Search for boosted diphoton resonances in the 10 to 70 GeV mass range using 138 fb-1 of 13 TeV pp collisions with the ATLAS detector

summary

The gist

A search was conducted for diphoton resonances in a mass range spanning from 10 to 70 GeV utilizing proton–proton collision data collected by the ATLAS experiment at the Large Hadron Collider (LHC).

In short

This search used ATLAS data from 13 TeV proton-proton collisions to find new particles decaying into diphotons between 10 and 70 GeV. The analysis found no significant evidence for a new particle, with global significance only reaching $1.5 extsigma, consistent with Standard Model expectations. Upper limits on the production cross-section were set at 4 to 17 fb.

Key concepts

Diphoton Resonance Search
The study looked for narrow peaks in the invariant mass spectrum of pairs of photons ($ ext{X} ightarrow ext{γγ}$). These peaks suggest a new, short-lived particle decaying into two high-energy photons. The goal is to find evidence for physics beyond the Standard Model.
Invariant Mass Spectrum
This is a plot showing the distribution of the combined energy and momentum of two particles, in this case, two photons. By analyzing this spectrum, researchers look for sharp 'resonances'—sudden spikes that could indicate a specific mass for a new particle.
Axion-Like Particle (ALP) Parameter Space
The search results are translated into parameters describing a hypothetical new particle, the ALP. This space is defined by its mass ($m_a$) and its coupling strength to gluons and photons ($f_a$). This method allows physicists to test specific theoretical models for new physics.
Background Modeling
The analysis had to carefully model all known Standard Model processes that mimic the signal, such as continuum photon production. A sophisticated approach using analytic functions and Gaussian Processes was used to accurately describe these complex background shapes.

Terminology used across episodes

This episode discusses

The paper

Search for boosted diphoton resonances in the 10 to 70 GeV mass range using 138 fb-1 of 13 TeV pp collisions with the ATLAS detector · Read on arXiv

EUROPEAN ORGANISATION FOR NUCLEAR RESEARCH (CERN)

A search for diphoton resonances in the mass range between 10 and 70 GeV with the ATLAS experiment at the Large Hadron Collider (LHC) is presented. The analysis is based on pp collision data corresponding to an integrated luminosity of 138 fb-1 at a centre-of-mass energy of 13 TeV recorded from 2015 to 2018. Previous searches for diphoton resonances at the LHC have explored masses down to 65 GeV, finding no evidence of new particles. This search exploits the particular kinematics of events with pairs of closely spaced photons reconstructed in the detector, allowing examination of invariant masses down to 10 GeV. The presented strategy covers a region previously unexplored at hadron colliders because of the experimental challenges of recording low-energy photons and estimating the backgrounds. No significant excess is observed and the reported limits provide the strongest bound on promptly decaying axion-like particles coupling to gluons and photons for masses between 10 and 70 GeV.

DOI: 10.1007/JHEP07(2023)155

Transcript

Introduction to the show: ident: High Energy Physics: Experiment. The day in high-energy experiment: detectors, searches and limits.

Kai: We're looking at the paper: "Search for boosted diphoton resonances in the 10 to 70 GeV mass range using 138 fb-1 of 13 TeV pp collisions with the ATLAS detector".

Mira: Here's the gist of the paper. A search was conducted for diphoton resonances in a mass range spanning from 10 to 70 GeV utilizing proton–proton collision data collected by the ATLAS experiment at the Large Hadron Collider (LHC).

Kai: This analysis employed an integrated luminosity of 138 fb-1 at a center-of-mass energy of 13 TeV, spanning the period from 2015 to 2018.

Mira: The methodology focused on exploiting the specific kinematics of events featuring pairs of closely spaced photons reconstructed within the detector, enabling an examination of invariant masses down to 10 GeV.

Kai: The search was designed to probe for new particles that might decay into diphotons, specifically focusing on axion-like particles coupling to gluons and photons.

Title and authors: Kai: So we're looking at this paper today, "Search for boosted diphoton resonances in the ten to seventy GeV mass range using one hundred thirty-eight fb-one of thirteen TeV pp collisions with the ATLAS detector <ref:2211.04172#pg1,Search for boosted diphoton resonances in the 10 to 70 GeV mass>." It’s a search for new particles called diphoton resonances in that specific mass window.

