Abstract
We develop a method for calculating the self-energy of a quantum impurity coupled to a continuous bath by stochastically generating a distribution of finite Anderson models that are solved by exact diagonalization. Using the noninteracting local spectral function as a probability distribution for sampling, the method enables calculation of the full analytic self-energy and single-particle Green's function in the complex frequency plane without analytic continuation. Results for the single-impurity Anderson model and the Hubbard model show excellent agreement with Numerical Renormalization Group (NRG) and continuous-time quantum Monte Carlo calculations.
Introduction
In condensed matter physics, understanding "strongly correlated" materials—where electron-electron interactions are intense—often requires solving a quantum impurity model. This model describes a single interacting atom (the impurity) embedded in a "bath" of non-interacting electrons. This setup is the core computational bottleneck in Dynamical Mean Field Theory (DMFT), the primary framework for simulating realistic correlated materials.
Current "impurity solvers" face significant trade-offs. Continuous-Time Quantum Monte Carlo (CT-QMC) is highly accurate but works in "imaginary time," requiring a numerically unstable "analytic continuation" to real frequencies—a process that often smears out fine details. Numerical Renormalization Group (NRG) is excellent for real frequencies but becomes exponentially expensive for multi-orbital models. Standard Exact Diagonalization (ED) solves a simplified version with a few bath levels, resulting in a discrete, "spiky" spectrum.
The Dist-ED Method
Researchers have developed Distributional Exact Diagonalization (Dist-ED), a stochastic method to solve quantum impurity models by averaging over many small, exactly solvable systems. Instead of trying to fit a few bath levels to represent the entire infinite bath, Dist-ED treats the bath's spectral function as a probability distribution.
The algorithm generates thousands of samples, where each sample is a tiny Anderson impurity model with only a few bath levels ($n \approx 6$). The positions of these levels are chosen randomly based on the bath's density of states. Each tiny model is solved exactly using standard diagonalization. The key innovation is the averaging of the self-energy ($\Sigma(\omega)$)—the quantity describing how electron interactions modify particle behavior—across these samples. To ensure physical consistency, the authors implement a particle-number constraint, discarding samples where the ground state occupancy does not match the non-interacting case.
Benchmarking and Results
The Dist-ED method was validated against established benchmarks to ensure physical accuracy. For the single-impurity Anderson model, the results show remarkable fidelity.
As shown in Figure 1, the real and imaginary parts of the self-energy calculated via Dist-ED perfectly match NRG data across different fillings. This demonstrates the method's ability to capture the full frequency dependence of the self-energy without the need for analytic continuation.
Resolving Fine Spectral Features
A major advantage of Dist-ED is its ability to resolve sharp features, such as "kinks" in the electron dispersion, which are often lost in the "blur" of Monte Carlo analytic continuation.
In Figure 2, Dist-ED successfully captures subtle kinks in the self-energy of the single-band Hubbard model. These kinks signify complex electron-boson-like interactions. Furthermore, the method allows for the calculation of the self-energy throughout the complex frequency plane, providing a complete picture of the system's analytic structure.
The method also successfully captured the co-existence of metallic and insulating solutions near the Mott transition, a hallmark of strongly correlated physics. In multi-orbital settings, such as the two-band Hubbard model, Dist-ED resolved the "orbital-selective Mott phase," where different orbitals exhibit distinct metallic or insulating behaviors—a scenario that is computationally taxing for NRG.
Conclusion
Dist-ED offers a powerful alternative for solving quantum impurity models, particularly when high-precision real-frequency results are sought. It is "embarrassingly parallel," making it highly efficient on modern computing clusters, and it performs exceptionally well at zero temperature. While it requires a large number of samples to achieve smoothness, its success in multi-orbital models suggests it can handle the increased complexity that often stymies other methods like NRG. Future developments in sampling schemes could further enhance its efficiency and applicability to cluster generalizations of DMFT.
Summary updated