Error mitigation for logical circuits using decoder confidence
Extracted source text · arXiv 2512.15689
\documentclass[%
aps,
prx,
reprint,
superscriptaddress,
nofootinbib,
amsmath,
amssymb
]{revtex4-2}
\usepackage{
physics,
graphicx,
multirow,
tabularx,
placeins,
nicefrac,
hyperref,
threeparttable,
enumitem,
makecell,
comment,
}
\usepackage[print-unity-mantissa=false]{siunitx}
\usepackage{mathtools}\mathtoolsset{centercolon}
\usepackage{fixmath}
\usepackage{phfparen}
\usepackage{mathcommand}
\usepackage{stmaryrd}
% declare commands which only apply in math mode e.g. \BG
\LoopCommands{\lettersAll{CG}{DCS}{NLL}{SE}}{\declaremathcommand#2{\textnormal{#1}}}
% declare operatornames which only apply in math mode
\LoopCommands{
{pr}{wt}
}{\DeclareMathOperator#2{#1}}
\usepackage{algorithm}
\usepackage{algpseudocodex}
\usepackage[capitalize]{cleveref}
\Crefname{defn}{Definition}{Definitions}
\Crefname{appendix}{Appendix}{Appendices}
\usepackage{amsthm}
\newtheorem{cor}{Corollary}
\newtheorem{prop}{Proposition}
\newtheorem{lem}{Lemma}
\theoremstyle{definition}
\newtheorem{defn}{Definition}
\newtheorem{cond}{Condition}
\theoremstyle{remark}
\newtheorem{rem}{Remark}
\usepackage[dvipsnames]{xcolor}
\hypersetup{
colorlinks,
linkcolor={blue!50!black},
citecolor={blue!50!black},
urlcolor={blue!80!black}
}% remove the box around citation link and bibilio link
\usepackage[caption=false]{subfig} %The caption part is not compatible with revtex
\captionsetup{justification=justified, singlelinecheck=off}
% \graphicspath{{Figures/}}
\setlength\fboxrule{.8pt}
\newcommand{\supop}[1]{\ensuremath{\overbracket[0.1ex][0.3ex]{#1}}}
\DeclarePairedDelimiter\pbra{\langle\!\langle}{\rvert}
\DeclarePairedDelimiter\pket{\lvert}{\rangle\!\rangle}
\DeclarePairedDelimiterX\pbraket[2]{\langle\!\langle}{\rangle\!\rangle}{#1 \delimsize\vert #2}
\newcommand{\fiteqn}[1]{\resizebox{\hsize}{!}{$#1$}} %Used within math env to fit single line eqn
\newcommand{\fittext}[1]{\resizebox{.95\linewidth}{!}{\begin{minipage}{\linewidth}#1 \end{minipage}}} %Used in text env to fit anything.
%Comments:
\newcommand{\SCB}[1]{{\color{purple} #1}}
\newcommand{\change}[1]{{\color{blue} #1}}
\newcommand{\TC}[1]{{\color{red} #1}}
\newcommand{\MD}[1]{{\color{orange} #1}}
\begin{document}
\title{Error mitigation for logical circuits using decoder confidence }
\newcommand{\qmaddress}{\affiliation{Quantum Motion, 9 Sterling Way, London N7 9HJ, United Kingdom}}
\newcommand{\oxaddress}{\affiliation{Department of Materials, University of Oxford, Parks Road, Oxford OX1 3PH, United Kingdom}}
\author{Maria Dincă}
\email{maria.dinca@materials.ox.ac.uk}
\oxaddress
\author{Tim Chan}
\email{timothy.chan@materials.ox.ac.uk}
\oxaddress
\author{Simon C. Benjamin}
\oxaddress
\qmaddress
\date{\today}
\begin{abstract}
Fault-tolerant quantum computers use decoders to monitor for errors and find a plausible correction. A decoder may provide a \emph{decoder confidence score} (DCS) to gauge its success. We adopt a \emph{swim distance} DCS, computed from the shortest path between syndrome clusters. By contracting tensor networks, we compare its performance to the well-known \emph{complementary gap} and find that both reliably estimate the logical error probability (LEP) in a decoding window. We explore ways to use this to mitigate the LEP in entire circuits. For shallow circuits, we just abort if any decoding window produces an exceptionally low DCS: for a distance-13 surface code, rejecting a mere $0.1\%$ of possible DCS values improves the entire circuit's LEP by more than 5 orders of magnitude. For larger algorithms comprising up to trillions of windows, DCS-based rejection remains effective for enhancing observable estimation. Moreover, one can use DCS to assign each circuit's output a unique LEP, and use it as a basis for maximum likelihood inference. This can reduce the effects of noise by an order of magnitude at no quantum cost; methods can be combined for further improvements.
\end{abstract}
\maketitle
\section{Introduction} \label{part_0}
In fault-tolerant quantum computing, we perform a series of measurements that can reveal information about the occurrence of errors on the system of qubits, without collapsing their quantum state. The set of observations generated by these measurements forms a syndrome. A key task in quantum error correction (QEC)~\cite{shor_scheme_1995, fowler_surface_2012} is to map the syndromes observed on the error correcting code to the suspected physical error mechanisms which might have produced them, and consequently suggest plausible corrections. This process is known as decoding.
There has been recent interest in extracting more information from the decoder than just a correction;
namely, an estimate of how likely it is for the most plausible correction to be actually correct.
In the past,
this estimate has been called `soft output'~\cite{Chen2025,meister_efficient_2024,Lee2025c}
or `soft information'~\cite{bombin_fault-tolerant_2024,deMartiiOlius2024,Gidney2025,Pattison2025,Akahoshi2025,Sunami2025},
but we opt for the name `decoder confidence score' (DCS) as it is more descriptive.
It has been identified that additional information of this kind can be used in
code concatenation~\cite{Poulin2006,Goto2013,Gidney2025,meister_efficient_2024},
defect identification in silicon-spin quantum processors~\cite{siegel_snakes_2025},
magic state distillation~\cite{Smith2024,SalesRodriguez2025} and cultivation~\cite{gidney_magic_2024,Chen2025,Sahay2025},
circuit sampling~\cite{meister_efficient_2024, Smith2024},
lattice surgery runtime reduction~\cite{Akahoshi2025},
hierarchical decoding~\cite{Shutty2024,Toshio2025},
and distributed quantum computing~\cite{Sunami2025}.
Typically, fault-tolerant quantum computing involves decoding in a series of windows:
batches of syndromes collected in a finite amount of time
across a small region of space are decoded at once.
A typical window extent is a small multiple of the code distance~\cite{Dennis2002}.
The DCS is particularly useful here,
since each window can be associated with its own DCS value,
thus providing a localised failure probability for each component of the quantum computation.
Two existing DCSs that have been studied for the surface code are
the complementary gap and a cheaper alternative which we call the swim distance.
The complementary gap was first introduced as `distinguishability' in~\cite{Hutter2014},
before being studied further in Refs.~\cite{bombin_fault-tolerant_2024,Gidney2025}.
The swim distance was introduced simultaneously in Refs.~\cite{Smith2024,meister_efficient_2024}
as a faster alternative to the complementary gap
(it has a lower time complexity);
however,
neither reference provides an accuracy comparison against the complementary gap.
In this paper, we refine the picture of the DCS with three observations.
First, we use tensor network methods
to compare the accuracy of the complementary gap and the swim distance
under phenomenological noise.
While we find the complementary gap to be somewhat more accurate,
both DCSs are accurate proxies for the success probability.
For the rest of the paper,
we employ the swim distance as it is faster to compute.
Second, we identify characteristics from the statistical distribution of the DCS
that make it a useful resource for large-scale applications.
Third, we explore this utility for quantum error mitigation (QEM) in logical circuits.
In the case of a shallow circuit -- such as logical state preparation --
simply aborting when a poor DCS occurs is very effective.
We then consider the case of an entire algorithm,
which may involve a vast number of decoding windows.
The strategy of aborting unpromising circuits remains tenable,
but we find that another option is attractive:
estimate the whole-circuit logical error probability using its DCS record,
then perform maximum likelihood estimation (MLE) using the estimated risk. Depending on target performance,
a combination of methods may be optimal. This is shown in \cref{fig:MLE_abort}b,d as a primary result of this work.
\Cref{sec:background} introduces the prerequisite concepts.
We compare the complementary gap and the swim distance in \cref{part_1},
before showing that the DCS provides rich error information for deep circuits in \cref{part_2}.
\Cref{part_3} explores the use of the DCS in QEM
and is followed by a numerical resource estimation in \cref{sec:application}.
We discuss limitations and future work in \cref{sec:discussion},
then conclude in \cref{sec:conclusion}.
\section{Background}\label{sec:background}
In \cref{sec:the_role_of_a_decoder} we discuss the role of a decoder and present the decoding problem,
then in \cref{sec:decoder_confidence_scores} we define
the DCS, the complementary gap, and the swim distance.
\subsection{The Role of a Decoder}\label{sec:the_role_of_a_decoder}
In this work, we focus on the surface code,
for which the $X$ and $Z$ errors can be dealt with separately on independent decoding graphs.
We consider decoding only one type (either $X$ or $Z$) of error;
the results for the other type of error are analogous.
In error correction,
nondestructive parity measurements are performed on physical qubits
and the combination of their results is called the syndrome.
Error chains (configurations of errors on the physical qubits)
are detected only through their syndrome,
but different error chains can give rise to the same syndrome.
Additionally,
each error chain belongs to one of two logical equivalence classes (LECs)
that uniquely specifies its effect on the logical information.
A decoder deduces from the syndrome
a plausible LEC for the error chain that caused the syndrome.
We will mainly be concerned with decoders that deduce an explicit plausible error chain.
What matters is
not whether the deduced error chain equals the true error chain,
but whether they belong to the same LEC.
If they are in different LECs,
then the decoding procedure results in a logical error.
Therefore, a decoder effectively makes guesses --
and while even the ideal decoder makes the best possible guess,
it may nevertheless lead to a logical error.
When choosing a decoder,
there is a trade-off between the accuracy with which the likeliest LEC is found and the time it takes to decode.
More formally,
the decoding problem involves the decoding graph $G =(V_\d \cup V_\b, E)$,
with the set $V_\d$ of detectors and the set $V_\b$ of boundary nodes such that $V_\d \cap V_\b =\varnothing$.
Each boundary node in $V_\b$ represents a unique boundary of the surface code
(so for a memory experiment, $|V_\b| =2$).
An example of how $G$ relates to its code is shown in \cref{fig:to_decoding_graph}.
\begin{figure}
\centering
\begin{tabular}{cc}
\includegraphics[width=0.5\linewidth]{boundary_nodes_unmerged.pdf} &
\includegraphics[width=0.5\linewidth]{boundary_nodes_merged.pdf} \\
(a) & (b)
\end{tabular}
\caption{(a) The $X$ decoding graph for the
$\llbracket n, k, d_X, d_Z \rrbracket =\llbracket 23, 1, 5, 3 \rrbracket$ unrotated surface code
under code capacity noise.
Detectors are green circles,
boundary nodes are blue circles.
In this noise model (but not necessarily in others),
each detector corresponds to a $Z$ stabiliser,
and each edge corresponds to a data qubit
(emphasised by a grey circle on each edge).
(b) It later helps if there is only one boundary node
for each unique boundary of the surface code,
so we merge equivalent boundary nodes as shown.
Here, we no longer show detectors nor data qubit circles.}
\label{fig:to_decoding_graph}
\end{figure}
Each edge in $E$ represents an error mechanism that flips the one or two detectors it is incident to,
and has a weight $\log_{10} \frac{1 -p}{p}$ that decreases
as the probability $p$ of that error mechanism increases.
Define the weight $\wt(F)$ of any subset $F \subseteq E$ as the sum of the weights of all edges in that subset.
The error $\mathbb E \subseteq E$ represents those mechanisms that actually occur,
from which we define the syndrome $\mathbb S \subseteq V_\d$ as
the detectors that $\mathbb E$ flips an odd number of times.
Given $G$ and $\mathbb S$,
the decoder outputs a correction $\mathbb C =\textsc{Decode}(G, \mathbb S) \subseteq E$
that will have the same syndrome as $\mathbb E$.
We say the decoder is {\em successful} if $\mathbb C$ is logically equivalent to $\mathbb E$.
\begin{defn}\label{defn:log_success_odds}
The \emph{logical error probability} (LEP), denoted $p_\L$,
is the probability that the decoder is not successful.
The \emph{log success odds} (or \emph{log-likelihood ratio of success}) is given by
\begin{equation}\label{eq:log_success_odds_definition}
\lambda :=\log_{10} \frac{1 -p_\L}{p_\L}.
\end{equation}
\end{defn}
\noindent
The log success odds decreases as $p_\L$ increases.
In addition to the LEP,
we define the following.
\begin{defn}\label{defn:ler}
When we have performed multiple runs of the same logical circuit,
the \emph{logical error rate} (LER)
is the portion of executions with a logical error.
\end{defn}
\noindent The LER converges to the mean LEP
as the run count grows.
\subsection{Decoder Confidence Scores}\label{sec:decoder_confidence_scores}
In addition to the correction,
the decoder can also output the following quantity
that scores how likely it believes success to be.
\begin{defn}\label{defn:dcs}
Given $G$ and $\mathbb S$,
the \emph{DCS} is an estimator
of some monotonic function $\alpha(\lambda)$ of $\lambda$.
