New LHC results 2015: Tuesday Dec 15 - interesting diphoton excess

[mfb]

After a slow start, the LHC and its detectors worked nicely and collected a lot of data this year (~3.5/fb). While many analyses are still ongoing, both ATLAS and CMS will report several results on Tuesday 3 pm (CET)*. The presentations will probably appear here, a https://webcast.web.cern.ch/webcast/play.php?event=442432 will be available.

The collision energy increased from 8 TeV (in 2012) to 13 TeV, so completely new mass regions could be probed. There are rumors about possible announcements already, but I won't comment them - we will know more in two days. Certainly an interesting event for particle physics.ALICE and LHCb don't focus on new heavy particles, so they don't profit so much from the higher energy. They will need more data to improve their earlier measurements significantly.

*to save some conversion issues: this post has been posted Sunday 4:38 pm CET, if you set your local time zone in the forum properly the event will start 1:38 earlier than this post, 2 days afterwards.

 

[Vanadium 50]

The 2015 run just ended (8 minutes ago).

 

[mfb]

Both ATLAS and CMS presented many analyses of the 2015 dataset today. The energy of the LHC proton collisions increased from 8 to 13 TeV. All the production rates of known particles changed, so many studies measured those. In addition, the higher energy allows to search for even heavier particles than before. The number of collisions was lower than in 2012 (“run 1” together with smaller datasets from 2010-2011), but for potential heavy particles the increased energy is more important. Therefore, analyses mainly looked for particles heavier than 1-2 TeV. Nothing significant was found there. Somewhere else, however…:

The diphoton spectrum
Among many other analyses, both experiments studied the production of two photons (“diphotons”) and their invariant mass spectrum. Particles that decay to two photons can lead to a clear peak in this spectrum. The ##\pi^0## particle was seen in that way in 2009 (it was known for decades, but measuring it helped to calibrate the detectors), and it was one of the two main discovery modes for the Higgs boson in 2012. Experimentally, it is a very clean measurement: if you do something wrong, you reduce the experimental precision, but it is hard to get a wrong result.

The diphoton results
The two experiments independently analyzed their data, and found more events than expected at a mass of around 750-760 GeV. There is no known particle at this mass, and no other known process that could lead to such an effect, apart from statistical fluctuations. The overall significance of the peak is hard to evaluate and depends on the model used (see below) - this was easier with the Higgs discovery, as the Higgs properties had clear predictions. If you consider just the ATLAS result or just the CMS result, a statistical fluctuation is certainly possible (and expected somewhere given the large number of analyses). Two statistical fluctuations of that size at the same place in two independent experiments? Still possible, but it is getting interesting.

Don’t look here, look elsewhere
In those searches, the look-elsewhere-effect is important: a fluctuation at a given mass might be unlikely, but there are many mass points where a fluctuation can happen. Therefore, experiments usually give two significance numbers: a local significance ("what is the probability that we see so many events at this specific point?”) and a global one ("what is the probability that we see such an excess at some place in the tested range?”). CMS gives 2.6 local and <1.2 global significance, ATLAS quotes 3.6 sigma local and 1.9 sigma global significance for a narrow signal and 3.9 sigma local / 2.3 sigma global for a broader signal.
2.6 sigma correspond to a probability of 0.47% for a random fluctuation, 3.6 sigma corresponds to 0.016%, 3.9 sigma to 0.005%.

Comparison to run 1
A possible particle at 750-760 GeV should have been produced at the lower energy in 2012 as well. With the much larger number of collisions, it gives an important cross-check. Both CMS and ATLAS re-investigated the old studies to check the compatibility, and the results are compatible. Both ATLAS and CMS had a bit more events than expected, but the deviation was not significant enough to get more attention.
CMS made a full combination of run 1 and run 2 data, giving 3 sigma local and <1.7 sigma global significance.
If the excess comes from a new particle, it would be a small hint that it is probably produced by gluon-gluon fusion, as this gives a better compatibility between run 1 and new data for ATLAS. No full combination from ATLAS (yet?).

Possible interpretations
To summarize: ?
The peaks are unexpected. Diphoton spectra are mainly investigated for Higgs- and Graviton-like particles. A Higgs at that mass should have a broader peak (on the other hand, the SM-like Higgs is at 125 GeV so new particles can behave differently), and Gravitons at that mass should be produced much more frequently.
I’m sure some theoreticians are writing explanations right now what this could be...

