Expansion formula of one-loop Einstein-Yang-Mills integrand
Extracted source text · arXiv 2512.08354
\documentclass[11pt,a4paper]{article}
\pdfoutput=1
\usepackage{cancel}
\usepackage{enumitem}
\usepackage{etex}
\usepackage{amsthm}
\usepackage{geometry}
\usepackage{dcolumn}
\usepackage{epsf}
\usepackage{mathrsfs}
\usepackage{multirow}
\usepackage{booktabs}
\usepackage{tabularx}
\usepackage{array}
\usepackage{slashed}
\usepackage{float}
\usepackage{leftidx}
\usepackage{setspace}
\usepackage{verbatim}
\usepackage{adjustbox}
\usepackage{tikz}
\usepackage[caption=false]{subfig}
\usepackage{arydshln}
\usepackage{jheppub}
\usepackage{shuffle}
\usepackage{amsmath}
\usepackage{cases}
\usepackage{lipsum}
\usepackage{tabularx}
\usepackage{graphicx}
\usepackage{lscape}
\numberwithin{equation}{section}
\usetikzlibrary{arrows,decorations.markings,shapes.arrows,patterns,positioning}
\setlength{\oddsidemargin}{0.75in}
\setlength{\evensidemargin}{0.75in} \setlength{\topmargin}{0.75in}
\setlength{\textwidth}{7.0in} \setlength{\textheight}{8.5in}
\renewcommand{\baselinestretch}{1.2}
\jot=2mm
\newcommand{\smallbullet}{\scalebox{0.5}{\text{$\bullet$~~}}}
\newcommand{\bea}{\begin{eqnarray}}
\newcommand{\eea}{\end{eqnarray}}
\newcommand{\bean}{\begin{eqnarray*}}
\newcommand{\eean}{\end{eqnarray*}}
\newcommand{\nn}{\nonumber\\}
\newcommand{\Sl}{\sum\limits}
\newcommand{\blue}{\color{blue}}
\newcommand{\red}{\color{red}}
\newcommand{\green}{\color{green}}
\newcommand{\black}{\color{black}}
\def\W #1{\widetilde{#1}}
\def\WH #1{\widehat{#1}}
\def\msout#1{\textrm{\sout{#1}}}
\def\mxout#1{\textrm{\xout{#1}}}
\def\Label#1{\label{#1}%
\smash{\hbox to0pt{\raise1ex\hbox{\tiny[#1]}\hss}}}
%\def\Label#1{\label{#1}}
\renewcommand{\eqref}[1]{eq.~(\ref{#1})}
%\newcommand{\[1]{eqs.~(\ref{#1})\xspace}
\newcommand{\figref}[1]{Fig.~\ref{#1}}
\newcommand{\tabref}[1]{table~\ref{#1}}
\newcommand{\secref}[1]{section~\ref{#1}}
\newcommand{\appref}[1]{appendix~\ref{#1}}
\newcommand{\tabincell}[2]{\begin{tabular}{@{}#1@{}}#2\end{tabular}}
\def\rightaction#1{\stackrel{\rightarrow}{\partial \over \partial #1}}
\def\leftaction#1{\stackrel{\lefttarrow}{\partial \over \partial #1}}
\def\func#1{\mathop{\rm #1}\nolimits}
\def\abs#1{\left| #1\right|}
\def\braket#1{\left\langle #1 \right\rangle}
\def\bra#1{\left\langle #1\right|}
\def\ket#1{\left| #1\right\rangle}
\def\gb #1{ \left\langle #1 \right]}
\def\tgb #1{ \left[ #1 \right\rangle}
%�����Զ�����%%%%%%%%%%%%%%%%%%%%%%%%%
%\newcommand{\tabincell}[2]{\begin{tabular}{@{}#1@{}}#2\end{tabular}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\Shuffle{{\text{\scriptsize{$\bigsqcup$}}}\mathchoice{\mkern-3mu}{\mkern-3mu}{\mkern-0mu}{\mkern-0mu}{\text{\scriptsize{$\bigsqcup$}}}}
\def\d{\partial}
\def\rmd{{\rm d}}
\def\la{\lambda}
\def\eps{\epsilon}
\def\lblb{({\bar\la}{\bar\la})}
\def\ald{{\dot\alpha}}
\def\bed{{\dot\beta}}
\def\gad{{\dot\gamma}}
\def\sid{{\dot\rho}}
\def\vev{\braket}
\def\tgb #1{ \left[ #1 \right\rangle}
\def\bket#1{\left| #1\right]}
\def\bvev#1{\left[ #1 \right]}
\def\Spaa{\vev}
\def\Spbb{\bvev}
\def\Spab{\gb}
\def\Spba{\tgb}
\def\Sl{\sum\limits}
\def\rchi{\raisebox{\depth}{$\chi$}}
\newcommand{\bbibitem}[1]{\bibitem{#1}\marginpar{#1}}
\newcommand{\ctobedelete}[1]{}
\renewcommand\arraystretch{1.3}
\allowdisplaybreaks
\title{Expansion formula of one-loop Einstein-Yang-Mills integrand}
\author[a,b]{Yi-Jian Du} \author[c]{Chongsi Xie}
\affiliation[a]{Department of Physics, School of Physics and Technology,
Wuhan University,\\
No.299 Bayi Road, Wuhan 430072, China}
\affiliation[b]{Hubei Key Laboratory of Nuclear Solid Physics, School of Physics and Technology, Wuhan University,\\
No.299 Bayi Road, Wuhan 430072, China}
\affiliation[c]{College of Science, Hainan Tropical Ocean University,\\
Sanya, Hainan 572022, P.R. China}
\emailAdd{yijian.du@whu.edu.cn}\emailAdd{chongsi.xie@whu.edu.cn}
\date{\today}
\abstract{
Building upon the algebraic consistency construction of one-loop Bern-Carrasco-Johansson (BCJ) numerators for Yang-Mills (YM) and Yang-Mills-scalar (YMS) theories, we explore the expansion formula of one-loop Einstein-Yang-Mills (EYM) integrands (with a gluon loop) in terms of conventional one-loop YM integrands with quadratic propagators. We first express the EYM integrand by tree-level amplitudes according to the forward limit approach. Employing a two-step expansion strategy, the gluon-loop EYM integrand is decomposed into tree-level YM amplitudes under the forward limit, which are subsequently expanded into tree-level bi-adjoint scalar (BS) ones. We then prove that when the kinematic coefficients in both expansion steps satisfy the one-loop consistency conditions, the EYM integrand is finally expanded as a combination of YM integrands with quadratic propagators.
The coefficients in this expansion formula coincide exactly with those in the expansion formula for YMS integrands. This correspondence highlights a shared kinematic structure, providing the proper foundation for constructing BCJ numerators in both YMS and EYM theories at one loop.
%Building upon the algebraic consistency construction of one-loop BCJ numerators for Yang-Mills (YM) and Yang-Mills-scalar (YMS) theories, we investigate the expansion formula of one-loop Einstein-Yang-Mills (EYM) integrands (with a gluon loop) in terms of conventional one-loop YM integrands with quadratic propagators. Employing a two-step expansion strategy, the gluon-loop EYM integrand is first decomposed into tree-level YM amplitudes under the forward limit, which are subsequently expanded into tree-level bi-adjoint scalar (BS) ones. We prove that when the kinematic coefficients in both expansion steps satisfy the one-loop consistency conditions, the coefficients relating EYM integrands to YM integrands coincide exactly with those derived earlier from the expansion of YMS integrands to BS integrands. This correspondence demonstrates a common kinematic structure that provides the proper basis for constructing BCJ numerators in both YMS and EYM theories at one loop.
}
\keywords{Scattering Amplitudes, Gauge Symmetry}
\begin{document}
\maketitle \flushbottom
%% Formal stuff
\section{Introduction}\label{sec:intro}
Colour-kinematic duality \cite{Bern:2008qj,Bern:2010ue} demonstrates that colour-dressed Yang-Mills (YM) amplitudes can be arranged as sums over trivalent graphs, each associated with a colour factor and a kinematic factor that shares the same algebraic identity as its colour counterpart. When each colour factor in a YM amplitude is replaced by another copy of the kinematic factor, the amplitude then turns into a gravitational (GR) amplitude.
This duality provides a deep insight into the connection between YM and GR.
How to construct the kinematic numerators (BCJ numerators) satisfying the algebraic identities is the main problem of the colour-kinematic duality.
