A portable implementation of one-loop Feynman integrals
This paper introduces a performance-portable implementation of the QCDLoop library for one-loop Feynman integrals, leveraging the Kokkos framework to enable efficient execution on modern CPUs and GPUs for high-energy physics precision calculations.
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
High-energy particle physics is the science of peering into the smallest building blocks of the universe, smashing them together at incredible speeds to see what they are made of and how they interact. To make sense of the debris from these collisions, scientists rely on complex mathematical tools called Feynman integrals. These are not simple calculations; they are the essential machinery used to predict the probability of specific events happening, such as a particle decaying or two particles scattering off one another. For decades, these predictions have been the backbone of experiments at facilities like the Large Hadron Collider, allowing researchers to compare what they see in the real world against what their theories say should happen. However, as the machines become more powerful and the data they produce grows massive, the computers needed to run these calculations are hitting a wall. The traditional processors that have long driven these simulations are becoming less efficient for the sheer volume of work required, forcing the scientific community to look toward new types of hardware, specifically graphics processing units, which are designed to handle thousands of tasks simultaneously.
A team of researchers has now taken a significant step toward solving this bottleneck by creating a new, flexible version of a critical calculation library that can run on both traditional computers and these modern graphics chips. The work, led by scientists at Argonne National Laboratory and Fermi National Accelerator Laboratory, focuses on a specific set of these mathematical tools known as scalar one-loop integrals. In the language of particle physics, these integrals represent the most common and fundamental ways particles interact in a single cycle of a collision. The researchers started with a well-established software package called QCDLoop, which has been the standard tool for these calculations for years but was built to run only on older, single-threaded computer processors. Their goal was to rebuild this package so it could run efficiently on the diverse and powerful graphics processors now found in modern supercomputers, without having to rewrite the entire code for every different type of machine.
To achieve this, the team used a software framework called Kokkos, which acts like a universal translator for computer code. This framework allows a single program to be written once and then run on almost any type of processor, whether it is a standard central processing unit or a specialized graphics card from different manufacturers like NVIDIA, AMD, or Intel. The researchers designed their new code to be "portable," meaning it does not care which specific hardware it is running on. They organized the calculations so that instead of processing one collision scenario at a time, the computer could handle thousands of them all at once. This approach is crucial because the new graphics processors have thousands of small cores that can work in parallel, but they struggle if asked to wait for one task to finish before starting the next. By removing the old method of storing previous results in a cache to save time, the new code embraces a strategy of pure parallelism, feeding the processors a continuous stream of data to keep them fully occupied.
The team then put this new system through a rigorous series of tests to ensure it was not only fast but also correct. They compared the results from their new graphics-based code against the original, trusted version running on a standard processor. They generated millions of random scenarios representing different particle collisions, covering every possible shape and complexity of interaction, from simple loops to more intricate four-point connections. In every case, the new code matched the old code with extreme precision, reproducing the results to the limit of what standard computer numbers can hold. They also tested the code on a wide variety of actual hardware, including the latest graphics cards from NVIDIA, AMD, and Intel, confirming that the software works reliably across all these different platforms. The results showed that for the most complex calculations, the new graphics-based approach was more than one hundred times faster than the traditional method running on a single processor core.
This speedup is a vital development for the future of particle physics, particularly as experiments move into an era of high-luminosity collisions where data will be generated at unprecedented rates. While the researchers noted that the speedup might be slightly less dramatic in real-world simulations where some calculations can be reused, the ability to run these essential tools on modern graphics hardware opens the door to faster and more detailed studies of the universe. The new code is designed to be accessible to other scientists, allowing them to integrate these high-speed calculations into their own software for analyzing data from current and future particle colliders. By making these complex mathematical tools available on the hardware that is becoming the standard for large-scale computing, the researchers have ensured that the theoretical tools needed to understand the next generation of experiments will be ready when the data arrives. The work stands as a bridge between the established methods of the past and the high-performance computing needs of the future, ensuring that the quest to understand the fundamental laws of nature can continue without being held back by the limits of older technology.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.