Edit: Jester found some toy model, a heavy scalar that couples to vector-like quarks.

What comes next?
I guess the experiments will refine their analyses where possible, make even more cross-checks and investigate the run 1 dataset in more detail. We might get new results at the Moriond conferences in March. The numbers can change a bit, but I don’t expect the message to change significantly. I guess the diphoton spectrum will be one of the key analyses with the 2016 dataset. Collisions are expected to start end of April, by July the dataset size could be sufficient to have an impact, at the end of the year it should be much larger than the dataset of this year. It will either show that the excess seen here was not new physics, or conclusively prove the existence of something new.

What else happened?
Various limits were set, many of them better than in run 1. The excesses seen in run 1 were re-investigated with run 2 data, and while more data is needed to fully rule them out, no excess was clearly seen again. ATLAS sees an excess in Z plus missing transverse momentum in run 1 and 2, but CMS does not see it in either run, and the significance is not very high (3 and 2 sigma local significance, respectively, with many places to look).
CMS had a very high-energetic electron/positron event earlier this year, which was surprising. It stayed the most high-energetic event of this type for the whole year, and the probability to find one event of at least this energy is about 3-4%, so one event is not too surprising.

 

[ohwilleke]

This is definitely a big deal. It is really the first credible evidence of beyond the Standard Model physics at the LHC.

Decays to diphoton events generally imply that the spin must be an even integer (e.g. scalar spin-0 or tensor spin-2), and that the electric charge is zero.

Nothing at this stage rules out the possibility of this resonance actually being a composite particle of existing fundamental particles (e.g. an excited state of a scalar or tensor glueball or a tetraquark of four top quarks), although there aren't good candidates in the Standard Model for composite particles this heavy either. Another possibility suggested by Marco Frasca, is that there are excited states of the ordinary Standard Model Higgs boson. https://marcofrasca.wordpress.com/2015/03/14/is-higgs-alone/

Most attention, of course, focuses on the heavily theorized possibility of a two-Higgs doublet model which includes a heavy scalar Higgs boson and a pseudo-scalar Higgs boson (A) in addition to a pair of charged Higgs bosons, which is found in all supersymmetry models and many non-supersymmetry models.

 

[ChrisVer]

If I recall well, in the diphoton decay wasn't it that 1 plot was shown with the "interesting" excess (of a glumpsy analysis) and three plots under it showing nothing important (more deeply studied)?
Unfortunately I had some problems and lost the very beginning of ATLAS presentation [and the whole CMS]...so it took me some time to "fit myself in what I was listening to"...

 

[mfb]

The beginning of the ATLAS presentation just had general things - data taking conditions, some performance plots and so on.

If I recall well, in the diphoton decay wasn't it that 1 plot was shown with the "interesting" excess (of a glumpsy analysis) and three plots under it showing nothing important (more deeply studied)?

No.

The conference notes are now available for ATLAS (diphotons here) and for CMS (diphotons here).

CMS has more plots, but a lower significance (those two could be related).

 

[ChrisVer]

Interesting, thanks @mfb
also do you know why the CMS and ATLAS have different background hierarchies on the same analysis?
Is it a luminosity thing or different MCs? I was looking at the parts that interest me for example (W') and there for ATLAS the top-bkg is above QCD dijet, while for CMS it's the opposite : QCD>tt.

 

[Vanadium 50]

I am going to be somewhat contrarian.

The Look Elsewhere Effect (or as it used to be called, a trials factor) is absolutely vital in understanding significance. If you toss a coin time times and ge ten heads in a row, you have a right to be surprised. If you toss ten thousand and you get a streak of ten, you don't. The diphoton search isn't really a search - it's really about 150 searches at 150 different masses, anyone of which (or all!) could turn up an excess. So the relevant number is what the experiments emphasized, the global significance, which is about 1/30 for ATLAS and 1/9 for CMS. Considering both experiments showed more than 30 results, it's hard to get too excited about this.

Jester found some toy model, a heavy scalar that couples to vector-like quarks.

Both experiments made the point that the global significance was the one to look at. I see Jester completely ignored this.

This is definitely a big deal. It is really the first credible evidence of beyond the Standard Model physics at the LHC.

I strongly disagree with this. At this stage, one cannot separate these bumps from a statistical fluctuation, and as these are both less than 2 sigma, they are far below any sensible threshold to even get excited about, much less claim evidence for BSM physics.

