A systematic literature review on large language models (LLMs) for automated program repair (APR) is presented, summarizing 127 relevant papers published between 2020 and 2024. The study analyzes the application of LLMs in APR, categorizes existing LLMs, and outlines three types of utilization strategies. It details specific repair scenarios, such as semantic bugs and security vulnerabilities, and discusses key aspects of integrating LLMs into APR, including input forms and open science. The paper highlights remaining challenges and potential future research directions. It provides a comprehensive overview of the research landscape, helping researchers understand achievements and promote future research. The artifacts are available at the GitHub repository: https://github.com/iSEngLab/AwesomeLLM4APR. The paper begins by introducing the problem of software bugs and the role of APR in software development and maintenance. It reviews existing APR techniques, including heuristic-based, constraint-based, and pattern-based approaches, and discusses the emergence of learning-based APR techniques using neural networks. The paper then explores the application of LLMs in APR, noting their success in various source code-related tasks, such as code generation, code summarization, and test generation. It discusses the integration of LLMs into APR, highlighting the complexity of this process and the diversity of research perspectives, repair phases, scenarios, and model utilization paradigms. The study presents a systematic literature review methodology, including research questions, search strategies, and study collection and selection processes. It analyzes the trends and distribution of LLM-based APR studies, the types of programming languages used, and the categories of publication contributions. The paper identifies the most popular LLMs, such as ChatGPT, GPT-4, CodeT5, and Codex, and discusses their applications in APR. It also explores the repair scenarios facilitated by LLMs, including semantic bugs, syntax errors, and static warnings, and highlights the challenges and opportunities in integrating LLMs into APR. The paper concludes with a summary of the findings, emphasizing the growing trend of using LLMs in APR and the potential for future research.A systematic literature review on large language models (LLMs) for automated program repair (APR) is presented, summarizing 127 relevant papers published between 2020 and 2024. The study analyzes the application of LLMs in APR, categorizes existing LLMs, and outlines three types of utilization strategies. It details specific repair scenarios, such as semantic bugs and security vulnerabilities, and discusses key aspects of integrating LLMs into APR, including input forms and open science. The paper highlights remaining challenges and potential future research directions. It provides a comprehensive overview of the research landscape, helping researchers understand achievements and promote future research. The artifacts are available at the GitHub repository: https://github.com/iSEngLab/AwesomeLLM4APR. The paper begins by introducing the problem of software bugs and the role of APR in software development and maintenance. It reviews existing APR techniques, including heuristic-based, constraint-based, and pattern-based approaches, and discusses the emergence of learning-based APR techniques using neural networks. The paper then explores the application of LLMs in APR, noting their success in various source code-related tasks, such as code generation, code summarization, and test generation. It discusses the integration of LLMs into APR, highlighting the complexity of this process and the diversity of research perspectives, repair phases, scenarios, and model utilization paradigms. The study presents a systematic literature review methodology, including research questions, search strategies, and study collection and selection processes. It analyzes the trends and distribution of LLM-based APR studies, the types of programming languages used, and the categories of publication contributions. The paper identifies the most popular LLMs, such as ChatGPT, GPT-4, CodeT5, and Codex, and discusses their applications in APR. It also explores the repair scenarios facilitated by LLMs, including semantic bugs, syntax errors, and static warnings, and highlights the challenges and opportunities in integrating LLMs into APR. The paper concludes with a summary of the findings, emphasizing the growing trend of using LLMs in APR and the potential for future research.