⚛️ quantum physics

Reducing the Complexity of Matrix Multiplication to $O(N^2log_2N)$ by an Asymptotically Optimal Quantum Algorithm

This paper proposes a quantum kernel-based matrix multiplication algorithm (QKMM) that achieves an asymptotically optimal complexity of $O(N^2 \log_2 N)$ , outperforming classical methods in both theoretical efficiency and practical performance.

Original authors: Jiaqi Yao, Ding Liu

Published 2026-02-10

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Jiaqi Yao, Ding Liu

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The "Super-Fast Librarian" Breakthrough: Making Math Move at Light Speed

Imagine you are a librarian in a massive, infinite library. Every day, your boss gives you two giant books, each containing millions of numbers arranged in grids (these are matrices). Your job is to perform a complex mathematical dance called "matrix multiplication"—essentially, you have to cross-reference every single number in the first book with every single number in the second book to create a third, brand-new book.

The Problem: The "Classical" Slowdown

For decades, the smartest mathematicians in the world (the "Classical Librarians") have been trying to find faster ways to do this. They’ve found clever shortcuts, but there is a mathematical wall they keep hitting. As the books get bigger, the time it takes to finish the job grows explosively. If you double the size of the books, the work doesn't just double; it grows by a massive factor. For modern AI (like ChatGPT), this "math homework" is the biggest bottleneck in the world. It’s like trying to empty an ocean with a teaspoon.

The Solution: The Quantum "Ghost" Librarian

This paper introduces a new way to do this work using Quantum Computing.

Think of a Classical Librarian as a person who has to read one line, then move to the next, then the next. Even if they are incredibly fast, they are still one person reading one line at a time.

The authors propose a Quantum Kernel-based Matrix Multiplication (QKMM). Instead of one person reading lines, imagine if the librarian could turn into a mist that fills the entire library at once. This "mist" doesn't read line by line; it touches every number in both books simultaneously.

How does it work? (The "Magic Mirror" Analogy)

The researchers use something called a "Quantum Kernel."

Imagine you have two complex patterns. Instead of measuring them with a ruler (which takes forever), you shine a special light through them onto a mirror. The way the light bounces off the patterns and hits the mirror tells you everything you need to know about how the patterns relate to each other, all in one single flash of light.

In this paper, the "light" is the quantum state, and the "mirror" is the quantum circuit. By using this "flash," they can calculate the results of the multiplication much faster than any classical human or computer ever could.

Why is this a big deal? (The "Speed Limit" Argument)

In math, there is a "speed limit"—a theoretical minimum amount of work required to solve a problem.

Classical computers are like cars driving on a winding mountain road; they can go fast, but they are limited by the curves.
This new Quantum algorithm is like a jet plane.

The paper mathematically proves that as the matrices (the books) get larger and larger, this quantum algorithm gets closer and closer to the absolute "speed limit" of physics. It is what scientists call "asymptotically optimal." In plain English: it is as close to perfect as math allows.

Will it work in the real world? (The "Stormy Weather" Test)

Quantum computers are notoriously "fussy." They are like delicate musical instruments that go out of tune if someone even breathes near them (this is called "noise").

The researchers didn't just say, "This works in theory!" They ran simulations to see how the algorithm would handle "noisy" or "stormy" conditions (simulating real-world, imperfect quantum hardware). They found that even when the "weather" was bad, their method was much more stable and reliable than previous quantum attempts. It’s like a ship that can still navigate even when the ocean is choppy.

Summary

The Old Way: Reading a massive book line-by-line, getting slower and slower as the book grows.
The New Way (QKMM): Turning the book into a cloud of light and seeing the answer in a single, massive flash.

This discovery paves the way for much faster AI, better scientific simulations, and a future where the most complex math problems on Earth can be solved in the blink of an eye.