 

[Vanadium 50]

was looking at the parts that interest me for example (W') and there for ATLAS the top-bkg is above QCD dijet, while for CMS it's the opposite : QCD>tt.

Where do you see this?

 

[ChrisVer]

Where do you see this?

http://cms-results.web.cern.ch/cms-results/public-results/preliminary-results/EXO-15-006/index.html
https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/CONFNOTES/ATLAS-CONF-2015-063/

 

[mfb]

What is a "background hierarchy"? The composition of the background? That depends on the chosen selection. It also depends on the detector (identification efficiencies, fake rates, ...)
CMS has just 2/3 to 3/4 the size of the ATLAS dataset due to problems with the magnet, but that scales everything in the same way.

Edit: wait, you are just talking about the plotting? Well, it does not matter what you plot on top. Usually smaller contributions are shown at the bottom in logarithmic plots as this makes them easier to see, but what do you do if the relative size changes in the plotting range?

 

[ChrisVer]

I strongly disagree with this. At this stage, one cannot separate these bumps from a statistical fluctuation, and as these are both less than 2 sigma, they are far below any sensible threshold to even get excited about, much less claim evidence for BSM physics.

I guess you are right... at least I don't see any other channel that would show the same peak at the same mass...

 

[JorisL]

I strongly disagree with this. At this stage, one cannot separate these bumps from a statistical fluctuation, and as these are both less than 2 sigma, they are far below any sensible threshold to even get excited about, much less claim evidence for BSM physics.

This is also what I got from the announcements.

But I'm by no means very well equipped to make definite calls about this.

One question I have is how much of their (relevant) data they analysed so far.
I don't believe I heard that piece of information. I do think I heard them talking about the total amount of data they recorded.

Edit:
It was nice to hear some physics talk during my break though.
Psychological approaches to the learning proces of humans aren't fun.

 

[ChrisVer]

What is a "background hierarchy"?

a word I just created, that's why I tried to explain what I meant with the W' example
In those cases that tt-evts>QCD-evts vs the tt-evts<QCD-evts for the different collaborations...

 

[ChrisVer]

Edit: wait, you are just talking about the plotting? Well, it does not matter what you plot on top. Usually smaller contributions are shown at the bottom in logarithmic plots as this makes them easier to see, but what do you do if the relative size changes in the plotting range?

The order of plotting makes it easier to see because what is on the top adds the most of events... if you did the same plots with W let's say on the bottom then the rest contributions would not even be obvious... So if you plot tops bellow QCD you do that because QCD is the dominant one vs the tops... for example the 1st bin of atlas you have ~900 multijets, while you have ~1100 tops. In cms you have ~200 tops vs approx ~500 qcd [hard to read that logarithmic scale]... I don't compare the two sizes of atlas-cms but the relative ones for each (QCD/tt)...

 

[mfb]

I am going to be somewhat contrarian.

The Look Elsewhere Effect (or as it used to be called, a trials factor) is absolutely vital in understanding significance. If you toss a coin time times and ge ten heads in a row, you have a right to be surprised. If you toss ten thousand and you get a streak of ten, you don't. The diphoton search isn't really a search - it's really about 150 searches at 150 different masses, anyone of which (or all!) could turn up an excess. So the relevant number is what the experiments emphasized, the global significance, which is about 1/30 for ATLAS and 1/9 for CMS. Considering both experiments showed more than 30 results, it's hard to get too excited about this.

Be careful to not account for the LEE twice.
There are two approximate approaches to estimate a combined significance:
(1) we can take the 1.9 sigma (or 2.3 if you prefer that number) global significance from ATLAS and ask "what is the probability to see something in CMS at exactly the same spot?" - where we get 2.6 sigma (3.0 with run 1).
(2) we can take the global significance of CMS (which is small) and ask "what is the probability to see something in ATLAS at exactly the same spot?" - where we get 3.6 sigma.
Either way, the probability to see two excesses that size at the same spot is small, and a combined global significance is above 3.5 sigma. Correlated systematics could reduce this number a bit but I don't think they are significant.
We still have an unaccounted LEE from the ~25 analyses done per experiment, of course.

 

[ohwilleke]

I am going to be somewhat contrarian.

