arXiv:2607.10550
Coherent Quantum Schrodinger Bridge: Two-Boundary Optimal Control for Quantum Algorithm Design
Masayuki Ohzeki
Abstract
Quantum algorithms are intrinsically two-boundary processes: an input state is prepared, and an output state or subspace is selected as the computational answer. We formulate this observation as a coherent Quantum Schrödinger Bridge (QSB), a pure-state Hamiltonian counterpart of Schrödinger bridge theory in which the endpoint constraint is imposed on state vectors and the transport cost is the quadratic control action. In this setting Aharonov's two-state vector becomes the natural optimal-control pair: a forward state from the input and a backward state from the target. Pontryagin's principle then yields a universal optimal Hamiltonian whose weak value is purely imaginary in the geodesic gauge. Thus weak values are not an auxiliary interpretation; they are the local response functions that quantify the drift of the pre-selected state toward the post-selected boundary. Applying this framework to unstructured search, periodicity finding, and matrix arithmetic, we reconstruct Grover's algorithm, the quantum Fourier transform underlying Shor's algorithm, and quantum singular value transformation (QSVT). The usual circuit components -- oracles, diffusion reflections, controlled phases, and signal-processing rotations -- emerge as Lie-algebraic syntheses of the optimal weak-value drift. This perspective unifies distinct algorithmic paradigms into a single geometric principle: algorithm design is the problem of choosing computational boundary conditions and realizing the corresponding optimal flow.
AI Summary
The paper asks whether quantum algorithm primitives can be organized as minimum-energy unitary transport between prescribed input and output states. It defines a coherent, pure-state “Quantum Schrödinger Bridge” by minimizing $J[H]=\frac12\int_0^T\mathrm{Tr}(H(t)^2)\,dt$, with the target specified up to global phase. Using Pontryagin’s principle, it derives the candidate Hamiltonian $H^*=\frac{i}{2}(|\chi\rangle\langle\psi|-|\psi\rangle\langle\chi|)$, where forward and adjoint states evolve under the same Hamiltonian. It shows that this generator is time-independent and connects it to imaginary weak values through the response identity $\partial_g\log P_A(g)|_{g=0}=2\operatorname{Im}A_w$. Boundary-projector commutators provide a proposed route from the continuous generator to circuit operations. Three applications reinterpret existing primitives. For Grover search, the construction identifies the two-dimensional rotation direction generated by oracle and diffusion reflections, with input–target overlap $1/\sqrt N$. For the QFT, it proposes deriving controlled-phase interactions from Fourier-basis boundary phases. For QSVT, it assigns a rotation angle $\arccos p(\sigma)$ to each singular-value subspace and invokes existing QSP/QSVT synthesis theorems to connect admissible polynomial profiles to phase sequences. No new query-complexity bound or phase-synthesis algorithm is provided. The scope is explicitly restricted to coherent pure-state control, not stochastic, channel, or mixed-state Schrödinger bridges. Direct implementation of the resulting generators is generally unavailable. Unknown target boundaries, including ground states sought by variational algorithms, remain outside the exact construction; intermediate-boundary methods are proposed as future work.
Central claim may not hold: The prescribed terminal alignment of the adjoint and target, together with their identical unitary dynamics, forces the proposed optimal Hamiltonian to generate only phase evolution rather than nontrivial state transport. The displayed optimal weak-value formula omits the adjoint norm, whose scale affects the Hamiltonian and cannot generally be discarded for a prescribed transfer time. Application II's claimed exact QFT Hamiltonian is diagonal and therefore cannot map computational-basis states to Fourier superpositions. Application II's assertion that product-state endpoints imply additive optimal cost and decoupled generators fails under the paper's unrestricted global Frobenius-cost optimization.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 7 pages