UrQMD-Agent: An LLM-Assisted Framework for Microscopic Tracing of Proton Azimuthal Correlations
This paper introduces UrQMD-Agent, an LLM-assisted framework that integrates large language models with deterministic scientific tools to perform microscopic transport studies, demonstrating its capability to trace and analyze negative proton azimuthal correlations in Au+Au collisions by revealing their origin in decay-fed proton pairs.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the subatomic world, matter behaves less like solid billiard balls and more like a chaotic, seething soup of particles. When scientists smash heavy atomic nuclei together at nearly the speed of light, they recreate conditions similar to those just moments after the Big Bang. These collisions generate a fireball of energy that cools and expands, transforming into a spray of new particles. To understand how this transformation happens, physicists rely on computer simulations that track the journey of every single particle, from the initial crash to the final moment they hit a detector. These simulations act as a microscope for the invisible, allowing researchers to reconstruct the history of a collision event by event. However, these models are incredibly complex, involving millions of lines of code and intricate rules for how particles interact, scatter, and decay. Making sense of the data they produce usually requires a team of experts manually sifting through the results, a process that is slow and prone to human error.
A new approach described in recent research combines the reasoning power of artificial intelligence with the raw calculation speed of these traditional physics simulations. The researchers built a system where a large language model—an advanced AI capable of understanding natural language and scientific literature—acts as a project manager. This AI does not perform the heavy math itself; instead, it plans the experiments, writes the necessary code, and interprets the results, while a separate layer of deterministic tools handles the actual number-crunching and data recording. This partnership allows for a level of detail previously difficult to achieve: the ability to trace the exact lineage of every particle in a collision, following it back through time to see where it came from and what influenced its path. By applying this method to gold nuclei collisions at various energies, the team uncovered a specific pattern in how protons—the positively charged particles inside atoms—move relative to one another, revealing a hidden signal that points directly to the origins of these particles.
The study focused on a specific type of collision between gold nuclei, conducted at six different energy levels ranging from 3.0 to 39.0 giga-electronvolts. The researchers were looking for a subtle statistical pattern in the way protons are distributed around the circle of the collision. Imagine a clock face where protons are scattered; if they were placed completely randomly, their distribution would follow a predictable baseline. However, the simulation showed that in the most violent, head-on collisions, the protons were not distributed randomly. Instead, they showed a distinct tendency to avoid certain angular arrangements, creating a "negative" fluctuation that stood out clearly from the background noise. This effect was strongest at lower energies, specifically between 3.0 and 27.0 giga-electronvolts, but it vanished at the highest energy tested, 39.0 giga-electronvolts, where the results were consistent with a random distribution.
To understand why this pattern existed, the team used their new framework to break down the collisions into pairs of protons. They discovered that the negative signal was not coming from all protons equally. Instead, it was carried almost entirely by protons that had been born from the decay of heavier, short-lived particles. In the language of the simulation, these are "decay-fed" protons. When the researchers separated the protons into groups based on their origins, they found that pairs where both protons came from decays were the primary drivers of the effect. At the lower energies, these specific pairs were responsible for the entire observed signal. At the very lowest energy of 3.0 giga-electronvolts, pairs involving one decay-fed proton and one "direct" proton also played a significant role, but at higher energies, the signal was dominated by the decay-fed pairs.
The most revealing part of the investigation was tracing the history of these decay-fed pairs backward through time. The simulation recorded the entire life story of every particle, including the moment a heavy particle was created and the moment it broke apart into protons. The researchers found that the negative correlation between the two protons was already present the moment their parent particles were formed. In other words, the protons were not influenced to move in this specific way by the chaotic environment of the collision as it expanded. Instead, the pattern was imprinted on them at their very birth, carried forward by their parent particles until they were released. As the particles traveled through the collision zone and eventually decayed, this initial correlation remained largely unchanged, suggesting that the source of the pattern lies in the initial creation of the parent particles rather than in the complex interactions that happen later.
This finding helps narrow down the microscopic origins of the behavior seen in these high-energy collisions. It suggests that the strange, non-random arrangement of protons is a direct inheritance from the particles that spawned them. The research also demonstrated the power of the new AI-assisted framework. By letting the AI manage the workflow and the deterministic tools handle the math, the researchers could perform a level of detailed, step-by-step tracing that would be difficult to coordinate manually. The system successfully connected the final results back to the earliest moments of the reaction, proving that such a hybrid approach can make complex microscopic transport models more transparent and easier to analyze. While the study did not identify the fundamental physical mechanism that causes the parents to be created with this specific correlation, it successfully ruled out the idea that the effect is caused by the particles interacting with each other after they are formed. The pattern is there from the start, waiting to be discovered by those who can trace the lineage of the particles with enough precision.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.