Accurate Computation of Quantum Excited States with Neural Networks

Overview

• Presents a variational Monte Carlo algorithm for estimating the lowest excited states of a quantum system
• Transforms the problem of finding excited states into finding the ground state of an expanded system
• Allows calculation of expected values of arbitrary observables, including off-diagonal expectations between different states
• Works well with recent neural network-based variational Ansätze for many-electron systems
• Accurately recovers vertical excitation energies and oscillator strengths on a range of molecules

Plain English Explanation

The paper introduces a new method for studying the excited states of quantum systems, which are the higher energy states that an atom or molecule can occupy. This is an important problem in fields like chemistry and physics, as understanding excited states is crucial for many applications.

The key idea is to transform the problem of finding the excited states into a simpler problem of finding the ground state of a larger system. This means the researchers don't need to explicitly calculate and keep track of the different excited states - instead, they just need to find the ground state of the expanded system.

The method also allows the researchers to calculate various properties of the excited states, such as their energies and the probability of transitions between them. Importantly, the method works well with recent advances in using neural networks to model the behavior of many-electron systems, like atoms and molecules.

The researchers show that by combining their new method with these neural network approaches, they can accurately predict the energies and other properties of excited states in a range of molecules. This is an important advance, as accurately modeling excited states has been a longstanding challenge in computational chemistry and physics.

Technical Explanation

The paper presents a variational Monte Carlo (VMC) algorithm for estimating the lowest excited states of a quantum system. This builds on previous work on using VMC to find the ground state of a system.

The key innovation is that the method transforms the problem of finding excited states into the problem of finding the ground state of an expanded system. This avoids the need for explicit orthogonalization of the different states, which can be computationally challenging.

The method allows the calculation of expected values of arbitrary observables, including off-diagonal expectations between different states, such as transition dipole moments. This enables the prediction of important spectroscopic properties.

The paper shows that the method works particularly well when combined with recent advances in using neural networks as variational Ansätze for many-electron systems. Specifically, the researchers demonstrate accurate predictions of vertical excitation energies and oscillator strengths on a range of molecules by integrating their method with the FermiNet and Psiformer Ansätze.

This represents the first deep learning approach to achieve accurate vertical excitation energies, including challenging double excitations, on benzene-scale molecules. The researchers expect the technique will be of great interest for applications in atomic, nuclear, and condensed matter physics, beyond the chemistry examples presented.

Critical Analysis

The paper presents a promising new approach for studying excited states of quantum systems, with several notable strengths:

• The transformation of the excited state problem into a ground state problem is a clever technical innovation that avoids the need for explicit orthogonalization.
• The ability to calculate a wide range of observables, including off-diagonal expectations, is a valuable capability.
• The successful integration with state-of-the-art neural network Ansätze demonstrates the flexibility and power of the method.

However, the paper also acknowledges some limitations and areas for further work:

• The method is entirely general, but the examples focus on molecular systems. Demonstrating the approach on a wider range of quantum systems would further validate its versatility.
• While the neural network Ansätze enable accurate predictions, the training of these models can be computationally expensive. Developing more efficient training techniques could broaden the applicability of the method.
• The paper does not provide a detailed analysis of the scaling of the algorithm with system size. Understanding the computational complexity is important for assessing the practical feasibility of the approach for larger-scale problems.

Overall, this work represents an important advance in the field of computational quantum mechanics, with the potential for significant impact across various domains of physics and chemistry. Further research to address the identified limitations and expand the range of applications would be valuable next steps.

Conclusion

The paper introduces a new variational Monte Carlo algorithm for estimating the lowest excited states of quantum systems. This method transforms the problem of finding excited states into the problem of finding the ground state of an expanded system, avoiding the need for explicit orthogonalization.

By combining this approach with recent advances in neural network-based variational Ansätze, the researchers demonstrate the ability to accurately predict vertical excitation energies and oscillator strengths for a range of molecules, including challenging cases like double excitations. This represents an important advancement in computational quantum mechanics, with potential applications across fields such as atomic, nuclear, and condensed matter physics.

While the paper highlights the strengths of the method, it also acknowledges areas for further research, such as expanding the range of systems studied and improving the computational efficiency of the neural network models. Addressing these limitations could lead to even broader impact of this innovative approach to studying excited states.