At tree level, a construction based on the recursive expansion formula of Yang-Mills-scalar (YMS) amplitudes was proposed in \cite{Fu:2017uzt,Chiodaroli:2017ngp,Teng:2017tbo,Du:2017kpo,Du:2017gnh}. By this approach, a colour-stripped YM amplitude is first expressed in terms of doubly colour-stripped YMS amplitudes, which are further expanded as a combination of doubly colour-stripped bi-adjoint scalar (BS) ones with the help of the recursive expansion formula\footnote{All YM amplitudes/integrands in this paper are colour-stripped ones and all YMS, BS amplitudes/integrands are doublely colour-stripped ones. More details about the relation between colour-dressed and colour-stripped ones can be found in the discussion of e.g., \cite{Porkert:2022efy,Du:2025yxz}.}. Once a YM amplitude has been expressed in terms of BS ones, the kinematic coefficients can generate all tree-level BCJ numerators satisfying the colour-kinematic duality, see e.g., \cite{ref-Kiermaier,Bern:2010yg,Bjerrum-Bohr:2012kaa,Fu:2012uy,Fu:2014pya}. The expansion relations satisfied by YM and YMS amplitudes have been generalised to GR and Einstein-Yang-Mills (EYM) amplitudes \cite{Fu:2017uzt,Chiodaroli:2017ngp,Teng:2017tbo,Du:2017kpo,Du:2017gnh}. More concretely, a GR/EYM amplitude is expanded in terms of YM ones, with the same kinematic coefficients in the expansion formulas of YM/YMS amplitudes.
Despite extensive study of tree-level numerators, a general construction of one-loop BCJ numerators remains an open problem. The forward limit approach based on worldsheet formalisms \cite{Mason:2013sva,He:2015yua,Cachazo:2015aol} provides an effective tool for understanding the loop integrands. Specifically, by this approach, an $n$-point one-loop integrand is expressed as a sum of $(n+2)$-point tree amplitudes, in which two particles carry the loop momentum $\pm\ell^{\mu}$ with opposite directions. When further expanding the tree amplitudes according to the tree-level expansion formulas, one then writes the one-loop YM/YMS (GR/EYM) integrands as a combination of $(n+2)$-point tree-level BS (YM) amplitudes under the forward limit \cite{He:2016mzd,He:2017spx,Geyer:2017ela,Porkert:2022efy}. Nevertheless, this forward limit expression of integrands is not the one in the conventional sense because the denominators of loop propagators are linear functions of the loop momentum $\ell^{\mu}$. Hence, the expansion coefficients cannot be directly used to generate one-loop BCJ numerators.
In recent work \cite{Cao:2025ygu,Du:2025yxz}, a consistency condition for the kinematic coefficients was proposed. As pointed out in \cite{Cao:2025ygu,Du:2025yxz}, once the kinematic coefficients in the forward limit expression of YM (YMS) integrands satisfy the consistency condition, they can be used to generate BCJ numerators of YM (YMS) at the one-loop level. Besides a general analysis of the existence of such coefficients, compact expansion formulas for scalar-loop YMS integrands with no more than three gluons and an arbitrary number of scalars were provided \cite{Du:2025yxz}, inspired by the localisation strategy established in \cite{Xie:2024pro,Xie:2025utp}. As observed in \cite{Du:2025yxz}, the expansion formula of the YMS integrand can be extended to EYM integrands with the same expansion coefficients. In this work, we provide solid proof of this statement. We express a one-loop EYM integrand in terms of tree-level colour-stripped YM amplitudes with forward limit, then further expand these YM amplitudes in terms of BS ones. We prove that once the coefficients in both steps obey the consistency conditions, the EYM amplitude can be finally expressed as a combination of quadratic-propagator-expressed one-loop YM integrands, with the same coefficients in the expansion formula of YMS integrands. While combining this expansion formula with the relationship between GR and EYM integrands \cite{He:2016mzd,He:2017spx,Geyer:2017ela,Porkert:2022efy,Xie:2025utp,Cao:2025ygu}, one then obtains an expansion formula of GR integrands, in terms of YM ones.
%While the double copy is firmly established for tree-level amplitudes, its generalization to loop level remains a significant algebraic and computational challenge. BCJ numerators at one loop level exist in principle, but their explicit construction is highly nontrivial. An efficient method for computing one loop amplitudes is the forward limit approach, which expresses the one loop integrand as tree level amplitudes with two additional legs carrying opposite loop momenta. However, as emphasized in recent work, coefficients from forward limit expansions cannot directly serve as BCJ numerators; they must satisfy additional consistency conditions to ensure cyclic symmetry under loop momentum shifts. In [1], the consistent construction of one loop BCJ numerators for Yang-Mills and Yang-Mills-scalar theories was systematically investigated through a two-step expansion strategy: first decomposing the Yang-Mills integrand into tree level Yang-Mills-scalar amplitudes, then expanding each YMS amplitude into bi-adjoint scalar amplitudes, with expansion coefficients constrained by algebraic consistency conditions. This method provides explicit one loop BCJ numerators for Yang-Mills theory with up to three external gluons.
%Building upon this foundation, the present work studies the extension to Einstein-Yang-Mills theory. We prove that the one loop EYM integrand can be expanded in terms of conventional Yang-Mills integrands with quadratic propagators, where the expansion coefficients precisely match those derived in [1] for the YMS-to-BS expansion.
The structure of this paper is given as follows. In \secref{sec:review}, we provide a brief review of the forward limit approach of YM and YMS integrands as well as the consistency conditions. The forward limit expressed EYM integrands, in terms of tree-level BS amplitudes, are proposed in \secref{sec:relation}. We prove the expansion formula of EYM integrand in \secref{sec:Proof} and summarise this work in \secref{sec:Conclusions}. Helpful formulas are included in the appendix.
\section{Forward limit, YM, YMS integrands, and consistency condition of numerators}\label{sec:review}
The forward limit approach states that the $n$-point integrand of one-loop amplitude can be expressed by $(n+2)$-point tree level amplitudes while two particles
carry the loop momentum $\pm\ell^{\mu}$ of opposite directions. For the Yang-Mills integrand, it is convenient to first express the tree-level $n$-point amplitude in terms of the Kleiss-Kuijf (KK) \cite{Kleiss:1988ne} basis of BS ones, and then take the forward limit as follows
%
{\bea
{\cal I}^{\text{YM}}(1,2,...,n)&=&{1\over \ell^2}\lim\limits_{k_{\pm}\to \pm\ell}\Sl_{h}A^{\text{YM}}(+,1,2,...,n,-)+\text{cyc} (1,...,n)\nn
%
&=&{1\over \ell^2}\lim\limits_{k_{\pm}\to \pm\ell}\Sl_{h}\left[\,\Sl_{\pmb\sigma\in \text{S}_n}N(+,\pmb\sigma,-)A^{\text{BS}}(+,\pmb\sigma,-|+,1,2,...,n,-)\,\right]+\text{cyc} (1,...,n)\nonumber\\
%
&\sim&\Sl_{h}\Biggl[\,\Sl_{\pmb\sigma\in \text{S}_n}{1\over (\ell-k_{\pmb\sigma_L})^2}N(+(\ell-k_{\pmb\sigma_L}),\pmb\sigma,-(\ell-k_{\pmb\sigma_L}))\nn
%
&&~~~~~\times
A^{\text{BS}}(+(\ell-k_{\pmb\sigma_L}),\pmb\sigma,-(\ell-k_{\pmb\sigma_L})|+(\ell-k_{\pmb\sigma_L}),1,2,...,n,-(\ell-k_{\pmb\sigma_L}))\Biggr]\nn
%
&\equiv&{\cal I}^{\text{YM}}(\ell;1,2,...,n).\label{Eq:ForwardYM}
%
\eea}
%
where we have expressed the tree-level YM amplitudes in terms of the KK basis of BS amplitudes. {In the above equation, we formally summed over gluon states $h$, while these are further reduced together with the polarisations $\epsilon_{\pm}$ as
%
\bea
\Sl_{h}\epsilon_{-}^{\mu}\epsilon_{+}^{\nu}\to\Delta^{\mu\nu},~~\eta_{\mu\nu}\Delta^{\mu\nu}\to D-2,~~V_{\mu}W_{\nu}\Delta^{\mu\nu}\to V\cdot W. \label{Eq:Polarizations}
\eea
%
The $D$ refers to the dimension of spacetime in which the loop momentum $\ell^{\mu}$ lives, when dimensional regularisation is taken. The $V_{\mu}$ and $W_{\nu}$ are two arbitrary vectors.
