Benchmarking Quantum Instruments

Darian McLaren, Matthew A. Graydon, Ali Mahmoud, Joel J. Wallman · 2025

Read Benchmarking Quantum Instruments by Darian McLaren, Matthew A. Graydon, Ali Mahmoud, Joel J. Wallman (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00179.txt.

Introduction

Darian McLaren and co-authors present a way to gauge errors in quantum measures that use feed forward, in a 2025 paper. Such measures are key parts of fault tolerant quantum computers, where later steps depend on earlier results. Their method fits the chance that a long run of mixed up measures all give the ideal result. All data come from one long run of measures in a row.

The authors also prove that key quality scores stay the same under the mixing step. Those scores can combine with the error rate to describe the faults at play, up to one known ambiguity about timing. Readers get a simpler test for a part of quantum hardware that is hard to check in other ways.

Catalog record

CREATORDarian McLaren, Matthew A. Graydon, Ali Mahmoud, Joel J. Wallman
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE9 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-00179

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Benchmarking Quantum Instruments
Darian McLaren, Matthew A. Graydon, Ali Mahmoud, Joel J. Wallman

Abstract
Quantum measurements with feed-forward are crucial components of fault-tolerant quantum computers. We show how the error rate of such a measurement can be directly estimated by fitting the probability that successive randomly compiled measurements all return the ideal outcome. Unlike conventional randomized benchmarking experiments and alternative measurement characterization protocols, all the data can be obtained using a single sufficiently large number of successive measurements. We also prove that generalized Pauli fidelities are invariant under randomized compiling and can be combined with the error rate to characterize the underlying errors up to a gauge transformation that introduces an ambiguity between errors happening before or after measurements.



Quantum instruments (
i.e
., nondestructive measurements with classical feed-forward
QI1
) are vital for detecting and correcting errors during quantum computation and communication. However, quantum instruments are also subject to noise, and so introduce errors into otherwise fault-tolerant schemes.
Therefore quantum instruments need to be characterized and any errors must be corrected to reliably process quantum information.

Historically, quantum characterization has primarily focused on quantum gates
eisert2020quantum
.
While complete characterizations are in principle possible using techniques such as process
Chuang1997
and gate-set tomography
Merkel2013
, the complexity of quantum error processes renders such characterizations untenable for even moderately sized quantum systems.
These techniques can be generalized to characterize quantum instruments
QIQND
;
Rudinger2022
;
Pereira2022
;
pereira2023
, but still suffer the same fundamental scaling issue.

An alternative approach is to partially characterize errors by estimating figures of merit.
The canonical example of this approach is randomized benchmarking
magesan2011scalable
, wherein a unitary 2-design
Dankert2009
is used to “twirl” an arbitrary error channel into a global depolarizing channel with the same process fidelity.
The depolarizing channel is repeated a variable number of times and the process fidelity is estimated by fitting the survival probability to an exponential curve.
The primary drawback of this approach is that many entangling operations are needed to produce a global depolarizing channel, so that the error rate is only loosely connected to the performance of individual gates in a general circuit.

This drawback was overcome by the combination of cycle benchmarking
erhard2019characterizing
and randomized compiling
Wallman2016a
, which use local twirling operations to produce stochastic channels
mclaren2023stochastic
.
Unlike global depolarizing channels, stochastic error channels are described by a probability distribution and the action of the error depends on the input state.
When the twirling operations are Pauli operations (or Weyl operations for qudits), preparations and measurements in local Pauli eigenstates can be used to isolate portions of the state space in which the action of the error channel is described by a single parameter that can be amplified and estimated by fitting to an exponential decay.
These parameters are referred to as Pauli fidelities, and their average is exactly the process fidelity, which determines the error probability in randomly compiled circuits.
Moreover, additional processing of these Pauli fidelities can be performed to estimate the probabilities of specific errors
flammia2020
.

Randomized compiling and cycle benchmarking have both recently been generalized to quantum instruments
beale2023randomized
;
zhang2024
;
hines2024
.
These schemes are designed to characterize syndrome measurements and thus allow nontrivial circuits to entangle measured and unmeasured qubits.
Though useful, allowing entangling circuits and a subsystem of unmeasured qubits introduces significant complexity as: not all generalized Pauli fidelities are learnable
zhang2024
;
the circuits depend on the Pauli fidelities to be learned; and benchmarking the unmeasured qubits increases the number of experiments.

In this paper, we present a simplified and more efficient protocol for characterizing instruments in the absence of other gates, and unmeasured qudits.
Specifically, we prove that the probability of obtaining the ideal outcome from the first
m
of
k
randomly compiled measurements decays exponentially in
k
, where, to first order, the base is the error rate.
We then show how additional post-processing can be used to learn more detailed properties of the noisy measurement.
As an example, we show that single-qubit instruments can be characterized up to a gauge ambiguity between errors occurring before or after the measurement.

The balance of this paper is structured as follows. In
section

I
, we define the multi-qudit Weyl operators and review how effective projectors can be introduced in a circuit. In
section

II
, we prove that the generalized Pauli infidelities
zhang2024
are invariant under randomized compiling and completely characterize the randomly compiled instrument. In
, we present our protocol for characterizing a noisy quantum instrument assuming no idling qudits along with a simulated example of its execution. We conclude with
. We also prove a bound on deviation from an ideal rank 1 projector in
that may be of independent interest.

