Research
Gutzwiller Quantum-Classical Embedding for Dynamics Simulations of Correlated Electron Systems
t-GQCE splits correlated lattice dynamics into a classical quasiparticle lattice and small quantum embedding solves, making long-time multi-orbital Hubbard quenches feasible.
- arXiv
- 2611.24592
- Published
- 2026-11-18
- Authors
- Joshua Aftergood, Daniele Guerci, Jacob Brunton, Feng Zhang, Thomas Iadecola, Peter P. Orth, Cai-Zhuang Wang, Nicola Lanatà, Yong-Xin Yao
AI summary
Overview
Research area: Quantum simulation of strongly correlated electron dynamics — specifically hybrid quantum-classical embedding methods that combine the Gutzwiller variational approach with quantum processors.
Technical level: Advanced. The paper is a methods/framework paper with substantial formalism (variational ansätze, equations of motion, quantum metric), though its central architecture can be grasped at an intermediate level.
Scope: The paper introduces t-GQCE, a hybrid framework in which the noninteracting quasiparticle lattice of a correlated material is evolved on classical processors while small interacting "embedding" problems are evolved on quantum processors, and demonstrates it on interaction quenches of the single-band Hubbard model on the Bethe lattice.
What This Paper Is About
Simulating how correlated electrons evolve in real time after a sudden change in their environment (an "interaction quench") is hard because the local electron correlations must be treated nonperturbatively over long times, which classical methods struggle with. Quantum embedding theories reduce the lattice problem to small local impurity models, and the ghost Gutzwiller approximation becomes exact in the limit of infinitely many ghost orbitals — but the resulting embedding problem grows exponentially, becoming the computational bottleneck. The authors propose to move exactly that bottleneck onto quantum processors, and test how well near-term quantum algorithms can follow the resulting embedding dynamics.
Key Contributions
-
A general hybrid framework (t-GQCE). The authors formulate time-dependent Gutzwiller quantum-classical embedding for generic multi-orbital lattice models with time-dependent Hamiltonians. The lattice dynamics is mapped onto (i) a noninteracting quasiparticle lattice evolved on classical processors and (ii) finite interacting embedding models evolved on quantum processors, with the two sides exchanging only the embedding Hamiltonian parameters and the single-particle density matrices of the embedding states.
-
Solver-agnostic design with a concrete near-term implementation. The framework does not depend on the choice of quantum dynamics solver — both near-term variational algorithms and fault-tolerant Hamiltonian simulation could be used. The authors demonstrate it with adaptive variational quantum dynamics simulation (AVQDS), with initial states prepared by adaptive variational quantum imaginary-time evolution (AVQITE).
-
Demonstration on the Hubbard model with and without ghost orbitals. They simulate interaction quenches of the single-band Hubbard model on the Bethe lattice, including a minimal single-qubit case (no ghost orbitals) run on IBM quantum hardware with two physical qubits total, and larger cases with B = 3, 5, and 7.
-
An assessment of measurement accuracy on near-term hardware. They evaluate how well embedding observables can be measured under a trapped-ion noise model and on the Quantinuum H2-Emulator, quantifying the effect of circuit compression, rotated-basis measurements with post-selection, and noise mitigation.
Main Findings
-
Ghost orbitals suppress spurious oscillations. Without ghost orbitals (B = 1), the conventional time-dependent Gutzwiller approximation produces coherent, undamped oscillations of local observables that are absent in exact dynamics. With B = 3, 5, and 7, t-GQCE suppresses these artificial oscillations and approaches the nonequilibrium dynamical mean-field theory (DMFT) results with increasing B, in both the weak and strong quench regimes.
-
A minimal hardware demonstration. For the B = 1 case, the embedding problem reduces to a single qubit. The authors simulate its dynamics on IBM quantum hardware using two physical qubits in total, where the second (ancilla) qubit is needed for Hadamard-test measurements of the quantum metric entering the equations of motion.
-
AVQDS tracks the embedding states accurately. Statevector simulations with 2, 4, and 6 ghost orbitals per spin — embedding problems of 8, 12, and 16 qubits — show infidelities below 5×10⁻⁴, using circuits ranging from a few hundred to about 3800 CNOT gates.
-
The demonstration is not fully self-consistent. For B = 3, 5, and 7, the AVQDS runs are decoupled from the self-consistent feedback loop: the time-dependent embedding Hamiltonians are first extracted from classical t-GQCE simulations that propagate the embedding states by exact diagonalization, and AVQDS is then run on those fixed Hamiltonians. This assesses AVQDS along the physical dynamical path but is not a quantum-in-the-loop simulation.
-
Noise remains the limiting factor. Under the benchmarked trapped-ion noise model, the residual noise bias exceeds the target accuracy of 0.01 adopted in earlier GQCE work, even after circuit compression, rotated-basis measurements with post-selection, and noise mitigation (which each reduce errors severalfold). Error rates three times lower than the benchmarked model are required to reach the 0.01 target for the single-particle density matrix and the double occupancy with 10⁵ shots per time step. The Quantinuum H2-Emulator gives errors of similar size.
