Variational methods based on physically-motivated ansatze have long been the cornerstone of simulating strongly-correlated quantum systems. However, the field experienced a paradigm shift in 2017 with the introduction of Machine Learning techniques, particularly Neural Networks and advanced optimization schemes, to this domain [1]. This innovation gave birth to Neural Quantum States (NQS), a powerful new class of variational ansatze that synergizes the physical insight of traditional approaches with the unparalleled function approximation capabilities of Neural Networks.

The impact of Neural-Network based ansatze on the field has been profound. They have consistently outperformed competing approaches such as tensor networks in simulating spin systems [2], achieved unprecedented numerical precision in ground-state searches for large benchmark problems [3], and enabled the study of long-time dynamics in systems of unprecedented scale [4,5]. The reach of these methods extends beyond spin systems, with state-of-the-art results also reported in Bosonic [6] and electronic systems [7].

This rapid progress can be attributed to two key factors. First, the development of robust, large-scale machine learning libraries such as PyTorch and JAX provided a solid foundation for implementing these complex models. Second, the early development of domain-specific, open-source software packages [8,9] played a crucial role in democratizing these techniques within the quantum physics community.

These open-source initiatives have been instrumental in accelerating research in the field of Machine Learning for quantum physics. They have provided accessible, well-documented libraries that have been widely cited in the literature, lowering the barrier to entry for young researchers and facilitating the reproduction and extension of published work. By centralizing efforts around a core set of libraries, the field as a whole has benefited from continuous algorithmic improvements, creating a virtuous cycle of innovation and implementation.

Objectives

Our primary objective is to bring together leading researchers, software developers, and newcomers to the field to chart the course for the next generation of Neural Quantum States research and supporting open-source infrastructure. In particular, this CECAM Flagship Workshop aims to:

• Identify and address in a collaborative manner challenges originating from software that hamper the development of the field of Machine Learning for Quantum Simulation and Neural Quantum States in particular;
• Kickstart new Open-Source Software efforts, in particular to satisfy the growing need of the community of reproducibility and sharing of Neural Network architectures and trained weights;
• Federate existing Open-Source Software efforts in the domain in an organic manner;
• Recruit new young researchers/students into this effort and make them feel part of the greater community;

We will begin by conducting a comprehensive review of the current state of Neural Quantum States and related open-source software. This will serve as a foundation for identifying critical areas for software improvement and algorithmic implementation. Through collaborative discussions and hands-on sessions, we aim to establish best practices for developing and maintaining open-source quantum physics software, with a particular focus on implementing recent advances and improving code sustainability.

A key focus of the workshop will be exploring the potential of pre-trained models for quantum physics problems [10]. We will work towards developing a framework for creating, sharing, and leveraging these models, potentially revolutionizing how we approach complex quantum systems.

The workshop will also serve as an incubator for new collaborative projects aimed at addressing identified gaps in the current software ecosystem. By bringing together experienced developers with newcomers to the field, we hope to foster knowledge transfer and cultivate the next generation of contributors to open-source quantum physics software.

Ultimately, we aim to produce a community-driven roadmap for the future of open-source software in Neural Quantum States and related fields. This roadmap will serve as a guide for future development efforts and help align the community's efforts towards common goals.

By addressing these objectives, we believe this workshop will catalyze the next phase of development in Neural Quantum States and open-source quantum physics software. It will accelerate progress in the field by enabling more researchers to contribute to and benefit from these powerful tools, ultimately advancing our understanding of complex quantum systems.