⚛️ quantum physics

Quantum phase estimation with optimal confidence interval using three control qubits

This paper presents an efficient method to prepare a discrete prolate spheroidal sequence state using a compact matrix product state representation and simple three-qubit operations, enabling optimal quantum phase estimation with high-confidence intervals on early fault-tolerant quantum computers using only three control qubits.

Original authors: Kaur Kristjuhan, Dominic W. Berry

Published 2026-08-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kaur Kristjuhan, Dominic W. Berry

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

In the realm of quantum computing, scientists are constantly trying to solve problems that are impossible for ordinary computers to tackle, such as simulating complex chemical reactions to discover new medicines or breaking down large numbers to test security systems. At the heart of many of these powerful algorithms lies a fundamental task called quantum phase estimation. Imagine trying to determine the exact pitch of a musical note that is being played inside a sealed box; you cannot hear the note directly, but you can interact with the box in specific ways to infer the pitch. In the quantum world, this "pitch" is a value called a phase, which is hidden inside the behavior of a particle. To find it, researchers use a special helper system, often called a control register, which acts like a set of tuning forks. By applying a series of operations controlled by these tuning forks, the hidden phase leaves a fingerprint that can be read out. The accuracy of this reading depends heavily on how the tuning forks are prepared before the experiment begins. If they are all set to the same volume, the result is often blurry, much like trying to hear a single instrument in a room where every sound is equally loud and unrefined. To get a sharp, precise answer, the volume of these tuning forks needs to be carefully shaped, tapering off smoothly at the edges to focus the energy on the true value.

For decades, the most effective way to shape these quantum tuning forks has been known, but it came with a steep price. The ideal shape, known as a discrete prolate spheroidal sequence, acts like a perfect filter that concentrates the measurement probability around the true answer, giving scientists the highest possible confidence that their result is correct. However, preparing this perfect shape on a quantum computer has been notoriously difficult. Previous methods required a massive amount of computational resources and a large number of quantum bits, or qubits, to be set up all at once. This made the technique impractical for the early generations of fault-tolerant quantum computers that are just beginning to emerge, which have very limited numbers of qubits available. The challenge was to find a way to create this optimal shape without needing a huge machine, allowing the technique to work on smaller, more accessible devices.

A team of researchers has now solved this problem by discovering a surprisingly simple way to prepare the necessary quantum state. They found that the complex pattern required for the optimal shape can be broken down into a chain of very simple, local interactions. Instead of trying to build the entire state at once, they demonstrated that it can be constructed step-by-step using a mathematical structure called a matrix product state. This approach revealed that the complex pattern does not need a massive, intricate machine to create; it can be approximated with extreme accuracy using a bond dimension of just four. In practical terms, this means the researchers can build the state using a sequence of operations that only involve three qubits at a time. This is a dramatic reduction in complexity, transforming a task that previously seemed to require a large, unwieldy setup into one that can be performed with a tiny, manageable number of resources.

The researchers tested their method on systems ranging from small setups up to dimensions as large as 2 to the power of 24, which corresponds to a control register of 24 qubits. In every case, they found that using a bond dimension of four provided a highly accurate approximation of the ideal state. The error introduced by this simplification was so small that it was negligible for practical purposes, even when aiming for extremely high confidence levels. This means that scientists can now achieve the best possible precision in their phase estimates without needing to sacrifice the number of qubits they have available. The method is so efficient that if the size of the system is a power of two, the entire process can be run using only three qubits for the control register. This is achieved by measuring a qubit, using the result to adjust the next step, and then recycling that qubit for the next part of the calculation, rather than keeping all qubits active simultaneously.

This breakthrough is particularly significant for the future of quantum chemistry and other applications where knowing the exact energy of a system is critical. In these fields, the goal is often to find the lowest energy state, or the ground state, of a molecule. If the measurement is not precise enough, the computer might accidentally report the energy of a higher, excited state, leading to incorrect conclusions. By using this new, efficient preparation method, researchers can ensure that their confidence intervals are as tight as possible, minimizing the chance of such errors. The study shows that the optimal performance previously thought to require large, complex machines is actually within reach of the smaller, early-generation quantum computers that are currently being built. The work provides a clear path forward, demonstrating that with the right approach, the most powerful tools of quantum estimation can be made to fit on the smallest available hardware, opening the door for more reliable and accessible quantum simulations in the near future.

