (and even the most rumor-happy phenomenologists in the world tend to think it isn’t), getting theorists and experimentalists to talk seriously (hopefully to each other) about jets is an important thing. Having watched the video of the KITP talk, it’s very much in the vein of “let’s try to figure out what the data are saying and what could be causing this,” not a premature celebration of new physics. 3. factotum says: CDF measurement has been scrutinized for 2 months. Why should we take D0’s word as final without subjecting it to the same level of scrutiny? 4. anon2 says: @factotum The D0 analysis is not new. They have been looking at this data as long as CDF has. This is, at least, what I’ve heard from a D0 member directly involved in this analysis. @Anonymous CDF does not need to tweak the Monte Carlo generator knobs to match the data control distributions because they make more strict kinematics cuts than D0. If D0 presents the results with an analysis strategy similar to CDF’s, the effect of the so-called “reweighing” (which is a really bad name) will be negligible. 5. Anonymous says: @anon2: then let’s hope D0 leaves the kinematic adjustment (=reweighing) issue out of this analysis. This will give us a clean comparison between CDF and D0. 6. Bernhard says: @ anon. “What can we do to convince ourselves something is new physics, if even high statistical significance isn’t a good guide?” Now that D0 is refuting the peak, I see two possibilities on the table. Either D0 screwed it up and is not seeing the peak because it overestimated the systematics and erased the signal in a sensitive high-background region or CDF is suffering from a subtle detector effect, so a subtle systematic effect that D0 does not suffer from and so does not see the peak. Either way, there is no chance (or a extremely tiny one) that the much better ATLAS and CMS detectors with better resolution eta coverage etc will suffer from both. Given enough data, it will be clear if there is something there, but to be sure this will take more than 1fb-1 I’m afraid. And imagine that both ATLAS and CMS could suffer from the same supposed systematic effect that CDF could be suffering from at the same time is very unlikely. So if they both see the peak, case closed. The apocalyptic scenario would be ATLAS seeing the peak and CMS not but I see no reason to imagine that such a bizarre scenario like that would happen. 7. tommaso says: I like a lot this thread -very informed. Interesting points are raised. In particular, the reweighting technique is indeed the crucial point IMO. Now all please stop and think. What do we do when we put a Monte Carlo simulation under our data ? We are comparing previously tuned simulations to newer data, nothing else. The parton shower simulation that is used to go from matrix element calculations to final state observables contains knobs that are turned to match the simulation output to large datasets of well-understood physics processes. Every time something matches poorly, one asks oneself if NP is the cause or a mismodeling issue; once the NP interpretation is disproven, the simulation receives a further tuning. In this case, what I think I can say is that W+jets production is a rather well-known and well-understood process, but its detailed kinematics had not been tested to this level of statistical precision before. So we are in a regime where mismodeling issues may arise. And they do. If a reweighting appears to be all what’s required to match the data, Occam’s razor comes a-slashing the pet signal that CDF published. Peter, thanks for the mention of the bet -but I guess that by now nobody else will take it. On the other hand, it is not a single$100 bet: I made available 20 tickets. 19 are still there, so if you believe the CDF signal will eventually survive, take one. Say so in my blog.

Cheers,
T.

8. Bernhard says:

Tommaso,

If this is the case, one should expect different mismodeling effects from different Monte Carlos, don’t you agree? I mean even if you have all generators needing some reweighting in that region to fit the SM one could expect the tunings not to be the same. And the shape and even how much the peak is pronounced would probably be generator dependent as well, right?

Also, I’m not sure I understand for example how could CDF and D0 apply such a different reweighting… The reweighting factors should be taken from a control region (not where the peak is or is not) where one would expect both detectors to agree very well so the final tuning (for each generator) would be similar. How can this, in the end cause, such a dramatic difference? Is hard to believe CDF would make such a fuzz on something that can be corrected by a simple reweight factor. What they are claiming is a genuine peak, something that cannot be corrected by scaling some curve up. So, a subtle detector effect seems more likely IMHO.

What did I miss?

9. Ken Lie says:

I want to share with you a rumor which is discussed in an informal way in Munich. The source of this information is in the theory group of CERN.
The LHC has seen a particle of energy 150 GEV which is not Higgs (!!)
Apparently, the result will be confirmed in 2-3 weeks.

10. Bernhard says:

Ken Lie,

“The LHC” didn’t see anything .

Which experiment?

11. Pingback: Another LHC Update « viXra log

12. DGP says:

I used to think I knew something about experiment design but I am awed by the difficulty of the background subtractions involved in these latest experiments.

I assume that all proper precautions have been taken to avoid a systematic error in the background: I know that more than one Monte Carlo code was used but are the algorithms and especially the pseudorandom number algrothims fully independent? And what dependence is there on assumptions about QCD? QCD is very good but it’s not quite as solid as QED.

I will be watching the contributions from the other experiments with great interest.

DGP.

13. Walter says:

@tommaso: exactly! However, at this point the question is not new physics or not.
Understanding this potential mismodeling in terms of the theory predictions will extend beyond CDF and D0. After all ATLAS and CMS use the same tools.
This point was very well made in anon’s first comment. If the deviations from the theory expectation are subtle, how can we conclude there is new physics? The CDF bump, no matter its origin, is a perfect case against which this can be discussed.

14. chris says:

DGP,

random number generators have matured substantially over the last 2 decades. with all the modelling of e.g. hadronization in there, RNGs are the last thing i would be concerned about.

15. DGP says:

Chris,

Thanks. I know RNGs have moved on since I modelled scattering but there was always a tendency for everyone to use the best, i.e. latest, RNG so independence was always an issue. I thought that might still happen.