In the square brackets on the second line, all permutations $\pmb\sigma\in \text{S}_n$ of external particles are summed over. The $\text{cyc} (1,...,n)$ means we also sum over all terms related by cyclic permutations. The expansion coefficients $N(+,\pmb\sigma,-)$ in (\ref{Eq:ForwardYM}) are functions of
Lorentz contractions of loop momentum $\ell^{\mu}$, external momenta $k^{\mu}_i$ and external polarisations $\epsilon^{\mu}_j$ of gluons. These functions form the Del Duca-Dixon-Maltoni (DDM) basis \cite{DelDuca:1999rs,Bern:2010yg} of BCJ numerators, which generate all other tree-level BCJ numerators via algebraic identities. On the third and the fourth lines of (\ref{Eq:Polarizations}), an overall shift $\ell^{\mu}\to\ell^{\mu}-k_{\pmb\sigma_L}^{\mu}$ (where the permutation $\pmb\sigma$ is supposed to have the form $\pmb\sigma_L,1,\pmb\sigma_R$) of loop momentum $\ell^{\mu}$ for each permutation $\pmb\sigma$ was performed. This shift does not affect the integral result in the framework of dimensional regularisation, thus the shifted objects, after summation, also produce an integrand of YM, denoted by ${\cal I}^{\text{YM}}(\ell;1,2,...,n)$. The notation means the loop momentum $\ell^{\mu}$ is always fixed as the one to the lhs. of $1$.}
{The scalar-loop YMS integrand} with scalars coupled to gluons also obeys a similar expression in terms of BS ones, based on the forward limit of the corresponding tree-level relation
%
{\bea
{\cal I}^{\text{YMS}}(1,2,...,r||\mathsf{G}\,|\,\pmb\rho)&=&\left[\,{1\over \ell^2}\lim\limits_{k_{\pm}\to \pm\ell}A^{\text{YMS}}(+,1,2,...,r,-||\mathsf{G}\,|\,+,\pmb\rho,-)+\text{cyc} (1,...,r)\,\right]+\text{cyc}(\pmb\rho)\nn
%
&=&\Biggl\{{1\over \ell^2}\lim\limits_{k_{\pm}\to \pm\ell}\bigg[\,\Sl_{\pmb\sigma\in \{1,...,r\}\shuffle\text{perms}\,\mathsf{G}}C(+,\pmb\sigma,-)A^{\text{BS}}(+,\pmb\sigma,-\,|\,+,\pmb\rho,-)\,\bigg]\nn
%
&&~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+\text{cyc} (1,...,r)\Biggr\}+\text{cyc}(\pmb\rho)\nn
%
&\sim&\Biggl\{\bigg[\,\Sl_{\pmb\sigma\in \{1,...,r\}\shuffle\text{perms}\,\mathsf{G}}{1\over (\ell-k_{\pmb\sigma_L})^2}C(+(\ell-k_{\pmb\sigma_L}),\pmb\sigma,-(\ell-k_{\pmb\sigma_L}))\nn
%
&&~~~~~~~~~~~\times A^{\text{BS}}(+(\ell-k_{\pmb\sigma_L}),\pmb\sigma,-(\ell-k_{\pmb\sigma_L})\,|\,+(\ell-k_{\pmb\sigma_L}),\pmb\rho,-(\ell-k_{\pmb\sigma_L}))\,\bigg]\nn
%
&&~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+\text{cyc} (1,...,r)\Biggr\}+\text{cyc}(\pmb\rho)\nn
%
&\equiv&{\cal I}^{\text{YMS}}(\ell;1,2,...,r||\mathsf{G}\,|\,\pmb\rho),\label{Eq:ForwardYMs}
%
\eea}
%
in which $C(+,\pmb\sigma,-)$ denotes the kinematic coefficients. On the second line, we introduced the shuffle $A\shuffle B$ of two ordered sets $A$ and $B$, which is defined by the permutations of all elements of the two sets so that the relative order in each ordered set is preserved. For example, $\{1,2\}\shuffle\{3,4\}$ is the collection of the following permutations
%
\bea
\{1,2,3,4\},~~\{1,3,2,4\},~~\{1,3,4,2\},~~\{3,1,2,4\},~~\{3,1,4,2\},~~\{3,4,1,2\}.
\eea
%
The $\text{perms}\,\mathsf{G}$ denotes all possible permutations of elements in $\mathsf{G}$.
In (\ref{Eq:ForwardYMs}), we have summed over all terms related by cyclic permutations of scalars $1$, ..., $r$, and the cyclic permutations of $\pmb\rho\in \text{perms}\,(\{1,...,r\}\cup\mathsf{G})$. As we have done in (\ref{Eq:ForwardYM}), we shifted the loop momentum $\ell^{\mu}$ and redefined the YMS integrand by fixing $\ell^{\mu}$ as the one attached to $1$ from left.
Though the expansion formulas (\ref{Eq:ForwardYM}) and (\ref{Eq:ForwardYMs}) generate the one-loop integrands of YM and YMS, these expressions are based on the forward limit of tree-level BS amplitudes, in which the loop propagators are linear propagators. Generally speaking, the expansion formulas (\ref{Eq:ForwardYM}) and (\ref{Eq:ForwardYMs}) cannot be directly extended to expansion formulas of YM, YMS integrands in terms of conventional loop BS integrands with quadratic propagators. In addition, though the coefficients in (\ref{Eq:ForwardYM}) and (\ref{Eq:ForwardYMs}) reflect the tree-level BCJ numerators in the DDM basis, they cannot be directly considered as the one-loop numerators.
As pointed out in \cite{Du:2025yxz}, in order to construct correct BCJ numerators, we should impose additional consistency conditions on the coefficients in (\ref{Eq:ForwardYM}) and (\ref{Eq:ForwardYMs})
%
\bea
N(\ell,\sigma_1=1,\sigma_2,...,\sigma_n,-\ell)&=&N(\ell+k_1,\sigma_2,...,\sigma_n,1,-(\ell+k_1))\nn
%
&=&...=N(\ell+k_{1,\sigma_{n-1}},\sigma_n,1,\sigma_2,...,\sigma_{n-1},-(\ell+k_{1,\sigma_{n-1}})),\label{Eq:ConsistencyYM}\\
%
C(\ell,\sigma_1=1,\sigma_2,...,\sigma_n,-\ell)&=&C(\ell+k_1,\sigma_2,...,\sigma_n,1,-(\ell+k_1))\nn
%
&=&...=C(\ell+k_{1,\sigma_{n-1}},\sigma_n,1,\sigma_2,...,\sigma_{n-1},-(\ell+k_{1,\sigma_{n-1}})),\label{Eq:ConsistencyYMs}
\eea
%
where $k_{1,\sigma_{n-1}}=k_1+k_{\sigma_2}+\dots+k_{\sigma_{n-1}}$. The $\pm$ particles are explicitly expressed by their momenta. Once the consistency conditions are satisfied, they can play as the consistency coefficients when expressing YM or YMS integrands in terms of the propagator matrix ${\cal I}^{\text{BS}}$ (i.e., the BS integrands with quadratic propagators)
%
\bea
{\cal I}^{\text{YM}}(\ell;1,2,...,n)&=&\Sl_{\{\sigma_2,...,\sigma_n\}\in\text{S}_{n-1}}N(\ell;1,\sigma_2,...,\sigma_n)\,{\cal I}^{\text{BS}}(\ell;1,\sigma_2,...,\sigma_n|1,2,...,n),\label{Eq:ExpYM}\\
%
{\cal I}^{\text{YMS}}(\ell;1,2,...,r||\mathsf{G}\,|\,\pmb\rho)&=&\Sl_{\{\sigma_2,...,\sigma_n\}\in\{2,...,r\}\shuffle\text{perms}\,\mathsf{G}}C(\ell;1,\sigma_2,...,\sigma_n)\,{\cal I}^{\text{BS}}(\ell;1,\sigma_2,...,\sigma_n\,|\,\pmb\rho),\label{Eq:ExpYMs}
\eea
%
In which the kinematic coefficients $N(\ell;1,\sigma_2,...,\sigma_n)$ and $C(\ell;1,\sigma_2,...,\sigma_n)$ are defined by those satisfying consistency conditions \eqref{Eq:ConsistencyYM} and \eqref{Eq:ConsistencyYMs}
%
\bea
N(\ell;1,\sigma_2,...,\sigma_n)\equiv N(\ell,\sigma_1=1,\sigma_2,...,\sigma_n,-\ell),~~C(\ell;1,\sigma_2,...,\sigma_n)=C(\ell,\sigma_1=1,\sigma_2,...,\sigma_n,-\ell).
\eea
%
These kinematic coefficients are now the proper ones for constructing BCJ numerators. In the coming sections, we show the expansion formula for EYM integrand (which involves gravitons, dilatons and B-fields) parallel with the YMS one (\ref{Eq:ExpYMs}), supposing we already have the consistency numerators $N(\ell;1,\sigma_2,...,\sigma_n)$ and $C(\ell;1,\sigma_2,...,\sigma_n)$ in (\ref{Eq:ExpYM}) and (\ref{Eq:ExpYMs}).
