Date: Fri, 11 Sep 1998 14:20:09 -0400 (EDT) From: PR/D-email To: toback@fnal.gov Cc: prd@aps.org Subject: dt6516 DT6516 Searches for new physics in diphoton events in $p p bar_ collisions at $sqrt s Abe,F./Akimoto,H./Akopian,A./Albrow,M.G./Amadon,A./Amendolia,S.R./Amidei Dear Dr. Toback: The above manuscript has been reviewed by one of our referees. Comments from the report are enclosed for your consideration. When you resubmit your manuscript, please include a summary of the changes made, and a brief response to all recommendations and criticisms. Sincerely yours, John Ripka Assistant to the Editor Physical Review D ------------------------------------------------------------------------------- REPORT OF THE REFEREE The authors have done a great deal of work examining this channel in general and their puzzling event in particular. I salute their endurance under the glare of publicity. There are a number of issues which should be addressed before publication. I place particular emphasis on the *'d items below. p 10 why must photons be central while electrons need not? p 17 measurement of photon efficiency with electron sample * what systematic error is associated with measuring photon efficiency with electrons? Was this cross-checked with other measurements from the direct photon analyses? This would appear to invite a pt dependence in the estimated vs true efficiency. p 18 what is the relative size of the sample in the left and right plots? They are subsets of each other, and as a result highly correlated (see for example the low Et photon efficiency, in which 3-sigma fluctuations are tracked point for point in the plots. * Why are the clear systematic dependencies on Et of the high-Et cuts ignored in the efficiency calculations? The efficiency near threshold of 22 GeV would appear to be 65% at most, not very close to the lower value of 80%. This choice needs further justification. p 20/21 Figure 7b (lower) does not represent an efficiency flat at 96% above 12 GeV. Clearly, the isolation requirement degrades efficiency above 19 GeV. That's why the requirements are changed above 22 GeV. p 22 * The estimated purity of the diphoton sample is only 15%. However, I saw no argument in the paper as to why SM expected rates should all be dominated by real photon pairs. This question is relevant particularly for the lepton, b, and gamma channels, as the SM rate is estimated using specific physics processes with 2 real photons, while the Etmiss and jet rates are estimated by more general arguments. The 15% purity figure would naively imply that SM rates are underestimated in these channels by a factor of 6-7. p 27 bottom: phrasing clumsy dominant mechanism ... is dominated by... p 50/51 4.8.3 The discussion left me confused. o Why should a photon be required to pass an isolation requirement in order to be considered as a source of the plug em cluster? Even QCD photons can be produced by radiation of a jet (quark bremsstrahlung) and thus not be isolated. o The end of the first paragraph seems to conclude that in this particular event a photon the plug region would have a 6-8% chance of having such a stub and VTX requirement, certainly not very unlikely. o Next the gamma-gamma sample is used to make an estimate independent of the rest of the event, which seems to argue that a photon out of the context of this event would have a stub and vtx requirement perhaps .2% of the time. * What are we to conclude? The second estimate does not change the circumstances of the actual event. Which probability best represents the probability that a photon could have manifested itself in this event in this way? Such a number is offered for the other interpretations, and should be offered for the photon interpretation. Even the .2% number for P(em|photon) is entirely comparable to the .3% number for P(em|electron). p 52 The fact that a jet passing the em selection criteria would be an unusual example of a jet is certainly no bar to that being the likeliest interpretation. Recall the central "gamma gamma" candidates. Jets faking central gammas are even less likely than in the forward case. Yet, after taking into account the large number of jets compared to direct photons, the bulk of the gamma gamma sample (85%) consists of jets! p 55/56 The summary focuses attention on the fact that the e candidate is unlikely to be a real e at the 10-3 or less level. It might be worth reiterating that the gamma-gamma is more likely than not to have a different origin than real directly produced photons, at a milder 15% level. section 5, p 56/57 The introduction of these section makes a strong distinction between the em plug candidate, in which the possibility of fakes are mentioned (since the probability of a real e giving the observation is estimated as .3%), and the diphotons, which are discussed as if they are certainly real. * There is something puzzling about the diphotons. The quoted purity is only 15%, yet perusing table 19 which does attempt a calculation of the fake rate, claims that real photons dominate the signal. This deserves an explanatory comment. Are the fakes badly underestimated, or is the 15% purity inappropriate here? p 58 * I did not see estimates for SM j+3 photon contributions, nor for e+3j (e.g. W + 3j with jets producing fakes) nor for Z + 2 photon or ZZ +2 photon (with missing Et from one of the Z decays). Are all really negligible wrt the computed processes? * For example, W+3j could be estimated from inputs of table 17 and 18 as 