Dynamical Mean-Field Theory for Real Materials using Variational Quantum Compressed Time Evolution on a Quantum Computer

AI summary

Overview

Research area: Quantum computing applied to condensed-matter and materials science — specifically quantum impurity solvers for DFT+DMFT simulations of strongly correlated electrons.

Technical level: Advanced. The paper assumes familiarity with dynamical mean-field theory, the Anderson impurity model, Green's functions, variational quantum algorithms, and Jordan–Wigner mapping.

Scope (one sentence): The authors build a hybrid density-functional-theory plus dynamical-mean-field-theory workflow in which a variational quantum-compressed time-evolution algorithm on an IBM Heron device computes the impurity Green's function of CaCuO₂, and they assess how far current noisy hardware can go toward a full self-consistent DMFT loop.

What This Paper Is About

Strongly correlated materials such as cuprate superconductors are hard to simulate because standard first-principles methods like density functional theory fail to capture strong local electron–electron repulsion, while the dynamical mean-field theory that does capture it still requires solving a quantum impurity problem that classical solvers handle only with difficulty (NRG and DMRG scale exponentially with bath size; quantum Monte Carlo suffers the fermionic sign problem and gives no direct spectral information). The authors ask whether a quantum computer can serve as that impurity solver for a real material, choosing CaCuO₂ — described in the paper as a parent structure of high-temperature cuprate superconductors — as the test case. Their goal is to compute the time-dependent Green's function on hardware, extract the density of states, and determine honestly whether the accuracy suffices for a fully self-consistent DMFT calculation.

Key Contributions

1. An end-to-end hybrid DFT+DMFT workflow with a quantum impurity solver for CaCuO₂, chaining a PBE density functional theory calculation (Quantum ESPRESSO), a Wannier90 downfolding to a single d_{x²−y²} band, a cRPA-derived Hubbard U of 3.11 eV, a CTHYB convergence at β = 30 eV⁻¹, an exact-diagonalization DMFT run at 0 K with a two-bath-site Anderson impurity model, and finally a quantum recomputation of the last ED-DMFT iteration.

2. A variational quantum compression scheme for the time-evolution operator that fits a Hamiltonian variational ansatz of fixed circuit depth to a single Trotter step iteratively in time, using a phase-sensitive cost function that evolves both the ground state and the X₀,σ-perturbed state, so that the Green's function (not just a state) is reproduced. The compression keeps circuits shallow enough to compile for the IBM Heron device.

3. A hardware-aware error-mitigation and post-processing pipeline: an exponential damping model tied to circuit depth and ground-state preparation, empirical rescaling using the analytically known non-interacting Green's function G⁰(t) and the constraint that the real part of G equals one at t = 0, followed by a Lehmann-representation fit seeded by Fourier and short-time Fourier transforms and constrained by sum rules and the impurity-model conditions G(ε_p) = 0 and ∂G/∂ω|_{ε_p} = 1/V_p².

4. An explicit diagnosis of NISQ hardware limits for DMFT, quantifying dephasing-induced damping and calibration-drift-induced temporal delay in the measured Green's function, and concluding that qualitative spectra are attainable but full DMFT self-consistency is not.

Main Findings

• Qualitative spectral features are recovered on hardware. The density of states reconstructed from IBM Heron measurements resolves the lower Hubbard band, the upper Hubbard band, and the central quasiparticle peak for CaCuO₂, using a broadening of η = 0.1. Individual peaks could not be resolved more accurately.

• Quasiparticle weight degrades by 15 % on hardware. The reported values are: Quantum Monte Carlo (QMC) 0.4453, Exact Diagonalization (ED) 0.3796, simulated compression with L = 3 up to t_max 0.3560, and IBM Heron 0.3027. The hardware value differs from the state-vector simulation by only 15 %, but all values sit below the QMC reference.