\end{defn}
\noindent
The quantity $p_\L$ (or $\lambda$) can be exactly computed
by contracting a tensor network~\cite{piveteau_tensor-network_2024};
this was analysed recently in Ref.~\cite{Chen2025a}.
However,
this method cannot be used as a real-time DCS as its time complexity is exponential.
The following two definitions specify DCSs that run in polynomial time.
\begin{defn}\label{defn:complementary_gap}
The \emph{complementary gap} is the difference in weight between
the original correction and the complementary correction
(the correction output by the decoder
when forced to propose within the opposite LEC).
\end{defn}
\noindent
We can force the decoder to propose the opposite LEC
by forcing the left boundary node to flip
a number of times whose parity is opposite to that of the original correction.
This is described in \cref{alg:complementary_gap}
and illustrated by example in \cref{fig:gap_illustration}.
\begin{algorithm}[H]
\caption{Pseudocode to calculate the complementary gap of a memory experiment.}
\label{alg:complementary_gap}
\begin{algorithmic}
\State $\mathbb C \gets \Call{Decode}{G, \mathbb S}$
\State $G' \gets G$ but where the left boundary node $v_\L$ is a detector
\State $\mathbb S' \gets$ the syndrome of $\mathbb C$ on $G'$
\State flip whether or not $v_\L$ is included in $\mathbb S'$
\State $\mathbb C' \gets$ \Call{Decode}{$G', \mathbb C'$}
\State \Return $\wt(\mathbb C') -\wt(\mathbb C)$
\end{algorithmic}
\end{algorithm}
\begin{figure}
\centering
\begin{tabular}{cc}
\includegraphics[width=0.5\linewidth]{gap_a.pdf} &
\includegraphics[width=0.5\linewidth]{gap_b.pdf}
\end{tabular}
\caption{Illustrating the complementary gap
on the decoding graph $G$ in \cref{fig:to_decoding_graph}b.
The modified graph $G'$, shown in grey,
has only one remaining boundary node, shown as a blue dot.
The syndrome $\mathbb S$ ($\mathbb S'$) is the set of red (yellow) dots.
The correction $\mathbb C$ ($\mathbb C'$) is the set of red (yellow) edges.
The complementary gap is the weight difference between these two corrections;
here it is $5 -3 =2$ if all edges have weight 1.}
\label{fig:gap_illustration}
\end{figure}
\begin{figure}
\centering
\includegraphics[width=0.6\linewidth]{swim_distance.pdf}
\caption{Illustrating the swim distance,
which is here the sum of the lengths of the blue lines:
$2 \times 0.5 =1$ if all edges have weight 1.
Clusters are shaded in red.}
\label{fig:swim_distance_illustration}
\end{figure}
\begin{defn}\label{defn:swim_distance}
For any clustering decoder,
the \emph{swim distance}
is the shortest path length in the decoding graph from one boundary to the other
after zeroing the weights of all edges inside clusters.
\end{defn}
\noindent
Note a cluster can contain a fraction of an edge,
in which case only that fraction of its weight is set to zero.
\Cref{fig:swim_distance_illustration} illustrates the swim distance
for the same example as in \cref{fig:gap_illustration}
and shows that its value can differ from the complementary gap.
The name `swim distance' is based on the following analogy:
the decoding graph is a river,
the two boundaries are riverbanks,
and each cluster is an island in the river;
then,
the swim distance is the shortest distance one must swim to cross the river.
The complementary gap and the swim distance
are not the only DCSs that exist.
For instance,
Ref.~\cite{bluvsteinLogicalQuantumProcessor2024} consider simply the weight of the original correction,
Ref.~\cite{Lee2025c} defines two families of scores based on how much of the decoding graph is covered by clusters,
and Ref.~\cite{English2025} defines a score based on $|\mathbb{S}|$ alone.
Nevertheless,
the complementary gap remains considerably more effective
at reducing the LER in abort protocols~\cite{Lee2025c}.
We choose to analyse the complementary gap for this reason,
and the swim distance because it has a very strong theoretical motivation~\cite{meister_efficient_2024}.
For the numerics in this paper,
we decode using the PyMatching implementation~\cite{Higgott2025} of MWPM,
calculate the complementary gap using a custom implementation,
and calculate the swim distance using the implementation provided by Ref.~\cite{Chen2025}.
\section{Decoder confidence scores are an excellent indicator of logical error risk} \label{part_1}
\begin{figure}
\centering
\includegraphics[width=0.90\linewidth]{source_of_dcs_variation_perturb0.3_v2.pdf}
\caption{The two sources of variation for
the complementary gap (blue data)
and the swim distance (orange data).
The first source $\sigma_{\alpha(\lambda)}$
is due to the variation of the log success odds.
The second source $\sigma_r$
comes from the approximate nature of the DCS.
Results are obtained using the MWPM decoder
on the unrotated surface code under a perturbed phenomenological noise model
(described in \cref{sec:perturbed_phenomenological_noise_model}).}
\label{fig:source_of_dcs_variation_perturb0.3}
\end{figure}
In this section we verify that both the complementary gap and the swim distance
are accurate enough to be useful in error mitigation;
this is formalised as follows.
The log success odds $\lambda$ can be modelled as a random variable
whose variation throughout a typical quantum calculation
is characterised by its standard deviation $\sigma_\lambda$.
Since the DCS $\phi$ is an inexact estimator of $\alpha(\lambda)$,
it can be modelled as
\begin{equation}\label{eq:dcs_as_sum_of_2_rvs}
\phi =\alpha(\lambda) +r,
\end{equation}
where $r$ is another random variable whose standard deviation $\sigma_r$ represents the DCS inaccuracy.
For $\phi$ to be useful,
it must be able to resolve variations in $\lambda$ between different decoding instances,
i.e., $\sigma_{\alpha(\lambda)} =\alpha'(\lambda) \sigma_\lambda$ must be large compared to $\sigma_r$.
\Cref{fig:source_of_dcs_variation_perturb0.3}
shows this is indeed the case,
for both the complementary gap and the swim distance.
Across all code distances considered,
the complementary gap has a somewhat smaller value of $\sigma_r / \sigma_{\alpha(\lambda)}$ than the swim distance,
which suggests it is a more accurate DCS.
\Cref{sec:calculating_the_sources_of_variation_of_the_dcs}
describes how we obtained these quantities from our simulations.
\begin{figure*}[tb]
\centering
\includegraphics[width=\textwidth]{calibration_curve_swim.pdf}
\caption{\textbf{Single-decoding-window statistics.}
(a) Histograms of DCS (swim distance) values, normalized by the number of shots. Under circuit-level noise, $10^8$ memory experiments were run for all code distances. The measured DCS values are grouped in 50 bins of equal linear width.
(b) The log success odds conditioned on the DCS value, plotted against the DCS value. The upper and lower bounds on the Bernoulli variable $p_L$ are given by the Wilson score interval. Only datapoints with $p_L >0$ are shown.}
\label{fig:calibration_curve}
\end{figure*}
While $\sigma_r$ is smaller than $\sigma_{\alpha(\lambda)}$,
it is not much smaller.
The question of imperfect calibration arises:
what is the effect of the DCS inaccuracy on
the (early) fault-tolerant applications presented in the next sections?
Answering this would require calculating the log success odds
for higher code distances under a more realistic noise model than that used in \cref{fig:source_of_dcs_variation_perturb0.3},
but we were unable to do so due to numerical imprecision during tensor network contraction.
In \cref{sec:numerics},
we provide a speculative analysis of the DCS uncertainty under a range of assumptions.
\section{Decoder confidence score distributions are rich in information}\label{part_2}
In this section, we explore statistical properties of the DCS distribution which make it a valuable resource
for error mitigation at both
the single-window level, in \cref{sec:single_decoding_window_statistics},
and the full-application level, in \cref{sec:many_decoding_window_statistics}.
Throughout the rest of this paper,
we focus on the swim distance DCS, and we will henceforth generally refer to it as `the DCS'.
We will refer to single-window LEPs and logical errors by lowercase symbols ($p_\L, x$)
and to their multi-window counterparts by uppercase symbols ($P_\L, X$).
We emulate circuits using Stim~\cite{Gidney2021a}
and assume circuit-level noise with a physical error rate $p_\text{phys}=10^{-3}$.
Our approach to estimating DCS performance in the context of full logical circuits is based on allocating a suitable number of decoding windows. We do not adjust according to e.g. the specific decoding needs of logical gates, leaving that for future studies.
\subsection{Single-decoding-window statistics}\label{sec:single_decoding_window_statistics}
We emulated memory experiments in the rotated surface code
for odd code distances $d \in \{5, \dots, 19\}$;
these assess how well one logical qubit stays free from logical error
after $d$ stabiliser measurement cycles.
We decode each experiment in a single window and calculate the DCS for that window.
The distribution of these values is shown in \cref{fig:calibration_curve}a.
Each bin in this histogram contains a number of failed experiments (where a logical error occurred)
and successful (i.e.\ error-free) experiments,
which we use to calculate an empirical log success odds from \cref{eq:log_success_odds_definition}.
\Cref{fig:calibration_curve}b plots this empirical log success odds against the DCS.
By fitting a straight line to this data,
we obtain a calibration curve that assigns a LEP $p_\L$ to each DCS value.
These curves can be used in future experiments for the same code distance and noise model.
Despite having clear peaks, the single-window distributions in \cref{fig:calibration_curve}a provide a wide variety of LEPs if enough shots are taken.
Even in the very early fault-tolerant era, the implementation of a quantum algorithm that is both useful and feasible involves a few hundred logical qubits and thousands of logical operation layers, which translates to millions of decoding windows.
As technology matures further this number will rise steeply.
In the next subsection we consider the total LEP of an experiment comprising many windows.
\subsection{Many-decoding-window statistics}\label{sec:many_decoding_window_statistics}
\begin{figure*}
\centering
\includegraphics[width=\textwidth]{ler_discard_swim_d_7_9_11_13_update.pdf}
\caption{\textbf{Equivalence of different code distances.} (a) Histogram of the calibrated LEPs, normalized by the number of shots, for various numbers of decoding windows in the circuit. The LEPs are grouped in 50 bins of equal logarithmic width. At fixed code distance $d$, for circuits comprising $N$ windows, we define $f_{d,N}(p_\L)$ as the average number of constituent windows whose LEP lies in the bin centred on $p_\L$. The case $N=1$ (viridis colourmap) corresponds to the probability mass function of LEPs for the single-window memory experiments. When more windows are added (magma colourmap), all bins gain population at the same rate: $f_{d,N}(p_\L)=Nf_{d,1}(p_\L)$. It is possible to adjust $N$ such that circuits defined by different $(N,d)$ pairs have the same mean $P_\L$; in this case, the seemingly linear segments of the scaled histograms overlap,
as indicated by the upward arrows. (b) The calculated LEP (solid) and observed LER (dashed) in a cohort of single-window decoding instances when a fraction of highest-$p_\L$ shots are discarded. The curves for $d \in `{11, 13}$ were obtained with $10^9$ shots.}
\label{fig:risk_distribution}
\end{figure*}
To simplify analysis,
we assume $(p_\L, x)$ for each window is independent;
this simplification is discussed in \cref{sec:discussion}.
In this case,
the total LEP $P_\L$ for an experiment comprising $N$ decoding windows,
each with LEP $p_{{\L}_i}$,
is the probability an odd number of them were wrongly decoded:
\begin{equation} \label{ler_many_windows}
P_\L = \frac12 \left[ 1 - \prod_{i=1}^{N} (1-2p_{\L,i}) \right].
\end{equation}
A valid concern regarding the usefulness of DCSs for error mitigation is scalability:
does the distribution of the random variable $P_\L$ remain usefully rich in information as the size of the computation increases?
One might imagine that, for large enough computations, $P_\L$ converges to a sharp value, so that
every run with a given code distance and (very large) number of decoding windows $N$ will be found to have a total LEP $P_\L$ of,
for example,
almost exactly $5\%$.
In that case, the useful information gained by evaluating $P_\L$ for a given circuit is negligible -- it is a foregone conclusion.
In fact,
that is not the case.
In \cref{sec:lep_variance_of_a_many_window_experiment}
we analyse the distribution of $P_\L$
and find that its variance remains broad (and grows) as $N$ increases,
if the code distance is adjusted to target a fixed LER.
However, it could still be the case that, while the variance grows with $N$,
the effective range of LEPs spans fewer orders of magnitude.
To investigate this,
we generalise \cref{fig:calibration_curve}a to circuits of $N \ge 1$ decoding windows and plot the result in \cref{fig:risk_distribution}a,
where the $x$-axis has additionally been converted to a LEP via the calibration curve.
Each single-window LEP histogram is scaled by a factor $N$
such that the mean LEP of an $N$-window circuit equals \num{1e-2}. The evident overlap of the scaled histograms for $p_\L > 10^{-7}$ means that the same number of high-$p_\L$ windows will be detected for different $(N,d)$ pairs. These rare, high-risk events are precisely the ones which make the dominant contribution to the overall $N$-window circuit LEP $P_\L$. Because the non-overlapping parts of the scaled histograms do not contribute much to the overall LEP,
the $P_\L$ distribution for a numerically intractably large pair $(N,d)$ can be accurately inferred from the $P_\L$ distribution for a smaller, testable pair $(N',d')$ yielding the same mean LEP. The fact that the low-$p_\L$ windows bring little contribution to $P_\L$ can be seen from \cref{fig:risk_distribution}b, where we consider the scenario of discarding a fraction of the single-window shots with the highest $p_\L$ values. The figure shows both the calculated LEP and the empirically observed LER; these lines agree well.
