arXiv:2609.26330
Effective-Hamiltonian Quantum Solvers for Differential Equations: Alternative Constructions and Function Encodings
Annie E. Paine
Abstract
Differential equations can be encoded as ground-state problems by constructing a positive-semidefinite effective Hamiltonian whose minimum-energy state represents the solution. We extend this framework in several directions. First, we show how multiple nonzero initial, boundary, and data conditions can be incorporated into the homogeneous formulation $A | f\rangle=0$ using a known nonzero reference condition. We then analyse an alternative formulation based on $A |f\rangle = | b\rangle$, with Hamiltonian $H_b=A^\dagger(I-|b\rangle\langle b|)A$, which incorporates source terms and nonzero constraints through an augmented system. For nonlinear differential equations, we examine the unphysical ground-state degeneracy introduced by tensor-product representations and consider both Hamiltonian constraints and ansatz-level restrictions for targeting physically valid product states. Finally, we develop grid-value amplitude-encoded versions of both Hamiltonian constructions and compare them with spectral coefficient encoding. Through linear and nonlinear examples, we assess how the choice of Hamiltonian formulation and function representation affects solution recovery, spectral gap, degeneracy, and readout. These results broaden the applicability of ground-state-based quantum differential-equation solvers while clarifying their principal practical trade-offs.
AI Summary
The paper studies how ground-state encodings of differential equations can accommodate nonzero constraints, alternative function representations, and polynomial nonlinearities. It extends the homogeneous residual formulation $A|f\rangle=0$ by expressing multiple initial, boundary, or data conditions relative to a known nonzero reference condition. It also adapts the linear-system Hamiltonian $H_b=A^\dagger(I-|b\rangle\langle b|)A$, augmenting the system with constraints to incorporate source terms and nonzero data. This construction requires a nonzero right-hand side and sufficient constraints to exclude unwanted null-space solutions. Both constructions are developed for spectral-coefficient and grid-value amplitude encodings and tested using noiseless statevector VQE simulations. Linear examples cover RC charging, the Airy equation, and composite-wall conduction using 3–4 qubits. For RC charging with eight coefficients or grid values, maximum errors are approximately $2\times10^{-6}$ for Chebyshev encoding versus $5\times10^{-3}$ for the grid discretisation. The wall example illustrates the need for distinct one-sided derivatives at material interfaces; domain decomposition resolves the incompatibility of a single smooth spectral representation. For a quadratic Fisher–KPP equation, tensor-product lifting produces ground-space degeneracies of 34–240. A shared-parameter repeated-product ansatz excludes unphysical states from the variational search; numerical continuation is additionally used to reach the desired solution branch. The paper explains why a Hamiltonian cannot select exactly all repeated-product states as its ground space. The simulations establish neither efficient ground-state preparation nor quantum advantage. Stated limitations include conditioning, spectral gaps, operator implementation, readout costs, variational convergence and noise, and increasing register requirements for higher-degree nonlinearities.
Error found: The RC example incorrectly excludes the general A0 framework when beta=0, because u'(0)=1 supplies the nonzero derivative reference explicitly allowed by that framework. The composite-wall Ab construction leaves B0 undefined and misdefines B1; the intended hot-boundary row is B0=(X_{s_0^-},0), while B1 should retain its previously defined cold-boundary row.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.