-
Cheap classical bookkeeping in simple cases. For the single-band Hubbard model with B = 1, the quasiparticle density matrix stays time-independent and the embedding Hamiltonian depends only on a single expectation value of the embedding state, so the full t-GQCE dynamics collapses to a single-qubit problem with an explicitly known ansatz.
Methodology in Plain English
The starting point is a standard quantum embedding idea: instead of treating every electron in a crystal at once, you treat one small group of strongly interacting orbitals exactly and fold everything else into an effective environment. In the ghost Gutzwiller approach, that environment is represented by B bath orbitals per correlated orbital, where B − 1 of them are "ghost" orbitals that add variational freedom; more ghosts means a better description, converging to dynamical mean-field theory as B goes to infinity. The catch is that the local embedding problem then involves (B+1)ν spin-orbitals and its Hilbert space grows exponentially.
The authors split the work by what each type of computer does best. The noninteracting quasiparticle lattice is a single-particle problem, so a classical processor evolves its density matrix. The small but strongly interacting embedding state is what a quantum computer can represent compactly, one qubit per spin-orbital.
At each time step the two sides exchange a small amount of information. The quantum processor measures single-particle density matrices of the embedding state, which the classical side uses to update the embedding Hamiltonian (the bath couplings and the Lagrange multipliers). The quantum processor then measures the matrix M and vector V that drive the variational update — these involve the quantum metric, obtained with Hadamard-test circuits — and the classical side integrates both the quasiparticle equations and the ansatz parameters forward in time.
For the quantum solver they use AVQDS, which builds the circuit adaptively: it starts from a shallow parameterized circuit and, whenever the residual McLachlan distance exceeds a threshold, appends the operator from a pool that most reduces that residual. The initial state comes from AVQITE run at equilibrium, and those angles keep evolving rather than being frozen. All gates use a pseudo-Trotterized product of exponentials of Pauli strings.
The testbed is the single-band Hubbard model on the Bethe lattice in the infinite-coordination limit, which gives a semicircular density of states with half-bandwidth W set to unity as the energy unit. The interaction U(t) is zero before the quench and switched on at t = 0, and the authors compare the resulting local observables against nonequilibrium DMFT.
Why This Matters
Impact on research. The work gives the time-dependent Gutzwiller family a concrete route to systematically improvable dynamics: instead of paying an exponential classical cost for ghost orbitals, the hard part is handed to a quantum processor whose qubit count scales linearly with the number of spin-orbitals. It also provides a realistic, quantitative accounting of what near-term hardware can and cannot do for this class of embedding simulation — notably that measurement noise, not circuit expressivity, is the binding constraint at the benchmarked error rates.
Real-world applications (as motivated in the paper):
- Interpreting ultrafast pump-probe experiments that drive correlated electron materials far from equilibrium on femtosecond time scales.
- Modeling light-induced transitions between metallic and insulating states.
- Studying transient superconductivity.
- Describing nonthermal control of competing orders in correlated materials.
Industry relevance. The framework sits at the intersection of quantum-hardware vendors and computational materials science: it is explicitly hardware-agnostic and was exercised on IBM superconducting hardware and on the Quantinuum H2-Emulator, and its accuracy targets are framed in terms of shot counts and gate error rates that hardware roadmaps track directly.
Future Directions
- Close the self-consistent loop. The B = 3, 5, and 7 demonstrations used embedding Hamiltonians precomputed by exact diagonalization; a fully self-consistent run with the quantum solver inside the feedback loop is left for future work.
- Reduce hardware error rates. Reaching the 0.01 accuracy target for the single-particle density matrix and double occupancy with 10⁵ shots per time step requires error rates three times lower than the benchmarked trapped-ion noise model, so improved gates, mitigation, or both are needed.
- Swap in other solvers. Because the framework is agnostic to the dynamics solver, fault-tolerant Hamiltonian simulation algorithms (product formulas, linear combinations of unitaries, truncated Taylor series, quantum signal processing, qubitization) can be substituted for AVQDS as hardware matures.
- Extend beyond the minimal model. The general formalism targets multi-orbital lattices with time-dependent Hamiltonians and arbitrary unit-cell structure; demonstrating it on multi-orbital or more realistic material models, and for longer times, is the natural next test.
Target Audience
Researchers in quantum embedding theory and strongly correlated electron physics, quantum-algorithm developers working on near-term dynamics simulation (particularly adaptive variational methods), and computational materials scientists interested in hybrid quantum-classical workflows. Readers assessing hardware requirements for practical quantum advantage in materials simulation will find the noise and shot-count analysis directly useful. Familiarity with the Gutzwiller approximation, DMFT, or variational quantum algorithms is helpful but the architectural argument is accessible to graduate students in condensed matter or quantum information.
Authors’ abstract
Simulating the real-time dynamics of correlated electron systems requires a nonperturbative treatment of local correlations over long times, a challenge for classical methods. We present the time-dependent Gutzwiller quantum-classical embedding framework, t-GQCE, for generic multi-orbital lattice models with time-dependent Hamiltonians. Within the time-dependent ghost Gutzwiller approximation, the lattice dynamics is mapped onto a noninteracting quasiparticle lattice,…
The full abstract on arXiv