Block-Product Gadget: Quantum Matrix Multiplication

• Block-product gadget is a suite of unitary circuits that compute matrix, Kronecker, and Hadamard products from quantum block-encodings with efficient resource usage.
• The construction leverages sophisticated ancilla management and permutation operators to achieve exponential ancilla-qubit savings with only a moderate gate complexity increase.
• Its application in quantum Hamiltonian simulation and quantum linear algebra improves algorithmic scaling and feasibility on near-term and future quantum devices.

A block-product gadget is a suite of unitary circuit constructions that enable resource-efficient computation of matrix products—specifically matrix-matrix, Kronecker, and Hadamard products—between quantum block-encodings. These gadgets enable significant reductions in required ancilla qubits, often yielding exponential savings for multipart products, with only a moderate increase in two-qubit gate complexity. Block-product gadgets form a foundational tool in quantum algorithms for Hamiltonian simulation and quantum linear algebra by providing a systematic framework for manipulating and combining block-encoded quantum operators (Dong et al., 19 Sep 2025).

1. Block-Encoding Framework

Block-encoding is a standard technique for embedding a (rectangular or square) complex matrix $A \in \mathbb{C}^{N \times N}$ (where $N=2^n$) into a higher-dimensional unitary $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$. An $(\alpha, a, \varepsilon)$-block-encoding of $A$ is a unitary $U$ such that

$$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$

i.e., the top-left $2^n\times2^n$ block of $U$ approximates $A/\alpha$ within error $N=2^n$0. This formalism allows quantum algorithms to manipulate large matrices via their encodings as unitaries, facilitating further product and functional constructions.

2. Matrix–Matrix Product Gadget

Given block-encodings $N=2^n$1 of $N=2^n$2 and $N=2^n$3 of $N=2^n$4, the block-product gadget constructs a block-encoding $N=2^n$5 of $N=2^n$6 using padded ancillas and permutations: $N=2^n$7 Here, $N=2^n$8 is a computational-basis permutation that aligns inner blocks for multiplication. The resulting encoding satisfies

$N=2^n$9

The gate complexity is $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$0, where the extra overhead arises from the realization of $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$1. The construction yields exponential ancilla-qubit savings for chains of products by avoiding naïve ancilla allocation per factor.

3. Kronecker and Hadamard Product Gadgets

(a) Kronecker Product

For $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$2 encoding $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$3 and $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$4 encoding $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$5, the natural tensor product $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$6 contains $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$7 among its blocks. Applying permutations $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$8 to the ancillas extracts $U \in \mathbb{C}^{2^{n+a} \times 2^{n+a}}$9 into the top-left block. The gadget has:

• Ancilla usage: $(\alpha, a, \varepsilon)$0 (no extra qubits).
• Gate overhead: $(\alpha, a, \varepsilon)$1, with $(\alpha, a, \varepsilon)$2, $(\alpha, a, \varepsilon)$3.
• For $(\alpha, a, \varepsilon)$4 powers of two: $(\alpha, a, \varepsilon)$5 CNOT gates.

(b) Hadamard Product

For $(\alpha, a, \varepsilon)$6 and $(\alpha, a, \varepsilon)$7 encoding $(\alpha, a, \varepsilon)$8, fan-out unitaries $(\alpha, a, \varepsilon)$9 and $A$0 are used to align entries: $A$1

$A$2

The resulting circuit yields an $A$3-block-encoding of $A$4. The two-qubit gate cost is $A$5 CNOTs.

4. Compression Gadgets and LCU-Based Constructions

(a) Compression Gadget

Given $A$6, each a $A$7-block-encoding of $A$8 (acting on a system register), the gadget recursively synthesizes a unitary $A$9 implementing a compressed block-encoding of the products $U$0. The circuit acts on:

• $U$1,
• $U$2,
• $U$3 ancillas plus the system.

Projecting onto $U$4, $U$5 encodes $U$6. Gate complexity is one query per $U$7 and $U$8 further gates, mainly for fan-out and multi-controlled operations. This matches the complexity of Lemma 13 of Low–Wiebe with simplified controls.

(b) LCU–Kronecker–Sum Block-Encoding

Given $U$9, with $$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$0 and $$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$1 block-encoding $$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$2, $$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$3, first Kronecker gadgets are used per term, then LCU with weights $$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$4. The final block-encoding has parameters

$$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$5

with total gate complexity

$$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$6

5. Ancilla and Gate Complexity Comparison

The following table summarizes the ancilla-qubit and gate overhead costs for block-product gadgets relative to naïve approaches:

Gadget	Ancillas (original)	Ancillas (gadget)	Gate Overhead
Mat–Mat	$$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$7	$$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$8	$$\left\| \left(\langle 0^a| \otimes I_n\right) U \left( |0^a\rangle \otimes I_n \right)- \frac{1}{\alpha} A \right\| \le \varepsilon,$$9
Kronecker	$2^n\times2^n$0	$2^n\times2^n$1	$2^n\times2^n$2
Hadamard	$2^n\times2^n$3	$2^n\times2^n$4	$2^n\times2^n$5
Compression	$2^n\times2^n$6	$2^n\times2^n$7	$2^n\times2^n$8
LCU–Kronecker–sum	$2^n\times2^n$9	$U$0	$U$1

“Ancillas (original)” reflects naïve resource use without the gadget; “Ancillas (gadget)” shows the improved count. “Gate Overhead” is the extra circuit cost relative to the original block-encodings themselves.

6. Applications in Quantum Algorithms

Block-product gadgets directly address resource bottlenecks in quantum Hamiltonian simulation, quantum linear algebra, and related algorithmic primitives. The exponential reduction in ancilla utilization enables practical implementation of deep matrix-product chains, time-dependent simulation (Dyson-series), and structured matrix decompositions (e.g., Kronecker sums). The generic interface accommodates rectangular matrices and allows efficient realization of complex block-encodings such as compressed Hamiltonians and sums of products with LCU.

A plausible implication is the improved feasibility and algorithmic scaling for large-scale quantum simulations on near-term and future quantum devices using block-encoded linear algebraic primitives (Dong et al., 19 Sep 2025).