通過基於評分的擴散模型實現快速HGCal探測器模擬 在雙光子衰變通道中搜尋頂夸克味變中性希格斯耦合

Abstract

隨著對撞機的不斷擴建和升級，物理學家面臨著越來越複雜的實驗需求，這導致對計算資源的需求急劇增加。現有的計算能力將難以持續支撐Geant4軟體完成精確且大規模的全套物理計算模擬，因此，尋求一種更加高效、快速的模擬方法已成為當前的研究重點。在此論文中，我們提出了使用擴散模型作為核心演算法，並結合transformer模型，嘗試模擬粒子能量在探測器內部的空間分佈。這一方法不僅能夠顯著加速模擬過程，還保持了與Geant4模擬結果相似的精度。本研究的最大特色在於其能夠生成與Geant4預測高度一致的三維能量分佈圖，而不僅僅是如同大多數類似研究所展示的在一維空間上的能量分佈。

除此之外，我也探討了頂夸克味變中性希格斯耦合（TopFCNH）的搜尋作為一個副專案。這種耦合在標準模型中被高度抑制，但在各種新物理理論中可以被顯著增強。通過分析CMS實驗中的雙光子衰變通道，本研究為探測罕見的頂夸克過程做出了貢獻。

As particle colliders continue to expand and upgrade, physicists face increasingly complex experimental demands, which in turn have led to a sharp rise in the need for computational resources. The current computational power will struggle to support full-scale and precise simulations using Geant4 software, especially as the scale of experiments grows. Therefore, finding a more efficient and fast simulation method has become a pressing priority in current research. In this thesis, we propose using a diffusion model as the core algorithm, coupled with a transformer model, to simulate the spatial distribution of particle energy within the detector. This approach not only significantly accelerates the simulation process but also maintains a level of accuracy comparable to Geant4 simulations. The key feature of this research lies in its ability to generate three-dimensional energy distributions that closely match those predicted by Geant4, rather than the one-dimensional energy distributions typical of most similar studies. Besides this machine learning-based simulation project, I also explore the search for top quark flavor-changing neutral Higgs (TopFCNH) interactions as a side project. These interactions are highly suppressed in the Standard Model but can be significantly enhanced in various new physics scenarios. By analyzing the H to r r decay channel at s = 13.6 TeV within the CMS experiment, this study contributes to the ongoing effort to probe rare top quark processes.