Mira: Exactly. The authors are focusing on a region where previous searches haven't looked as hard, specifically between ten and seventy giga electron volts, using data from the ATLAS experiment at the Large Hadron Collider.

Lev: It’s interesting because they are targeting new physics scenarios, like axion-like particles that couple to gluons and photons. That coupling is what makes them abundant in proton-proton collisions.

Kai: Right, so what's the main experimental setup they are using here? They aren't just looking at any photon pair; they are exploiting specific kinematics where those two photons are very close together inside the detector.

Mira: That’s key. This allows them to reconstruct invariant masses down to ten giga electron volts, which is what makes this search sensitive in that lower mass region, especially near ten giga electron volts.

Lev: From an engineering standpoint, reconstructing those low-energy photons and estimating the backgrounds is tough because of the experimental challenges at the LHC.

Kai: That’s where they tackled a big problem: managing that challenging background shape, particularly in the lower mass ranges below twenty giga electron volts. They used a transverse momentum for photon pairs greater than fifty giga electron volts to help flatten that background distribution, making it easier to describe with simple analytic functions.

Mira: And when they do fit the spectrum, they use a model made of analytic functions to describe both the signal and the background components when looking for those narrow resonances. They even used a combination of two analytic pieces to model both the turn-on shape and the smoothly decreasing part of that spectrum.

Lev: That sounds like a lot of modeling work just to get a handle on what’s actually happening in the data. How does that modeling choice affect what we're actually looking for?

Kai: The primary parameter they extract from the likelihood fit is this fiducial production cross-section times branching ratio, which is written as sigma times B of X into diphotons. This is related to how many signal events they fit in the data by dividing by a correction factor, C X, which accounts for how well the signal actually gets detected.

Title and authors: Mira: That correction factor C X varies between zero point two and zero point five depending on the resonance mass, so it introduces some uncertainty in that final cross-section value. They present these limits both as a function of the mass and also recast them in terms of an axion-like particle, defined by its mass m a and its decay constant f a.

Lev: Recasting it into the ALP parameter space is helpful because it connects this search to a broader theory about these new particles. It lets us see what kind of coupling strength we're constraining, which is important for theorists.

Kai: The results showed that the data were generally consistent with the Standard Model background expectations across most of that mass range they looked at. However, they found a local significance of three point one sigma for a hypothesized resonance mass at nineteen point four giga electron volts.

Mira: But then they noted that the global significance only reached one point five sigma, which is what leads them to conclude that there are no significant deviations from the Standard Model found so far.

Lev: So, even with a local bump suggested by the data, the overall picture remains consistent with what we expect from standard processes. That's a cautious result when you're hunting for new physics.

Kai: The resulting upper limits they put on that cross-section times branching ratio range from four to seventeen femtobarns, and those variations are mostly due to statistical fluctuations and the uncertainties in their background modeling. They also used Gaussian Processes to smooth out the simulated background-only sample, which helps mitigate some of those uncertainties.

Mira: It’s important for us to remember that these limits have some wiggle room because of those modeling uncertainties, even with the smoothing techniques they employed. That wiggle room is something we need to keep in mind when interpreting these bounds on axion-like particles.

Lev: From an error correction viewpoint, if you were trying to run this analysis on real hardware, you’d have to account for the overhead of getting those photon measurements right at that level of precision. The modeling uncertainty here is a bit like needing a very stable cryostat just to measure the temperature accurately enough.

Kai: Exactly, because the method they used, relying on those analytic functions and the Gaussian Processes, is a way they handled some of those statistical fluctuations without having to rely on just one simple fit across the whole spectrum.

Title and authors: Mira: And that approach of using two analytic pieces for the background shape—one for the turn-on region and one for the smoothly decreasing part—is what makes this analysis more complex than some previous diphoton resonance searches two twenty-six <ref:2211.04172#pg3>. It shows how much better they had to get at modeling these low-mass features.

Lev: So, while they found no significant deviation, the fact that they had to use such a detailed background model tells us a lot about the difficulty of isolating a signal when you're looking for something very narrow in a noisy environment.

Kai: Before we wrap up this look at "Search for boosted diphoton resonances in the ten to seventy GeV mass range using one hundred thirty-eight fb-one of thirteen TeV pp collisions with the ATLAS detector," I just want to mention that they were really pushing the experimental limits on where these searches can go, extending coverage down to ten giga electron volts <ref:2211.04172#pg1,Search for boosted diphoton resonances in the 10 to 70 GeV mass>.