6E4(6E-4)(6E-4)(2E-5) = 4E-7, which is considerably larger than the estimate in table 19 of 2E-9, which uses only the "fake" photon probability of 3E-5 rather than the radiation probability of 6E-4. Even correcting down by a factor of 4 for double counting as suggested in Table 18 makes this channel one of the largest, and is a factor of 6 larger than the "non-electron source" rate quoted in the overall conclusions on p 62 and p 70. * Further, although photon and tau causes of the plug em cluster are discussed in section 4.8, I see nothing in table 19 which appears to estimate a tau contribution, and am unable to ascertain whether it includes a real forward photon contribution. What are the SM rates for each of the listed possible sources of the em plug cluster? Section 4.8 gives P(em|x) where x is the hypothetical cause. To complete the calculation P(x|em) = P(em|x) * P(x) * const , where P(x) is the relative rate of the potential sources (at least in the SM). Section 5 attempts to calculate P(x) for the most part, since it applies only the nominal cuts, not the SVX/VTX information selections used in section 4.8. * However, section 5 is incomplete, as P(x) is not evaluated at all for tau sources of the plug cluster, and as mentioned above, it is unclear whether the estimates actually include real-photon -> em plug contributions to the rate. * If, in the opinion of the authors, information weighing on the origin of the em plug cluster is important, then it should all be available, preferably in the form of a table giving x P(em|x) P(x) e,tau, sect 4.8 sect 5 summarized by source x j,gamma p 61, 3rd line from bottom capitalize Standard at start of sentence p 67 don't you mean "less restrictive than" instead of "comparable" when discussing the limits which are not shown on your plots? p 70 * see discussion earlier about rates for real photon as source of em cluster, which would appear to raise the "fake electron" rate well above 6E-8. To be sure, the event remains pretty striking no matter what its origin, and SM backgrounds remain small. p 75 [19] which version of pythia/spythia? Various versions have various bugs, and people may wish to evaluate this. p 77 [ 32] does the Etmiss change in the same direction as alternative vertices are chosen for the photons and plug? p 78 [39] What does this footnote mean? That you don't understand the shower shape after all? Or that you do and shouldn't have called it unusual? p 88 where does the factor of 6 come from? From 4!/(2!)(2!)? There is nothing intrinsically important about the actual event multiplicity. Does considering only multiplicity 4 sources underestimate the overlap rate? ** Author's Response: We would like to thank the Referee for their remarks. We hope they find our responses satisfactory. We will answer them by page number for easy reference. P 10 There are a number of reasons why we only look at central photons. While we could have used photons in the plug, the primary reason in this particular analysis is the trigger. In order to have two photons pass the trigger, one in the plug, the threshold would have been substantially raised making the additional sensitivity small compared to the additional work needed to fully understand the photons in the plug and forward area. The reasons that electrons are allowed to be in the plug area is that they have been studied for a great deal of time in W and Z physics as well as in top physics, and the identification of these objects is very standardized. Furthermore the purity of electrons happens to be higher than for diphotons with one photon in the plug calorimeter. P 17 As the referee clearly knows there are indeed systematic differences between the way electrons and photons interact with the calorimeter and tracking chambers. Thus, one must indeed be careful when estimating efficiencies for photons using an electron sample. However, these differences have been checked and understood in detail for the CDF detector as well as in the detector simulations. These results have been used to do the cross section measurements as well as the gamma+1jet and higgs->gammagamma analyses. We should also point out that the efficiencies are only used when we are setting limits. As described later in the text (and this addresses a future point of the referee as well), we take into account the differences between electron and photons interaction dependencies when setting limits. Specifically, we have used the detector simulation to model the photon Et dependencies for the signal simulation and corrected this using the ratio of efficiencies (see Figure 6) as measured from electrons using similar samples of Z-> ee events from data and the Pythia Monte Carlo. Since these two efficiency measurements track each other well and the differences are well understood we use this correction as an overall correction factor on a per photon basis. There is very little systematic bias in the ratio of efficiencies which is the value used in the calculations. However, to be conservative we have taken the systematic uncertainty to be equal to half the range of the efficiencies (not the ratio of efficiencies which is smaller) which is about 5% and added it in quadrature with the statistical uncertainties. We should also point out that this error is small compared to all the other errors which are inherent in this analysis (luminosity, theoretical cross section etc.). Thus, we have concluded that the additional systematic errors due to the method of measuring