As mfb cogently explains, the Look Elsewhere Effect (or as it used to be called, a trials factor) is absolutely vital in understanding significance. If you toss a coin time times and ge ten heads in a row, you have a right to be surprised. If you toss ten thousand and you get a streak of ten, you don't. The diphoton search isn't really a search - it's really about 150 searches at 150 different masses, anyone of which (or all!) could turn up an excess. So the relevant number is what the experiments emphasized, the global significance, which is about 1/30 for ATLAS and 1/9 for CMS. Considering both experiments showed more than 30 results, it's hard to get too excited about this. . . .
I strongly disagree with this. At this stage, one cannot separate these bumps from a statistical fluctuation, and as these are both less than 2 sigma, they are far below any sensible threshold to even get excited about, much less claim evidence for BSM physics.

The reduction in significance due to the Look Elsewhere Effect is significantly reduced by the fact that the result is replicated in two independent experiments. The odds of the same anomaly showing up in the same place is vastly lower than the odds of either experiment producing an anomaly like this on their own. There is still a look elsewhere effect ding for the combined result, but instead of greatly suppressing the significance of the individual findings, it only modestly reduces it, because the likelihood of a random replication of the same bump in two experiments is much lower than the likelihood of a random bump in a single experiment. Even considering the look elsewhere effect that removes the likelihood of a random replication of the same bump in two experiments, the combined significance of the result is in excess of 3 sigma. It is not correct statistically to calculate a look elsewhere effect separately for each experiment and then to combine the reduced probabilities from the two experiments to estimate significance.

This finding certainly doesn't meet the criterion of five sigma for a "discovery" but it is the most significant BSM data that the LHC has identified to date (assuming that no Standard Model explanation can be found for the resonance).

 

[Orodruin]

This is definitely a big deal. It is really the first credible evidence of beyond the Standard Model physics at the LHC.

Now let's be careful with what we call evidence ... Generally physicists do not like to quote evidence before 3 sigma (including the look elsewhere effect). Of course, it is possible that a combination of CMS and ATLAS would give that, but that will not be known until they make such an analysis.

Also, let us be careful with what we call behond the standard model. There are already credible evidence for this. Apart from the fact that gravity technically is BSM, this year's Nobel prize is too.

Even considering the look elsewhere effect that removes the likelihood of a random replication of the same bump in two experiments, the combined significan

Well, you really cannot tell without doing the combined analysis. Simply taking the local significance at the best fit of the other experiment is naive at best. Of course it helps that the locations are the same, but you still need to do the combined analysis in order to claim evidence.

 

[Vanadium 50]

While I agree that you can set up (a posteriori, of course) a search where you let one experiment define a window and then use the data from the other, I don't think that completely eliminates the trials factor. Several people have written the words "exact same mass", but there is no evidence of that. CMS says 760 GeV, and ATLAS says 750 GeV. Now, you can say "well, that's close enough", but that opens up the question how close is close enough, and I would estimate that gives you a trials factor of maybe 3 or 4. Not 150, but not 1 either.

Perhaps more importantly, though, if you are going to look at the combination at 13 TeV, you also need to look at the combination at 8 TeV. And now the story "the old data doesn't quite exclude the new data" becomes more problematic.

 

[mfb]

We don't have a proper combination, sure. The ATLAS mass resolution was said to be 6 GeV I think, I guess the CMS one is similar. Especially if we have a non-narrow width, the 10 GeV difference is quite small. I don't want to start scanning the p-value profiles now, and I don't think we get a combination as 2016 is not far away... we'll see what happens with more data.

8 TeV data from CMS increases the significance. I don't see the combination in the conference note - I would be interested in how they combined it, as the cross-section ratio depends on the production process.
Not sure about the ATLAS result. We know it is a bit below the run 2 result, but that does not necessarily have to reduce the significance.

 

[MathematicalPhysicist]

I am going to be somewhat contrarian.

The Look Elsewhere Effect (or as it used to be called, a trials factor) is absolutely vital in understanding significance. If you toss a coin time times and ge ten heads in a row, you have a right to be surprised. If you toss ten thousand and you get a streak of ten, you don't. The diphoton search isn't really a search - it's really about 150 searches at 150 different masses, anyone of which (or all!) could turn up an excess. So the relevant number is what the experiments emphasized, the global significance, which is about 1/30 for ATLAS and 1/9 for CMS. Considering both experiments showed more than 30 results, it's hard to get too excited about this.
Both experiments made the point that the global significance was the one to look at. I see Jester completely ignored this.
I strongly disagree with this. At this stage, one cannot separate these bumps from a statistical fluctuation, and as these are both less than 2 sigma, they are far below any sensible threshold to even get excited about, much less claim evidence for BSM physics.

