Abstract:Representations of multivariate functions with low-dimensional functions that depend on subsets of original coordinates (corresponding of different orders of coupling) are useful in quantum dynamics and other applications, especially where integration is needed. Such representations can be conveniently built with machine learning methods, and previously, methods building the lower-dimensional terms of such representations with neural networks [e.g. Comput. Phys. Comm. 180 (2009) 2002] and Gaussian process regressions [e.g. Mach. Learn. Sci. Technol. 3 (2022) 01LT02] were proposed. Here, we show that neural network models of orders-of-coupling representations can be easily built by using a recently proposed neural network with optimal neuron activation functions computed with a first-order additive Gaussian process regression [<a class="link-https" data-arxiv-id="2301.05567" href="https://arxiv.org/abs/2301.05567">arXiv:2301.05567</a>] and avoiding non-linear parameter optimization. Examples are given of representations of molecular potential energy surfaces.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to construct a low - dimensional representation of the high - dimensional function \( f(\mathbf{x}) \), especially in cases where integration is required. Specifically, the author aims to achieve the orders - of - coupling representation of multivariate functions by using a single neural network with an optimal neuron activation function (without nonlinear parameter optimization). This method can effectively handle high - dimensional data and reduce computational complexity. ### Specific Problem Description 1. **Representation and Integration of High - Dimensional Functions**: - In theoretical chemistry, especially when solving the Schrödinger equation, it is necessary to integrate high - dimensional functions. As the dimension \( D \) increases, the grid size of numerical integration grows exponentially, leading to a sharp rise in computational cost. - Traditional methods such as Monte Carlo integration or sparse grids (such as Smolyak grids) can alleviate but not completely solve the problem. 2. **Low - Dimensional Approximation**: - To solve the problem of representing high - dimensional functions, researchers have proposed various low - dimensional approximation methods, such as sums - of - products (SOP) and orders - of - coupling representation. These methods allow only low - dimensional integration to be processed, thereby reducing computational complexity. 3. **Application of Machine Learning Methods**: - Previous studies have proposed using neural networks and Gaussian Process Regression (GPR) to construct low - dimensional approximations. However, these methods usually require nonlinear parameter optimization, which not only increases computational cost but may also lead to problems such as overfitting. ### The Solution in the Paper This paper proposes a new method that uses a single neural network with an optimal neuron activation function to construct the orders - of - coupling representation without performing nonlinear parameter optimization. The specific steps include: - **Define Weights by Rules**: Define the weight matrix \( W \) of the neural network by rules, so that the input of each neuron is a linear combination of the original coordinates \( \mathbf{x} \). - **Optimize the Activation Function by Gaussian Process Regression**: Use first - order additive Gaussian process regression to construct the optimal activation function \( \sigma_i(y_i) \) for each neuron. - **Application Examples**: The effectiveness of this method was tested through the potential energy surfaces (PES) of water molecules (H₂O) and formaldehyde molecules (H₂CO). ### Conclusion This method significantly simplifies the model construction process while maintaining similar performance to existing high - dimensional model representation (HDMR) techniques, avoiding the computational burden and overfitting problems brought by nonlinear optimization. In addition, this method shows good stability and accuracy when dealing with high - dimensional data.

Orders-of-coupling representation with a single neural network with optimal neuron activation functions and without nonlinear parameter optimization

Discovering the Representation Bottleneck of Graph Neural Networks from Multi-order Interactions

Effective Many-body Interactions in Reduced-Dimensionality Spaces Through Neural Network Models

Linearization Turns Neural Operators into Function-Valued Gaussian Processes

Formation of Representations in Neural Networks

Neural network representation of quantum systems

Function-Space Optimality of Neural Architectures With Multivariate Nonlinearities

Simulating Quantum Many-Body States with Neural-Network Exponential Ansatz

Probabilistic graphs using coupled random variables

Quantum optical neural networks with programmable nonlinearities

Measure, Manifold, Learning, and Optimization: A Theory Of Neural Networks

Neural network solutions of bosonic quantum systems in one dimension

Constructing exact representations of quantum many-body systems with deep neural networks

Symmetries and many-body excited states with neural-network quantum states

Transformer neural networks and quantum simulators: a hybrid approach for simulating strongly correlated systems

Solving the Quantum Many-Body Problem with Artificial Neural Networks

A simple linear algebra identity to optimize Large-Scale Neural Network Quantum States

Linking Gaussian process regression with data-driven manifold embeddings for nonlinear data fusion

Empowering deep neural quantum states through efficient optimization

Approximating quantum many-body wave-functions using artificial neural networks

Neural network enhanced hybrid quantum many-body dynamical distributions