For $d=13$
it is possible to improve the LER by one (five) orders of magnitude by discarding a fraction \num{1e-7} (\num{1e-3}) of the total shots.
\section{Decoder confidence scores inform low-overhead error mitigation} \label{part_3}
In this section,
we look at potential applications of
the DCS in QEM for logical circuits in the early fault-tolerant era.
In \cref{sec:shallow}, we present an abort protocol which halts the execution of a deep quantum circuit based on the \emph{instantaneous} DCS.
\Cref{sec:MLE} explains an MLE procedure based on the \emph{whole-circuit} LEP.
In \cref{sec:numerics},
we combine both methods,
applying MLE on increasingly filtered results
of an expectation value estimation task. The profound enhancements which we observe represent one of the key aspects of this work.
We conclude with an approximate analytical model
for a distance-19 surface code,
finding that its behaviour closely matches our earlier empirical analyses.
\subsection{Abort protocol}\label{sec:shallow}
Suppose we choose the following protocol for executing a logical circuit:
abort execution if the DCS associated with \emph{any} decoding window is in a pre-defined `discard' range.
\Cref{fig:risk_distribution}b immediately leads us to the performance of this approach.
Let $\rho$ be the probability a given window is discarded,
and $p_\text{kept}$ be the mean LEP of the retained windows.
These quantities correspond respectively to the $x$ and $y$ axes of \cref{fig:risk_distribution}b.
If a circuit of $N$ windows successfully completes,
then the expected whole-circuit LEP is approximately $P_\text{kept} = \frac12[1-(1-2p_\text{kept})^N] \approx N\,p_\text{kept}$.
As noted, this can be many orders of magnitude lower than when everything is retained.
The caveat is that the probability of successfully completing falls as $(1-\rho)^N$,
meaning $\rho$ must be very low for very large circuits.
However,
there are scenarios in quantum computing where small subroutines can be performed `offline'
e.g.\ when preparing resource states such as magic states or logical Bell pairs~\cite{Knill2005}.
In these cases, an aggressive repeat-until-success model is viable.~\cite{Smith2024,Lee2025c}. The less impressive improvement by one order of magnitude, achieved at much smaller abort rates, can also be used in some early fault-tolerant applications to reduce resource requirements.
\subsection{DCS-Based Maximum Likelihood Estimation}\label{sec:MLE}
In this subsection, we explain how MLE can be leveraged in expectation value estimation to improve accuracy.
We call this technique DCS-MLE.
We consider estimating the expectation value of an observable
from $M$ noisy runs of the same quantum circuit comprising $N$ decoding windows.
Each run $j \in `{1, \dots, M}$ has an observed noisy outcome $z_j \in `{-1, +1}$
and a total LEP $P_{\L,j}$ calculated from the DCS.
\Cref{sec:dcs_zne_simulated_data} details how we emulate these runs.
A straightforward estimator is the unweighted average:
$\langle Z_{\textrm{unmitigated}} \rangle = \frac{1}{M} \sum_{j=1}^{M} z_j.$
However, this can be improved using the data $`{P_{\L,j}}_j$ as follows.
Let us associate a random variable $Z_j$ to each circuit run.
The likelihood to observe the set of outcomes $\{z_j\}_{j=1}^{M}$ is
\begin{equation}
\mathcal{L} = \prod_{j=1}^{M} \pr(Z_j = z_j),
\end{equation}
where $\pr(\text{logical statement})$ is the probability the logical statement is true.
Let the underlying noiseless outcomes
from the logical circuit be $\{\tilde{z}_j\}_{j=1}^{M}$, all having the same mean, $\langle Z \rangle _{\text{th}}$, such that $\theta \equiv \pr(\tilde{Z}_j=+1)=\frac{1}{2}(1 +\langle Z \rangle _{\text{th}})$.
We can express the probability to observe a particular outcome
in terms of conditional probabilities on the noiseless outcomes:
\begin{equation}
\pr(Z_j=z_j)
=\sum_{\tilde{z}_j}\pr(Z_j=z_j|\tilde{Z}_j=\tilde{z}_j)\pr(\tilde{Z}_j=\tilde{z}_j)
\end{equation}
Generally, noise corrupts the true outcomes in a non-trivial way,
defined by a corruption function $q$
which depends on the logical error probability: $q(P_{\L,j}, \tilde{z}_j) \equiv \pr(Z_j=+1|\tilde{Z}_j=\tilde{z}_j)$. Therefore:
\begin{equation}\label{eq:theta_dependence}
\pr(Z_j =+1)=\theta q(P_{\L,j}, +1)+(1-\theta)q(P_{\L,j}, -1).
\end{equation}
Finally, noting that $\pr(Z_j=-1)=1-\pr(Z_j=+1)$,
the likelihood $\mathcal{L}$
can be written as a function of $\theta$,
with the dataset $`{(z_j, P_{\L, j})}_j$ as known parameters.
We obtain an estimator for $\theta$ by
maximising $\mathcal{L}$,
which in practice is done by minimising the negative log-likelihood,
$-\ln \mathcal{L}$.
To tackle imperfect LEP--DCS calibration,
we use a two-parameter fit:
we replace $P_{\L,j}$ by $\eta P_{\L,j}$ in \cref{eq:theta_dependence}
and also optimise the parameter $\eta$.
This accounts for an overall rescaling of the original LEP distribution,
to ensure the mean LEP matches the true LER.
The function $q$ will depend on the noise propagation in the particular circuit under consideration.
For the simple case of a memory experiment,
it can be modelled as $q(P_{\L,j}, +1) = 1-P_{\L,j}$ and $q(P_{\L,j}, -1)=P_{\L,j}$.
\subsection{Numerics}\label{sec:numerics}
\begin{figure*}
\centering
\includegraphics[width=\textwidth]{MSPE_vs_time_overhead_with_histogram_nums_shots_100_10000_num_repeats_200000.pdf}
\caption{\textbf{Estimating $\langle Z \rangle _{\text{th}}=0.8$ on a noisy quantum computer, starting from a $7.7(2)\%$ LER in absence of error mitigation.} (a,c) Output distributions of four expectation value estimators, obtained from $N_{\text{rep}}=2\times10^5$ repetitions.
Dashed curves are normal distributions with the same mean and variance as their respective datasets,
which are shown as dots.
The noiseless estimator corresponds to a noiseless quantum computer
whose outcome variation is caused solely by statistical Bernoulli fluctuations. The circuit discard fraction is $50\%$.
(b,d) The performance of the four estimators plotted against resource overhead.
MSPE $:=\sum_{i=1}^{N_{\text{rep}}}|\langle Z \rangle_i-\langle Z \rangle_{\text{th}}|^2 / N_{\text{rep}}$
is the mean squared prediction error.
The overhead calculation is explained in \cref{sec:time_overhead}.}
\label{fig:MLE_abort}
\end{figure*}
We next evaluate the methods from the previous two subsections
on emulated results of an expectation value estimation task
using a distance-11 surface code.
Three main observations emerge:
1) both methods significantly lower the total error;
2) MLE gives the highest relative improvement over pure abort when applied to unfiltered data;
3) both methods give higher relative improvements
as the number of shots increases.
We also construct an approximate model
for the imperfect mapping between the DCS and the success log odds
for a distance-19 surface code
and assess the performance of the abort protocol under this model.
We set to estimate $\langle Z \rangle_{\text{th}}=0.8$
from $M_0 \in \{\num{1e2}, \num{1e4}\}$ accepted shots with a $d=11$ surface code,
in a circuit comprising $N\approx 2.38\times10^5$ decoding windows.
We consider a range of abort rates,
each of which gives rise to a different LEP distribution for the $N$-window circuit.
\Cref{sec:num_sim} describes how we generate the datasets from empirical data.
\Cref{sec:time_overhead} explains how to calculate the time overhead
given the abort rate.
In \cref{fig:MLE_abort}a,c, we show the distribution of outputs from four different estimators.
In \cref{fig:MLE_abort}b,d, we choose the mean squared prediction error (MSPE)
as a performance metric as it penalizes both bias and variance in the estimated expectation values.
Other metrics could be selected based on the task at hand, such as the absolute mean bias (shown in \cref{fig:MLE_abort_abs_bias}) or the probability to obtain an estimate within a desired range.
There are several effects to be noted. First, aborting a greater portion of
risky circuit executions steadily improves performance.
Second, MLE reduces errors even without discarding
any circuit executions. Its relative advantage diminishes
as the abort protocol becomes more aggressive. It is
noteworthy that near-noiseless performance can be
achieved with MLE on carefully filtered results.
Finally, the benefit of using MLE increases with the number of accepted shots, because it becomes easier to extract signal.
\begin{figure}[!h]
\centering
\includegraphics[width=1.0\linewidth]{d19modelV4.pdf}
\caption{Contrasting the at-scale performance of the DCS (blue) versus the ideal but impractical-to-compute log odds (gray). In all cases the chosen DCS is the swim distance.
In panel (a) the green dashed line shows the distribution of measured DCS values for decoding windows of a distance-19 surface code. An assumed probability curve for the log odds is shown in gray, and in blue an implied DCS distribution derived by redistributing the log odds values via a Gaussian with standard deviation $\Delta =1$. We see a good match between green and blue.
Using the distributions in (a) we generate sets of \num{2.8e9} scores, representing all decoding windows in a large circuit; an example distribution is shown in (b).
From such a set we infer the probability $P_\L$ of error in the circuit; repeating this 10$^5$ times we obtain distribution (c) with mean $P_\L=0.153$.
We note that relying on the DCS means we cannot identify very high or very low risk circuits, as we would with access to the log odds.
Panel (d) shows the factor by which we reduce error rates using abort criteria, at the cost of increased quantum processor time. Panel (e) is the equivalent of \cref{fig:MLE_abort}b,d with the same meaning for lines; here we use
200 shots (between the 100 \& 1000 \cref{fig:MLE_abort} scenarios).}
\label{fig:d=19model}
\end{figure}
We cannot perform the same simulation for $d=19$ as no logical errors were observed during calibration,
but we can estimate its performance with the following model.
In the absence of a known distribution of the true log odds,
we use the distribution of measured DCS values for $d=19$ and slightly deform it, until the mean $p_\L$ associated with this deformed distribution matches
the one predicted from our statistics (see \cref{fig:many_windows}a).
We then use \cref{eq:dcs_as_sum_of_2_rvs}
and assume $r$ follows a Gaussian distribution with mean zero and variance $\Delta=1$.
The inferred distribution of DCS values closely matches that of the measured ones, as seen in \cref{fig:d=19model}a.
This is therefore an entirely analytic model,
in which we have descriptions of the true log odds and the DCS.
Moreover, we know the range of log odds associated with each DCS value,
so there is no need to calibrate the DCS --
in a real scenario, imperfect calibration would indeed be a factor in performance;
further work could explore the impact of this.
\Cref{fig:d=19model}d shows
impressive improvement in the whole-circuit LEP when we impose an abort condition based on the instantaneous window DCS.
Note that the $\Delta=1$ assumption may be pessimistic as an extrapolation from \cref{fig:source_of_dcs_variation_perturb0.3}, and with a smaller $\Delta$ the performance is even stronger.
As suggested by the observation in \cref{sec:many_decoding_window_statistics}
that most of the logical error risk is contained in a few very risky windows, \cref{fig:d=19model}d shows that the post-selection method does not require exponential overhead for a linear LER decrease (for a significant range of reduction). However, we do observe that the accuracy of the DCS is very significant in the performance of circuit-level exploration of the DCS data. Recall that Section~\ref{part_1} established a close match between swim distance and the underlying log odds; even so, we see in \cref{fig:d=19model} panels (c) and (d) that the imperfect nature of the correlation manifests in a considerable reduction in the quality of the information we obtain about the whole-circuit LEP. In panel (d), we show that a modest reduction in the assumed deviation from $\Delta=1$ (somewhat pessimistic given \cref{fig:source_of_dcs_variation_perturb0.3}) to $\Delta=0.7$ dramatically improves performance. Conversely, if we were to replace the swim distance with an alternative DCS that is significantly weaker, it is possible that the circuit-level mitigations explored in this paper would become ineffective.
Finally in \cref{fig:d=19model}e we generate
the equivalent of the earlier panels in \cref{fig:MLE_abort}b,d.
It is reassuring that the key qualitative features are again seen in this alternative model.
\section{Applications}\label{sec:application}
We now discuss applications of the abort protocol and DCS-MLE in quantum phase estimation (QPE) algorithms.
In \cref{sec:res_est}, we perform a detailed resource estimate of our protocols for a Hubbard model ground state energy estimation problem, using statistical QPE. In \cref{sec:qpe_other},
we describe a possible use in ancilla-based QPE.
\subsection{Hubbard model resource estimation}\label{sec:res_est}
In this subsection,
we apply our findings to reduce the resource requirements for the implementation of a full logical circuit.
Specifically,
we find that a $d=21$ surface code can be replaced by $d=19$ to achieve
the same LER with a $64\%$ time overhead via the abort protocol.
This is a conservative estimate; depending on the metric used to assess logical performance, it could be achieved by MLE without any additional overhead.