I assume that by “hadronization” modelling you mean the sort of consideration I meant when I referred to the assumptions in QCD.

Regards,

DGP

16. Peter Woit says:

anon.,

I didn’t intend to suggest that phenomenologists at the KITP or elsewhere were optimistic about the CDF signal, just thought it was an interesting sign of the times (the times being this month…) that at a workshop supposedly devoted to the LHC results, the hot topic was a Tevatron topic. I did listen to some of the talk last night, and noticed that if the speaker or those in the audience knew what the D0 result was about to be, they were keeping quiet about it…

Ken Lie,

Are you sure this isn’t a garbled version of the Tevatron story about a supposed 150 GeV state discussed here? I haven’t heard any rumors about either of the LHC experiments seeing a 150 GeV state. If there are more details though, pray tell…

17. factotum says:

@anon2: CDF looked internally for over a year, then their result has been subject to 2 months of public scrutiny

Sure, D0 may have had the internal studies for a long time. But they should be subject to the same questioning and testing from the broader community, just as CDF has, and not just be taken immediately at their word..

18. Philip Gibbs says:

Peter, the new LHC results are being presented at PLHC2011 this week. There are ten talks from ATLAS and one from CMS that cover new results using 2011 data. All the plots appeared in conference notes last week so we already know they just show new exclusions, not new physics. D0 are also presenting Wjj there on Saturday.

19. tommaso says:

Dear Bernhard,
not necessarily different MC would give different outputs. If you take Alpgen matrix element calculations and add a Herwig parton shower evolution, and then take Sherpa and add a Herwig parton shower evolution, the parton shower is the same.
Further, the reweighting issue involves quite a few choices that require craftmanship. Experimentalists are proud of this part of their job, which is in fact what distinguishes them from a piece of computer code (a simulation of the typical particle experimentalist is not so hard to write, anyway). I think this is enough to understand why different experiments can see and do different things.
As for a detector artifact, I doubt it. The detectors are very, very well understood by now.
Cheers,
T.

20. Peter Woit says:

Philip,

I noticed that the most interesting of these talks, the Friday ones with some results from 2011 data, have now been posted, see

http://indico.cern.ch/conferenceOtherViews.py?view=standard&confId=100963

The bottom line: SUSY and extra dimensions are in even more trouble. No hint of the Higgs yet, but they’re getting closer. If it’s there it could easily start to show up in the data they’re analyzing now, trying to get ready for conferences in late July.

21. Bernhard says:

Dear Tommaso,

Thanks. I agree with you that if one has the same parton shower, the output can be almost the same.

As for reweighting, you are probably talking about something more sophisticated than I thought. I naively thought you were just talking about the procedure of choosing a control region (let´s say a dijet QCD region for a final state with no leptons) that you know well the MC should reproduce it and if the data gives say 100 events and your MC just 85, one multiplies this MC and all other curves by a factor of 1.17… This should be fairly simple to do…

As for the detector effect, you maybe right, but we never know for sure… 🙂

22. anon says:

to ken lie :
i can say that there are quite a lot analyses which have some sort of large disagreement with expectation. mc modeling sometimes is not reliable which people might not realize in some special cases. 2 or 3 sigma is easy to see, what’s interesting is if this is a modeling problem, people would see significance growing as more data come in, which resembles what people expect if there’s a real signal. if this happens to a *benchmark* analysis, rumor may spread easily.

23. model and sim says:

I was curious, for the modellers out there, as the means and distributions are changed in the monte carlo model, I assume that someone is recording those changes in assumptions. Has anyone done analysis on those changes as part of a dual solution check?

24. M says:

there is a new rumor from LHC…

25. Anon says:

Tomasso says: Further, the reweighting issue involves quite a few choices that require craftmanship. Experimentalists are proud of this part of their job, which is in fact what distinguishes them from a piece of computer code (a simulation of the typical particle experimentalist is not so hard to write, anyway). I think this is enough to understand why different experiments can see and do different things.

Hmm, as a theoretician reading the last sentence, it doesn’t sound like something to be that proud of 🙂

26. Peter Woit says:

M,

I’m hoping there’s finally enough luminosity for rumors to start about seeing something with a plausible cross-section to actually be the Higgs. Or, at least there should be a supersymmetry signal rumor, I’m surprised there haven’t been any of those yet. Let’s hear some more details…

27. tulpoeid says:

Oooooh, this is starting to border begging, guys…

28. d0 compared to cdf says:

This is still a little prelimary, but after pulling apart the distributions, the principle reason why cdf sees a particle and d0 does not appears to be correlated with cdf favorings top quark events over w events. Because the top event distribution has a different peak than the w event distribution, the difference in proportionality appears to cause a deficit in the region were cdf claims to see a new particle. There also appears to be a shift in distributions related to choice of bin size which is also having a lesser effect.

The underlying distributions for the different types of events appear to be largely in agreement between the two teams, but CDF definitely appears to favors more events as being top quark events as compared to D0

29. Peter Woit says:

d0 compared to cdf,

Thanks for the informative comment. People interested in this should also look at Michael Schmitt’s blog entry here

http://muon.wordpress.com/2011/06/12/top-background-shapes-and-the-cdf-mjj-anomaly/

30. d0 compared to cdf says:

Thanks,
One of the things I am hoping to figure out is why there is a deficit at 65 GeV in the D0 analysis that appears to have harmonics. One can enhance the harmonics and transform into the t-domain and see two distinct events each half cycle. The same deficit at 65 GeV is apparent in the CDF data, but the CDF error interferes with the harmonics. If I can isolate the nature of the error, it might be possible to see if the harmonics where actually enhanced in the 7.3 fb^-1 data.
I am sure there is a logical explanation for the harmonics, it could just be a fluke.