Equilibrium learning trains a physical system using observables of its relaxed states. We extend this principle to gate-based quantum circuits by embedding a parametrized circuit in a Feynman-Kitaev (FK) Hamiltonian and imposing the compilation target either through an output clamping term or through a target-dependent cyclic condition on the clock register. The ground state energies and the overlaps of the corresponding ground states with the free history state define faithful cost functions for unitary and state compilation. We show that these costs are nonlinear functions of the compilation infidelity and therefore reshape the optimization landscape. We show that starting from the free history state, quantum phase estimation can prepare the relevant ground states and estimate the cost functions with polynomial resources. In two-qubit benchmarks, optimization with the equilibrium cost functions reaches a fixed fidelity threshold in fewer sweeps than standard compilation infidelity cost function for many instances. The results demonstrate the potential for encoding quantum learning problems as computation of equilibrium observables.
AI Summary
Can equilibrium learning be formulated for gate-based quantum compilation? This paper embeds a parametrized circuit in a Feynman–Kitaev Hamiltonian and encodes the target through either an output-state clamp or a target-dependent cyclic clock connection. Ground-state energies and infidelities between the resulting ground states and the free history state serve as compilation objectives. Unitary compilation is encoded using maximally entangled input/reference registers.
Spectral analysis using finite-rank perturbations and explicit secular equations expresses these objectives as nonlinear functions of compilation infidelity $\epsilon$. For equal unit input/output couplings, the clamped ground energy is $E=1-\cos[(T+1)^{-1}\arcsin\sqrt{\epsilon}]$. The energy objectives, equal-coupling clamped overlap objective, and cyclic overlap objective are faithful and increasing in $\epsilon$, so their gradients follow the infidelity gradient while changing its magnitude. Unequal clamping strengths can instead produce an anti-aligned overlap gradient near $\epsilon=1$. Depth-independent history-state overlaps, including a clamped squared-overlap bound of $1/4$, support phase-estimation-based evaluation. Gaps scale as $\Theta(T^{-2})$ at fixed interior infidelity; the clamped gap deteriorates as $\Theta(\sqrt{F}/T^2)$ at small fidelity $F$.
Noiseless two-qubit compilation benchmarks with exact gradients often reach $F\geq0.99$ in fewer coordinate-descent sweeps than tuned QAQC, but gains depend on the task. No complete gate/query/sampling resource comparison or end-to-end advantage is established. The objectives can also be evaluated by classically transforming measured fidelity; larger-scale trainability and faithful, effective objectives directly accessible through local observables remain unresolved.
Little sign of AI-generated textMissing comparisonStrong assumptions
Central claim may not hold: The cyclic Hamiltonian retains a gauge-invariant overlap phase discarded in Theorem 2, so its ground energy can be positive even when the circuit perfectly compiles the target up to global phase.
The proposed clamped QPE ground-state preparation and overlap-cost measurement lack a uniform polynomial-resource guarantee because the ground-state gap can be exponentially small at exponentially small compilation fidelity.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
Equilibrium learning trains a physical system using observables of its relaxed states. We extend this principle to gate-based quantum circuits by embedding a parametrized circuit in a Feynman-Kitaev (FK) Hamiltonian and imposing the compilation target either through an output clamping term or through a target-dependent cyclic condition on the clock register. The ground state energies and the overlaps of the corresponding ground states with the free history state define faithful cost functions for unitary and state compilation. We show that these costs are nonlinear functions of the compilation infidelity and therefore reshape the optimization landscape. We show that starting from the free history state, quantum phase estimation can prepare the relevant ground states and estimate the cost functions with polynomial resources. In two-qubit benchmarks, optimization with the equilibrium cost functions reaches a fixed fidelity threshold in fewer sweeps than standard compilation infidelity cost function for many instances. The results demonstrate the potential for encoding quantum learning problems as computation of equilibrium observables.
Can equilibrium learning be formulated for gate-based quantum compilation? This paper embeds a parametrized circuit in a Feynman–Kitaev Hamiltonian and encodes the target through either an output-state clamp or a target-dependent cyclic clock connection. Ground-state energies and infidelities between the resulting ground states and the free history state serve as compilation objectives. Unitary compilation is encoded using maximally entangled input/reference registers.
