New Algorithms at LHC Enhance Particle Collision Analysis

At the Large Hadron Collider (LHC), things move fast—like, 40 million times per second fast! That's how often protons collide, and it’s the job of the Compact Muon Solenoid (CMS) experiment to study these Collisions. Why? To look for new and exciting things in the universe that we don’t fully understand yet. If that sounds like a treasure hunt, you’re right! But it’s a really complicated one with loads of Data.

#The Challenge of Too Much Data

With nearly 100 million channels of data pouring in from the CMS detectors during each collision, it’s like trying to find a needle in a data haystack. To make sense of it all, only a small slice of this data—about 1,000 events each second—can actually be stored for deeper analysis later. To do this, scientists use a two-tier trigger system.

The first tier, called the Level-1 (L1) trigger, is made up of special hardware processors. These clever gadgets pick out the most interesting events from the chaotic dance of protons and filter out the boring stuff. The second tier, known as the high-level trigger (HLT), takes the selected data and processes it further, aiming to save even more by the time it stores events for future study.

But there's a catch! The L1 trigger is not perfect. It’s like a bouncer at a club who only lets in people wearing a certain outfit. Sometimes, it might miss someone who’s dressed differently but could be a superstar. In the world of physics, this means that unexpected signals from new particles might be overlooked if they don't fit the usual patterns.

#A New Trick Up Our Sleeves

To tackle this issue, the CMS collaboration is trying something new—machine learning! Think of machine learning as a really smart assistant that helps scientists spot unusual events without relying on old rules. They’ve developed two different algorithms for spotting these Anomalies: AXOL1TL and Cicada. They sound fancy, but they basically aim to catch events that stand out from the crowd.

#What is AXOL1TL?

AXOL1TL stands for Anomaly eXtraction Online Level-1 Trigger Algorithm. It uses some neat tricks from a type of neural network called an autoencoder. An autoencoder is like a magician that learns to compress data and then recreates it. It’s trained to know what normal collisions look like. When it sees something weird, like a potential new particle, it struggles to recreate that input accurately, raising a red flag.

#And What About CICADA?

CICADA, on the other hand, uses convolutional autoencoders. It’s like AXOL1TL, but with an added twist—imagine having a magic eye that looks at pictures. CICADA analyzes images of energy deposits to detect unusual events. It’s a fresh perspective that allows things to be spotted while they’re happening.

Both methods work together, helping scientists monitor live data from the collisions without interrupting the main processes. So, if AXOL1TL and CICADA were superheroes, they’d be the dynamic duo capturing the action at the LHC.

#How They Work Together

Both algorithms were trained using a special dataset of proton-proton collisions, specifically zero-bias data collected at the LHC during 2023. Half of this data was used to teach the algorithms how to recognize what’s normal, and half was saved for testing their skills.

In tests, AXOL1TL showed a whopping 46% increase in efficiency compared to traditional methods when looking for exotic decay patterns from the Higgs boson. It’s like finding a golden ticket in a box of chocolates!

#Inputs Used for Anomaly Detection

Now, let’s talk about what these algorithms look for. AXOL1TL uses data about jets (think of them as bursts of energy), electrons, muons, and a measurement called missing transverse energy (MET). These inputs come from the L1 trigger and help AXOL1TL figure out what’s happening.

CICADA, however, processes energy deposits in a way that resembles analyzing images. By treating energy data like a picture, it can spot unusual patterns in the data, leading to a different kind of anomaly that AXOL1TL might miss.

#The Technical Side – But Not Too Technical!

AXOL1TL employs a special type of autoencoder called a Variational Autoencoder. Imagine it as adjusting the autoencoder’s settings to make it even better at spotting anomalies. It learns to understand the data better while making sure it sticks to a normal pattern, avoiding any wild surprises.

Similarly, CICADA uses convolutional layers in its autoencoder to work with the image-like inputs of the collisions. It tells the algorithms to be aware of any strange happenings that could indicate something out of the ordinary.

Both algorithms are trained to be efficient, and they are coded into hardware. Why? Because speed is crucial! When you’re dealing with data at the rate of 40 million collisions per second, you need systems that can keep up. They are implemented on special chips called FPGAs, which help perform calculations lightning-fast!

#Performance in Action

AXOL1TL has been fully tested in real-time with the CMS system. During testing, it operated at different sensitivity levels to catch anomalies, from very strict to quite relaxed. This flexible approach allows it to gather data that might reveal new discoveries.

Interestingly, the data flagged by AXOL1TL often showed different patterns than the regular L1 trigger events. This is important because it helps scientists collect novel events that might hint at new physics beyond what we currently know.

When looking at the types of events it captures, AXOL1TL is particularly good at spotting events with multiple jets—something traditional methods might miss. This gives it an edge when searching for new particles or phenomena that the old methods might overlook.

#Looking Ahead for New Discoveries

As scientists continue to analyze the data flagged by AXOL1TL, they will examine different properties to see if there are clues hiding within. They’ll study patterns in the mass distributions of various particles, such as jets and photons, to ensure there aren’t any biases that come from the selection process itself.

In the end, what we have here are two innovative algorithms, AXOL1TL and CICADA, bringing fresh techniques to help in the search for new physics at the LHC. They aim to find the unusual and unexpected, making the life of physicists a bit easier in their quest to unlock the secrets of the universe. Who knows what they’ll discover next? Maybe a particle that throws all the textbooks out the window!