A team including Google Quantum AI researchers published a quantum algorithm for Sylvester matrix equations in npj Quantum Information on August 21, 2026. The equation has the form AX + XB = C, where A, B, and C are known matrices and X is unknown.
Sylvester equations appear in control theory, physics, signal analysis, and numerical linear algebra. The paper targets a quantum output format called block-encoding instead of preparing the solution matrix as a quantum state.
What the algorithm produces
A block-encoding places a scaled version of a matrix inside a larger unitary operation. The algorithm prepares a block-encoding of X / x, where x is a positive normalization factor. Quantum circuits then estimate selected properties of the solution matrix through operations on this unitary.
This output choice changes the comparison with classical and earlier quantum linear-algebra methods. State-preparation algorithms encode solution data into amplitudes. Extracting individual entries or selected properties from those states often requires additional sampling. A block-encoding gives direct access to matrix operations suited to specific downstream tasks.
The authors assume block-encoding access to the input matrices. This assumption describes how data enters the algorithm. Building these input oracles for a real application remains part of the end-to-end cost.
Claimed scaling
The paper reports query and gate complexity almost linear in a condition number derived from A and B. Complexity grows logarithmically with matrix dimension and inverse error under the stated access model.
The condition number matters. Poorly conditioned matrices make the quantum procedure more expensive. Efficient scaling appears when the relevant condition number grows slowly with problem size, such as a polylogarithmic relationship in the paper's favorable cases.
The authors also show circuits for BQP-complete problems. BQP is the class of decision problems solvable by a quantum computer with bounded error in polynomial time. This complexity result establishes broad computational reach for the construction. It does not establish a practical advantage for every Sylvester equation.
Where the equation appears
Control systems use matrix equations to describe stability, feedback, and state evolution. Physics uses related equations for linear response and dynamical models. The paper also discusses connections to Riccati equations, which appear in optimal control and signal processing.
Those applications give the algorithm a clear target. A useful demonstration still needs a complete workflow covering matrix loading, circuit construction, fault-tolerant execution, output extraction, and comparison with the best classical solver for a named problem.
Theory, not a hardware result
The paper presents quantum circuits and complexity analysis. No current quantum processor ran a large Sylvester equation through this work. The favorable scaling depends on matrix access, conditioning, precision, and the cost of extracting the desired result.
The result therefore belongs to the fault-tolerant algorithm layer of quantum computing. It extends quantum linear algebra beyond ordinary systems of equations and gives researchers a new block-encoding construction for matrix-valued outputs.
The next useful step is an end-to-end application benchmark. Researchers need to select a control or physics problem, build its input oracles, estimate total fault-tolerant resources, and compare output quality and cost against a classical implementation.