Spectral analysis using finite-rank perturbations and explicit secular equations expresses these objectives as nonlinear functions of compilation infidelity ϵ. For equal unit input/output couplings, the clamped ground energy is E=1−cos[(T+1)−1arcsinϵ]. The energy objectives, equal-coupling clamped overlap objective, and cyclic overlap objective are faithful and increasing in ϵ, so their gradients follow the infidelity gradient while changing its magnitude. Unequal clamping strengths can instead produce an anti-aligned overlap gradient near ϵ=1. Depth-independent history-state overlaps, including a clamped squared-overlap bound of 1/4, support phase-estimation-based evaluation. Gaps scale as Θ(T−2) at fixed interior infidelity; the clamped gap deteriorates as Θ(F/T2) at small fidelity F.
Noiseless two-qubit compilation benchmarks with exact gradients often reach F≥0.99 in fewer coordinate-descent sweeps than tuned QAQC, but gains depend on the task. No complete gate/query/sampling resource comparison or end-to-end advantage is established. The objectives can also be evaluated by classically transforming measured fidelity; larger-scale trainability and faithful, effective objectives directly accessible through local observables remain unresolved.
Parametric quantum circuits are usually trained by classical, local optimizers, which use the quantum computer only to evaluate the cost function, or its gradient, one point at a time. Quantum Hamiltonian Descent (QHD) takes a different route: in this quantum algorithm for continuous optimization, the parameters become the position of a quantum particle, with the cost function as its potential, and two time schedules control kinetic and potential energy so that the wavefunction explores the whole landscape and tunnels towards the global minimum. QHD, however, assumes a classically computable cost function, and its digital implementation is poorly understood: standard error bounds grow exponentially with the number of qubits per parameter, and the schedules are borrowed from classical optimization. Here we propose to train parametric circuits with QHD entirely on the quantum computer, through an online step in which the qubits encoding the parameters coherently control the circuit on a second register. We prove that, for cost functions with a smooth periodic extension, including those of parametric circuits, the simulation error admits a bound independent of the grid resolution, and that for circuits the bound grows at most linearly with the number of parameters. We also show that the ideal dynamics depends on the two schedules only through a single curve, which controls the adiabaticity of the search, and that the remaining freedom can be used to minimize the gate count. Simulations on benchmark functions and parametric circuits show that, at equal cost, schedules designed in this way markedly increase the probability of finding the global minimum.
We propose a method for finding approximate compilations of quantum unitary transformations, based on techniques from policy gradient reinforcement learning. The choice of a stochastic policy allows us to rephrase the optimization problem in terms of probability distributions, rather than variational gates. In this framework, the optimal configuration is found by optimizing over distribution parameters, rather than over free angles. We show numerically that this approach can be more competitive than gradient-free methods, for a comparable amount of resources, both for noiseless and noisy circuits. Another interesting feature of this approach to variational compilation is that it does not need a separate register and long-range interactions to estimate the end-point fidelity, which is an improvement over methods which rely on the Hilbert-Schmidt test. We expect these techniques to be relevant for training variational circuits in other contexts.