Mira: And they did this by focusing on those specific kinematic regions and improving how they describe the background shape using those two analytic pieces, which is a pretty detailed description of what's happening in that spectrum.

Lev: For real hardware testing, if we wanted to push beyond these limits, we'd need to address the systematic uncertainties tied to trigger efficiency turn-on more directly rather than relying solely on those smooth analytic fits.

Kai: So for the final word, this paper shows that while no significant deviations from the Standard Model were found in this mass range, they have put some of the strongest bounds on these new diphoton resonances yet.

Mira: It’s a good piece of work because it addresses a previously unexplored region and uses sophisticated modeling to constrain axion-like particles in a way that connects directly to particle physics theory.

Lev: It’s valuable for error correction research because it shows how detailed background modeling is absolutely necessary when you're looking for narrow features in high-energy data.

Kai: That's all we have time for this one, focusing on the search performed in "Search for boosted diphoton resonances in the ten to seventy GeV mass range using one hundred thirty-eight fb-one of thirteen TeV pp collisions with the ATLAS detector <ref:2211.04172#pg1,Search for boosted diphoton resonances in the 10 to 70 GeV mass>."

The paper's summary: Kai: So we just looked at how they set up this search for diphoton resonances between ten giga electron volts and seventy giga electron volts at the Large Hadron Collider using data from a one hundred thirty eight femtobarns of proton-proton collisions.

Mira: That’s right. The core thing they did was focus on exploiting those specific kinematics where two photons are very close together inside the detector to look for these new particles, especially axion-like particles coupling to gluons and photons.

Lev: From an error correction standpoint, that means they're looking at a region where the signal should be relatively narrow in mass, which makes it hard to separate from the continuum background noise.

Kai: Exactly. And when they tried to handle that challenging background shape below twenty giga electron volts, they used a specific technique involving requiring those photon pairs to have a transverse momentum greater than fifty giga electron volts to flatten things out.

Mira: That helps them describe the shape using analytic functions, and it’s interesting because they had to use two different analytic pieces—one for the turn-on region and one for the smoothly decreasing part—to model that background accurately.

Lev: Modeling like that sounds heavy on assumptions, so I wonder how much confidence we should really put in those two pieces when you're trying to pin down a new physics signal.

Kai: The main thing they pull out of all that fitting is this fiducial production cross-section times branching ratio, which they relate directly to the number of signal events they fit in the data by dividing by a correction factor.

Mira: That correction factor changes depending on the resonance mass, so it introduces some uncertainty into that final cross-section value. They also recast these limits into a different space defined by an axion-like particle's mass and its decay constant, which is important for theorists to see what kind of coupling strength they are constraining.

Lev: Recasting it into the mass and the decay constant space is smart because it gives physicists a clearer picture of what this measurement actually tells them about the underlying theory.

Kai: The results showed that most of their data was just consistent with what we expect from standard model background processes across most of that mass range they looked at.

Mira: But they did find a local significance, three point one sigma, for a hypothesized resonance mass around nineteen point four giga electron volts.

Lev: A three point one sigma is suggestive, but it doesn't automatically mean we found something new physics-related; it just means the data has a peak there that might be statistical fluctuation.

Kai: So they concluded that the global significance only hit one point five sigma, meaning no significant deviation from the standard model was found overall.

Mira: The final upper limits they set on that production cross-section times branching ratio range from four to seventeen femtobarns, and those variations are mostly due to statistical fluctuations and uncertainties in their background modeling.

Lev: They used Gaussian Processes to smooth out the background-only sample, which helps reduce those statistical fluctuations, but I wonder if that smoothing hides some real features we might be missing.

Kai: It's a good point about the smoothing; it helps manage the noise, but it’s still just a template based on what they expect the background to look like.

Mira: This search is valuable because it covers a region where previous searches haven't looked as hard, extending coverage down to ten giga electron volts, which is crucial for constraining these axion-like particles.

Lev: So while they didn't find a clear signal, they’ve put some of the strongest upper limits on these new diphoton resonances yet in that specific mass range.

Kai: And it shows how detailed background modeling has to be when you're looking for something very narrow in high-energy data.

Mira: It’s a good piece of work because it addresses a previously unexplored region and uses sophisticated modeling to constrain axion-like particles in a way that connects directly to particle physics theory.