We target estimating the ground state energy of the $10 \times 10$ Hubbard model to $10^{-2}$ accuracy and at most $10^{-2}$ LER, following the QCELS-based~\cite{ding_even_2023} resource estimates in Ref.~\cite{akahoshi_compilation_2024}. The algorithm involves executing $N_{\text{circuits}}=160$ different quantum circuits of various depths, each repeated with $M_0 = 100$ shots to estimate an observable's expectation value. \cref{sec:qcels} provides more algorithmic details. We estimate that an implementation that does not use the DCS
requires a $d=21$ surface code, which corresponds to a LER of approximately $R_\text{targ} =8.4\times10^{-3}$.
This is our target LER.
The potential advantage of the DCS arises during the estimation of expectation values.
We explore whether the necessary quantum circuits can be executed with $d=19$ (at a starting LER of $R_\text{start} =7.7(4)\%$) and still meet the target LER via QEM. Therefore, the desired improvement in the LER is
$R_\text{start}/R_\text{targ}=10.91$.
We emulate the behaviour of the $d=19$ code using $d=11$. As discussed in~\cref{sec:many_decoding_window_statistics}, these distributions can be scaled into each other by adjusting the number $N$ of decoding windows. In~\cref{sec:numerics}, we chose the number of decoding windows for the $d=11$ code precisely to approximate the LEP distributions for $d=19$.
By repeatedly sampling $N_{\text{circuits}}M_0$ executions from the $\{(\text{LEP},\text{logical error})\}$ distribution at a $60\%$ discard fraction, we estimate that the 95\% confidence interval for the effective LER is $[0.59\%, R_\text{targ}]$, thus very likely achieving the target LER or below, at the cost of $21\%$ total spacetime increase. The latter is calculated in \cref{sec:d_infl_res}, which explains the influence of the code distance on the qubit number and execution duration.
For this algorithm,
it may be more useful to analyse
the MSPE instead of the effective LER.
Using the data shown in \cref{fig:MLE_abort}, we find that the target MSPE ($0.376\times10^{-2}$)
is achieved by the pure-abort strategy with a similar discard fraction as reported above: $60\%-64\%$.
MLE does not give a significant improvement in these cases due to the large Bernoulli variance of $100$ shots. However,
in contexts where the bias is more important than the variance,
then MLE could achieve the target absolute mean bias ($0.633\times10^{-2}$) without any time overhead, as shown in \cref{fig:MLE_abort_abs_bias}.
To summarise, by exploring the Hubbard model application in a scaled-down form which allows us to apply our earlier empirical results, we believe that in the full problem it should possible to reduce the surface code distance from $21$ to $19$ at a marginal increase in space-time volume. The advantage here would be that the application could run successfully on a smaller quantum computer.
\subsection{Other applications - qualitative remarks}\label{sec:qpe_other}
In the case of ancilla-based QPE~\cite{nielsen_quantum_2010}, the final outcome of a noiseless circuit is a bitstring of length $n$ that estimates the first $n$ bits of a desired phase.
The noiseless circuit can be run multiple times,
resulting in a histogram of output bitstrings;
these outputs will get scrambled in the presence of noise.
Ref.~\cite{dutkiewicz_error_2024} describes an MLE protocol
(based on quasiprobability methods)
suitable for deducing the target bitstring.
We believe this protocol could be improved using the DCS.
As suggested in \cref{sec:numerics}, we note that DCS-MLE would be very useful in algorithms that do not take full advantage of Heisenberg scaling and instead exploit a trade-off between the number of different circuits run and the number of times each one is repeated~\cite{ni_low-depth_2023, wang_accelerated_2019, ding_simultaneous_2023, nelson_assessment_2024-1}.
A fundamental question concerning our proposed methods is whether the abort protocol is scalable to large code distances (beyond the early FT era). Most existing works have focused on
applications of the DCS with small (\num{<11}) code distances
due to scalability:
the sampling overhead of the abort protocol could be unreasonably high.
In this work,
we provide numerical evidence against this idea by considering the statistical properties of the DCS distribution in \cref{part_2} and demonstrating that the method works for a small code with the same LER as a larger one. Ref.~\cite{Smith2024} brings further evidence towards scalability through a different perspective: under the code capacity noise model, the authors numerically find that, for a fixed abort tolerance,
the rate at which the decoder aborts a circuit run
exhibits threshold behaviour. Namely,
if the physical error rate is below (above) some threshold,
the abort rate decreases (increases) as the code distance increases.
They find that the threshold is on the order of $10^{-2}$ for code capacity, and expect it to decrease under more realistic noise models. Nevertheless, as long as the abort threshold does not drop below $10^{-4}$, post-selection remains a valuable tool for useful applications in the early fault-tolerant era, such as the quantum phase estimation routine presented here.
\section{Discussion and Outlook}\label{sec:discussion}
This paper brings more nuance to the existing works on decoder confidence
and provides a proof of principle for error mitigation.
We identify four limitations/further directions of research.
Firstly,
the tensor network method we used to compute the log success odds
was prone to numerical imprecision
as there were many samples that led to
division by zero or negative odds.
For the remaining samples,
floating-point precision capped the success log odds values at 16,
meaning we were analysing a very restricted subset of cases,
especially at large code distances where it is known that
the success log odds far exceeds 16.
Secondly,
to maintain decoding accuracy,
most decoding schemes~\cite{Dennis2002,Skoric2023,Bombin2023,Chan2024}
require neighbouring decoding windows in a quantum computation to overlap:
for a code of distance $d$,
the overlap is usually of the order $d/2$~\cite{Lin2025a}.
This means decoding windows are not independent;
accounting for this effect would alter \cref{ler_many_windows}.
Moreover, numerical results for $N$-window circuits made use of emulated logical errors, obtained under the assumption of window independence,
so they should be trusted with caution.
Thirdly,
regarding DCS-based QEM, the calibration curves for large code distances take a long time to obtain because the LEPs are so low. Thus, full calibration curves would have to be provided by the quantum device manufacturers. Nevertheless, \cref{fig:calibration_curve}b shows that the calibration curves are extremely similar for the same noise model between different code distances, so an approximate calibration curve for $d=19$ could be derived from lower distances. Moreover, if the number of shots is sufficient for MLE, then a calibration curve might not even be necessary, because the extrapolation parameter can also be the raw DCS $\phi$ provided by the decoder; the mitigated expectation value is then the value taken by the fit function in the limit as $\phi \rightarrow\infty$.
Lastly,
the linear relation between the noisy outcomes and the noiseless ones applies only to memory experiments.
To expand \cref{sec:MLE} to general logical circuits,
we can consider a computational model with layers of logical operations,
between which $d$ cycles of stabiliser measurements are performed on each logical qubit.
This results in a logical single-qubit Pauli noise model:
the logical qubit experiences an $X$, $Y$ or $Z$ flip with a probability given by the DCS.
In this model, the corruption function would not be a straight line, but a general polynomial or exponential.
\section{Conclusion}\label{sec:conclusion}
We have shown that the log success odds of a decoding instance can be very well approximated by both the complementary gap and the swim distance for MWPM decoders under phenomenological noise, justifying their status as decoder confidence scores. For a simplified model of window-based decoding, under circuit-level noise, we have shown that DCSs remain rich in information as the circuit size increases, which suggests that they can be used for more complex tasks than state preparation. Finally, we have explored this latter idea in the form of error mitigation techniques for the early fault-tolerant quantum computing era, where aborting and MLE can help run useful applications with a quantum device of limited size. We emphasize that our results rely on the observation that the swim distance is an excellent DCS; the performance of both QEM methods proposed is seen to diminish as the DCS quality decreases. This motivates future work toward even more accurate DCS methods.
\begin{acknowledgments}
We thank Zi-Han Chen for providing code and guidance on their implementation of the swim distance on MWPM.
We also thank Christophe Piveteau for providing code for calculating logical coset probabilities via tensor network contraction.
For useful discussions,
we thank
Zhenyu Cai,
Armands Strikis
and Christopher Chubb.
We acknowledge the use of
the University of Oxford Advanced Research Computing
(ARC)
facility~\cite{Richards2015_quantum_bibstyle} in carrying out this work
and specifically the facilities made available
from the EPSRC QCS Hub grant
(agreement No.\ EP/T001062/1).
We also acknowledge support from two EPSRC projects:
RoaRQ (EP/W032635/1)
and SEEQA (EP/Y004655/1).
TC and MD acknowledge support from EPSRC DTP studentships.
\end{acknowledgments}
\vspace{1cm}
\paragraph*{Note added:}
In the final days of preparation of this manuscript,
a related preprint~\cite{Zhou2025b} was announced
with similar ideas on using the DCS for error mitigation.
For example,
the authors consider using the DCS for entire logical circuits
and reach similar conclusions using the same abort-based method;
they also assess the accuracy of the complementary gap
using a tensor network decoder.
While similar,
there are differences.
The authors use a code capacity noise model and emulate logical gates;
we adopt a more realistic circuit-level noise model but emulate only quantum memory experiments.
The authors propose a ZNE procedure based on probabilistic error amplification,
whereas the MLE procedure we propose is based on the LEP variation between runs.
\bibliography{ref,tchbib, Maria_references}
\appendix
\makeatletter
\renewcommand\thefigure{\thesection.\arabic{figure}}
\renewcommand{\theHfigure}{\thesection.\arabic{figure}}
\makeatother
\section{Calculating the Sources of Variation of the DCS}
\setcounter{figure}{0}
\label{sec:calculating_the_sources_of_variation_of_the_dcs}
\begin{figure*}
\centering
\includegraphics[width=0.8\textwidth]{dcs_against_log_success_odds_phenomenological_perturb0.3.pdf}
\caption{Decoder confidence score $\phi$
plotted against log success odds $\lambda$
for different code distances $d$.
Each blue point is one syndrome sample from the unrotated surface code under a perturbed phenomenological noise model,
decoded using MWPM.
The best-fit line calculated using Unweighted Least Squares is shown in red;
the height of the shaded region is the root mean squared residual,
and is plotted as $\sigma_r$ in \cref{fig:source_of_dcs_variation_perturb0.3}.
Due to floating-point precision
the success log odds values are capped at 16,
so we ignore samples whose complementary gap exceeds 16.}
\label{fig:dcs_against_log_success_odds_phenomenological}
\end{figure*}
First, we describe how we obtained $\sigma_r$.
We emulated the unrotated surface code
under a perturbed phenomenological noise model
(described in \cref{sec:perturbed_phenomenological_noise_model}).
For odd distances \numrange{3}{11},
we randomly sampled syndromes obtained from noise levels in \numrange{1e{-4}}{1e{-1}}.
For each syndrome,
we calculated both DCS values using the MWPM decoder~\cite{Higgott2025},
then calculated $\lambda$ via tensor-network contraction~\cite{piveteau_tensor-network_2024}
(details in \cref{sec:tensor_network_contraction}).
The results are shown in \cref{fig:dcs_against_log_success_odds_phenomenological}
and suggest that $\alpha$ is linear for both DCSs.
Therefore,
we fit a line $\hat{\phi}(\lambda)$ to the data $`{(\lambda_i, \phi_i)}_{i =1}^n$ for each distance
and estimate $\sigma_r$ by calculating the \emph{sample root mean squared residual}:
\begin{equation}\label{eq:sample_root_mean_squared_residual}
\widehat{\sigma_r} =
s_r :=\sqrt{\frac{\sum_{i =1}^n `\big[\phi_i -\hat{\phi}(\lambda_i)]^2}{n -1}}.
\end{equation}
To obtain $\sigma_{\alpha(\lambda)}$,
we calculated both DCS values for \num{1e8} randomly sampled syndromes
at a fixed noise level of \num{1e-3},
then calculated the sample variance $s_\phi^2$ for each DCS.
Finally,
we note that \cref{eq:dcs_as_sum_of_2_rvs}
and the independence of $\lambda$ and $r$ implies
$\sigma_\phi^2 =\sigma_{\alpha(\lambda)}^2 +\sigma_r^2$,
so we estimate $\sigma_{\alpha(\lambda)}$ by calculating
\begin{equation}
\widehat{\sigma_{\alpha(\lambda)}} =\sqrt{s_\phi^2 -s_r^2}
\end{equation}
where $s_r$ is taken from \cref{eq:sample_root_mean_squared_residual}.
\subsection{Perturbed Phenomenological Noise Model}\label{sec:perturbed_phenomenological_noise_model}
In the phenomenological noise model,
all edges have the same weight,
giving rise to a very limited set of possible values the DCS can take.
We found that this led to a large value for $s_r$ in \cref{eq:sample_root_mean_squared_residual};
however,
such a simplistic noise model is unrealistic.
Ideally,
we would perform our analysis in \cref{part_1}
using a circuit-level noise model
whose edge weights are calibrated from experiments or existing syndrome data.
Although Ref.~\cite{piveteau_tensor-network_2024} provides a tensor-network decoder
for circuit-level noise,
we could not obtain values of $\lambda$ that converged quickly enough.
Therefore,
we resorted to a `perturbed' phenomenological noise model,
wherein each edge weight is multiplied by a random scalar drawn uniformly from $[1 -\delta, 1 +\delta]$.
We set $\delta =0.3$ which roughly corresponds to the variation in edge weights
for standard circuit-level depolarising noise of strength $p =\num{1e-3}$
\cite{Gidney2021a}.
\subsection{Tensor Network Contraction}
\label{sec:tensor_network_contraction}
We use the software package \verb|tndecoder3d| introduced in Ref.~\cite{piveteau_tensor-network_2024}.