We introduce an quantum-classical hybrid algorithm, named automatic quantum circuit encoding (AQCE), that is designed to encode an arbitrary quantum state | Ψ 〉 onto an optimal quantum circuit C ̂ composed of a finite set of single- and two-qubit quantum gates. The algorithm employs an objective function based on the absolute value of fidelity, F = 〈 0 | C ̂ † | Ψ 〉 , which is iteratively maximized to construct an optimal quantum circuit C ̂ with controlled accuracy. Here, | 0 〉 denotes a trivial product state in the computational basis of a quantum computer. The core of this algorithm lies in the sequential determination of an optimal set of two-qubit unitary operators, identified one by one through the singular value decomposition of the fidelity tensor. Once an optimal set of operators is determined, including the location of qubits on which each operator acts, elementary quantum gates are assigned algebraically to these two-qubit unitary operators. Importantly, these procedures are deterministic without assuming any quantum circuit ansatz and therefore eliminate the need for parameter optimization of parametrized quantum gates. Through noiseless numerical simulations, we demonstrate the effectiveness of the AQCE algorithm in encoding ground states of quantum many-body systems, including the spin-1/2 antiferromagnetic Heisenberg model and the spin-1/2 X Y model. We also compare these results with quantum circuit encoding employing predefined circuit structures such as Trotter-like and MERA-like circuit ansatze. Furthermore, our algorithm extends to classical data, for instance, classical images represented as quantum states using amplitude encoding. The adaptability enables us to adjust the quantum resource requirement, i.e., the number of qubits, by partitioning classical data into multiple distinct segments. This feature holds potential for near-term applications in quantum machine learning, such as a state preparation of classical data for an input quantum state to be processed. Finally, using a real quantum device provided by IBM Quantum, we experimentally validate that a quantum circuit generated by the AQCE algorithm can indeed reasonably represent the original quantum state. Published by the American Physical Society 2024
Quantum-circuit compilation aims at finding an optimized realization of a target circuit under given constraints, e.g., the minimization of hardware-induced errors and the unitary-equivalence of the circuit. We connect the compilation process with the thermodynamics of a many-body spin system: circuit infidelity plays the role of the energy function and low-temperature states correspond to compiled circuits. In the paradigmatic case where crosstalk between parallel gates is present, we find a phase transition between a disordered phase and an antiferromagnetic brick-wall phase, compatible with the Ising universality class. At larger crosstalk, we observe a Z3-ordered regime, suggesting that increasingly serial compiled circuits are associated with emergent Zn-ordered phases. When the unitary-equivalence constraint is removed, these phases disappear, showing that the equivalence between circuits underlies the emergent criticality and constitutes a source of complexity in quantum-circuit compilation and, more generally, in equivalence-constrained optimization. Finally, we observe that the Kolmogorov complexity of the circuit enhances the emergence of ordered phases.
Quantum compiling, where a parameterized quantum circuit is trained to learn a target unitary, is an important primitive for quantum computing that can be used as a subroutine to obtain optimal circuits or as a tomographic tool to study the dynamics of an experimental system.While much attention has been paid to quantum compiling on discrete-variable hardware, less has been paid to compiling in the continuousvariable paradigm.Here we motivate several, closely related, short-depth continuous-variable algorithms for quantum compilation.We analyze the trainability of our proposed cost functions and numerically demonstrate our algorithms by learning arbitrary Gaussian operations and Kerr nonlinearities.We further make connections between this framework and quantum learning theory in the continuous-variable setting by deriving no-free-lunch theorems.These generalization bounds demonstrate a linear resource reduction for learning Gaussian unitaries using entangled coherent-Fock states and an exponential resource reduction for learning arbitrary unitaries using two-mode-squeezed states.
Quantum Hamiltonian descent (QHD) formulates continuous optimization as time-dependent quantum dynamics, where a kinetic term drives exploration and a potential term encodes the objective function. Digital implementations of QHD require encoding the search space into qubits, and this choice can shift the dominant cost among logical qubits, circuit depth, non-Clifford rotations, and potential synthesis. In this work, we present an encoding-aware resource analysis comparing one-hot and binary amplitude encodings for QHD. We derive gate-count scalings, construct and validate circuits against classical \Sch-equation solvers, and estimate Clifford+Rz and fault-tolerant Clifford+T resources on benchmark optimization problems. Binary encoding reduces the data register from O(dN) to O(dlogN) qubits and gives comparable asymptotic scaling for both kinetic and potential evolutions. Across all benchmark problems studied, binary encoding also uses fewer Rz rotations than one-hot encoding, making it the preferred option for fault-tolerant implementations where arbitrary rotations dominate the cost. Kinetic approximations based on low-momentum spectra and approximate QFTs can further reduce the binary kinetic cost to polylogarithmic scaling. However, for targets such as Ackley, potential synthesis can dominate the total cost and reduce the benefit of kinetic approximations. These results suggest that exploiting the analytic structure of the target function to compile the potential evolution in QHD more efficiently is needed for further resource reductions.