• The compressed ansatz beats a fixed Trotterization of equal gate count. For real and imaginary components of iG(t), circuits with L = 2 and L = 3 outperform a first-order fixed Trotterization with the same gate count as L = 3, both in curvature and amplitude; L = 1 follows the curvature better than the Trotter baseline despite diverging quickly. Even with L = 3, G(t) is only approximated correctly until t ≈ 4, and the authors deliberately ran to t_max = 8 to observe the effect on the spectral shape.

• Short time windows limit frequency resolution. Because of the small t_max, the Fourier transformation resolves only four ω_j; the fitting procedure supplies the rest.

• Damping scales with circuit depth as modeled. The non-interacting reference G⁰(t), computed analytically and measured on hardware for L ∈ {1, 2, 3}, is approximated correctly by all three cases at least up to t_max, with damping varying only slightly over time because entanglement is confined to each spin sector. This validates using G⁰(t) to extract the damping coefficients α_L and α_GS.

• Two distinct hardware error signatures appear in the Green's function envelope. The envelope |iG(t)|² shows Poincaré recurrences (revivals from the finite bath size). The hardware result captures the revival dynamics but shows amplitude damping and a horizontal shift of the peaks to later times. The shift is attributed to device calibration drift during the execution window, since real and imaginary components were run sequentially; drift alters qubit frequencies and causes systematic under-rotation of two-qubit entangling gates, weakening effective couplings and retarding correlation propagation.

• Error mitigation only partially works. The rescaling corrects early-time amplitudes but overcompensates or undercompensates at later times, and it cannot correct frequency shifts.

• Full DMFT self-consistency is out of reach on current hardware. The authors conclude that the numerical accuracy required to feed the Green's function back into the bath-fitting equations and converge the DMFT loop exceeds NISQ capabilities, and that post-NISQ hardware will be needed. They also note that training the parametrized circuits directly on hardware is infeasible because gradient estimates would suffer the same calibration drift observed for consecutive time steps.

• Resource footprint of the executed circuits. After compilation to the CZ-native gate set of ibm_torino, the trained circuits contain 65 two-qubit gates, of which 35 prepare the non-trivial ground state and the Hadamard test. The G⁰ circuits use 61 instead of 65 CZ gates for L = 3 at later times, because small parameter values led the compiler to filter certain gates. Each time point used 8192 shots.

Methodology in Plain English

1. Get the material's electronic structure from standard classical methods. A conventional density functional theory calculation with the PBE functional produces the band structure of CaCuO₂, and a downfolding step extracts a single-band Hubbard model for the copper–oxygen d_{x²−y²} band that crosses the Fermi level. A constrained random-phase-approximation calculation supplies the interaction strength U = 3.11 eV.

2. Converge DMFT classically first. The DMFT equations are solved self-consistently with the CTHYB impurity solver (10⁸ sweeps, cycle length 400, β = 30 eV⁻¹), then re-converged with an exact-diagonalization solver at 0 K that uses a two-bath-site Anderson impurity model obtained by fitting a Lehmann representation to the hybridization function.

3. Replace only the final impurity solve with a quantum computation. The impurity model is encoded on qubits via the Jordan–Wigner transformation. The ground state is prepared with a variational quantum eigensolver, and the time evolution is compressed into a shallow Hamiltonian variational ansatz trained classically in state-vector simulation. The Green's function is then read out with a Hadamard test at discrete times nΔt for L ∈ {1, 2, 3} layers.

4. Clean up and fit. Damping is characterized by comparing the non-interacting Green's function G⁰(t) — known analytically — measured on the same hardware with equivalent-depth circuits. Amplitudes are rescaled, then a Lehmann representation is fitted to the measured G(t), with initial frequency guesses from a Fourier transform (η = 0.1) and a windowed short-time Fourier transform.

5. Extract observables. The density of states follows from the imaginary part of the fitted Green's function, and the quasiparticle weight is obtained — because differentiating the self-energy is numerically unstable — by integrating the density of states over the quasiparticle peak, with boundaries at the lowest local minima below and above the Fermi level.