the efficiency using electrons is small compared to the error in the measurement itself and so have not mentioned it explicitly. p 18 The efficiencies in the right-hand plots are made using about 1600 events with the full sample in the left hand plot being about a factor of 1.2 larger. The referee is quite correct in pointing out that the efficiencies are measured using the same samples and are therefore highly correlated. However, we should point out that the left-hand side plots are shown for illustration purposes only as the purity of the sample is lower. The systematic dependencies are in the individual efficiencies. Again, we point out that to account for these dependencies, we have used the detector simulation to model the Et dependencies for the signal simulation and used the ratio of efficiencies as an overall correction. There is very little systematic bias in the ratio of efficiencies which is the value used in the calculations and we have taken a conservative estimate of the systematic error. The 22 GeV choice is based on the trigger efficiencies, not the identification efficiencies which we think the detector simulation models well enough. Finally, to address the issue of picking an efficiency for the plots. Since the low threshold plots are flat we believe there is no further justification needed. For the high-threshold plots the only real variation is in the Z/gamma -> ee samples where there is less purity. In the purer samples (Z -> ee with a mass peak requirement) there really is very little variation outside the one low data point at low Et. Even this point is certainly consistent, within its errors as well as that of the stated uncertainty of the overall efficiency method, with the efficiency stated. We have added a sentence which makes it clear that the efficiencies are the average measure efficiency of the sample. Again, we point out that the only number used in a calculation anywhere in the paper is the correction factor which reflects figure 6, where the ratio is clearly flat as a function of Et. We believe no further justification is needed on this issue. p 20/21 Within statistics the efficiency of the lower threshold trigger is flat above 12 GeV. Where the statistics run out the high-threshold trigger kicks in. The referee is quite correct that as photons become more energetic more of their energy leaks into the hadronic calorimeter with the effect that they become less isolated at the trigger level. Thus, the isolation requirement is why the new trigger is used at 22 GeV and is why the efficiency for diphoton events to pass the trigger is degraded. However, we require isolated photons (i.e, photons which pass the offline cuts which mimic the trigger requirements) so the efficiency of photons which pass the offline cuts is not degraded and is consistent with 96%. p 22 In the paper we are trying to estimate the number of events we should see in the data from SM sources which would pass the specified cuts. The estimated rates are not for only real diphoton events. In some of the cases, the contributions from real diphoton events dominates, however I don't think the paper implies that this is always the case. As is explicit in the text, the methods in each case use the data (both real diphotons and fakes) and standard fake rate estimation methods. Specifically, they look at the jets within each event, multiply by the fake rate per jet and sum over jets. This is described in the text and the methods are discussed ad nauseum in references 13, 14 and 15. Thus, the rates include events with both real and fake photons (in appropriate weighting) in association with b's, leptons and/or photons. The rates are therefore not underestimated, nor do we feel the reader has been misled. p 27 Fixed. p 50/51 4.8.3 I think the confusion comes from the referee's desire for this section to be a discussion that it is not. The bottom line is that we cannot prove or disprove the isolated photon hypothesis, and putting a probability on this being a photon would be very misleading when we have no way to estimate the probability that the nearby track is related to it or not. Thus, we have done what we believe is the most appropriate thing which is to give the facts of the event as they relate to the isolated photon hypothesis, as well as study a sample of photons to see if they would look like the cluster we have. The answer, as discussed in the text, is indeed mixed. If this were an photon, and we aren't saying it isn't, it is a non-isolated one, and hence wouldn't pass photon standard identification criteria, or, alternatively, it's an unusual example of an isolated one in that it would not pass any reasonable set of photon cuts. Basically, on the face of it as it has all the identifying characteristics of a track nearby (VTX occupancy and an SVT track). However, there are extenuating circumstances which say that because this is a busy event there are reasons why there might be charged particles near it, either because this is part of a jet, or a stray particle from the underlying event (or different event) which might make this "isolated" photon non isolated. Alternatively, this could be not a single 'object', but a photon with a tau nearby. We believe that just stating the experimental facts is much better than quoting 'probabilities' on hypotheses that are dependent on the (unknown) larger sample. The referee is quite correct in saying that a photon need not be isolated to be a photon. However, if this is a photon