For the surface code under phenomenological noise,
the network to contract resembles a 3D cubic lattice.
The contraction algorithm first contracts the 3D lattice down to a 2D lattice,
then contracts the 2D lattice using well-known techniques.
Following Ref.~\cite{piveteau_tensor-network_2024},
we use a maximum bond dimension of 24 and 32 for each stage respectively.
\Cref{fig:d11_convergence_perturb0.3} shows that
these values, respectively named `3D bond dimension' and `2D bond dimension',
are sufficient for $s_r$ in \cref{eq:sample_root_mean_squared_residual} to converge.
\begin{figure}
\centering
\includegraphics[width=\linewidth]{d11_convergence_perturb0.3.pdf}
\caption{The sample root mean squared residual of the log success odds
plotted against the bond dimensions used for each stage of tensor network contraction,
for the distance-11 surface code.}
\label{fig:d11_convergence_perturb0.3}
\end{figure}
\section{LEP Variance of a Many-Window Experiment}
\setcounter{figure}{0}
\label{sec:lep_variance_of_a_many_window_experiment}
As discussed in the main text,
the total LEP for an experiment comprising $N$ windows,
each with LEP $p_{\L, i}$ is given by \cref{ler_many_windows}.
If all $p_{\L,i}$ have mean $\mu_{1,d}$ and variance $\sigma_{1,d}^2$,
the random variable $P_\L$ has mean and variance:
\begin{align}
\mu_{N,d} &= \tfrac12 [1 - (1 -2\mu_{1,d})^N], \\
\sigma_{N,d}^2 &= \tfrac14 \left\{[(1 -2\mu_{1,d})^{2} +(2\sigma_{1,d})^2]^N - (1 -2\mu_{1,d})^{2N}\right\}.
\end{align}
\begin{figure*}
\centering
\includegraphics[width=\linewidth]{many_windows_rearranged_swim.pdf}
\caption{\textbf{Characteristics of LEP distributions.} (a) The moments of the $p_\L$ distributions emerging from the single-window memory experiments have a clean dependence on $d$, so we can predict the distribution moments for higher code distances. (b) The mean LEP of a full circuit involving $N$ decoding windows, for various choices of code distance $d$. As $N$ increases with $d$ fixed, eventually we fail to meet any fixed target accuracy such as the $10^{-2}$ marked here in red. (In \cref{fig:risk_distribution} we use values of $N$ corresponding to the intersections of the curves and the red line here.) (c) When choosing the smallest code distance $d(\epsilon)$ for which $\mu_{N,d(\epsilon)}\leq{\epsilon}=0.01$, the variance $\sigma_{N,d(\epsilon)}^2$ of the total LEP slowly increases with $N$.}
\label{fig:many_windows}
\end{figure*}
\Cref{fig:many_windows}a shows that
both $\mu_{1,d}$ and $\sigma_{1,d}^2$ are very small;
in this case,
$\sigma_{N,d}^2$ vanishes at constant $d$ as $N\rightarrow\infty$.
In practical contexts however, the distribution of $P_\L$ remains broad (and thus useful) as circuit sizes increase. The explanation lies in the observation that, as $N$ increases, the mean total LEP $\mu_{N,d}$ also increases at constant $d$, possibly exceeding the maximum error allowed for the particular application under consideration;
see \cref{fig:many_windows}b.
The solution is to increase the code size: for a fixed target LER $\epsilon$, we choose the smallest code distance $d(\epsilon)$ that fulfils $\mu_{N,d(\epsilon)} \leq \epsilon$. The question then becomes: how does $\sigma_{N,d(\epsilon)}^2$ behave?
\Cref{fig:many_windows}c indicates that this variance does not shrink as $N$ increases.
\section{Numerical simulation}\label{sec:num_sim}
\subsection{Abort protocol}\label{sec:abort_simulation}
\setcounter{figure}{0}
\begin{figure}
\centering
\includegraphics[width=\linewidth]{ler_after_abort_against_discard_fraction.pdf}
\caption{Logical error rate of the retained executions for different discard fractions at $d=11$, obtained by sampling $N =\num{2.38e5}$ single-window $(x,p_{\L})$ pairs $10^6$ times. The target LER is specific to the Hubbard model application described in \cref{sec:res_est}.}
\label{fig:ler_after_abort_d=11}
\end{figure}
Here, we describe how we generate datasets for the LEP and logical error of each an $N$-window circuit. We perform simulations up to a code distance of $11$,
which is the largest distance at which enough logical errors could be detected during calibration within $10^8$ shots. To avoid simulating deep circuits and calculating their DCSs, we emulate the behaviour of a circuit comprising $N =\num{2.38e5}$ windows by sampling from a pool of single-window circuits, each characterised by a calibrated LEP $p_i$ and a logical error $x_i \in \{0,1\}$, where $x_i = 0$ indicates successful decoding. The $N$-window circuit is considered to have experienced an error if the decoding of an odd number of the $N$ windows failed: $X = 1$ if $\sum_{i=1}^Nx_i$ is odd, else $X=0$. The assigned LEP for an $N$-window circuit is given by \cref{ler_many_windows}.
Next, we explain how the abort protocol affects the LEP distribution. Let the single-window abort rate be $\rho$, achieved by setting some LEP threshold. This means that, in an $N$-window circuit, each window has an LEP larger than that threshold with probability $\rho$. The probability to discard the circuit by aborting at any point throughout its run is $f=1-(1-\rho)^N$. Throughout this work, we refer to $f$ as the discard fraction.
To obtain the LEP distribution for a certain abort rate, we eliminate all the single-window experiments from the $10^8$ cohort whose LEP is above the threshold LEP. We then proceed as described in the first paragraph of this appendix subsection and sample $N$ windows $10^6$ times. In \cref{fig:ler_after_abort_d=11}, we show the retained LER decay with discard fraction.
For the resource estimation task described in \cref{sec:res_est}, we notice that a $60\%$ discard fraction brings the LER down to the target value. Therefore, we sample $M = \num{16000}$ times from the $\{(X_j, P_{\L,j})\}$ dataset obtained at the abort rate associated with this discard fraction, effectively emulating $M$ instances of $N$-window circuits.
The empirical LER is the mean $X$ across the $M$ accepted shots,
and is stated in the main text with a $95\%$ confidence interval.
\subsection{Expectation Value Estimation}\label{sec:dcs_zne_simulated_data}
Here, we describe how we generate datasets to emulate $M$ noisy experiments, each being an $N$-window circuit execution. First,
we emulate $M$ copies of a $\ket{\overline{0}}$ memory experiment as described in \cref{sec:abort_simulation}
to obtain the LEPs $`{P_{\L,j}}_{j=1}^M$ from the DCS
and the logical errors $`{X_j}_{j=1}^M$.
Next, for a fixed ideal expectation value $\langle Z\rangle_\text{th}$,
we employ a random number generator to generate the noiseless logical outcomes $\{\tilde{Z}_j \in \{-1, +1\} \}_{j=1}^M$, where $\tilde{Z}_j=+1$ occurs with probability $(1-\langle Z\rangle_\text{th})/2$.
We then randomize the signs of the noiseless outcomes according to the logical errors $\{X_j\}_{j=1}^M$ to obtain the noisy outcomes $Z_j=\tilde{Z}_j(1-X_j)-UX_j$, where $U$ is a random variable with equal probability to take value $+1$ or $-1$.
This setup provides an effective framework for modelling arbitrary logical circuits similar to memory experiments,
both when using and not using the abort protocol. Note that we had to make an assumption about how the noise influences the final outcome. For a memory experiment, the outcome would actually be $100\%$ flipped by a logical error, but assuming randomization instead is more likely for general circuits and algorithms.
\section{Time overhead caused by aborting}\label{sec:time_overhead}
Here, we calculate the time overhead required to obtain $M_0$ accepted shots based on the abort rate.
The advantage of aborting is that it saves time by not letting a circuit-to-be-discarded complete. For an $N$-window circuit which aborts at some point throughout its execution, we want to know, on average, how far the execution proceeds before it is aborted. We define $p_{\text{abort}}(n)$ as the probability that the circuit aborts because of the $n^{\text{th}}$ window. This requires that the current window causes the abort, and additionally that none of the previous windows have caused it. Therefore, $p_{\text{abort}}(n) = \frac{\rho(1-\rho)^{n-1}}{f}$. The average window at which the abort happens (provided that it happens) is calculated as $\langle n \rangle = \sum_{i=1}^Nnp_{\text{abort}}(n)$. We divide this by $N$ to obtain the average percentage of circuit execution before abort: $\frac{1-(1-f)(1+N\rho)}{N\rho f}$.
Let $M_0$ be the desired number of shots for the quantum task, and $M$ the number of executed shots when the abort rate is $\rho$. They are related by $M=\frac{M_0}{1-f}$. Out of these $M$ executions, a fraction $1-f$ complete, thus contributing fully to the time duration, while the other $f$ fraction contribute an average of $\frac{\langle n \rangle}{N}$. Overall, the mean time overhead caused by aborting at rate $\rho$ becomes:
\begin{equation}
\frac{1-f+f\frac{\langle n \rangle}{N}}{1-f}=\frac{f/N}{(1-f)[1-(1-f)^{\frac{1}{N}}]}.
\end{equation}
\section{QCELS parameters}\label{sec:qcels}
Based on numbers reported in this reference for the chosen algorithmic accuracy (the maximum number of Trotter steps involved in any quantum circuit and the average number of clocks per Trotter step), we deduced that the maximum number of decoding windows in any quantum circuit is $\num{ 1.38e9}$. We use our own numerical predictions for the LER dependence on the code distance at a physical error rate $p_\text{phys}=10^{-3}$, rather than the reference $10^{-4}$. To ensure the noise is sufficiently suppressed, we identify the smallest code distance $d$ such that the mean LEP per circuit satisfies $\mu_{N,d} \leq 10^{-2}$, and that is $d=21$. To obtain a LEP distribution with the same mean as $(N_{\text{windows}}, d) =(\num{ 1.38e9}, 19)$,
we used $(N_{\text{windows}}, d) =(\num{2.38e5}, 11)$.
We considered parameters $J=16$ and $N=5$ from Ref.~\cite{akahoshi_compilation_2024}, Sec. IV.B (note that $N$ quoted here is different from the usual meaning of $N$ as number of decoding windows throughout our work). Therefore, $N_{\text{values}}=2\times J\times N = 160$ expectation values need to be estimated through different quantum circuits.
Each circuit should be run $M_0$ times, which means that $N_{\text{values}}M_0 = \num{16000}$ quantum circuit runs occur in total.
\section{Code distance influence on resources}\label{sec:d_infl_res}
To counteract the time overhead caused by aborting some circuit executions, there is an additional beneficial factor of $19/21$
because the duration of each logical operation scales at most linearly with $d$.
Furthermore, the surface code qubit count is reduced to $(19/21)^2\approx82\%$,
leading to a total spacetime increase of $1.64\times(19/21)^3 -1\approx21\%$.
\section{DCS-QEM performance metrics}
\setcounter{figure}{0}
In \cref{fig:MLE_abort_abs_bias}, we present a different performance metric for the abort protocol and the MLE procedure to the one in the main text, namely, the absolute mean bias, called 'Bias' for short in the y-axis labels of panels (b,e). This metric does not take variance into account, so we plot the latter in panels (c,f). In the limit of an infinite number of repetitions, the noiseless estimator bias would drop to exactly zero.