Technical Summary: Reducing the Complexity of Matrix Multiplication to $O(N^2 \log^2 N)$

1. Problem Statement

Matrix multiplication is a cornerstone of modern computational science, particularly for training deep neural networks and performing large-scale linear algebra. While classical algorithms have progressively reduced the complexity from the standard $O(N^3)$ toward the current best of $O(N^{2.371552})$ , reaching the theoretical lower bound of $O(N^2)$ remains an unsolved challenge.

Existing Quantum Matrix Multiplication (QMM) research lacks a unified complexity metric. Current approaches often rely on "circuit stacking," which leads to cubic scaling ( $O(N^3)$ ) in quantum resources (qubits and gates), making them impractical for the Noisy Intermediate-Scale Quantum (NISQ) era. There is a critical need for a QMM algorithm that is both asymptotically efficient and resource-optimized.

2. Methodology

The authors propose the Quantum Kernel-based Matrix Multiplication (QKMM) algorithm. The methodology is built upon the principle of the quantum kernel inner product modulus method, which measures the similarity between quantum states via their inner products.

The algorithm follows a hierarchical, parallelized structure:

Vector-to-Vector (V2V): The fundamental building block. It uses amplitude encoding to map normalized vectors into quantum states and extracts the inner product modulus directly from the ground state $|0\rangle^{\otimes n}$ , eliminating the need for ancillary qubits or complex measurements.
Vector-to-Matrix (V2M): An extension of V2V that uses an index register and multi-controlled amplitude encoding to perform matrix-vector multiplication in a single circuit.
Matrix-to-Matrix (M2M): A dual-parallel extension using two index registers. It decomposes matrices $A$ and $B$ into rows and columns, respectively, allowing the entire product $C = AB$ to be computed in one quantum circuit.
Multi-Matrix Multiplication (M-MM): A further extension that allows a single circuit to compute the product of one matrix with multiple other matrices simultaneously.

Complexity Analysis: The authors define complexity using basic quantum gate counts (the fundamental unit of quantum computation). By analyzing the decomposition of multi-controlled $RY$ gates required for amplitude encoding, they mathematically derive a total gate complexity of $O(N^2 \log^2 N)$ .

3. Key Contributions

Asymptotic Optimality: The paper provides a rigorous mathematical proof that as $N \to \infty$ , the complexity $O(N^2 \log^2 N)$ approaches the theoretical lower bound of $O(N^2)$ , outperforming all known classical algorithms.
Unified Metric: It establishes "quantum gate complexity" as a standardized metric for evaluating QMM, providing a more granular and consistent comparison than previous query-based or qubit-based metrics.
Resource Efficiency: The QKMM design minimizes qubit overhead and avoids the "circuit stacking" problem, making it more suitable for NISQ-era hardware.
Comprehensive Framework: The work provides a complete suite of algorithms (V2V, V2M, M2M, M-MM) and detailed circuit implementations.

4. Results

The authors validated their algorithm through extensive simulations using the pyQPanda library on the Origin Quantum platform under two conditions:

Noiseless Simulation:
- The QKMM demonstrated significant parallel acceleration.
- As the level of parallelism increased (from V2V to M-MM), the computational time decreased, confirming the benefits of quantum superposition.
Noisy Simulation (NISQ-like environment):
- The algorithm was tested against realistic noise parameters ( $T_1, T_2$ relaxation and gate error rates).
- Robustness: At low dimensions (2–4), QKMM maintained high fidelity ( $>0.95$ ).
- Error Analysis: The study found that while environmental noise ( $T_1, T_2$ ) has a negligible effect on fidelity, quantum gate noise is the dominant factor in performance degradation as dimensions increase. Under combined noise, fidelity drops sharply, highlighting the need for future improvements in data encoding efficiency.

5. Significance

This research marks a significant step toward "quantum-intensive computing." By providing an asymptotically optimal algorithm for matrix multiplication, the authors offer a potential pathway to accelerate the training of massive AI models and solve complex scientific computing problems. The work bridges the gap between theoretical quantum advantage and practical implementation by addressing the specific resource constraints of current quantum hardware.