\section{The relation between EYM and YM integrands}\label{sec:relation}
According to the worldsheet approach \cite{He:2016mzd,He:2017spx,Geyer:2017ela,Porkert:2022efy}, the EYM integrand, like the YM and YMS ones, also arises from the forward limit of the tree-level expansion formula, i.e.,
%
\bea
\mathcal{I}^{\,\text{EYM}}(1...,r||\mathsf{H})&=&\Sl_{\pmb\sigma\in\{1,...,r\}\shuffle \text{perms}\,\mathsf{H} }{1\over (\ell-k_{\pmb\sigma_L})^2}C(\ell-k_{\pmb\sigma_L},\pmb\sigma,-(\ell-k_{\pmb\sigma_L}))\label{eq:gravity-half-ladder-welded0}\\
%
&&~~~~~~~~~~~~~~~~~~~~~~~~~~~\times A^{\text{YM}}(+(\ell-k_{\pmb\sigma_L}),\pmb\sigma_L,1,\pmb\sigma_R,-(\ell-k_{\pmb\sigma_L}))+\text{cyc}(1,...,r),\nonumber
%
\eea
%%
where we have applied the fact that tree-level EYM amplitudes share the same expansion formula as the YMS amplitudes \cite{Fu:2017uzt}. The summation over $h$ in this formula and the coming discussions are absorbed for brevity. For a given permutation $\pmb\sigma$, we again use $\pmb\sigma_L$ and $\pmb\sigma_R$ to denote the ordered subsets separated by the scalar $1$, i.e., $\pmb\sigma=\pmb\sigma_L,1,\pmb\sigma_R$. When the tree-level amplitudes $A^{\text{YM}}(+(\ell-k_{\pmb\sigma_L}),\pmb\sigma_L,1,\pmb\sigma_R,-(\ell-k_{\pmb\sigma_L}))$ are further expanded in terms of BS ones, the integrand (\ref{eq:gravity-half-ladder-welded0}) turns into
%
\bea
\mathcal{I}^{\,\text{EYM}}(1...,r||\mathsf{H})
&=&\Sl_{\pmb\sigma\in\{1,...,r\}\shuffle \text{perms}~\mathsf{H} }{1\over (\ell-k_{\pmb\sigma_L})^2}C(\ell-k_{\pmb\sigma_L},\pmb\sigma,-(\ell-k_{\pmb\sigma_L}))\nn
%
&&\times\Sl_{\pmb\rho\in\text{perms}~(\{1,...,r\}\cup\mathsf{H}) }~ A^{\text{BS}}(\ell-k_{\pmb\sigma_L},\pmb\sigma,-(\ell-k_{\pmb\sigma_L})|\ell-k_{\pmb\sigma_L},\pmb\rho,-(\ell-k_{\pmb\sigma_L}))\nn
&&\times\W N(\ell-k_{\pmb\sigma_L},\pmb\rho,-(\ell-k_{\pmb\sigma_L}))
+\text{cyc}(1,...,r).\label{EYMeg0}
\eea
%
In the above expression, we have used the notation $\W N(+,\pmb\rho,-)$ instead of $N(+,\pmb\rho,-)$ in (\ref{Eq:ForwardYM}), to emphasise that the polarisations in $\W N(+,\pmb\rho,-)$ are different from those in $C(+,\pmb\sigma,-)$. In the next section, we prove that if all coefficients $C(+,\pmb\sigma,-)$ and $\W N(+,\pmb\rho,-)$ obey the consistency conditions (\ref{Eq:ConsistencyYM}) and (\ref{Eq:ConsistencyYMs}), we can express the EYM integrand in terms of YM integrands, associated with the same coefficients $C(\ell;1,\sigma_2,...,\sigma_n)$ in (\ref{Eq:ExpYMs})
%
\bea
\boxed{{\cal I}^{\text{EYM}}(\ell;1,2,...,r||\mathsf{H}\,)=\Sl_{\{\sigma_2,...,\sigma_n\}\in\{2,...,r\}\shuffle\text{perms}\,\mathsf{H}}C(\ell;1,\sigma_2,...,\sigma_n)\,{\cal I}^{\text{YM}}(\ell;1,\sigma_2,...,\sigma_n)}.\label{Eq:ExpEYM}
\eea
For the integrands with no more than three gravitons, the coefficients $C(\ell;1,\sigma_2,...,\sigma_n)$ can be provided by the compact expressions (\ref{Eq:CoefficientEG1a}), (\ref{Eq:YMSExp2GravitonsCF}), and (\ref{Eq:YMSExp3GravitonsCF}).
\section{Proof of the expansion formula of EYM integrand}\label{sec:Proof}
In this section, we present a proof of \eqref{Eq:ExpEYM}, based on the forward limit expression (\ref{EYMeg0}) and the consistency conditions (\ref{Eq:ConsistencyYM}), (\ref{Eq:ConsistencyYMs}).
\subsection{An example study}
Without loss of generality, let us consider the simple example with two gluons $1$, $2$ and two gravitons $p$, $q$. We concentrate on the contribution of left permutation $\pmb\sigma=\{1,p,q,2\}\in \{1,\{2\}\shuffle\text{perms}\{p,q\}\}$ and right permutation $\pmb\rho=\{p,1,2,q\}\in\text{perms}\,(\{1,2\}\cup\{p,q\})$. For this choice of $\pmb\rho$ and $\pmb\sigma$, the forward limit of BS tree amplitude in (\ref{EYMeg0}), i.e., $A^{\text{BS}}(\ell,1,p,q,2,-\ell|\ell,p,1,2,q,-\ell)$ is expressed by the sum of all possible Feynman diagrams, in which tree-level BS currents are attached to the linear propagator line between $+$ and $-$. More concretely,
\bea
A^{\text{BS}}(\ell,1,p,q,2,-\ell|\ell,p,1,2,q,-\ell)=\begin{minipage}{1.8cm}\includegraphics[width=1.8cm]{EYM1}\end{minipage}+\begin{minipage}{2.3cm}\includegraphics[width=2.3cm]{EYM2}\end{minipage}+...,\label{Eq:BSFeyn}
\eea
where we only displayed two Feynman diagrams explicitly and included other ones in the ellipsis. {\it Generally speaking, there are (i) nonvanishing Feynman diagrams where all subcurrents are nonvanishing ones, i.e., the first diagram in (\ref{Eq:BSFeyn}), and (ii) vanishing diagrams which involve subcurrents containing distinct elements in the left and right ordered subsets, i.e., the second diagram in (\ref{Eq:BSFeyn}).} Apparently, diagrams in case (ii) do not contribute to the result. Hence, we only need to consider the nonvanishing diagrams, in which any subcurrent contains the same elements in the left and right subsets.
When BS amplitude (\ref{Eq:BSFeyn}) is associated with the coefficients $C(\ell,1,p,q,2,-\ell)$ and $\W N(\ell,p,1,2,q,-\ell)$, the first diagram in (\ref{Eq:BSFeyn}) contributes a term
%
\bea
C(\ell,1,p,q,2,-\ell)\begin{minipage}{1.8cm}\includegraphics[width=1.8cm]{EYM1}\end{minipage}\W N(\ell,p,1,2,q,-\ell).\label{Eq:EYMeg1}
\eea
%
The loop propagators along the line between $+$ and $-$ are linear ones. A crucial observation is that the summation over $\pmb\sigma$, $\pmb\rho$ in (\ref{EYMeg0}) and over the cyclic permutations of gluons together permit the following term
%
\bea
C(\ell-k_q-k_2,q,2,1,p,-(\ell-k_q-k_2))\begin{minipage}{4.4cm}\includegraphics[width=4.4cm]{EYM3}\end{minipage}\W N(\ell-k_q-k_2,2,q,p,1,-(\ell-k_q-k_2)),\label{Eq:EYMeg2}\nn
\eea
%
which comes from the BS amplitude with $\pmb\sigma=\{q,2,1,p\}$, $\pmb\rho=\{2,q,p,1\}$.
% The two terms
%
%\bea
%&&C(\ell+k_1,p,q,2,1,-(\ell+k_1))~\begin{minipage}{2.8cm}\includegraphics[width=2.8cm]{EYM4}\end{minipage}~\W N(\ell+k_1,1,2,q,p,-(\ell+k_1)),\nn
%&&C(\ell-k_2,2,1,p,q,-(\ell-k_2))~\begin{minipage}{2.8cm}\includegraphics[width=2.8cm]{EYM5}\end{minipage}~\W N(\ell-k_2,q,p,1,2,-(\ell-k_2)),\label{Eq:EYMeg3}
%\eea
%
%which respectively come from the BS amplitudes with $\pmb\sigma=\{p,q,2,1\}$, $\pmb\rho=\{1,2,q,p\}$, and $\pmb\sigma=\{2,1,p,q\}$, $\pmb\rho=\{q,p,1,2\}$ are also provided.