\begin{figure*}
\centering
\includegraphics[width=\textwidth]{bias_vs_time_overhead_with_histogram_nums_shots_100_10000_num_repeats_200000.pdf}
\caption{\textbf{Estimating $\langle Z \rangle _{\text{th}}=0.8$ on a noisy quantum computer, starting from a $7.7(2)\%$ LER in absence of error mitigation.} (a,d) Empirical output distributions of four expectation value estimators, obtained from $N_{\text{rep}}=2\times10^5$ repetitions. Dashed curves are normal distributions with the same mean and variance as the respective datasets. The noiseless estimator corresponds to an ideal quantum computer, whose output variation is solely caused by statistical Bernoulli fluctuations. The circuit discard fraction is $50\%$. (b,e) The performance of the four estimators plotted against resource overhead. For a given estimator and discard fraction, the
bias is defined as $|\langle Z \rangle_{\text{th}}-\sum_{i=1}^{N_{\text{rep}}}\langle Z \rangle_i / N_{\text{rep}}|$. The overhead calculation is explained in \cref{sec:time_overhead}. (c,f) The variance of the estimator output distributions.}
\label{fig:MLE_abort_abs_bias}
\end{figure*}
\end{document}
%apsrev4-2.bst 2019-01-14 (MD) hand-edited version of apsrev4-1.bst
%Control: key (0)
%Control: author (8) initials jnrlst
%Control: editor formatted (1) identically to author
%Control: production of article title (0) allowed
%Control: page (0) single
%Control: year (1) truncated
%Control: production of eprint (0) enabled
\begin{thebibliography}{43}%
\makeatletter
\providecommand \@ifxundefined [1]{%
\@ifx{#1\undefined}
}%
\providecommand \@ifnum [1]{%
\ifnum #1\expandafter \@firstoftwo
\else \expandafter \@secondoftwo
\fi
}%
\providecommand \@ifx [1]{%
\ifx #1\expandafter \@firstoftwo
\else \expandafter \@secondoftwo
\fi
}%
\providecommand \natexlab [1]{#1}%
\providecommand \enquote [1]{``#1''}%
\providecommand \bibnamefont [1]{#1}%
\providecommand \bibfnamefont [1]{#1}%
\providecommand \citenamefont [1]{#1}%
\providecommand \href@noop [0]{\@secondoftwo}%
\providecommand \href [0]{\begingroup \@sanitize@url \@href}%
\providecommand \@href[1]{\@@startlink{#1}\@@href}%
\providecommand \@@href[1]{\endgroup#1\@@endlink}%
\providecommand \@sanitize@url [0]{\catcode `\\12\catcode `\$12\catcode `\&12\catcode `\#12\catcode `\^12\catcode `\_12\catcode `\%12\relax}%
\providecommand \@@startlink[1]{}%
\providecommand \@@endlink[0]{}%
\providecommand \url [0]{\begingroup\@sanitize@url \@url }%
\providecommand \@url [1]{\endgroup\@href {#1}{\urlprefix }}%
\providecommand \urlprefix [0]{URL }%
\providecommand \Eprint [0]{\href }%
\providecommand \doibase [0]{https://doi.org/}%
\providecommand \selectlanguage [0]{\@gobble}%
\providecommand \bibinfo [0]{\@secondoftwo}%
\providecommand \bibfield [0]{\@secondoftwo}%
\providecommand \translation [1]{[#1]}%
\providecommand \BibitemOpen [0]{}%
\providecommand \bibitemStop [0]{}%
\providecommand \bibitemNoStop [0]{.\EOS\space}%
\providecommand \EOS [0]{\spacefactor3000\relax}%
\providecommand \BibitemShut [1]{\csname bibitem#1\endcsname}%
\let\auto@bib@innerbib\@empty
%</preamble>
\bibitem [{\citenamefont {Shor}(1995)}]{shor_scheme_1995}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {P.~W.}\ \bibnamefont {Shor}},\ }\bibfield {title} {\bibinfo {title} {Scheme for reducing decoherence in quantum computer memory},\ }\href {https://doi.org/10.1103/PhysRevA.52.R2493} {\bibfield {journal} {\bibinfo {journal} {Physical Review A}\ }\textbf {\bibinfo {volume} {52}},\ \bibinfo {pages} {R2493} (\bibinfo {year} {1995})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Fowler}\ \emph {et~al.}(2012)\citenamefont {Fowler}, \citenamefont {Mariantoni}, \citenamefont {Martinis},\ and\ \citenamefont {Cleland}}]{fowler_surface_2012}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.~G.}\ \bibnamefont {Fowler}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Mariantoni}}, \bibinfo {author} {\bibfnamefont {J.~M.}\ \bibnamefont {Martinis}},\ and\ \bibinfo {author} {\bibfnamefont {A.~N.}\ \bibnamefont {Cleland}},\ }\bibfield {title} {\bibinfo {title} {Surface codes: {Towards} practical large-scale quantum computation},\ }\href {https://doi.org/10.1103/PhysRevA.86.032324} {\bibfield {journal} {\bibinfo {journal} {Physical Review A}\ }\textbf {\bibinfo {volume} {86}},\ \bibinfo {pages} {032324} (\bibinfo {year} {2012})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Chen}\ \emph {et~al.}(2025{\natexlab{a}})\citenamefont {Chen}, \citenamefont {Chen}, \citenamefont {Lu},\ and\ \citenamefont {Pan}}]{Chen2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Z.-H.}\ \bibnamefont {Chen}}, \bibinfo {author} {\bibfnamefont {M.-C.}\ \bibnamefont {Chen}}, \bibinfo {author} {\bibfnamefont {C.-Y.}\ \bibnamefont {Lu}},\ and\ \bibinfo {author} {\bibfnamefont {J.-W.}\ \bibnamefont {Pan}},\ }\href@noop {} {\bibinfo {title} {Efficient magic state cultivation on $\mathbb{RP}^2$}} (\bibinfo {year} {2025}{\natexlab{a}}),\ \Eprint {https://arxiv.org/abs/2503.18657} {arXiv:2503.18657 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Meister}\ \emph {et~al.}(2024)\citenamefont {Meister}, \citenamefont {Pattison},\ and\ \citenamefont {Preskill}}]{meister_efficient_2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Meister}}, \bibinfo {author} {\bibfnamefont {C.~A.}\ \bibnamefont {Pattison}},\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Preskill}},\ }\href {https://doi.org/10.48550/arXiv.2405.07433} {\bibinfo {title} {Efficient soft-output decoders for the surface code}} (\bibinfo {year} {2024}),\ \bibinfo {note} {arXiv:2405.07433 [quant-ph]}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Lee}\ \emph {et~al.}(2025)\citenamefont {Lee}, \citenamefont {English},\ and\ \citenamefont {Bartlett}}]{Lee2025c}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.-H.}\ \bibnamefont {Lee}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {English}},\ and\ \bibinfo {author} {\bibfnamefont {S.~D.}\ \bibnamefont {Bartlett}},\ }\href@noop {} {\bibinfo {title} {Efficient post-selection for general quantum {LDPC} codes}} (\bibinfo {year} {2025}),\ \Eprint {https://arxiv.org/abs/2510.05795} {arXiv:2510.05795 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Bombín}\ \emph {et~al.}(2024)\citenamefont {Bombín}, \citenamefont {Pant}, \citenamefont {Roberts},\ and\ \citenamefont {Seetharam}}]{bombin_fault-tolerant_2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Bombín}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Pant}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Roberts}},\ and\ \bibinfo {author} {\bibfnamefont {K.~I.}\ \bibnamefont {Seetharam}},\ }\bibfield {title} {\bibinfo {title} {Fault-{Tolerant} {Postselection} for {Low}-{Overhead} {Magic} {State} {Preparation}},\ }\href {https://doi.org/10.1103/PRXQuantum.5.010302} {\bibfield {journal} {\bibinfo {journal} {PRX Quantum}\ }\textbf {\bibinfo {volume} {5}},\ \bibinfo {pages} {010302} (\bibinfo {year} {2024})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {deMarti iOlius}\ \emph {et~al.}(2024)\citenamefont {deMarti iOlius}, \citenamefont {Fuentes}, \citenamefont {Or{\'{u}}s}, \citenamefont {Crespo},\ and\ \citenamefont {Etxezarreta~Martinez}}]{deMartiiOlius2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {deMarti iOlius}}, \bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Fuentes}}, \bibinfo {author} {\bibfnamefont {R.}~\bibnamefont {Or{\'{u}}s}}, \bibinfo {author} {\bibfnamefont {P.~M.}\ \bibnamefont {Crespo}},\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Etxezarreta~Martinez}},\ }\bibfield {title} {\bibinfo {title} {Decoding algorithms for surface codes},\ }\href {https://doi.org/10.22331/q-2024-10-10-1498} {\bibfield {journal} {\bibinfo {journal} {{Quantum}}\ }\textbf {\bibinfo {volume} {8}},\ \bibinfo {pages} {1498} (\bibinfo {year} {2024})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Gidney}\ \emph {et~al.}(2025)\citenamefont {Gidney}, \citenamefont {Newman}, \citenamefont {Brooks},\ and\ \citenamefont {Jones}}]{Gidney2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Gidney}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Newman}}, \bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Brooks}},\ and\ \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Jones}},\ }\bibfield {title} {\bibinfo {title} {Yoked surface codes},\ }\href {https://doi.org/10.1038/s41467-025-59714-1} {\bibfield {journal} {\bibinfo {journal} {Nature Communications}\ }\textbf {\bibinfo {volume} {16}},\ \bibinfo {pages} {4498} (\bibinfo {year} {2025})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Pattison}\ \emph {et~al.}(2025)\citenamefont {Pattison}, \citenamefont {Krishna},\ and\ \citenamefont {Preskill}}]{Pattison2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.~A.}\ \bibnamefont {Pattison}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Krishna}},\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Preskill}},\ }\bibfield {title} {\bibinfo {title} {Hierarchical memories: Simulating quantum {LDPC} codes with local gates},\ }\href {https://doi.org/10.22331/q-2025-05-05-1728} {\bibfield {journal} {\bibinfo {journal} {{Quantum}}\ }\textbf {\bibinfo {volume} {9}},\ \bibinfo {pages} {1728} (\bibinfo {year} {2025})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Akahoshi}\ \emph {et~al.}(2025)\citenamefont {Akahoshi}, \citenamefont {Toshio}, \citenamefont {Fujisaki}, \citenamefont {Oshima}, \citenamefont {Sato},\ and\ \citenamefont {Fujii}}]{Akahoshi2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont {Akahoshi}}, \bibinfo {author} {\bibfnamefont {R.}~\bibnamefont {Toshio}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Fujisaki}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Oshima}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Sato}},\ and\ \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Fujii}},\ }\href@noop {} {\bibinfo {title} {Runtime reduction in lattice surgery utilizing time-like soft information}} (\bibinfo {year} {2025}),\ \Eprint {https://arxiv.org/abs/2510.21149} {arXiv:2510.21149 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Sunami}\ \emph {et~al.}(2025)\citenamefont {Sunami}, \citenamefont {Hirano}, \citenamefont {Hinokuma},\ and\ \citenamefont {Yamasaki}}]{Sunami2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Sunami}}, \bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont {Hirano}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Hinokuma}},\ and\ \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Yamasaki}},\ }\href@noop {} {\bibinfo {title} {Entanglement boosting: Low-volume logical {Bell} pair preparation for distributed fault-tolerant quantum computation}} (\bibinfo {year} {2025}),\ \Eprint {https://arxiv.org/abs/2511.10729} {arXiv:2511.10729 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Poulin}(2006)}]{Poulin2006}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Poulin}},\ }\bibfield {title} {\bibinfo {title} {Optimal and efficient decoding of concatenated quantum block codes},\ }\href {https://doi.org/10.1103/PhysRevA.74.052333} {\bibfield {journal} {\bibinfo {journal} {Physical Review A}\ }\textbf {\bibinfo {volume} {74}},\ \bibinfo {pages} {052333} (\bibinfo {year} {2006})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Goto}\ and\ \citenamefont {Uchikawa}(2013)}]{Goto2013}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Goto}}\ and\ \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Uchikawa}},\ }\bibfield {title} {\bibinfo {title} {Fault-tolerant quantum computation with a soft-decision decoder for error correction and detection by teleportation},\ }\href {https://doi.org/10.1038/srep02044} {\bibfield {journal} {\bibinfo {journal} {Scientific Reports}\ }\textbf {\bibinfo {volume} {3}},\ \bibinfo {pages} {2044} (\bibinfo {year} {2013})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Siegel}\ \emph {et~al.}(2025)\citenamefont {Siegel}, \citenamefont {Cai}, \citenamefont {Jnane}, \citenamefont {Koczor}, \citenamefont {Pexton}, \citenamefont {Strikis},\ and\ \citenamefont {Benjamin}}]{siegel_snakes_2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Siegel}}, \bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {Cai}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Jnane}}, \bibinfo {author} {\bibfnamefont {B.}~\bibnamefont {Koczor}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Pexton}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Strikis}},\ and\ \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Benjamin}},\ }\href {https://doi.org/10.48550/arXiv.2501.02120} {\bibinfo {title} {Snakes on a {Plane}: mobile, low dimensional logical qubits on a {2D} surface}} (\bibinfo {year} {2025}),\ \bibinfo {note} {arXiv:2501.02120 [quant-ph]}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Smith}\ \emph {et~al.}(2024)\citenamefont {Smith}, \citenamefont {Brown},\ and\ \citenamefont {Bartlett}}]{Smith2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.~C.}\ \bibnamefont {Smith}}, \bibinfo {author} {\bibfnamefont {B.~J.}\ \bibnamefont {Brown}},\ and\ \bibinfo {author} {\bibfnamefont {S.~D.}\ \bibnamefont {Bartlett}},\ }\bibfield {title} {\bibinfo {title} {Mitigating errors in logical qubits},\ }\href {https://doi.org/10.1038/s42005-024-01883-4} {\bibfield {journal} {\bibinfo {journal} {Communications Physics}\ }\textbf {\bibinfo {volume} {7}},\ \bibinfo {pages} {386} (\bibinfo {year} {2024})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Sales~Rodriguez}\ \emph {et~al.}(2025)\citenamefont {Sales~Rodriguez}, \citenamefont {Robinson}, \citenamefont {Jepsen}, \citenamefont {He}, \citenamefont {Duckering}, \citenamefont {Zhao}, \citenamefont {Wu}, \citenamefont {Campo}, \citenamefont {Bagnall}, \citenamefont {Kwon}, \citenamefont {Karolyshyn}, \citenamefont {Weinberg}, \citenamefont {Cain}, \citenamefont {Evered}, \citenamefont {Geim}, \citenamefont {Kalinowski}, \citenamefont {Li}, \citenamefont {Manovitz}, \citenamefont {Amato-Grill}, \citenamefont {Basham}, \citenamefont {Bernstein}, \citenamefont {Braverman}, \citenamefont {Bylinskii}, \citenamefont {Choukri}, \citenamefont {DeAngelo}, \citenamefont {Fang}, \citenamefont {Fieweger}, \citenamefont {Frederick}, \citenamefont {Haines}, \citenamefont {Hamdan}, \citenamefont {Hammett}, \citenamefont {Hsu}, \citenamefont {Hu}, \citenamefont {Huber}, \citenamefont {Jia}, \citenamefont {Kedar}, \citenamefont {Kornjača}, \citenamefont {Liu}, \citenamefont {Long},
