Quantum Signal Processing (QSP)

Overview

Quantum signal processing, or QSP, is a framework for building quantum circuits that apply a chosen polynomial function to a quantum system. It alternates a fixed operation that encodes a number x with a sequence of single-qubit rotations whose angles are carefully chosen. The result is a circuit that implements a polynomial in x of degree equal to the number of repetitions, using only a single extra qubit.

QSP gives the best known asymptotic algorithms for Hamiltonian simulation, amplitude amplification, and several other tasks, and it is the foundation for quantum singular value transformation. Martyn, Rossi, Tan, and Chuang described it in 2021 as a grand unification of quantum algorithms, because Grover search, phase estimation, and Hamiltonian simulation all appear as special cases.

The problem it solves

Designing optimal quantum circuits for applying complex mathematical functions to quantum states, used throughout quantum simulation and quantum linear algebra.

History and background

The idea grew out of composite pulses in nuclear magnetic resonance, where sequences of carefully timed rotations are used to build robust operations. Low, Yoder, and Chuang described a methodology for resonant equiangular composite quantum gates in Physical Review X in 2016, which contained the QSP structure. In 2017, Low and Chuang used it to give an optimal Hamiltonian simulation algorithm in Physical Review Letters, matching known lower bounds in both time and error.

Low and Chuang extended the approach to qubitization in 2019. In the same year, Gilyén, Su, Low, and Wiebe generalized QSP to matrices with quantum singular value transformation, and Haah and later groups gave methods for computing the rotation angles. Chao and coauthors in 2020 and Dong, Meng, Whaley, and Lin in 2021 developed numerically stable angle-finding methods, which made the approach usable in software such as pyqsp.

How it works

Alternates a fixed quantum operation with carefully chosen single-qubit rotations to effectively apply a target polynomial function to the operation's eigenvalues, achieving near-optimal circuit depth.

Consider a single qubit and an operation that rotates it by an angle that encodes a number x between minus 1 and 1. By itself, one such rotation gives an output amplitude equal to x. If you interleave several of these rotations with fixed rotations about a different axis, the amplitude becomes a polynomial in x, whose degree equals the number of signal rotations.

The surprising fact is that the family of reachable polynomials is rich and precisely characterized. Any polynomial of degree d with the right parity, even or odd, and with magnitude at most 1 on the interval from minus 1 to 1 can be realized with d plus 1 phase angles. The angles do not depend on x, so the same circuit works for every value of x.

In practice, x stands for an eigenvalue of a larger operator, accessed through a block-encoding. Applying QSP to the block-encoded operator transforms every eigenvalue by the polynomial at once. Choosing the polynomial as an approximation to a function, such as the complex exponential for time evolution or the inverse function for matrix inversion, gives algorithms for those tasks.

Step by step

1. Choose the function f to apply, and find a polynomial P of degree d that approximates f well on the interval from minus 1 to 1, with the right parity and a magnitude bounded by 1. Chebyshev expansions are often used.
2. Compute the phase angles phi zero through phi d that make the QSP circuit implement P. This is a classical computation done once, using numerical methods.
3. Build the signal operator W of x, a rotation that encodes the value x, or a controlled block-encoding of the matrix of interest.
4. Alternate W with single-qubit phase rotations about the z axis by the computed angles, for a total of d applications of W.
5. Apply the circuit to the target and measure or project the ancilla. The system now has the polynomial of the operator applied to it.
6. Use amplitude amplification if needed to boost the success probability.

A worked example

To simulate time evolution under a Hamiltonian H with norm at most 1 for time t, use the Jacobi-Anger expansion, which writes e to the minus i x t as a sum of Bessel functions times Chebyshev polynomials. Truncating at degree d of order t plus a logarithmic term gives an accurate polynomial approximation. QSP then implements the real and imaginary parts as two polynomials, which combine to e to the minus i H t.

The resulting algorithm uses about t plus log of 1 over epsilon over log log of 1 over epsilon calls to the block-encoding of H, which matches the lower bound. Trotter-based methods, in comparison, need a number of steps that grows polynomially with 1 over epsilon.

Complexity and resources

What hardware it needs

• Fault-tolerant hardware for the deep circuits that arise for high degree polynomials.
• A block-encoding of the operator, which is often built with linear combination of unitaries.
• High-precision single-qubit rotations about the z axis, which are cheap on most platforms.

Applications

• Optimal Hamiltonian simulation for chemistry and materials.
• Amplitude amplification and search with fixed-point guarantees.
• Matrix inversion and linear systems, with improved dependence on condition number.
• Ground-state preparation, Gibbs sampling, and eigenvalue filtering.

Limitations and caveats

• Phase-angle computation can be numerically delicate, especially for polynomials of very high degree.
• The method needs a block-encoding of the operator, which can be expensive to construct.
• It is a fault-tolerant technique, and noise on current devices limits depth.

Common misconceptions

• QSP does not apply a function to the data stored in a state. It transforms the operator's eigenvalues, which acts on all eigencomponents simultaneously.
• It is not limited to Hamiltonian simulation. The same circuit structure handles many tasks.

How it compares

• Trotter-Suzuki product formulas are simpler but do not achieve optimal scaling.
• Linear combination of unitaries gives high precision but a larger cost from the sum of weights.
• Quantum singular value transformation extends QSP from eigenvalues to singular values of non-Hermitian matrices.

Real-world impact

Considered one of the most important theoretical advances in quantum algorithm design in the past decade, unifying and improving many previously separate quantum algorithms under one framework.

Why it matters

QSP changed how quantum algorithms are designed. Instead of inventing a separate circuit for each task, designers choose a polynomial and let QSP provide the circuit, which gave optimal or improved algorithms for many problems at once.

For people learning the field, it is also the clearest illustration of the idea that a few primitives, block-encoding and polynomial transformation, underlie most of the best quantum algorithms.

Frequently asked questions

Keep learning

• Related algorithm: Quantum Singular Value Transformation (QSVT)
• Related algorithm: Linear Combination of Unitaries (LCU)
• Related algorithm: Trotter-Suzuki Decomposition
• Related algorithm: Quantum Phase Estimation (QPE)

This page is written for educational purposes. Complexity statements describe the best known results at the time of review (September 30, 2026), and research in this area moves quickly, so check recent papers before relying on a specific bound.

← Back to Algorithms Database