Loop-induced at NLO+PS for LHC simulations: exact, fast scale and PDF event weights, theory uncertainties, and the limits of rescaled LO+PS
This paper presents an efficient NLO+PS simulation of loop-induced production at LHC energies, introducing an exact loop-amplitude cache to accelerate event generation and demonstrating that machine-learning-based corrections outperform simple inclusive rescaling for differential distributions, though they still face limitations in fully restoring depleted high-momentum tails and accounting for all theoretical uncertainties.
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 author. For technical accuracy, refer to the original paper. Read full disclaimer
In the heart of the Large Hadron Collider, where protons smash together at nearly the speed of light, physicists are hunting for the subtle fingerprints of the Higgs boson. One of the most interesting ways this particle appears is when it is born alongside a Z boson, a heavy carrier of the weak force. While most of these pairs are created by quarks colliding, a smaller but significant fraction comes from a much more complex process: two gluons, the particles that bind quarks together, fusing to create the pair. This specific pathway is a loop-induced process, meaning the gluons must briefly transform into a heavy top quark and back again before producing the final result. Because this process is so rare and mathematically intricate, calculating its behavior with high precision has long been a challenge. Understanding exactly how often these pairs form and how they move is crucial for testing the Standard Model of particle physics and searching for new, hidden laws of nature.
To tackle this, researchers at the University of Maryland have developed a new method to simulate these collisions with unprecedented speed and accuracy. They focused on the loop-induced gluon fusion process at the energy levels currently used by the collider, specifically 13 and 13.6 trillion electron volts. The team created a sophisticated computer simulation that calculates the probability of these events occurring, including the messy details of how the particles interact and fly apart. A major hurdle in such simulations is that calculating the complex quantum loops for every single event is incredibly slow, especially when physicists need to test how the results change under different theoretical assumptions. The authors solved this by building an exact "cache," or a digital memory bank, that stores the results of these difficult calculations. When the simulation needs the same information again, it simply retrieves it instead of recalculating it. This innovation reduced the time required to generate a full set of events by a factor of twenty-six, making it feasible to produce the massive datasets needed for modern experiments.
With this faster engine, the team generated detailed predictions for how the Higgs and Z bosons behave, paying close attention to events where the particles are moving very fast, or "boosted." They found that the standard way of estimating these events, which involves taking a rough calculation and scaling it up, fails dramatically in these high-speed scenarios. The old method, which relies on a simple multiplier to adjust the numbers, cannot reproduce the complex shapes of the particle distributions seen in the more accurate, high-level simulation. In fact, for the fastest-moving events, the simple method overestimates the number of collisions by a factor of nearly seventeen compared to the precise calculation. This discrepancy arises because the simple method does not correctly account for how the particles recoil against each other when a third particle, like a jet of quarks, is thrown into the mix.
To fix this without having to re-run the expensive high-level simulations for every experiment, the researchers trained a machine-learning system to act as a translator. They used this system to learn the difference between the rough, low-level simulations and the precise, high-level ones. Once trained, this system could take existing, cheaper simulation data and apply a correction that made it look like the expensive, accurate version. For the specific types of data used by the CMS and ATLAS experiments, this correction worked remarkably well in most regions, reducing the uncertainty in the predictions from about twenty-eight percent down to seventeen percent. However, the method has a clear limit. In the most extreme cases, where the Higgs boson recoils against a very hard jet, the original low-level simulations simply do not contain enough events for the machine learning to learn from. In these rare, high-energy tails, the corrected predictions still fall short of the true value, leaving a gap that cannot be filled without generating new, high-quality data from scratch.
The study also provided a detailed map of the uncertainties involved in these calculations. They discovered that while the uncertainty from the strength of the strong nuclear force is the largest overall, a different source of error becomes dominant in the high-speed regions. This error comes from how the mass of the top quark is defined in the mathematical formulas. In the bins of data where the Higgs boson has a transverse momentum greater than 150 gigaelectron volts, the uncertainty related to the top quark's mass definition grows to between thirty-two and fifty percent. This finding suggests that future improvements in our understanding of the Higgs boson will depend heavily on refining how we treat the mass of the top quark in these calculations.
Ultimately, this work provides a dual path forward for the physics community. For new experiments, the authors recommend generating fresh, high-precision data using their fast, cached simulation method, which can now handle the heavy computational load efficiently. For existing data that was generated with older, simpler methods, the machine-learning corrections offer a powerful way to upgrade the results, provided the analysis does not focus on the rarest, hardest-to-reach corners of the data. By combining these tools, physicists can now probe the loop-induced production of the Higgs boson with a clarity that was previously out of reach, ensuring that the search for new physics is built on the most solid theoretical foundation possible.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.