If you toss a coin 10 times with a fair coin any streak what so ever is likely even 10 heads. Sure if you keep tossing then also every streak is likely.

But then again you might as well get in every toss in 10,000 tosses a head, but in the 10,001 toss tails then what?

I don't understnad what there is to being surprised that after ten tosses you get a streak of 10 heads if the coin is fair, every sequence of heads and tails is equally likely to occur.

Your puzzlement reminds mean the puzzlement of when I played backgammon with a friend that I needed to get
the two dice to be 3 and got it.

 

[Orodruin]

I don't understnad what there is to being surprised that after ten tosses you get a streak of 10 heads if the coin is fair, every sequence of heads and tails is equally likely to occur.

Because this happens only one in ##2^{10}## times. Granted, any other ordered series of outcomes also has that probability, but this is likely not what you are comparing with.

Your puzzlement reminds mean the puzzlement of when I played backgammon with a friend that I needed to get
the two dice to be 3 and got it.

Well, it is a 1 in 36 and the fact that it was the exact result you needed to get.

 

[MathematicalPhysicist]

Because this happens only one in ##2^{10}## times. Granted, any other ordered series of outcomes also has that probability, but this is likely not what you are comparing with.Well, it is a 1 in 36 and the fact that it was the exact result you needed to get.

So what? is it time to believe in ghosts, devils, fairies and santa claus?

I am just saying that statistically these things happen, and quite often (if we increase the measurements).

 

[Orodruin]

So what? is it time to believe in ghosts, devils, fairies and santa claus?

I honestly do not see how you can read that into my post.

I am just saying that statistically these things happen, and quite often (if we increase the measurements).

Which was the entire point of V50's post, which you were seemingly arguing against!

 

[jerromyjon]

I am just saying that statistically these things happen, and quite often (if we increase the measurements).

Just that it coincides between experiments or we'd have never heard about it.
If it is new physics what could it be? Any wild guesses what might fit?

 

[ChrisVer]

Which was the entire point of V50's post, which you were seemingly arguing against!

not really... from V50's post I got that he said if you toss it 10 times and get 10 heads that will be surprising... it will not be surprising however if you toss it 1000 times and get 10 sucessive heads.

The probabilities are not the same...
The first
[itex] P(n=10, k=10) = \frac{f(n=10,k=10)}{\sum_{i=0}^{10} f(n=10,i)}\approx 0.001[/itex]
While for ##n=1000## you can work finely with a normal distribution and get that the probability for x>10 is quite large (almost 1). I have some problem figuring out how to add the "successive" results to that

 

[Orodruin]

not really... from V50's post I got that he said if you toss it 10 times and get 10 heads that will be surprising... it will not be surprising however if you toss it 1000 times and get 10 sucessive heads.

Which was exactly his point! LHC has tossed a lot of coins - which is why he is saying "it is hard to be excited about this".

 

[MathematicalPhysicist]

I am just saying, that even if you tossed only 10 times this streak of 10 heads may still occur which I don't understnad why does it surprise anyone.

I mean if you were to restrict somehow this streak to occur then I'd be surprised.

Even if you stop at 10 tosses, such a scenario of 10 heads is still possible why should it surprise anyone?
Just because you didn't see such a streak before doesn't mean it's not possible, like the streak of Golden-State in basketball, theoretically they can be losseless (if the games aren't fixed).

 

[MathematicalPhysicist]

not really... from V50's post I got that he said if you toss it 10 times and get 10 heads that will be surprising... it will not be surprising however if you toss it 1000 times and get 10 sucessive heads.

The probabilities are not the same...
The first
[itex] P(n=10, k=10) = \frac{f(n=10,k=10)}{\sum_{i=0}^{10} f(n=10,i)}\approx 0.001[/itex]
While for ##n=1000## you can work finely with a normal distribution and get that the probability for x>10 is quite large (almost 1). I have some problem figuring out how to add the "successive" results to that

Each sequence of possible outcomes has the same liklihood to occur, which is (0.5)^10.

This is why I say I cannot be suprised by ten streak of heads out of 10 tosses, since you might as well be suprised also of any other sequence of heads and tails out of ten tosses.

 

[ChrisVer]

This is why I say I cannot be suprised by ten streak of heads out of 10 tosses, since you might as well be suprised also of any other sequence of heads and tails out of ten tosses.