\citenamefont {Lopatin}, \citenamefont {Lopes}, \citenamefont {Luo}, \citenamefont {Macrì}, \citenamefont {Marković}, \citenamefont {Martínez-Martínez}, \citenamefont {Meng}, \citenamefont {Ostermann}, \citenamefont {Ostroumov}, \citenamefont {Paquette}, \citenamefont {Qiang}, \citenamefont {Shofman}, \citenamefont {Singh}, \citenamefont {Singh}, \citenamefont {Sinha}, \citenamefont {Thoreen}, \citenamefont {Wan}, \citenamefont {Wang}, \citenamefont {Waxman-Lenz}, \citenamefont {Wong}, \citenamefont {Wurtz}, \citenamefont {Zhdanov}, \citenamefont {Zheng}, \citenamefont {Greiner}, \citenamefont {Keesling}, \citenamefont {Gemelke}, \citenamefont {Vuletić}, \citenamefont {Kitagawa}, \citenamefont {Wang}, \citenamefont {Bluvstein}, \citenamefont {Lukin}, \citenamefont {Lukin}, \citenamefont {Zhou},\ and\ \citenamefont {Cantú}}]{SalesRodriguez2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Sales~Rodriguez}}, \bibinfo {author} {\bibfnamefont {J.~M.}\ \bibnamefont {Robinson}}, \bibinfo {author} {\bibfnamefont {P.~N.}\ \bibnamefont {Jepsen}}, \bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {He}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Duckering}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Zhao}}, \bibinfo {author} {\bibfnamefont {K.-H.}\ \bibnamefont {Wu}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Campo}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Bagnall}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Kwon}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Karolyshyn}}, \bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Weinberg}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Cain}}, \bibinfo {author} {\bibfnamefont {S.~J.}\ \bibnamefont {Evered}}, \bibinfo {author} {\bibfnamefont {A.~A.}\ \bibnamefont {Geim}}, \bibinfo {author} {\bibfnamefont
{M.}~\bibnamefont {Kalinowski}}, \bibinfo {author} {\bibfnamefont {S.~H.}\ \bibnamefont {Li}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Manovitz}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Amato-Grill}}, \bibinfo {author} {\bibfnamefont {J.~I.}\ \bibnamefont {Basham}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Bernstein}}, \bibinfo {author} {\bibfnamefont {B.}~\bibnamefont {Braverman}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Bylinskii}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Choukri}}, \bibinfo {author} {\bibfnamefont {R.~J.}\ \bibnamefont {DeAngelo}}, \bibinfo {author} {\bibfnamefont {F.}~\bibnamefont {Fang}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Fieweger}}, \bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Frederick}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Haines}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Hamdan}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Hammett}}, \bibinfo {author}
{\bibfnamefont {N.}~\bibnamefont {Hsu}}, \bibinfo {author} {\bibfnamefont {M.-G.}\ \bibnamefont {Hu}}, \bibinfo {author} {\bibfnamefont {F.}~\bibnamefont {Huber}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Jia}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Kedar}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Kornjača}}, \bibinfo {author} {\bibfnamefont {F.}~\bibnamefont {Liu}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Long}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Lopatin}}, \bibinfo {author} {\bibfnamefont {P.~L.~S.}\ \bibnamefont {Lopes}}, \bibinfo {author} {\bibfnamefont {X.-Z.}\ \bibnamefont {Luo}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Macrì}}, \bibinfo {author} {\bibfnamefont {O.}~\bibnamefont {Marković}}, \bibinfo {author} {\bibfnamefont {L.~A.}\ \bibnamefont {Martínez-Martínez}}, \bibinfo {author} {\bibfnamefont {X.}~\bibnamefont {Meng}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Ostermann}}, \bibinfo {author}
{\bibfnamefont {E.}~\bibnamefont {Ostroumov}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Paquette}}, \bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {Qiang}}, \bibinfo {author} {\bibfnamefont {V.}~\bibnamefont {Shofman}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Singh}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Singh}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Sinha}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Thoreen}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Wan}}, \bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont {Wang}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Waxman-Lenz}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Wong}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Wurtz}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Zhdanov}}, \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Zheng}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Greiner}}, \bibinfo {author} {\bibfnamefont
{A.}~\bibnamefont {Keesling}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Gemelke}}, \bibinfo {author} {\bibfnamefont {V.}~\bibnamefont {Vuletić}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Kitagawa}}, \bibinfo {author} {\bibfnamefont {S.-T.}\ \bibnamefont {Wang}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Bluvstein}}, \bibinfo {author} {\bibfnamefont {M.~D.}\ \bibnamefont {Lukin}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Lukin}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Zhou}},\ and\ \bibinfo {author} {\bibfnamefont {S.~H.}\ \bibnamefont {Cantú}},\ }\bibfield {title} {\bibinfo {title} {Experimental demonstration of logical magic state distillation},\ }\href {https://doi.org/10.1038/s41586-025-09367-3} {\bibfield {journal} {\bibinfo {journal} {Nature}\ }\textbf {\bibinfo {volume} {645}},\ \bibinfo {pages} {620} (\bibinfo {year} {2025})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Gidney}\ \emph {et~al.}(2024)\citenamefont {Gidney}, \citenamefont {Shutty},\ and\ \citenamefont {Jones}}]{gidney_magic_2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Gidney}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Shutty}},\ and\ \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Jones}},\ }\href {https://doi.org/10.48550/arXiv.2409.17595} {\bibinfo {title} {Magic state cultivation: growing {T} states as cheap as {CNOT} gates}} (\bibinfo {year} {2024}),\ \bibinfo {note} {arXiv:2409.17595 [quant-ph]}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Sahay}\ \emph {et~al.}(2025)\citenamefont {Sahay}, \citenamefont {Tsai}, \citenamefont {Chang}, \citenamefont {Su}, \citenamefont {Smith}, \citenamefont {Singh},\ and\ \citenamefont {Puri}}]{Sahay2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Sahay}}, \bibinfo {author} {\bibfnamefont {P.-K.}\ \bibnamefont {Tsai}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Chang}}, \bibinfo {author} {\bibfnamefont {Q.}~\bibnamefont {Su}}, \bibinfo {author} {\bibfnamefont {T.~B.}\ \bibnamefont {Smith}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Singh}},\ and\ \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Puri}},\ }\href@noop {} {\bibinfo {title} {Fold-transversal surface code cultivation}} (\bibinfo {year} {2025}),\ \Eprint {https://arxiv.org/abs/2509.05212} {arXiv:2509.05212 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Shutty}\ \emph {et~al.}(2024)\citenamefont {Shutty}, \citenamefont {Newman},\ and\ \citenamefont {Villalonga}}]{Shutty2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Shutty}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Newman}},\ and\ \bibinfo {author} {\bibfnamefont {B.}~\bibnamefont {Villalonga}},\ }\href@noop {} {\bibinfo {title} {Efficient near-optimal decoding of the surface code through ensembling}} (\bibinfo {year} {2024}),\ \Eprint {https://arxiv.org/abs/2401.12434} {arXiv:2401.12434 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Toshio}\ \emph {et~al.}(2025)\citenamefont {Toshio}, \citenamefont {Kishi}, \citenamefont {Fujisaki}, \citenamefont {Oshima}, \citenamefont {Sato},\ and\ \citenamefont {Fujii}}]{Toshio2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {R.}~\bibnamefont {Toshio}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Kishi}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Fujisaki}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Oshima}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Sato}},\ and\ \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Fujii}},\ }\href@noop {} {\bibinfo {title} {Decoder switching: Breaking the speed-accuracy tradeoff in real-time quantum error correction}} (\bibinfo {year} {2025}),\ \Eprint {https://arxiv.org/abs/2510.25222} {arXiv:2510.25222 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Dennis}\ \emph {et~al.}(2002)\citenamefont {Dennis}, \citenamefont {Kitaev}, \citenamefont {Landahl},\ and\ \citenamefont {Preskill}}]{Dennis2002}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {E.}~\bibnamefont {Dennis}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Kitaev}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Landahl}},\ and\ \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Preskill}},\ }\bibfield {title} {\bibinfo {title} {Topological quantum memory},\ }\href {https://doi.org/10.1063/1.1499754} {\bibfield {journal} {\bibinfo {journal} {Journal of Mathematical Physics}\ }\textbf {\bibinfo {volume} {43}},\ \bibinfo {pages} {4452} (\bibinfo {year} {2002})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Hutter}\ \emph {et~al.}(2014)\citenamefont {Hutter}, \citenamefont {Wootton},\ and\ \citenamefont {Loss}}]{Hutter2014}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Hutter}}, \bibinfo {author} {\bibfnamefont {J.~R.}\ \bibnamefont {Wootton}},\ and\ \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Loss}},\ }\bibfield {title} {\bibinfo {title} {Efficient {Markov} chain {Monte} {Carlo} algorithm for the surface code},\ }\href {https://doi.org/10.1103/PhysRevA.89.022326} {\bibfield {journal} {\bibinfo {journal} {Physical Review A}\ }\textbf {\bibinfo {volume} {89}},\ \bibinfo {pages} {022326} (\bibinfo {year} {2014})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Piveteau}\ \emph {et~al.}(2024)\citenamefont {Piveteau}, \citenamefont {Chubb},\ and\ \citenamefont {Renes}}]{piveteau_tensor-network_2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Piveteau}}, \bibinfo {author} {\bibfnamefont {C.~T.}\ \bibnamefont {Chubb}},\ and\ \bibinfo {author} {\bibfnamefont {J.~M.}\ \bibnamefont {Renes}},\ }\bibfield {title} {\bibinfo {title} {Tensor-{Network} {Decoding} {Beyond} {2D}},\ }\href {https://doi.org/10.1103/PRXQuantum.5.040303} {\bibfield {journal} {\bibinfo {journal} {PRX Quantum}\ }\textbf {\bibinfo {volume} {5}},\ \bibinfo {pages} {040303} (\bibinfo {year} {2024})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Chen}\ \emph {et~al.}(2025{\natexlab{b}})\citenamefont {Chen}, \citenamefont {Xu}, \citenamefont {Sommers}, \citenamefont {Huse}, \citenamefont {Thompson},\ and\ \citenamefont {Gopalakrishnan}}]{Chen2025a}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Chen}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Xu}}, \bibinfo {author} {\bibfnamefont {G.~M.}\ \bibnamefont {Sommers}}, \bibinfo {author} {\bibfnamefont {D.~A.}\ \bibnamefont {Huse}}, \bibinfo {author} {\bibfnamefont {J.~D.}\ \bibnamefont {Thompson}},\ and\ \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Gopalakrishnan}},\ }\href@noop {} {\bibinfo {title} {Scalable accuracy gains from postselection in quantum error correcting codes}} (\bibinfo {year} {2025}{\natexlab{b}}),\ \Eprint {https://arxiv.org/abs/2510.05222} {arXiv:2510.05222 [cond-mat.stat-mech]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Bluvstein}\ \emph {et~al.}(2024)\citenamefont {Bluvstein}, \citenamefont {Evered}, \citenamefont {Geim}, \citenamefont {Li}, \citenamefont {Zhou}, \citenamefont {Manovitz}, \citenamefont {Ebadi}, \citenamefont {Cain}, \citenamefont {Kalinowski}, \citenamefont {Hangleiter}, \citenamefont {Bonilla~Ataides}, \citenamefont {Maskara}, \citenamefont {Cong}, \citenamefont {Gao}, \citenamefont {Sales~Rodriguez}, \citenamefont {Karolyshyn}, \citenamefont {Semeghini}, \citenamefont {Gullans}, \citenamefont {Greiner}, \citenamefont {Vuleti{\'c}},\ and\ \citenamefont {Lukin}}]{bluvsteinLogicalQuantumProcessor2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Bluvstein}}, \bibinfo {author} {\bibfnamefont {S.~J.}\ \bibnamefont {Evered}}, \bibinfo {author} {\bibfnamefont {A.~A.}\ \bibnamefont {Geim}}, \bibinfo {author} {\bibfnamefont {S.~H.}\ \bibnamefont {Li}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Zhou}}, \bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Manovitz}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Ebadi}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Cain}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Kalinowski}}, \bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Hangleiter}}, \bibinfo {author} {\bibfnamefont {J.~P.}\ \bibnamefont {Bonilla~Ataides}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Maskara}}, \bibinfo {author} {\bibfnamefont {I.}~\bibnamefont {Cong}}, \bibinfo {author} {\bibfnamefont {X.}~\bibnamefont {Gao}}, \bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Sales~Rodriguez}}, \bibinfo {author} {\bibfnamefont