produced as a part of a jet, then the source of the cluster should probably be referred to as a jet. The idea is that we are trying to uncover the true source of the event. If a photon were produced as part of the partonic process then it should, in general, be isolated. The referee is also correct in assessing that between 6% and 8% of phi space would have the VTX occupancy or an SVX track. The reader can judge for themselves whether they want to call the occurrence of a track near an isolated photon at the 6%-8% unlikely or very unlikely. Ultimately, we believe the referee had the correct response in that it is unclear whether the cluster in the plug calorimeter is due to a photon. However, we think the final sentence of the paragraph is the correct conclusion. We believe the reader should conclude, as we did, that we can neither prove nor disprove the photon hypothesis and that if this is an isolated photon (at the parton level), then it is an unusual example. We do not give a probability that this is a photon since we feel it would be misleading. We do not offer probabilities that any of the interpretations are correct, rather we discuss studies which tell us how likely an object is to produce the cluster we observed. We have added text at the beginning of the discussion to make this more clear. p 52 The referee is correct that if the cluster is due to a jet, then this would be an unusual example of a jet. The referee is also correct that the there is no reason why the jet interpretation cannot be the likeliest solution. However, there is no reason to believe that it IS the likeliest interpretation. It is probably the least interesting (in some peoples estimation) however this doesn't make it the most likely. We don't know anything about the ensemble of events which produced this event therefore we cannot say anything about what is most likely to have produced part of it. To use an example of the referee's logic, the same argument would predict that a second cluster produced in association with a 40 GeV electron from a Z is most likely a jet. To take it a step further, let us consider two other simple, and applicable, ensemble cases: Isolated high Et photons, and isolated high Et electrons. In the data there are about 325K events with at least one high Et isolated photon, 82K have at least one jet (~25%), vs. 12K with missing Et>25 GeV (~4%). All other objects are even less likely (electrons, muons, taus, b-jets etc). Thus, for this sample the most likely thing to find in addition to a single photon is indeed a jet. However, if we look at the case of isolated, high Et electrons the case is radically different. There are roughly 70K events with at least one high Et isolated electron, ~50K have missing Et (W production), ~10K have a jet and ~5K have a second electron (Z production). Thus, in this sample one is MOST likely to find an additional neutrino, and electrons are only half as likely as jets. The point is that without knowing what the sample is, one cannot assume anything about the production of the event and therefore the relative production probabilities of the extra objects in the event. What we know how to do as experimentalists (without an ensemble of events) is answer the question "if this were a jet/electron/tau/photon, what is the probability that it would pass a given set of cuts." This is what we have estimated, and we find that the probabilities of taus, jets and photons to pass the cuts are small and on the few percent level or less. Each interpretation requires this to be an unusual example of whatever the parent particle is. Whatever it is, it easily passes the standard electron cuts, and would easily fail any reasonable photon or tau cut. Because we don't know what type of object this is, and it easily passes the electron cuts, we have tried do to the most reasonable thing we could think of which is to estimate the backgrounds without requiring the object to be an electron, rather just that it pass the standard electron cuts which this object easily passes. p 55/56 We have chosen not to dwell on the 15% purity number because it doesn't tell us anything about this particular event. Again we do not have an ensemble of these events so we have no indication that it has the same purity as the full sample. For example, the gamma+gamma + electron and muon sample as far as we can tell seems to be pure photons (certainly not at the 15% although the statistics are bad). Furthermore, we have studied the photon clusters and there is no reason to believe they are not photons. As mentioned in the text they easily pass much more stringent photon cuts which would be far too inefficient to be used in a search. section 5, p 56/57 The statement of the referee is not correct. We have estimated the rate at which events (both real and fake) pass the photon, electron and missing Et requirements. The purity of the full diphoton sample is irrelevant here. What is relevant is the rate at which real and fake objects are produced in association with each other to contribute to the eeggmet production rate. We have made the introduction of the chapter clearer to reflect the fact that we have not estimated the rate at which real photons are part of the topology, rather that "objects which pass the photon cuts" are. We do not find it unusual for the purity of diphoton samples to have be different under different additional topology requirements. p 58 All the sources the referee have mentioned are already incorporated in