Why This Matters

Impact on research. The paper is a rare full-stack reality check: it takes a genuine material, a genuine DFT+DMFT pipeline, and a genuine cloud quantum processor, and reports where the pipeline breaks rather than only where it succeeds. It supplies the community with concrete diagnostics — depth-dependent exponential damping, calibration-drift-induced time shifts, a 15 % quasiparticle-weight deviation — that future algorithm and hardware work can target. It also demonstrates that variational time-evolution compression can make real-time Green's function measurements fit within current gate budgets.

Real-world applications (as framed by the paper):

• High-temperature superconductors. The authors describe finding a scalable high-temperature superconductor as the most profound motivation, and treat CaCuO₂ as the parent structure of the cuprate family, where conventional electron–phonon coupling fails to explain superconductivity.
• Stimulus-responsive materials. The paper notes that strongly correlated materials are extraordinarily sensitive to temperature, pressure, and electromagnetic fields and are therefore candidates for a wide array of advanced technologies.
• Cuprate and nickelate oxide families. The authors explicitly name both oxide families as targets whose strong correlations must be decoded.
• Correlated-electron modeling generally. The workflow is presented as a template for DFT+DMFT studies of correlated materials beyond the specific compound studied.

Industry relevance. The author list includes researchers from Robert Bosch GmbH (Corporate Sector Research and Advance Engineering), Robert Bosch LLC's Research and Technology Center in Sunnyvale, and Quint Computing GmbH, alongside Friedrich-Alexander-Universität Erlangen-Nürnberg. The paper explicitly calls decoding strong electronic correlations "an industrial imperative for the future of technology," and the work was funded by the German Federal Ministry of Research, Technology and Space through the MANIQU (Grant No. 13N15577) and EQUAHUMO (Grant No. 13N16067) projects and by Munich Quantum Valley.

Future Directions

• Post-NISQ hardware for self-consistent DMFT. The central open problem is achieving the numerical accuracy needed to feed the quantum-computed Green's function back into the bath-fitting equations; the authors state that hardware beyond the NISQ era is required.
• Longer simulation times. Extending t_max would let the Fourier transform resolve more Lehmann frequencies and sharpen the density of states; the current run to t_max = 8 already includes points where G(t) is not correctly approximated (accurate only to t ≈ 4).
• Hardware-robust training and drift suppression. Because calibration drift corrupts gradient estimates, the authors suggest that training parametrized circuits on hardware is infeasible for long times; reordering time points is offered as a workaround for evaluating G(t) but not for training. Improved calibration stability or drift-aware protocols are implied needs.
• Larger systems and cluster DMFT. The authors anticipate that once hardware matures, quantum DMFT for larger systems and cluster impurities could become competitive with classical methods such as DMRG. The paper notes that its own CaCuO₂ study is one-shot DMFT without charge-correction feedback, and points to a separate reference for cluster DMFT and antiferromagnetic treatments.

Target Audience

Researchers working at the intersection of quantum computing and materials science: quantum algorithm developers interested in impurity solvers and real-time Green's function methods; DFT+DMFT practitioners evaluating whether quantum hardware is ready for their workflows; quantum-chemistry and condensed-matter method developers; and industrial R&D groups in quantum technology and advanced materials, particularly those assessing hardware readiness honestly. Graduate students with a background in many-body physics and variational quantum algorithms will find the methodology sections and appendices a useful practical reference for error-mitigation strategies on superconducting devices.

Authors’ abstract

We design a hybrid quantum-classical computational workflow, combining density functional theory and dynamical mean-field theory, to investigate the electronic properties of CaCuO2, a parent structure of high-temperature cuprate superconductors. Our approach uses a quantum algorithm as an impurity solver that is based on variational quantum compressed time evolution to compute the time-dependent Greens function on an IBM Heron quantum device. Utilizing error mitigation techniques and a Lehmann representation fitting procedure, we extract the density of states from the signal in the accessible time regime. Despite the significant challenges rooted in the residual imprecision of currently accessible NISQ devices, we successfully resolve the characteristic features of the spectral function. However, we find that the required numerical accuracy for full DMFT self-consistency exceeds current hardware capabilities, indicating that practical real-material DMFT applications will necessitate post-NISQ hardware developments.