Technical Summary: Quantum Phase Estimation with Optimal Confidence Interval Using Three Control Qubits

Problem Statement
Quantum Phase Estimation (QPE) is a fundamental subroutine for algorithms such as integer factorization and ground state energy estimation in quantum chemistry. The standard textbook approach prepares an equal superposition (rectangular window) on a control register. However, this uniform weighting causes spectral leakage, spreading energy across neighboring frequency bins and degrading the precision of the phase estimate. While classical signal processing utilizes window functions (e.g., Discrete Prolate Spheroidal Sequences, or DPSS) to suppress leakage and maximize the concentration of probability around the true frequency, implementing these optimal states on quantum hardware is resource-intensive.

Existing methods for preparing DPSS states, such as Grover-Rudolph or Quantum Singular Value Transformation, typically require a number of non-Clifford gates that scales polynomially with the dimension of the state (exponentially with the number of qubits). Furthermore, these methods often require preparing all qubits simultaneously and may rely on Quantum Read-Only Memory (QROM), necessitating a large number of ancilla qubits. These requirements are prohibitive for early-generation fault-tolerant quantum computers, which are characterized by limited logical qubit counts and slow gate operations. The challenge is to prepare a DPSS state efficiently to minimize the width of the confidence interval for the phase estimate while adhering to strict resource constraints.

Methodology
The authors propose a resource-efficient state preparation scheme based on Matrix Product State (MPS) representations. The core of the methodology involves:

MPS Approximation of DPSS: The authors demonstrate that the DPSS state, which solves the spectral concentration problem to minimize confidence interval width, can be accurately approximated by an MPS with a very low bond dimension ( $\chi$ ).
Circuit Construction: They derive a quantum circuit to prepare these MPS states using a sequence of simple three-qubit operations. By utilizing the right-canonical form of the MPS, the circuit prepares the state starting from the most significant qubit.
Semi-Classical Fourier Transform and Qubit Recycling: When the dimension of the DPSS state is a power of two ( $D=2^n$ ), the authors integrate the MPS preparation with a semi-classical inverse Quantum Fourier Transform (QFT). This approach allows for the sequential measurement and recycling of control qubits. As each qubit is measured to extract a bit of the phase estimate, it is reset to $|0\rangle$ and reused for the next operation in the MPS preparation sequence.
Arbitrary Dimension Handling: For dimensions that are not powers of two, the authors propose padding the DPSS state with amplitudes derived from the continuous prolate spheroidal wave function. An inequality test is then applied to post-select the desired $D$ -dimensional subspace, ensuring the state remains an accurate approximation of the DPSS.

Key Results

Bond Dimension Sufficiency: Numerical analysis across dimensions up to $2^{24}$ reveals that an MPS with a bond dimension of $\chi=4$ is sufficient to provide a highly accurate approximation of the DPSS state. The infidelity ( $1-F$ ) for $\chi=4$ is on the order of $10^{-11}$ to $10^{-14}$ , which is negligible for practical applications. Increasing the bond dimension to 8 or 16 yields no significant improvement in fidelity.
Confidence Interval Performance: The relative increase in error probability ( $\delta$ ) when using an MPS approximation with $\chi=4$ instead of the exact DPSS is well below 0.1% for confidence levels up to $1-10^{-4}$ . The approximation maintains the optimal confidence interval width up to a sharp cutoff determined by the bond dimension.
Resource Efficiency: The proposed circuit requires only three qubits for the control register at any given time, regardless of the total dimension $D$ (provided $D$ is a power of two). This is achieved by recycling qubits through sequential measurement. The number of non-Clifford gates scales linearly with the number of qubits in the control register, a significant improvement over prior polynomial scaling methods.
Padding Strategy: For non-power-of-two dimensions, the paper demonstrates that padding the state with values from the prolate spheroidal wave function (rather than zero-padding) allows the MPS to approximate the target state with high fidelity while maintaining a high success probability for the post-selection step.

Significance and Claims
The paper claims that this work provides a practical pathway for implementing optimal QPE on early-generation fault-tolerant quantum computers. By reducing the control register requirement to just three logical qubits and utilizing a low-bond-dimension MPS, the method circumvents the qubit count limitations that currently hinder the use of optimal window functions like DPSS.

The authors emphasize that their approach enables the user to perform the minimal number of controlled applications of $U$ to achieve a target confidence interval width, which is critical for ground state energy estimation where the cost of Hamiltonian simulation scales linearly with the evolution time. The work establishes that high-precision phase estimation with optimal confidence intervals is feasible without the massive overhead of previous state preparation techniques, making it suitable for systems with limited logical qubit availability. The paper does not claim to solve the general problem of arbitrary state preparation but specifically targets the DPSS state for QPE, showing that its complex structure can be captured efficiently by simple, local quantum operations.

Technical Summary: Quantum Phase Estimation with Optimal Confidence Interval Using Three Control Qubits

More like this