Abstract
Computing the excited states of a Hamiltonian is a fundamental challenge in quantum chemistry and materials science. While the ground state is often accessible via the Variational Quantum Eigensolver (VQE), excited states typically require "deflation" techniques that are computationally expensive and sensitive to noise. This paper introduces Quantum EigenGame, a game-theoretic approach where eigenvectors are treated as independent "players" seeking a Nash equilibrium. This formulation allows for the simultaneous discovery of multiple energy levels without explicit Hamiltonian deflation. We provide a rigorous analysis of derivative-free optimization in this context and demonstrate that Quantum EigenGame converges accurately on benchmarks like the $H_2$ molecule, outperforming standard methods like Variational Quantum Deflation (VQD) in hyperparameter robustness.
The Challenge of Excited States
In quantum simulations, finding the ground state is only half the battle. Understanding chemical dynamics, electronic transitions, and material properties requires knowledge of excited states—the higher energy levels of a system's Hamiltonian.
The conventional approach, Variational Quantum Deflation (VQD), finds these states sequentially. After finding the ground state, the algorithm "deflates" the Hamiltonian by adding a penalty term to prevent the next optimization from collapsing back to the ground state. This process is iterative and requires careful tuning of penalty weights ($\beta$). If $\beta$ is too small, the algorithm fails to find the excited state; if too large, the optimization landscape becomes difficult to navigate. Furthermore, on Noisy Intermediate-Scale Quantum (NISQ) hardware, calculating gradients is often impractical, necessitating robust derivative-free optimization (DFO).
Methodology: A Game-Theoretic Approach
Quantum EigenGame reimagines the search for eigenvectors as a competitive game. Instead of a single global objective function, each eigenvector $|\psi(\theta^{(i)})\rangle$ is assigned to a "player" $i$.
The Rules of the Game
Each player $i$ aims to maximize their own utility function $u_i$. The utility is designed such that:
- Energy Maximization: Players are rewarded for finding states with high variance (corresponding to energy levels).
- Implicit Orthogonality: Players are penalized for aligning with "parent" players (those representing lower energy states).
As shown in the figure above, the game follows a hierarchy. Player 1 finds the ground state. Player 2 then finds the first excited state by maximizing energy while remaining orthogonal to Player 1's result. This continues until a Nash equilibrium is reached, where no player can improve their utility by changing their parameters $\theta^{(i)}$.
Quantum Implementation
To compute the necessary inner products and penalty terms on a quantum computer, the authors utilize a modified Hadamard Test. This circuit allows for the evaluation of complex overlaps between different parameterized states, which is essential for the game's penalty mechanism. Additionally, the paper provides a theoretical framework for 0th-order optimization, proving that the algorithm can converge even when only noisy function values (shots) are available, rather than exact gradients.
Results and Performance
The algorithm was benchmarked against the $H_2$ molecule using a 2-qubit mapping. The primary comparison was made against Variational Quantum Deflation (VQD).
Key Findings:
- Hyperparameter Robustness: Unlike VQD, which requires manual tuning of the penalty parameter $\beta$, Quantum EigenGame uses an "energy-aware" regularization that scales automatically. This makes it significantly easier to deploy without extensive pre-trial tuning.
- Accuracy in Noise: In simulations with 10,000 shots (modeling realistic device noise), Quantum EigenGame successfully identified excited state energy levels with high precision.
- Scalability: Ablation studies on synthetic Hamiltonians up to 1024 dimensions showed that the number of iterations required scales predictably, maintaining performance even with derivative-free approximations.
Conclusion and Future Outlook
Quantum EigenGame provides a scalable, robust alternative to traditional deflation-based methods for excited state calculation. By leveraging game theory, it avoids the pitfalls of manual hyperparameter tuning and offers a clear path toward parallelized execution on multiple quantum processors.
Future work will focus on testing the algorithm on larger, more complex chemical systems and implementing a fully parallel version of the game. As NISQ hardware continues to evolve, the ability of Quantum EigenGame to handle noisy, derivative-free environments makes it a promising candidate for practical quantum chemistry applications.
Summary updated