Real-Time Iteration Scheme for Dynamical Mean-Field Theory: A Framework for Near-Term Quantum Simulation

We present a time-domain iteration scheme for solving the Dynamical Mean-Field Theory (DMFT) self-consistent equations using retarded Green’s functions in real time. Unlike conventional DMFT approaches that operate in imaginary time or frequency space, our scheme operates directly with real-time quantities. This makes it particularly suitable for near-term quantum computing hardware with limited Hilbert spaces, where real-time propagation can be efficiently implemented via Trotterization or variational quantum algorithms. We map the effective impurity problem to a finite one-dimensional chain with a small number of bath sites, solved via exact diagonalization as a proof-of-concept. The hybridization function is iteratively updated through time-domain fitting until self-consistency. We demonstrate stable convergence across a wide range of interaction strengths for the half-filled Hubbard model on a Bethe lattice, successfully capturing the metal-to-insulator transition. Despite using limited time resolution and a minimal bath discretization, the spectral functions clearly exhibit the emergence of Hubbard bands and the suppression of spectral weight at the Fermi level as interaction strength increases. This overcomes major limitations of two-site DMFT approximations by delivering detailed spectral features while preserving efficiency and compatibility with quantum computing platforms through real-time dynamics.

💡 Research Summary

The authors introduce a novel real‑time iteration scheme for solving the self‑consistent equations of Dynamical Mean‑Field Theory (DMFT) that works directly with retarded Green’s functions in the time domain. Conventional DMFT solvers operate in imaginary time or frequency space, which makes them ill‑suited for near‑term quantum computers that naturally evolve quantum states in real time. By reformulating the DMFT self‑consistency condition as Δ_R(t) = t² G_R^loc(t) (with t the scaled hopping amplitude), the authors eliminate the need for analytic continuation or Fourier transforms and open a direct pathway to quantum‑hardware implementation.

To bridge the gap between the infinite bath required by DMFT and the limited qubit resources of NISQ devices, the impurity problem is mapped onto a finite one‑dimensional chain consisting of a single interacting impurity (site 0) coupled to five non‑interacting bath sites (sites 1‑5). This “5‑site bath” construction is a systematic improvement over the minimal two‑site DMFT, providing enough variational freedom to approximate the continuous hybridization function Δ(t) with high spectral resolution while keeping the total system size to six qubits. The chain Hamiltonian H_chain = H_imp + H_hyb + H_bath is fully characterized by hopping amplitudes {t_i} and on‑site energies {ε_i}, which are constrained by particle‑hole symmetry at half‑filling.

The iteration proceeds as follows: (i) an initial hybridization Δ⁽⁰⁾(t) is obtained analytically from the non‑interacting Bethe lattice Green’s function (a Bessel‑function expression); (ii) at each iteration k the bath parameters P = {t_i, ε_i} are optimized by minimizing a χ² error between the target Δ⁽ᵏ⁾(t) and the discrete-chain hybridization Δ_R^chain(t;P) = t₀² g₁₁(t;P), where g₁₁(t) is the Green’s function of the first bath site computed by exact diagonalization (ED); (iii) with the fitted parameters the full chain Hamiltonian is solved to obtain the interacting impurity retarded Green’s function G_R^imp(t). In the proof‑of‑concept study the authors use noise‑free ED, but they explicitly discuss how this step would be carried out on quantum hardware via a two‑stage procedure: (a) preparation of the ground state of H_chain using a variational quantum eigensolver (VQE) or adiabatic evolution, and (b) real‑time propagation of excitations using Trotterized circuits or variational time‑evolution algorithms. To suppress long‑time numerical instabilities (especially relevant for Trotterized evolution on noisy devices) a damping factor e^{−ηt} is applied to G_R^imp(t), equivalent to a small imaginary shift in frequency. (iv) The new hybridization is updated via the DMFT self‑consistency relation Δ_R^new(t) = t*² G_R^imp(t); (v) linear mixing Δ⁽ᵏ⁺¹⁾(t) = α Δ⁽ᵏ⁾(t) + (1−α) Δ_R^new(t) with α≈0.8 stabilizes convergence. The loop repeats until the change in Δ(t) falls below a preset tolerance, typically within 10–15 iterations.

Benchmark calculations are performed for the half‑filled Hubbard model on the Bethe lattice with interaction strengths U/t* ranging from 0 to 8. In the weak‑coupling regime (U≲2) the spectral function exhibits a sharp quasiparticle peak at the Fermi level, characteristic of a correlated metal. As U increases, the quasiparticle weight diminishes and the peak splits into two Hubbard bands centered at ±U/2, signaling the Mott insulating state. Importantly, even with only five bath sites the method reproduces these features with quantitative accuracy: the Hubbard bands are clearly resolved, and the suppression of spectral weight at ω=0 follows the expected trend. This level of detail surpasses the two‑site DMFT, which can only capture a coarse redistribution of weight and fails to resolve distinct Hubbard bands.

The paper’s contributions are threefold: (1) a fully real‑time DMFT iteration scheme that avoids analytic continuation and is naturally compatible with quantum hardware; (2) a minimal yet sufficiently expressive bath discretization (5 bath sites) that captures essential spectral features of the Mott transition; (3) a concrete blueprint for implementing each step on NISQ devices, emphasizing low circuit depth, modest qubit counts, and error‑mitigation strategies (damping, linear mixing). The authors also discuss scalability: increasing the number of bath sites systematically improves the approximation of the continuous hybridization function, and the same algorithmic structure can be extended to multi‑orbital or cluster DMFT.

Future work suggested includes (i) extending the bath to 10–20 sites to approach the continuum limit, (ii) performing full quantum‑hardware demonstrations with Trotterized evolution and VQE‑based ground‑state preparation, assessing the impact of gate errors and decoherence, (iii) integrating advanced error‑mitigation techniques such as zero‑noise extrapolation or dynamical decoupling, and (iv) applying the framework to realistic material models where multi‑orbital interactions and spin‑orbit coupling are essential. By establishing a practical, quantum‑ready DMFT workflow, this study paves the way for leveraging near‑term quantum processors to tackle strongly correlated electron problems that remain out of reach for classical methods.