Not really. In fact it's obvious that if your coin is fair, you expect a 50-50 out of it. If you get 100% success on H, then you should study what's the possibilities for this to be a real outcome and you got it due to some statistical fluctuations, else conclude that your coin is not fair.

The (0.5)^10 works for 10-trials. It doesn't for 1000. What's the likelihood to get 10 successive Heads when you flip the coin 1000 times? If for example I would say that for the 10 successes I would need (.5)^10, then I should also count the "multiplicity" of those successes within 1000 trials.
In other words the HHHHHHHHHHT , THHHHHHHHHH both belong to the 10 successive heads in 11 trials. This grows I think pretty fast, so that's why 10 successive Hs in 1000 trials can be quiet probable..

 

[mfb]

Even if you stop at 10 tosses, such a scenario of 10 heads is still possible why should it surprise anyone?

It is possible that a meteorite will hit your house tomorrow. Would you be surprised? I'm quite sure! Would you be surprised if a meteorite would hit some house somewhere in the world? While this is still a rare event: probably not.
Where is the difference? You have categories of different size: "my house", "some random other house".

We have the same with the coin: We expect it to be fair, but we also consider the options "the coin gives head all the time" and "the coin gives tails all the time". Getting 10 times head in 10 attempts strongly favors the (previously unlikely) "all head" hypothesis compared to the fair coin.
A result of HTTHTHHHTHTHTTTHHTTTH is not surprising normally. If I predict exactly this pattern in advance, on the other hand, it stands out, and if we get exactly this pattern it is surprising.

There is a massive number of options how the thousands of events in the analyses can be distributed, but only a few unlikely of them give some peak-like structure. It is surprising to find one of those.

 

[Orodruin]

Even if you stop at 10 tosses, such a scenario of 10 heads is still possible why should it surprise anyone?

Because it is a very extreme result which would happen only one out of 2^10 times. Managing to get this result with only 10 tosses is very unlikely to happen. Sure, it could happen, but it is not likely to. It is much much more likely to obtain 5 heads and 5 tails in any given set if 10 tosses.

Let me ask you this: Would you take an even money bet where you win if a coin lands face up 10 times in a row? If so I have a betting proposal for you. I will even give you twice as good odds!

 

[MathematicalPhysicist]

It is possible that a meteorite will hit your house tomorrow. Would you be surprised? I'm quite sure! Would you be surprised if a meteorite would hit some house somewhere in the world? While this is still a rare event: probably not.
Where is the difference? You have categories of different size: "my house", "some random other house".

We have the same with the coin: We expect it to be fair, but we also consider the options "the coin gives head all the time" and "the coin gives tails all the time". Getting 10 times head in 10 attempts strongly favors the (previously unlikely) "all head" hypothesis compared to the fair coin.
A result of HTTHTHHHTHTHTTTHHTTTH is not surprising normally. If I predict exactly this pattern in advance, on the other hand, it stands out, and if we get exactly this pattern it is surprising.

There is a massive number of options how the thousands of events in the analyses can be distributed, but only a few unlikely of them give some peak-like structure. It is surprising to find one of those.

I must say that I still don't see the reason to be surprised.
Do you say that in a fair coin after tossing 10 times, the outcome of always heads is less likely than any other outcome?
Show me how do you calculate your probabilities.
What I am trying to say that after tossing ten times a fair coin we have (0.5)^10 chance of getting any sequence of outcomes.
The liklihood of getting heads 1000 times in a row is less likely than 10 times in a row, but it doesn't mean it can never happen.

As for the metorite, I wouldn't be suprised if it should hit something on Earth then it can happen to hit my house, it depends on its trajectory though.

 

[Orodruin]

Do you say that in a fair coin after tossing 10 times, the outcome of always heads is less likely than any other outcome?

This depends on your definition of "any other outcome". Of course, taking "any other outcome" as the complement of that one outcome, it is obvious that it is less likely. Taken as any other fixed series, of course not.

In physics you will often be dealing with macroscopic states where the observables are composed of several of the microstates. Even if each microstate is equally likely, the macrostate containing more microstates will be more likely and a macrostate containing only one microstate will be very unlikely. This is the situation here, as it was when you rolled 3-3 - the macro states are "you win" and "you lose". The "you lose" state has a 1/36 probability and you should therefore be more surprised if you win than if you lose based on the previous knowledge.

 

[nikkkom]

Please stop this bickering about how much exactly are you (not) surprised by the slight diphoton excess. It's not productive.