Supposing the left and right kinematic coefficients satisfy the consistency conditions (\ref{Eq:ConsistencyYMs}) and (\ref{Eq:ConsistencyYM}) respectively, we have
%
\bea
C(\ell-k_q-k_2,q,2,1,p,-(\ell-k_q-k_2))&=&C(\ell,1,p,q,2,-\ell),\nn
%
\W N(\ell-k_q-k_2,2,q,p,1,-(\ell-k_q-k_2))&=&\W N(\ell,p,1,2,q,-\ell),\label{Eq:EYMeg4}
\eea
%
where the equation on the last line can be obtained by redefining the loop momentum in (\ref{Eq:ConsistencyYM}) via $\ell\to \ell'=\ell+k_p$.
Now we dress (\ref{Eq:EYMeg1}) and (\ref{Eq:EYMeg2}) by $1\over \ell^2$ and $1\over (\ell-k_q-k_2)^2$ respectively and then sum them together. Considering the consistency conditions (\ref{Eq:EYMeg4}) and noting that the Feynman diagrams in each sum are related by cyclic permutations of the BS subcurrents, we get the following term
%
\bea
&&C(\ell,1,p,q,2,-\ell)\left[\,{1\over\ell^2}\begin{minipage}{1.8cm}\includegraphics[width=1.8cm]{EYM1}\end{minipage}+{1\over(\ell-k_q-k_2)^2}\begin{minipage}{4.2cm}\includegraphics[width=4.2cm]{EYM3}\end{minipage}\,\right]\W N(\ell,p,1,2,q,-\ell)\nn
%
&=& C(\ell;1,p,q,2)~\begin{minipage}{1.8cm}\includegraphics[width=1.8cm]{EYM6}\end{minipage}~\W N(\ell;p,1,2,q),\label{Eq:EYMeg05}
%
%&&C(\ell,1,p,q,2,-\ell)\Bigg[{\W N(\ell,p,1,2,q,\ell)\over\ell^2}\begin{minipage}{2.3cm}\includegraphics[width=2.3cm]{EYM2}\end{minipage}+{\W N(\ell+k_1,1,2,q,p,-(\ell+k_1))\over(\ell+k_1)^2}\begin{minipage}{2.8cm}\includegraphics[width=2.8cm]{EYM4}\end{minipage}\nn
%&&~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+{\W N(\ell-k_2,q,p,1,2,-(\ell-k_2))\over(\ell-k_2)^2}\begin{minipage}{2.8cm}\includegraphics[width=2.8cm]{EYM5}\end{minipage}\Bigg]\nn
%&=&C(\ell,1,p,q,2,-\ell)\Bigg[{1\over\ell^2}\begin{minipage}{2.1cm}\includegraphics[width=2.1cm]{EYM2}\end{minipage}+{1\over(\ell+k_1)^2}\begin{minipage}{2.3cm}\includegraphics[width=2.3cm]{EYM4}\end{minipage}+{1\over(\ell-k_2)^2}\begin{minipage}{2.3cm}\includegraphics[width=2.3cm]{EYM5}\end{minipage}\Bigg]\W N(\ell,p,1,2,q,\ell)\nn
%
%&=& C(\ell;1,p,q,2)\begin{minipage}{2cm}\includegraphics[width=2cm]{EYM7}\end{minipage}\W N(\ell;p,1,2,q). \label{Eq:EYMeg5}
\eea
%
where the coefficients are reexpressed by
\bea
C(\ell;1,p,q,2)\equiv C(\ell,1,p,q,2,-\ell),~~~~\W N(\ell;p,1,2,q,-)\equiv\W N(\ell,p,1,2,q,-\ell).
\eea
In (\ref{Eq:EYMeg05}), we have applied the identity (\ref{Eq:IDPartialFraction1}), through which the sum of linear-propagator Feynman diagrams related by cyclic permutations of BS subcurrents produces a quadratic-propagator Feynman diagram.
%The kinematic numerators in the first line of (\ref{Eq:EYMeg5}) are distinct because they do not satisfy the consistency condition, yet the BS subcurrents $\phi_{pq|12}$ with mismatched left– and right-elements vanish, so the second line still follows. Moreover, terms receiving nonvanishing contributions from BS subcurrents, such as those in (\ref{Eq:EYMeg05}) can always be precisely related by the consistency conditions and cyclic structure.
For the concrete $\pmb\sigma=\{1,p,q,2\}$ and $\pmb\rho=\{p,1,2,q\}$, there is another nonvanishing diagram, the tadpole diagram. In this case, the tadpole term in (\ref{Eq:BSFeyn}), together with the kinematic coefficients, is presented by
%
\bea
C(\ell;1,p,q,2)\,\left[\,{1\over \ell^2}\,\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG1}\end{minipage}~\W N(\ell;p,1,2,q)\,\right].
\eea
%
When considering other $\pmb\rho$ in (\ref{EYMeg0}), we get tadpole terms with other right permutations. This boundary case is naturally a quadratic propagator term. As demonstrated in \cite{Xie:2025utp}, the sum of all such tadpole diagrams that are related via cyclic permutations of gluons vanishes. Now we formally retain these tadpole terms and recollect them with other diagrams that share the same left and right permutations up to cyclicity. These related diagrams come from the sum over $\{1,\{2\}\shuffle \text{perms}\{p,q\}\}$, the cyclic sum of 1, 2, and the sum over right permutations in (\ref{EYMeg0}). For our choice of $\pmb\sigma=\{1,p,q,2\}$ and $\pmb\rho=\{p,1,2,q\}$, we get
%
\bea
&&C(\ell;1,p,q,2)\,\left[\,\begin{minipage}{1.8cm}\includegraphics[width=1.8cm]{EYM6}\end{minipage}~\W N(\ell;p,1,2,q)+\begin{minipage}{2.6cm}\includegraphics[width=2.6cm]{EYM6a}\end{minipage}~\W N(\ell-k_2;1,2,q,p)+\dots\right.\nn
%
&&~~~~~~~~~~~~~~~~~~\,\left.+\,\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG2}\end{minipage}~\W N(\ell;p,1,2,q)+\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG3}\end{minipage}~\W N(\ell;q,p,1,2)+\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG4}\end{minipage}~\W N(\ell;2,q,p,1)\right.\nn
%
&&~~~~~~~~~~~~~~~~~~\,\left.+\,\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG5}\end{minipage}~\W N(\ell;1,2,q,p)+\begin{minipage}{2.1cm}\includegraphics[width=2.1cm]{TadpoleEG2a}\end{minipage}~\W N(\ell-k_2;p,1,2,q)+\dots\,\right],\label{Eq:EYMeg06}
\eea
%
where we have applied the consistency condition of $C(\ell;1,p,q,2)$ for terms in the square brackets. For example, in the second term, we have
\bea
C(\ell;1,p,q,2)\begin{minipage}{2.2cm}\includegraphics[width=2.2cm]{EYM6a}\end{minipage}\W N(\ell-k_2;1,2,q,p)=C(\ell-k_2;2,1,p,q)\begin{minipage}{2.2cm}\includegraphics[width=2.2cm]{EYM6a}\end{minipage}\W N(\ell-k_2;1,2,q,p).
\eea
The second line of (\ref{Eq:EYMeg06}) and the first term on the third line come from the summation over right permutations, while there also exist terms with $\phi_{21pq|\pmb\rho}$, $\phi_{q21p|\pmb\rho}$, $\phi_{pq21|\pmb\rho}$ (where $\pmb\rho\in\text{cyc}(p12q))$. We only explicitly display one term here, i.e., the second term on the third line. Now we denote the summation inside the square brackets by $\mathsf{T}(\ell)$:
%
\bea
\mathsf{T}(\ell)=\mathsf{T}^{(1)}+\mathsf{T}^{(2)}+\dots+\mathsf{T}^{(3)}+\mathsf{T}^{(4)}+\mathsf{T}^{(5)}+\mathsf{T}^{(6)}+\mathsf{T}^{(7)}+\dots.
