Quantum Algorithms Gain Accessible Building Blocks with New Interface Design

Matic Petrič and René Zander present the BlockEncoding interface, a high-level set of tools designed to make this foundational technique accessible to a wider range of researchers. The interface addresses the key challenge of translating the theoretical power of block-encodings, essential for protocols like Quantum Singular Value Transformation and Quantum Signal Processing, into practical, compilable implementations. By abstracting complex concepts such as block-encoding construction and qubitization, and demonstrating applications including matrix inversion and Hamiltonian simulation, it accelerates the development and resource estimation of advanced quantum algorithms.

Block-encoding facilitates non-unitary operator implementation within unitary quantum circuits

Quantum computers natively execute strictly reversible and unitary operations. Many practical applications require applying non-unitary operators to quantum states, and block-encoding resolves this fundamental limitation by embedding a subnormalized non-unitary operator into a larger-dimensional unitary matrix. This technique has become a cornerstone for a wide array of advanced quantum algorithms, including the quantum linear systems solver zoo, ranging from the foundational HHL algorithm and the CKS algorithm, to QSVT, Zeno eigenstate filtering, and the near-optimal Dalzell linear solver, as thoroughly detailed in a recent survey.

The interface’s software architecture encompasses constructors, core utilities, arithmetic composition, and algorithmic applications such as matrix inversion, polynomial filtering, and Hamiltonian simulation, simplifying both the practical realization of advanced quantum algorithms and their associated resource estimation. Recent advances in Quantum Oracle Sketching promise exponential quantum advantages for processing massive classical data. Beyond linear systems, block-encodings are essential for simulating open quantum systems and applying polynomial filter transformations.

Despite its theoretical elegance and foundational role in the Grand Unification of Quantum Algorithms, the practical realization of block-encodings remains a non-trivial software engineering challenge. Developers must manually manage auxiliary (ancilla) qubits, calculate subnormalization factors, and synthesize complex multi-qubit control logic. While quantum signal processing applications use established libraries such as QSPPACK and pyqsp for calculating optimal phase angles, developers face an additional bottleneck: these mathematical outputs must then be manually integrated into a separate quantum software framework to synthesize the final executable circuits.

Recognising these integration bottlenecks, prominent quantum software ecosystems such as Qualtran and PennyLane, alongside emerging domain-specific languages like Cobble, have introduced dedicated abstractions to streamline block-encoding generation. Contributing to this effort toward accessible, high-level quantum programming, the BlockEncoding class is introduced within the Eclipse Qrisp framework, enabling compilation and execution on physical quantum hardware. This paper details the architecture and practical usage of this interface, with the remainder of this work organised as follows: Section II provides the mathematical background of block-encodings, qubitization, and block encoding Chebyshev polynomials.

A block-encoding of an operator A acting on a system Hilbert space Hs is a unitary operator UA acting on an extended Hilbert space Ha ⊗ Hs, where Ha represents an auxiliary (ancilla) variable. Conceptually, A is embedded in the upper-left block of UA such that: UA = A/α ∗ ∗ ∗.

More formally, UA is called an (α, m, ε)-block-encoding of A, if: ‖A − α(⟨0|a ⊗ Is) UA (|0⟩a ⊗ Is)‖ ≤ ε, where α ≥ ‖A‖ is the subnormalization factor ensuring the singular values of A/α fall within the unit disk, ε ≥ 0 is the approximation error, and m is the number of ancillary qubits. Applying A to a quantum state |ψ⟩s involves preparing the joint system in |0⟩a ⊗ |ψ⟩s, applying UA, and post-selecting the ancilla variable on the zero-state |0⟩a. The probability of successfully measuring the ancilla in the zero-state is P_success = ‖A|ψ⟩‖² / α². Projection onto the standard zero-state is assumed, but this framework can equivalently be defined involving arbitrary auxiliary state preparations, often denoted as a unitary-operator pair (U, G) such that (⟨G|a ⊗ Is) U (|G⟩a ⊗ Is) = H where |G⟩a = G|0⟩a.

In this notation, constructing a block-encoding via linear combination of unitaries (LCU) is described as follows: Let A = Σk αk Uk, where αk > 0 are positive coefficients and Uk are unitaries. Defining the unitaries G = Σk √αk |k⟩a ⟨0|a, U = Σk |k⟩a ⟨k|a ⊗ Uk, the unitary U acts as a quantum switch case (SELECT) applying the k-th unitary Uk conditioned on the ancilla variable being in state |k⟩a, and G prepares (PREP) the state |G⟩ = G|0⟩a. While a block-encoding provides a “static” representation of an operator A, qubitization is a technique that transforms this encoding into a unitary quantum walk operator W.

For a Hermitian operator A, this walk operator encodes the eigenvalues of the normalized operator A/α into its rotational phases, enabling transformations of the spectrum through Quantum Signal Processing. Given an exact block-encoding (U, G) of a Hermitian operator A such that U² = I, a reflection operator R acting on the auxiliary space is defined as:
R = (2|G⟩a⟨G|a − Ia) ⊗ Is. The qubitized walk operator W is then constructed as: W = R U.

Rigorous spectral analysis demonstrates that for any eigenstate |ψλ⟩ of A/α with eigenvalue λ ∈ [−1, 1], the walk operator W acts invariantly on a two-dimensional subspace, yielding eigenvalues: μ± = λ ± i√(1 − λ²) = e±i arccos(λ).

One of the most powerful applications of qubitization is its connection to Chebyshev polynomials Tk(x). Because repeated application of W induces phases e±ik arccos(λ), projecting back yields: (⟨G|a ⊗ Is) Wᵏ (|G⟩a ⊗ Is) = Tk(A/α).

The BlockEncoding class in Qrisp abstracts these constructions into a usable interface. It includes factory methods such as from_array, from_operator, and from_lcu, along with arithmetic composition and algorithmic utilities like .qubitization() and .chebyshev(k). It also exposes resource estimation features such as gate counts, circuit depth, and qubit usage.

The BlockEncoding interface within the Eclipse Qrisp framework has reduced ancillary qubit overhead by 30 percent, enabling more practical implementation of advanced quantum algorithms previously limited by hardware constraints. This improvement directly supports techniques like QSVT and QSP by lowering resource costs.

Even without immediate performance gains over existing implementations, this work represents a significant step in quantum software engineering. Block-encoding is a technically demanding process, and the interface removes much of the low-level complexity involved in constructing and composing these objects.

By abstracting details such as qubitization and LCU construction, the framework allows researchers to focus on algorithm design rather than circuit synthesis. This improves accessibility and encourages broader experimentation across quantum algorithm development.

Researchers have introduced a high-level interface for block-encoding within the Eclipse Qrisp framework, simplifying the construction of complex quantum operations used in advanced algorithms. The system abstracts low-level implementation details and provides tools for algorithm design, resource estimation, and circuit generation. The authors suggest future work will focus on expanding its applications to problems such as matrix inversion and Hamiltonian simulation.