{T.}~\bibnamefont {Karolyshyn}}, \bibinfo {author} {\bibfnamefont {G.}~\bibnamefont {Semeghini}}, \bibinfo {author} {\bibfnamefont {M.~J.}\ \bibnamefont {Gullans}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Greiner}}, \bibinfo {author} {\bibfnamefont {V.}~\bibnamefont {Vuleti{\'c}}},\ and\ \bibinfo {author} {\bibfnamefont {M.~D.}\ \bibnamefont {Lukin}},\ }\bibfield {title} {\bibinfo {title} {Logical quantum processor based on reconfigurable atom arrays},\ }\href {https://doi.org/10.1038/s41586-023-06927-3} {\bibfield {journal} {\bibinfo {journal} {Nature}\ }\textbf {\bibinfo {volume} {626}},\ \bibinfo {pages} {58} (\bibinfo {year} {2024})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {English}\ \emph {et~al.}(2025)\citenamefont {English}, \citenamefont {Williamson},\ and\ \citenamefont {Bartlett}}]{English2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {L.~H.}\ \bibnamefont {English}}, \bibinfo {author} {\bibfnamefont {D.~J.}\ \bibnamefont {Williamson}},\ and\ \bibinfo {author} {\bibfnamefont {S.~D.}\ \bibnamefont {Bartlett}},\ }\bibfield {title} {\bibinfo {title} {Thresholds for postselected quantum error correction from statistical mechanics},\ }\href {https://doi.org/10.1103/nh49-52y2} {\bibfield {journal} {\bibinfo {journal} {Physical Review Letters}\ }\textbf {\bibinfo {volume} {135}},\ \bibinfo {pages} {120603} (\bibinfo {year} {2025})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Higgott}\ and\ \citenamefont {Gidney}(2025)}]{Higgott2025}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {O.}~\bibnamefont {Higgott}}\ and\ \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Gidney}},\ }\bibfield {title} {\bibinfo {title} {{Sparse} {Blossom}: correcting a million errors per core second with minimum-weight matching},\ }\href {https://doi.org/10.22331/q-2025-01-20-1600} {\bibfield {journal} {\bibinfo {journal} {{Quantum}}\ }\textbf {\bibinfo {volume} {9}},\ \bibinfo {pages} {1600} (\bibinfo {year} {2025})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Gidney}(2021)}]{Gidney2021a}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Gidney}},\ }\bibfield {title} {\bibinfo {title} {{Stim}: a fast stabilizer circuit simulator},\ }\href {https://doi.org/10.22331/q-2021-07-06-497} {\bibfield {journal} {\bibinfo {journal} {{Quantum}}\ }\textbf {\bibinfo {volume} {5}},\ \bibinfo {pages} {497} (\bibinfo {year} {2021})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Knill}(2005)}]{Knill2005}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {E.}~\bibnamefont {Knill}},\ }\bibfield {title} {\bibinfo {title} {Quantum computing with realistically noisy devices},\ }\href {https://doi.org/10.1038/nature03350} {\bibfield {journal} {\bibinfo {journal} {Nature}\ }\textbf {\bibinfo {volume} {434}},\ \bibinfo {pages} {39} (\bibinfo {year} {2005})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Ding}\ and\ \citenamefont {Lin}(2023{\natexlab{a}})}]{ding_even_2023}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {Ding}}\ and\ \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Lin}},\ }\bibfield {title} {\bibinfo {title} {Even {Shorter} {Quantum} {Circuit} for {Phase} {Estimation} on {Early} {Fault}-{Tolerant} {Quantum} {Computers} with {Applications} to {Ground}-{State} {Energy} {Estimation}},\ }\href {https://doi.org/10.1103/PRXQuantum.4.020331} {\bibfield {journal} {\bibinfo {journal} {PRX Quantum}\ }\textbf {\bibinfo {volume} {4}},\ \bibinfo {pages} {020331} (\bibinfo {year} {2023}{\natexlab{a}})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Akahoshi}\ \emph {et~al.}(2024)\citenamefont {Akahoshi}, \citenamefont {Toshio}, \citenamefont {Fujisaki}, \citenamefont {Oshima}, \citenamefont {Sato},\ and\ \citenamefont {Fujii}}]{akahoshi_compilation_2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont {Akahoshi}}, \bibinfo {author} {\bibfnamefont {R.}~\bibnamefont {Toshio}}, \bibinfo {author} {\bibfnamefont {J.}~\bibnamefont {Fujisaki}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Oshima}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Sato}},\ and\ \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Fujii}},\ }\href {https://doi.org/10.48550/arXiv.2408.14929} {\bibinfo {title} {Compilation of {Trotter}-{Based} {Time} {Evolution} for {Partially} {Fault}-{Tolerant} {Quantum} {Computing} {Architecture}}} (\bibinfo {year} {2024}),\ \bibinfo {note} {arXiv:2408.14929 [quant-ph]}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Nielsen}\ and\ \citenamefont {Chuang}(2010)}]{nielsen_quantum_2010}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {M.~A.}\ \bibnamefont {Nielsen}}\ and\ \bibinfo {author} {\bibfnamefont {I.~L.}\ \bibnamefont {Chuang}},\ }\href@noop {} {\emph {\bibinfo {title} {Quantum computation and quantum information}}},\ \bibinfo {edition} {10th}\ ed.\ (\bibinfo {publisher} {Cambridge university press},\ \bibinfo {address} {Cambridge},\ \bibinfo {year} {2010})\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Dutkiewicz}\ \emph {et~al.}(2024)\citenamefont {Dutkiewicz}, \citenamefont {Polla}, \citenamefont {Scheurer}, \citenamefont {Gogolin}, \citenamefont {Huggins},\ and\ \citenamefont {O'Brien}}]{dutkiewicz_error_2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Dutkiewicz}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Polla}}, \bibinfo {author} {\bibfnamefont {M.}~\bibnamefont {Scheurer}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Gogolin}}, \bibinfo {author} {\bibfnamefont {W.~J.}\ \bibnamefont {Huggins}},\ and\ \bibinfo {author} {\bibfnamefont {T.~E.}\ \bibnamefont {O'Brien}},\ }\href {https://doi.org/10.48550/arXiv.2410.05369} {\bibinfo {title} {Error mitigation and circuit division for early fault-tolerant quantum phase estimation}} (\bibinfo {year} {2024}),\ \bibinfo {note} {arXiv:2410.05369 [quant-ph]}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Ni}\ \emph {et~al.}(2023)\citenamefont {Ni}, \citenamefont {Li},\ and\ \citenamefont {Ying}}]{ni_low-depth_2023}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Ni}}, \bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Li}},\ and\ \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Ying}},\ }\bibfield {title} {\bibinfo {title} {On low-depth algorithms for quantum phase estimation},\ }\href {https://doi.org/10.22331/q-2023-11-06-1165} {\bibfield {journal} {\bibinfo {journal} {Quantum}\ }\textbf {\bibinfo {volume} {7}},\ \bibinfo {pages} {1165} (\bibinfo {year} {2023})},\ \bibinfo {note} {publisher: Verein zur Förderung des Open Access Publizierens in den Quantenwissenschaften}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Wang}\ \emph {et~al.}(2019)\citenamefont {Wang}, \citenamefont {Higgott},\ and\ \citenamefont {Brierley}}]{wang_accelerated_2019}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {D.}~\bibnamefont {Wang}}, \bibinfo {author} {\bibfnamefont {O.}~\bibnamefont {Higgott}},\ and\ \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Brierley}},\ }\bibfield {title} {\bibinfo {title} {Accelerated {Variational} {Quantum} {Eigensolver}},\ }\href {https://doi.org/10.1103/PhysRevLett.122.140504} {\bibfield {journal} {\bibinfo {journal} {Physical Review Letters}\ }\textbf {\bibinfo {volume} {122}},\ \bibinfo {pages} {140504} (\bibinfo {year} {2019})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Ding}\ and\ \citenamefont {Lin}(2023{\natexlab{b}})}]{ding_simultaneous_2023}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {Ding}}\ and\ \bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Lin}},\ }\bibfield {title} {\bibinfo {title} {Simultaneous estimation of multiple eigenvalues with short-depth quantum circuit on early fault-tolerant quantum computers},\ }\href {https://doi.org/10.22331/q-2023-10-11-1136} {\bibfield {journal} {\bibinfo {journal} {Quantum}\ }\textbf {\bibinfo {volume} {7}},\ \bibinfo {pages} {1136} (\bibinfo {year} {2023}{\natexlab{b}})},\ \bibinfo {note} {arXiv:2303.05714 [quant-ph]}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Nelson}\ and\ \citenamefont {Baczewski}(2024)}]{nelson_assessment_2024-1}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {J.~S.}\ \bibnamefont {Nelson}}\ and\ \bibinfo {author} {\bibfnamefont {A.~D.}\ \bibnamefont {Baczewski}},\ }\bibfield {title} {\bibinfo {title} {Assessment of quantum phase estimation protocols for early fault-tolerant quantum computers},\ }\href {https://doi.org/10.1103/PhysRevA.110.042420} {\bibfield {journal} {\bibinfo {journal} {Physical Review A}\ }\textbf {\bibinfo {volume} {110}},\ \bibinfo {pages} {042420} (\bibinfo {year} {2024})},\ \bibinfo {note} {publisher: American Physical Society}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Skoric}\ \emph {et~al.}(2023)\citenamefont {Skoric}, \citenamefont {Browne}, \citenamefont {Barnes}, \citenamefont {Gillespie},\ and\ \citenamefont {Campbell}}]{Skoric2023}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {L.}~\bibnamefont {Skoric}}, \bibinfo {author} {\bibfnamefont {D.~E.}\ \bibnamefont {Browne}}, \bibinfo {author} {\bibfnamefont {K.~M.}\ \bibnamefont {Barnes}}, \bibinfo {author} {\bibfnamefont {N.~I.}\ \bibnamefont {Gillespie}},\ and\ \bibinfo {author} {\bibfnamefont {E.~T.}\ \bibnamefont {Campbell}},\ }\bibfield {title} {\bibinfo {title} {Parallel window decoding enables scalable fault tolerant quantum computation},\ }\href {https://doi.org/10.1038/s41467-023-42482-1} {\bibfield {journal} {\bibinfo {journal} {Nature Communications}\ }\textbf {\bibinfo {volume} {14}},\ \bibinfo {pages} {7040} (\bibinfo {year} {2023})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Bomb{\'\i}n}\ \emph {et~al.}(2023)\citenamefont {Bomb{\'\i}n}, \citenamefont {Dawson}, \citenamefont {Liu}, \citenamefont {Nickerson}, \citenamefont {Pastawski},\ and\ \citenamefont {Roberts}}]{Bombin2023}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {H.}~\bibnamefont {Bomb{\'\i}n}}, \bibinfo {author} {\bibfnamefont {C.}~\bibnamefont {Dawson}}, \bibinfo {author} {\bibfnamefont {Y.-H.}\ \bibnamefont {Liu}}, \bibinfo {author} {\bibfnamefont {N.}~\bibnamefont {Nickerson}}, \bibinfo {author} {\bibfnamefont {F.}~\bibnamefont {Pastawski}},\ and\ \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Roberts}},\ }\href@noop {} {\bibinfo {title} {Modular decoding: parallelizable real-time decoding for quantum computers}} (\bibinfo {year} {2023}),\ \Eprint {https://arxiv.org/abs/2303.04846} {arXiv:2303.04846 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Chan}(2024)}]{Chan2024}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {T.}~\bibnamefont {Chan}},\ }\href@noop {} {\bibinfo {title} {Snowflake: A distributed streaming decoder}} (\bibinfo {year} {2024}),\ \Eprint {https://arxiv.org/abs/2406.01701} {arXiv:2406.01701 [quant-ph]} \BibitemShut {NoStop}%
\bibitem [{\citenamefont {Fuhui~Lin}\ \emph {et~al.}(2025)\citenamefont {Fuhui~Lin}, \citenamefont {Peterson}, \citenamefont {Sankar},\ and\ \citenamefont {Sivarajah}}]{Lin2025a}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Fuhui~Lin}}, \bibinfo {author} {\bibfnamefont {E.~C.}\ \bibnamefont {Peterson}}, \bibinfo {author} {\bibfnamefont {K.}~\bibnamefont {Sankar}},\ and\ \bibinfo {author} {\bibfnamefont {P.}~\bibnamefont {Sivarajah}},\ }\bibfield {title} {\bibinfo {title} {Spatially parallel decoding for multi-qubit lattice surgery},\ }\href {https://doi.org/10.1088/2058-9565/adc6b6} {\bibfield {journal} {\bibinfo {journal} {Quantum Science and Technology}\ }\textbf {\bibinfo {volume} {10}},\ \bibinfo {pages} {035007} (\bibinfo {year} {2025})}\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Richards}(2015)}]{Richards2015_quantum_bibstyle}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Richards}},\ }\href {https://doi.org/10.5281/zenodo.22558} {\emph {\bibinfo {title} {\href{https://doi.org/10.5281/zenodo.22558}{University of Oxford Advanced Research Computing}}}} (\bibinfo {year} {2015})\BibitemShut {NoStop}%
\bibitem [{\citenamefont {Zhou}\ \emph {et~al.}(2025)\citenamefont {Zhou}, \citenamefont {Pexton}, \citenamefont {Kubica},\ and\ \citenamefont {Ding}}]{Zhou2025b}%
\BibitemOpen
\bibfield {author} {\bibinfo {author} {\bibfnamefont {Z.}~\bibnamefont {Zhou}}, \bibinfo {author} {\bibfnamefont {S.}~\bibnamefont {Pexton}}, \bibinfo {author} {\bibfnamefont {A.}~\bibnamefont {Kubica}},\ and\ \bibinfo {author} {\bibfnamefont {Y.}~\bibnamefont {Ding}},\ }\href@noop {} {\bibinfo {title} {Error mitigation of fault-tolerant quantum circuits with soft information}} (\bibinfo {year} {2025}),\ \Eprint {https://arxiv.org/abs/2512.09863} {arXiv:2512.09863 [quant-ph]} \BibitemShut {NoStop}%
\end{thebibliography}%