\eea
%
The seven terms in turn denote the expressions of the seven terms inside the square brackets of (\ref{Eq:EYMeg06}). When we shift the momentum in these terms
%
\bea
\mathsf{T}(\ell)\to \mathsf{T}'(\ell)&=&\mathsf{T}^{(1)}|_{\ell\to\ell-k_p}+\mathsf{T}^{(2)}|_{\ell\to\ell+k_2}+\dots+\mathsf{T}^{(3)}|_{\ell\to\ell-k_p}+\mathsf{T}^{(4)}|_{\ell\to\ell-k_p-k_q}\nn
&&+\mathsf{T}^{(5)}|_{\ell\to\ell+k_1}+\mathsf{T}^{(6)}+\mathsf{T}^{(7)}|_{\ell\to\ell+k_2-k_p}+\dots,
\eea
%
and applying the consistency condition (\ref{Eq:ConsistencyYM}), the $\mathsf{T}'(\ell)$ immediately turns into
%
\bea
\mathsf{T}'(\ell)&=&\left[\,\begin{minipage}{2.3cm}\includegraphics[width=2.3cm]{EYM8}\end{minipage}~\W N(\ell-k_p;p,1,2,q)+\begin{minipage}{2.1cm}\includegraphics[width=2.1cm]{TadpoleEG3a}\end{minipage}~\W N(\ell;1,2,q,p)+\dots\right.\nn
&&~~\,\left.+\,\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG6}\end{minipage}~\W N(\ell-k_p;p,1,2,q)+\begin{minipage}{2.4cm}\includegraphics[width=2.3cm]{TadpoleEG7}\end{minipage}~\W N(\ell-k_q-k_p;q,p,1,2)\right.\nn
%
&&~~\,\left.+\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG8}\end{minipage}\W N(\ell+k_1;2,q,p,1)+\begin{minipage}{1.5cm}\includegraphics[width=1.5cm]{TadpoleEG5}\end{minipage}\W N(\ell;1,2,q,p)+\begin{minipage}{1.7cm}\includegraphics[width=1.7cm]{TadpoleEG4a}\end{minipage}\W N(\ell-k_p;p,1,2,q)+\dots\,\right]\nn
%
&=&\left[\,\begin{minipage}{2.3cm}\includegraphics[width=2.3cm]{EYM8}\end{minipage}+\begin{minipage}{2.1cm}\includegraphics[width=2.1cm]{TadpoleEG3a}\end{minipage}+\dots+~~\,\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG6}\end{minipage}~~+~~\begin{minipage}{2.4cm}\includegraphics[width=2.3cm]{TadpoleEG7}\end{minipage}\right.\nn
&&~~~~~~~~~~~~~~~~~~~~~~~~\,\left.+~~\begin{minipage}{1.6cm}\includegraphics[width=1.6cm]{TadpoleEG8}\end{minipage}~~+~~\begin{minipage}{1.5cm}\includegraphics[width=1.5cm]{TadpoleEG5}\end{minipage}~~+~~\begin{minipage}{1.7cm}\includegraphics[width=1.7cm]{TadpoleEG4a}\end{minipage}~~+\dots\,\right]\W N(\ell;1,2,q,p)\nn
%
&=&{\cal I}^{\text{BS}}(1,p,q,2|\ell;1,2,q,p)\W N(\ell;1,2,q,p).
\eea
This is nothing but just one term of the expansion formula (\ref{Eq:ExpYM}) of YM (with exchanging the roles of left and right and defining the position of loop momentum via the $1$ in the right permutation). Since the $\mathsf{T}(\ell)$ and $\mathsf{T}'(\ell)$ are related by shifts of loop momentum, when integrating over loop momentum, they result in the same YM integrand. Therefore, the $\mathsf{T}(\ell)$ is also a term of the YM integrand, but the loop momentum is defined via the position of $1$ in the left permutation. Inspired by the above discussions, we rewrite the expression in (\ref{Eq:EYMeg06}) as follows
\bea
&&C(\ell;1,p,q,2)\big[T(\ell;1p|p1,q2|2q)~\W N(\ell;p,1,2,q)+T(\ell-k_2;21|12,pq|qp)~\W N(\ell-k_2;1,2,q,p)+\dots\nn
&&+T(\ell;1pq2|p12q)~\W N(\ell;p,1,2,q)+T(\ell;1pq2|qp12)~\W N(\ell;q,p,1,2)+T(\ell;1pq2|2qp1)~\W N(\ell;2,q,p,1)\nn
&&+T(\ell;1pq2|12qp)~\W N(\ell;1,2,q,p)+T(\ell-k_2;21pq|p12q)~\W N(\ell-k_2;p,1,2,q)+\dots\big]\nn
&=&C(\ell;1,p,q,2)\left[\Sl_{T}T(\ell-k_A;A1B|P1Q,\dots)~\W N(l-k_A+k_P;1,2,q,p)\right],
\eea
where all possible Feynman diagrams $T(l-k_A;A1B|P1Q,\dots)$ (which implies the subcurrents attached to the loop are $\phi_{A1B|P1Q}$,...) with respect to the permutations $\pmb\sigma=\{1,p,q,2\}$ and $\pmb\rho=\{p,1,2,q\}$ up to cyclicity are summed over. Specifically, suppose the position of the $\ell$ is fixed by $1$ in the left permutation and the subcurrent contains $1$ is $\phi_{A1B|P1Q}$, where $A$, $B$, $P$, $Q$ are four ordered sets satisfying $A\cup B=P\cup Q$.
When applying the above discussions to terms with all possible left and right permutations
\bea
&&\pmb\sigma\in\Big\{\{1,2\}\shuffle\text{perms}\{p,q\},\text{cyc} (1,2)\Big\}=\Big\{\pmb{\alpha}=\{1,\{2\}\shuffle\text{perms}\{p,q\}\},\text{cyc}~\pmb{\alpha}\Big\},\nn
&&\pmb\rho\in\text{perms}\,(\{1,2\}\cup\{p,q\})=\Big\{\pmb\beta=\{1,\text{perms}\{2,p,q\},\text{cyc}~\pmb{\beta}\Big\},
\eea
we transform (\ref{EYMeg0}) into
%
\bea
&&\mathcal{I}^{\,\text{EYM}}(\ell;1,2||\{p,q\})\label{EYMeg7}\nn
%
&=&\Sl_{\pmb\sigma_{2,4}\in\{2\}\shuffle\text{perms}\{p,q\}}\,C(\ell;1,\pmb\sigma_{2,4}){\left[\,\Sl_{\pmb\rho_{2,4}\in\text{perms}\{2,p,q\} }\Sl_{T}T(\ell-k_A;A1B|P1Q,\dots)\W N(l-k_A+k_P;1,\pmb\rho_{2,4})\right]}\nn
%
&=&\Sl_{\pmb\sigma_{2,4}\in\{2\}\shuffle\text{perms}\{p,q\}}\,C(\ell;1,\pmb\sigma_{2,4})~\mathcal{I}^{\text{YM}}(\ell;1,\pmb\sigma_{2,4}),
\eea
%
where the expression in the square brackets on the second line is just the YM integrand, with the loop momentum $\ell$ fixed to the left of $1$ in the left permutation.
\subsection{The general proof}
The above discussions can be straightforwardly extended to a general proof. We first express the forward limit of tree-level BS amplitude $A^{\text{BS}}(\ell-k_{\pmb\sigma_L},\pmb\sigma,-(\ell-k_{\pmb\sigma_L})|\ell-k_{\pmb\sigma_L},\pmb\rho,-(\ell-k_{\pmb\sigma_L}))$ in (\ref{EYMeg0}) by the sum of all possible Feynman diagrams, which can be collected as follows.
\begin{itemize}
\item Divide the left permutation $\pmb\sigma$ into ordered subsets $\pmb\sigma\to \pmb\sigma_1|\pmb\sigma_2|...|\pmb\sigma_I$.
\item For a given division of $\pmb\sigma$, divide the right permutation $\pmb\rho\to \pmb\rho_1|\pmb\rho_2|...|\pmb\rho_I$ such that $\pmb\rho_i$ contains {\it exactly the same elements as $\pmb\sigma_i$}.
\item Associate each pair of ordered subsets $\pmb\sigma_i$, $\pmb\rho_i$ with a BS subcurrent $\phi_{\pmb\sigma_i|\pmb\rho_i}$ attached to the linear propagator line between $+$ and $-$.
\end{itemize}
The tree-level BS amplitude $A^{\text{BS}}$ with forward limit is then given by
%
\bea
A^{\text{BS}}(\ell-k_{\pmb\sigma_L},\pmb\sigma,-(\ell-k_{\pmb\sigma_L})|\ell-k_{\pmb\sigma_L},\pmb\rho,-(\ell-k_{\pmb\sigma_L}))=\Sl_{\substack{\pmb\sigma\to \pmb\sigma_1|\pmb\sigma_2|...|\pmb\sigma_I\\ \pmb\rho\to \pmb\rho_1|\pmb\rho_2|...|\pmb\rho_I }}\begin{minipage}{3.2cm}\includegraphics[width=3.2cm]{GenProofFeyn}\end{minipage}.\label{Eq:GenFeynBSForward}
\eea
%
Substituting (\ref{Eq:GenFeynBSForward}) into (\ref{EYMeg0}), one rewrites the EYM integrand as
%
\bea
\mathcal{I}^{\,\text{EYM}}(\ell;1...,r||\mathsf{H})&=&\Sl_{\substack{\pmb\sigma\in\{1,...,r\}\shuffle \text{perms}~\mathsf{H} \\ \pmb\rho\in\text{perms}~(\{1,...,r\}\cup\mathsf{H})}}\,\Sl_{\substack{\pmb\sigma\to \pmb\sigma_1|\pmb\sigma_2|...|\pmb\sigma_I\\ \pmb\rho\to \pmb\rho_1|\pmb\rho_2|...|\pmb\rho_I }}{1\over (\ell-k_{\pmb\sigma_L})^2}C(\ell-k_{\pmb\sigma_L},\pmb\sigma,-(\ell-k_{\pmb\sigma_L}))\label{EYMeg00}\nn
%
&&~~~~~~~\times \begin{minipage}{3.2cm}\includegraphics[width=3.2cm]{GenProofFeyn}\end{minipage}\W N(\ell-k_{\pmb\sigma_L},\pmb\rho,-(\ell-k_{\pmb\sigma_L}))+\text{cyc}(1,...,r).
