A Fast and Smart Foundation Model for Particle Transport Simulation

Descripción del proyecto

Reimaginar la física de partículas con simulaciones más rápidas e inteligentes

Comprender los misterios más fundamentales del universo, desde el bosón de Higgs hasta la materia oscura, depende de simulaciones precisas de partículas. Aunque estas simulaciones son esenciales para diseñar experimentos y analizar grandes cantidades de datos, se están volviendo demasiado lentas y consumen demasiados recursos para seguir el ritmo de las necesidades científicas. En este sentido, el proyecto LEGO, financiado por el Consejo Europeo de Investigación, pretende introducir un método innovador de predicción de la partícula siguiente. Inspirado en la forma en que los modelos lingüísticos predicen las palabras, este método combina velocidad e inteligencia para hacer simulaciones miles de veces más rápidas manteniendo la precisión. Se espera que LEGO abra nuevas posibilidades en los campos de la física, la ingeniería y la medicina.

Objetivo

"Answering the deepest questions in particle physics, from the laws governing the Higgs boson to the true nature of dark matter, relies crucially on particle transport simulation. Simulation is the best way to create the required pattern recognition algorithms to make sense of our data and to design precision experiments.

But the field is at an impasse: Physics insight scales with simulation, but after allocating billions of hours to this task, we are reaching a resource limit. Missing discoveries due to suboptimal algorithms or detector designs is a real danger for two reasons: Simulators are not only excruciatingly slow but also ""probabilistic black boxes"" that fail to extract the best information when it comes to algorithm training.

With LEGO, I will lead an interdisciplinary team of researchers in Munich to re-imagine particle transport. Just as AI language models can create any text with ""next-word prediction,"" I will establish a new simulation model around ""next-particle prediction."" By predicting instantly what otherwise would take thousands of steps, LEGO will be the first ultra-fast simulation that retains the general-purpose nature of traditional simulators. LEGO will also be the first ""smart simulator"" in particle physics that provides access to probability and gradient data of the internal evolution of particle showers. This data enables us to extract more information without more simulation and is the key to reaching ultimate physics performance.

Smarter and faster simulations combine for an orders-of-magnitude more effective modeling of particle detectors. I will demonstrate the impact of such an advance for key applications for future colliders. In light of the European Strategy Update and AI advances, I exploit a crucial time window to create the best foundation for algorithm and detector design. As particle transport is crucial for fields far beyond particle physics, such as medicine, LEGO will open the door to a broad set of new applications."