Lev: For real hardware testing, if we wanted to push beyond these limits, we'd need to address the systematic uncertainties tied to trigger efficiency turn-on more directly than just relying on those smooth analytic fits.

The paper's improvements: Tom: So we're talking now about how they improved their search for those diphoton resonances between ten giga electron volts and seventy giga electron volts in this paper.

Kai: The authors suggest a few specific upgrades to make the search more robust, especially concerning how they model that signal shape.

Mira: They propose using a double-sided Crystal Ball function instead of whatever modeling they used before, which involves a Gaussian core with power-law tails whose parameters evolve linearly with respect to the resonance mass.

Lev: That sounds like a lot of work because you have to define those power-law tails precisely so you don't introduce new biases into the fit when you're trying to find that tiny signal.

Kai: Right, it’s about getting a better description of what a real signal might look like in the mass distribution, even if it’s just theoretical input guiding the fit.

Mira: Then they also enhanced their background estimation by using a data-driven method where they parameterize the continuum shape with an analytic function built from two pieces: one for the turn-on region and one for the smoothly decreasing part.

Lev: That two-piece approach sounds like it’s trying to capture both the sharp start and the slow fall of that background, which is where I think a lot of real hardware overhead comes into play.

Kai: They also introduced Gaussian Processes to mitigate background modeling uncertainty by fitting a Gaussian Process to the representative background sample histogram to get a smoothed background template.

Mira: That means instead of fitting the raw data directly, they’re using this smoothed template for their signal-plus-background fits, which should reduce those statistical fluctuations in the final result.

Lev: Using that smoothing is smart because it reduces the noise in the measurement record before you even start fitting for new physics.

Kai: So these improvements are all about making sure that when you look at those limits, they aren't just artifacts of how we chose to model the background shape.

Mira: The implications here is that by using these more sophisticated tools, they’ve pushed the limits of what kind of constraints we can put on axion-like particles in this specific diphoton channel.

Lev: It shows that even with limited data and a complex background, you need those advanced techniques to get meaningful results instead of just noise.

Kai: These tools are pushing the experimental limits on where these searches can go, extending coverage down to ten giga electron volts in a way that was much more rigorous.

Mira: It’s a good piece of work because it addresses a previously unexplored region and uses these advanced modeling techniques to constrain axion-like particles in a way that connects directly to particle physics theory.

Lev: We need those kinds of detailed models if we’re going to get any meaningful signal out of the data, even with just upper limits.

Conclusion: Kai: So we’ve gone through the whole thing on "Search for boosted diphoton resonances in the ten to seventy GeV mass range using one hundred thirty-eight fb-one of thirteen TeV pp collisions with the ATLAS detector." It really boils down to them trying to find these new particles, like axion-like particles, by looking at photon pairs that are very close together.

Mira: Exactly. They used a specific method—requiring a transverse momentum greater than fifty giga electron volts—to make that background shape easier to handle when they were fitting the spectrum.

Lev: It shows how critical those kinematic cuts are for making the math manageable, even though I’d have to worry about the trigger efficiency turn-on messing up those numbers on real hardware.

Kai: The main conclusion is that while they saw a local bump at nineteen point four giga electron volts, the overall global significance was only one point five sigma, meaning they didn't find anything significant.

Mira: That means their final upper limits on the production cross-section times branching ratio are between four and seventeen femtobarns.

Lev: Those limits are pretty tight for that mass range, so even if there's no signal, it sets a very clear boundary for what we expect to see from these new physics models.

Kai: The implication is that this search is pushing the limits of sensitivity down into the ten giga electron volt range, which was a bit of a blind spot before.

Mira: It’s a good piece of work because it addresses that gap and uses rigorous modeling to constrain axion-like particles in a way that connects directly to particle physics theory.

Lev: For error correction research, it's important because it shows how much detailed background modeling is absolutely necessary when you're looking for narrow features in high-energy data.

Kai: So they’ve shown that even with sophisticated tools, there are still gaps we need to fill to see a definite signal.

Mira: This search provides a solid baseline, and the next step is figuring out how to extend this coverage or look for different types of decay modes in those same mass windows.

Lev: We're going to talk about those extension strategies next, because you can’t just stop searching when you hit a limit; you gotta figure out where the next experiment needs to be focused.

More episodes

← Home