\eea
%
On the rhs. of the above expression, for a given left permutation $\pmb\sigma\in\{1,...,r\}\shuffle \text{perms}~\mathsf{H}$, a right permutation $\pmb\rho\in\text{perms}~(\{1,...,r\}\cup\mathsf{H})$ and a specified pair of divisions of them, each pair of ordered subsets has the general form
%
\bea
\pmb\sigma_i= \mathsf {S}_i\shuffle\text{perms}\,\mathsf{H}_i\equiv\left\{s^i_1,s^i_2,...,s^i_{j_i}\right\}\shuffle \text{perms}\,\left\{h^i_{1},...,h^{i}_{l_i}\right\},~~~~\pmb\rho_i=\left\{\rho^{i}_1,\rho^{i}_2,...,\rho^{i}_{j_i+l_i}\right\},
\eea
%
where $\left\{s^i_1,s^i_2,...,s^i_{j_i}\right\}$ denotes the ordered subset of scalars which agrees with the ordering $\{1,2,...,r\}$, $\left\{h^i_{1},...,h^{i}_{l_i}\right\}$ is the graviton set, and $\left\{\rho^{i}_1,\rho^{i}_2,...,\rho^{i}_{j_i+l_i}\right\}$ stands for the ordered subset of scalars and gluons agrees with the full right permutation $\pmb\rho$. For a given pair of divisions in \eqref{EYMeg0}, we can always find other related divisions
%
\bea
&&\pmb\sigma_2|...|\pmb\sigma_I|\pmb\sigma_1,~~~~~~\,\,\pmb\rho_2|...|\pmb\rho_I|\pmb\rho_1,\nn
%
&&\pmb\sigma_3|...|\pmb\sigma_1|\pmb\sigma_2,~~~~~~\,\,\pmb\rho_3|...|\pmb\rho_1|\pmb\rho_2,\nn
%
&&\dots\nn
%
&&\pmb\sigma_I|\pmb\sigma_1|...|\pmb\sigma_{1-1},~~~~\pmb\rho_I|\pmb\rho_{1}|...|\pmb\rho_{i-1}.
\eea
%
Collecting these contributions together, we get
%
\bea
&&C(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}},\pmb\sigma_1,...,\pmb\sigma_I,-(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}}))\begin{minipage}{3.5cm}\includegraphics[width=3.5cm]{GenProofFeyn1}\end{minipage}{\W N(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}},\pmb\rho_1,...,\pmb\rho_I,-(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}}))\over (\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}})^2}\label{EYMeg1}\nn
%
&+&C(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}},\pmb\sigma_2,...,\pmb\sigma_1,-(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}}))\begin{minipage}{3.5cm}\includegraphics[width=3.5cm]{GenProofFeyn2}\end{minipage}{\W N(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}},\pmb\rho_2,...,\pmb\rho_1,-(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}}))\over (\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}})^2}\nn
%
&+&...\nn
%
&+&C(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}},\pmb\sigma_I,...,\pmb\sigma_{I-1},-(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}}))\begin{minipage}{3.3cm}\includegraphics[width=3.3cm]{GenProofFeyn3}\end{minipage}{\W N(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}},\pmb\rho_I,...,\pmb\rho_{I-1},-(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}}))\over (\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}})^2}\nn
\eea
%
where $k_{\pmb\sigma_i,\pmb\sigma_{jL}}=k_{\pmb\sigma_i}+k_{\pmb\sigma_{i+1}}+\dots+k_{\pmb\sigma_{j-1}}+k_{\pmb\sigma_{jL}}$. We supposed the scalar $1$ is involved in $\pmb\sigma_j=\{\pmb\sigma_{jL},1,\pmb\sigma_{jR}\}$ and $\pmb\rho_j=\{\pmb\rho_{jL},1,\pmb\rho_{jR}\}$. The consistency conditions (\ref{Eq:ConsistencyYMs}), (\ref{Eq:ConsistencyYM}) allow us to express these coefficients as
\bea
&&C(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}},\pmb\sigma_1,...,\pmb\sigma_I,-(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}}))=C(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}},\pmb\sigma_2,...,\pmb\sigma_1,-(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}}))=...\nn
%
&=&C(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}},\pmb\sigma_I,...,\pmb\sigma_{I-1},-(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}}))=C(\ell,1,\pmb\sigma_{jR},\pmb\sigma_{j+1},...,\pmb\sigma_{j-1},\pmb\sigma_{jL},-\ell),\nn
%
&&\W N(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}},\pmb\rho_1,...,\pmb\rho_I,-(\ell-k_{\pmb\sigma_1,\pmb\sigma_{jL}}))=\W N(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}},\pmb\rho_2,...,\pmb\rho_1,-(\ell-k_{\pmb\sigma_2,\pmb\sigma_{jL}}))=...\nn
%
&=&\W N(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}},\pmb\rho_I,...,\pmb\rho_{I-1},-(\ell-k_{\pmb\sigma_I,\pmb\sigma_{jL}}))=\W N(\ell-k_{\pmb\sigma_{jL}},\pmb\rho_j,...,\pmb\rho_{j-1},-(\ell-k_{\pmb\sigma_{jL}})).
%
\eea
Thus, the coefficients in each term of (\ref{EYMeg1}) can be extracted as overall coefficients. According to the identity (\ref{Eq:IDPartialFraction1}), the sum of diagrams in (\ref{EYMeg1}) produces a quadratic-propagator BS Feynman diagram. Therefore, (\ref{EYMeg1}) turns into
%
\bea
C(\ell;1,\pmb\sigma_{jR},\pmb\sigma_{j+1},...,\pmb\sigma_{j-1},\pmb\sigma_{jL})\begin{minipage}{3.9cm}\includegraphics[width=3.9cm]{GenProofFeyn4}\end{minipage}\W N(\ell-k_{\pmb\sigma_{jL}};\pmb\rho_{j},\pmb\rho_{j+1},...,\pmb\rho_{j-1}),
\eea
%
characterized by a pair of {\it cyclic divisions} (i.e., the cyclic equivalence has been quotiented out) of $\pmb\sigma$ and $\pmb\rho$.
Putting all the above terms with respect to divisions of a given pair of $\pmb\sigma\in\{1,...,r\}\shuffle \text{perms}~\mathsf{H}$ and $\pmb\rho\in\text{perms}~(\{1,...,r\}\cup\mathsf{H})$ together, we just obtain the sum of diagrams corresponding to all cyclic divisions of $\pmb\sigma$ and $\pmb\rho$. When dressing the two kinematic factors, this sum becomes
%
\bea
C(\ell;1,\pmb\sigma_{jR},\pmb\sigma_{j+1},...,\pmb\sigma_{jL})\left[\Sl_{T}T(\ell-k_{\pmb\sigma_{jL}};\pmb\sigma_{j}|\pmb\rho_{j},\dots)~\W N(l-k_{\pmb\sigma_{jL}}+k_{\pmb\rho_{jL}};1,\pmb\rho_{jR},\pmb\rho_{j+1},...,\pmb\rho_{jL})\right],
\eea
%
where, after suitable shifts of the loop momenta, the terms in square brackets are precisely those of the YM integrand ${\cal I}^{\text{YM}}(\ell;1,\pmb\sigma_{jR},\pmb\sigma_{j+1},...,\pmb\sigma_{jL})$ defined by (\ref{Eq:ExpYM}). Note that the loop momentum $\ell^{\mu}$ is defined as the one to the left of $1$ in $\pmb \sigma$ in the above expression, but to the left of $1$ in $\pmb \rho$ with respect to the definition (\ref{Eq:ExpYM}). When all permutations
\bea
&&\pmb\sigma\in\Big\{\{1,2,\dots,r\}\shuffle\text{perms}~\mathsf{H},\text{cyc} (1,\dots,r)\Big\}=\Big\{\pmb{\alpha}=\{1,\{2,\dots,r\}\shuffle\text{perms}~\mathsf{H}\},\text{cyc}~\pmb{\alpha}\Big\},\nn
&&\pmb\rho\in\text{perms}\,(\{1,2,...,r\}\cup\mathsf{H})=\Big\{\pmb\beta=\{1,\text{perms}\,(\{2,...,r\}\cup\mathsf{H})\},\text{cyc}~\pmb{\beta}\Big\},
\eea
are considered, we reexpress the integrand as
%
\bea
\mathcal{I}^{\,\text{EYM}}(\ell;1...,r||\mathsf{H})&=&\Sl_{\pmb\sigma_{2,n}\in\{2,...,r\}\shuffle\text{perms}~\mathsf{H}}C(\ell;1,\pmb\sigma_{2,n})\\
%
&&~~~~\times\left[\Sl_{\pmb\rho_{2,n}\in\text{perms}(\{2,...,r\}\cup\mathsf{H})}\Sl_{T}T(\ell-k_{\pmb\sigma_{jL}};\pmb\sigma_{j}|\pmb\rho_{j},\dots)~\W N(l-k_{\pmb\sigma_{jL}}+k_{\pmb\rho_{jL}};1,\pmb\rho_{2,n})\right].\nonumber
\eea
%
The expression in the square brackets is nothing but the YM integrand ${\cal I}^{\text{YM}}(\ell;1,\pmb\sigma_{2,n})$, hence we have completed the proof of (\ref{Eq:ExpEYM}).