I
PRELIMINARIES

We briefly set out notation and define the
n
-qudit Weyl operators and channels.
We will use the convention that operators acting on pure states will be denoted using capital Roman font (e.g.,
X
,
Z
), whilst channels will be denoted using capital calligraphic letters (e.g.,
\mathcal{M}
,
\mathcal{X}
,
\mathcal{Z}
).
For any operator
A
acting on pure states, we use the same letter in calligraphic font to denote the corresponding channel that acts by conjugation, i.e.,
\mathcal{A}(\rho)=A\rho A^{\dagger}
.

Let
d
be a positive integer and
\oplus
denote addition modulo
d
.
The characters of
\mathbb{Z}_{d}^{n}
are the functions

\displaystyle\chi_{z}:\mathbb{Z}_{d}^{n}\to\mathbb{C}::\chi_{z}(j)=\exp\left(\frac{2\pi iz\cdot j}{d}\right),

(1)

for
z\in\mathbb{Z}_{d}^{n}
.
These functions satisfy Schur’s orthogonality relations,

\displaystyle d^{-n}\sum_{z}\chi_{j}^{*}(z)\chi_{k}(z)=\delta_{j,k}.

(2)

Note that we will frequently use the identities

\displaystyle\chi_{a}(b)=\chi^{*}_{a}(-b)=\chi^{*}_{-a}(b)=\chi_{b}(a).

(3)

The generalized single-qudit
X
and
Z
operators are

\displaystyle X=\sum_{j\in\mathbb{Z}_{d}}\left|j\oplus 1\middle\rangle\!\middle\langle j\right|,\quad Z=\sum_{j\in\mathbb{Z}_{d}}\chi_{1}(j)\left|j\middle\rangle\!\middle\langle j\right|.

(4)

We use the vectorized power notation

\displaystyle A^{a}=\otimes_{i\in\mathbb{Z}_{n}}A^{a_{i}}

(5)

for an operator
A
and a vector
a\in\mathbb{Z}_{d}^{n}
to describe tensor products of powers of the
X
and
Z
operators, so that for
x,z\in\mathbb{Z}_{d}^{n}
,

\displaystyle X^{x}=\sum_{j\in\mathbb{Z}_{d}^{n}}\left|j\oplus x\middle\rangle\!\middle\langle j\right|,\quad Z^{z}=\sum_{j\in\mathbb{Z}_{d}^{n}}\chi_{z}(j)\left|j\middle\rangle\!\middle\langle j\right|.

(6)

We thus have the braiding relation

\displaystyle Z^{z}X^{x}=\chi_{z}(x)X^{x}Z^{z}

(7)

and, from
eq.

2
, the inversion formula

\displaystyle\left|j\middle\rangle\!\middle\langle j\right|=d^{-n}\sum_{z\in\mathbb{Z}_{d}^{n}}\chi_{j}^{*}(z)Z^{z}.

(8)

The
n
-qudit projective Weyl group is the set

\displaystyle\{Z^{z}X^{x}:x,z\in\mathbb{Z}_{d}^{n}\}.

(9)

Quantum operations are completely positive trace non-increasing maps acting between the spaces of positive semi-definite operators.
We will use the vectorization map

\displaystyle\left.\left|A\right\rangle\!\right\rangle=\sum_{a,b\in\mathbb{Z}_{d}^{n}}\langle b|A|a\rangle\bigotimes_{j\in\mathbb{Z}_{n}}|a_{j}b_{j}\rangle,

(10)

so that
\left.\left|A\otimes B\right\rangle\!\right\rangle=\left.\left|A\right\rangle\!\right\rangle\otimes\left.\left|B\right\rangle\!\right\rangle
and
\langle\!\langle A|B\rangle\!\rangle=\tr A^{\dagger}B
.
For clarity, we will use
\left.\left|k\right\rangle\!\right\rangle=\left.\left|\left|k\middle\rangle\!\middle\langle k\right|\right\rangle\!\right\rangle
.
If we wish to restrict to considering (normalized) quantum states we define a quantum channel to be any completely positive trace preserving map between spaces of density operators.
Of vital importance are the Weyl channels
\mathcal{X}^{x}
and
\mathcal{Z}^{z}
, which satisfy

\displaystyle\mathcal{X}^{x}

\displaystyle=d^{-n}\sum_{a,b}\chi^{*}_{x}(a)|Z^{a}X^{b}\rangle\!\rangle\!\langle\!\langle Z^{a}X^{b}|

\displaystyle\mathcal{Z}^{z}

\displaystyle=d^{-n}\sum_{a,b}\chi_{z}(b)|Z^{a}X^{b}\rangle\!\rangle\!\langle\!\langle Z^{a}X^{b}|.

(11)

We can then construct effective projectors by inserting random Weyl operators, since by
eq.

2
,

\displaystyle\sum_{x}\chi_{x}(c)\mathcal{X}^{x}

\displaystyle=\sum_{b}|Z^{c}X^{b}\rangle\!\rangle\!\langle\!\langle Z^{c}X^{b}|

\displaystyle\sum_{z}\mathcal{Z}^{z}

\displaystyle=\sum_{a}|Z^{a}\rangle\!\rangle\!\langle\!\langle Z^{a}|.

(12)

(Note that
(
beale2023randomized
, Lemma 1)
incorrectly differs from a normalization factor as the
d^{-n}
normalization factor in
section

I
was omitted.)

II

Read the complete text