estimates in Table 19. Before we go through the sources individually, we again point out that that events must pass all the selection criteria. Thus, one has to specify where all the objects come from be they real or fake. SM j+3 photon contributions. This source does not have intrinsic missing Et, and one of the photons must have passed the electron selection criteria. The rate at which this process produces 4 objects which pass the two electron and two photon cuts is covered as part of the estimation in the second row of the Table 19 to be 3E-5. The probability of finding fake missing et in such an event is 4E-3. Thus, we expect less than 1E-7 from this source. SM Wgammagamma + jet where the jet fakes a plug electron is covered in row 3. Unfortunately, the referee's calculation is incorrect because the referee has used some numbers incorrectly. More specifically, the referee has made a calculation where data already exists in that the referee have taken 6E4 W's and multiplied by 6E-4 for real photon production which predicts 36 Wgamma events in the data, there are only 4. The more correct version is what we have done which is take the 4 Wgamma events multiplied by the real+fake rate of an additional gamma (2E-3) and multiply by the rate of seeing a fake plug electron (2E-5). Thus the total contribution is 5E-8 which is smaller than the dominant backgrounds. e+3 jet is dominated by W+3 jet and is in the next to the bottom row of Table 19. There are 58000 events in the data with an object which passes the central electron cut and have Met>25 (note that this include real W's as well as sources which fake the electron and missing et part). The rate at which three addition jets are produced and fake the two photons and plug electron put the total rate at 2E-9. The rate for when central electrons are faked is even smaller and is not included in the table. Z+2 photon is covered in row 2, as it requires eegg production and fake missing et, and is less than or equal to 1E-7. ZZ+2 photon is not included explicitly as it is tiny, there is no fake component and the two electron candidates do not look like they are from a Z at the many sigma level. To make the numbers more concrete, the ratio of Sigma(WW->eenunu)/Sigma(ZZ->eenunu) is of order 20. The rate at which Z->ee with Mee>110 is less than 2% as opposed to WW which is about 50%. So, ZZ+2 photon should be at least a factor of 500 smaller than WWGammaGamma. In table 19, the tau and photon contributions have been added in implicitly in that they are required to pass the electron cuts. For example take the first in table 19. The 4 events in the data which pass the ee+missing et requirements include the contributions from taus and photons which pass the electron cuts. This includes contributions to both central and plug electrons. We have made this more explicit in the introductory paragraph of section 5. We understand the disappointment of the referee that we do not give relative probabilities on the origin of the cluster in the plug calorimeter. Ultimately, we must again point out that one cannot assess the probability of an object being due to any one source without knowing the source of the ensemble. Given that we have one unusual event we certainly do not know the source. What we know how to do as experimentalists is answer the question "if this were a jet/electron/tau/photon, what is the probability that it would pass a given set of cuts." This is what we have estimated as best we can given extenuating circumstances and the particularity of the event itself. We find that the probabilities of taus, jets and photons to pass the cuts are small: on the few percent level or less. The estimation methods are subtle and the reader is referred to more of the details in the references. We also understand the importance of including the fact that we do not know the origin of the cluster into our SM estimates of the rate at which the event is produced. This is what we have done. The important point to remember is that we have asked the question "what is the rate at which SM sources pass the given cuts." Our methods include jet, electron, tau and photon contributions in very conservative methods. The fact that the cluster would be a very unusual example of each of the hypotheses is included directly. p 61 Done. p 67 No. We mean comparable. The models compared are not identical. We use a full NLO calculation and uses all particle production. DZero uses only gaugino production. p 70 See earlier discussion about real photons as the source of em clusters. There is no justification for believing that the "fake electron" rate is well above 6E-8. p 75 [19] We have added the version numbers to the text. For the referee's edification, we used Pythia 5.710 and Spythia 2.08. p 77 [32] The missing et for the different vertices always rises and is approximately 68, 81 and 84 GeV for the vertices at Z=-8.9, -33.7 and -38.9 respectively. p 78 [39] Our statement is correct. The shape is unusual and we understand why it is unusual. p88 The factor of six comes from the 6 different ways the event could have come from any two of the 4 vertices. The event multiplicity is important as it gives us an indication of how many collisions the individual objects could have come from. There are only 2.5 expected vertices in this event so we have, perhaps, overestimated the expected contribution from this source. Considering it is by far the smallest type of source we are not concerned about it being overestimated.