\section{Conclusions}\label{sec:Conclusions}
In this work, we provided a concrete proof of the statement that the gluon-loop single-trace EYM integrands at one-loop level (with quadratic propagators) satisfy expansion relations, by which an EYM integrand is expanded in terms of quadratic propagator expressed one-loop YM integrands. The expansion coefficients in this formula have the same form as those in the corresponding expansion formula for scalar-loop YMS integrands. This proof was based on the forward limit approach of the EYM integrand as well as the consistency conditions of one-loop expansion coefficients. Since the existence of consistent coefficients (for YM and YMS) has been proven in \cite{Du:2025yxz}, any expansion formula of YMS integrands with consistent coefficients implies the corresponding EYM expansion formula. Hence, the explicit formula for YMS integrands \cite{Du:2025yxz} with up to three gluons precisely results in the EYM expansion formula with up to three gravitons (the coefficients are displayed in \appref{app:Coefficients}). This work may provide more hints for understanding the BCJ duality at one-loop level.
\begin{acknowledgments}
We are grateful to Chih-Hao Fu and Yihong Wang for helpful discussions. YD is supported by NSFC under Grant No. 11875206.
\end{acknowledgments}
\appendix
\section{The Berends-Giele currents of BS theory}
The Berends-Giele current $\phi_{A|\W A}$, where $A$ and $\widetilde{A}$ are ordered sets with the same number of elements, is defined as follows (\cite{Mafra:2016ltu}):
%
\bea
\phi_{A|\widetilde{A}}={1\over s_A}\Sl_{\substack{A\to A_L,A_R\\ \widetilde{A}\to\widetilde{A}_L,\widetilde{A}_R}}\Bigl[\,\phi_{A_L|\widetilde{A}_L}\phi_{A_R|\widetilde{A}_R}-\phi_{A_R|\widetilde{A}_L}\phi_{A_L|\widetilde{A}_R}\,\Bigr],~\label{Eq:BScurrent}
\eea
%
where $s_A\equiv k_A^2$, $k_A^{\mu}$ denotes the total momentum of elements in $A$. We summed over divisions $A\to A_L,A_R$ and $\widetilde{A}\to\widetilde{A}_L,\widetilde{A}_R$ in (\ref{Eq:BScurrent}) so that in the first term $|A_L|=|\W A_L|$, $|A_R|=|\W A_R|$, or in the second term $|A_R|=|\W A_L|$, $|A_L|=|\W A_R|$. The starting point of this definition is $\phi_{a|a}=1$, $\phi_{a|b}=0$ $(a\neq b)$. An important property is $\phi_{A|\widetilde{A}}=0$ when $A\setminus\W A\neq 0$. In other words, for nonvanishing $\phi_{A|\widetilde{A}}$ the elements of the ordered set $A$ must be identical with those of $\widetilde{A}$. Another property is $\phi_{A|\widetilde{A}}= \phi_{\W A|{A}}$.
\section{Forward limit of tree-level BS amplitude and one-loop BS integrand}
The color-stripped BS amplitude with forward limit is expressed by
summing all possible Feynman diagrams where tree-level BS subcurrents are planted to the linear propagator line between $+$ and $-$:
%
\bea
&&{1\over {\ell'}^2}A^{\text{BS}}(\ell',\pmb\sigma,-\ell'|\,\ell',\pmb\rho,-\ell')={1\over {\ell'}^2}\Sl_{\substack{\pmb\sigma\to \pmb\sigma_1|\pmb\sigma_2|...|\pmb\sigma_I\\ \pmb\rho\to \pmb\rho_1|\pmb\rho_2|...|\pmb\rho_I \\ |\pmb\sigma_i|=|\pmb\rho_i|}}\begin{minipage}{3.2cm}\includegraphics[width=3.2cm]{GenProofFeyn}\end{minipage}\\
%
&=&\Sl_{\substack{\pmb\sigma\to \pmb\sigma_1|\pmb\sigma_2|...|\pmb\sigma_I\\ \pmb\rho\to \pmb\rho_1|\pmb\rho_2|...|\pmb\rho_I \\ |\pmb\sigma_i|=|\pmb\rho_i|}}{1\over {\ell'}^2}\,\phi_{\pmb\sigma_1|\pmb\rho_1}\,{1\over 2 k_{\pmb\sigma_1}\cdot \ell'+k^2_{\pmb\sigma_1}}\,\phi_{\pmb\sigma_2|\pmb\rho_2}\,\dots\,{1\over 2 (k_{\pmb\sigma_1}+...+k_{\pmb\sigma_{I-1}})\cdot \ell'+(k_{\pmb\sigma_1}+...+k_{\pmb\sigma_{I-1}})^2}\,\phi_{\pmb\sigma_I|\pmb\rho_I}.\nonumber
\eea
%
In the above expression $\ell'=\ell-k_{\pmb\sigma_1}-...-k_{\pmb\sigma_{j-1}}-k_{\pmb\sigma_{jL}}=\ell-k_{\pmb\sigma_{L}}$, where we suppose $1$ belongs to
$\pmb\sigma_j=\{\pmb\sigma_{jL},1,\pmb\sigma_{jR}\}$.
The one-loop colour-stripped integrand of BS with quadratic propagators is defined by
%
\bea
&&{\cal I}^{\text{BS}}(\pmb\sigma|\pmb\rho)=\Sl_{\substack{\text{CycDiv}\,\pmb\sigma\to \pmb\sigma_1,...,\pmb\sigma_I\\ \text{CycDiv}\,\pmb\rho\to \pmb\rho_1,...,\pmb\rho_I\\ |\pmb\sigma_i|=|\pmb\rho_i|}}\,\begin{minipage}{3.9cm}\includegraphics[width=3.9cm]{GenProofFeyn4}\end{minipage}\nn
%
&=&\Sl_{\substack{\text{CycDiv}\,\pmb\sigma\to \pmb\sigma_1,...,\pmb\sigma_I\\ \text{CycDiv}\,\pmb\rho\to \pmb\rho_1,...,\pmb\rho_I\\ |\pmb\sigma_i|=|\pmb\rho_i|}}\,{\phi_{\pmb\sigma_j|\pmb\rho_j}\over (\ell-k_{\pmb\sigma_{jL}})^2}\,{\phi_{\pmb\sigma_{j+1}|\pmb\rho_{j+1}}\over (\ell+ k_{\pmb\sigma_{jR}})^2}\,\dots\,{\phi_{\pmb\sigma_{j-1}|\pmb\rho_{j-1}}\over (\ell+ k_{\pmb\sigma_{jR}}+k_{\pmb\sigma_{j+1}}+...+k_{\pmb\sigma_{j-1}})^2}\,.
\eea
%
Note that the ${\cal I}^{\text{BS}}(\pmb\sigma|\pmb\rho)$ and ${\cal I}^{\text{BS}}(\pmb\rho|\pmb\sigma)$ correspond to the same BS amplitude, due to the symmetry of BS subcurrents.
Partial fraction identity induces the following identity between tree-level Feynman diagrams with forward limit and a one-loop Feynman diagram with quadratic propagators
%
\bea
{1\over {\ell'}^2}\begin{minipage}{3.2cm}\includegraphics[width=3.2cm]{GenProofFeyn}\end{minipage}+\text{cyc}(\phi_{\pmb\sigma_1|\pmb\rho_1},\phi_{\pmb\sigma_2|\pmb\rho_2},...,\phi_{\pmb\sigma_I|\pmb\rho_I})=\begin{minipage}{3.9cm}\includegraphics[width=3.9cm]{GenProofFeyn4}\end{minipage}\label{Eq:IDPartialFraction1}.
\eea
%
\section{A concrete construction of coefficients}\label{app:Coefficients}
In \cite{Du:2025yxz}, concrete expressions of the coefficients for YMS expansion formula (\ref{Eq:ExpYMs}) have been presented:
%
\bea
C(\ell;1,\pmb\sigma)\cong\epsilon_p\cdot X_p(\ell;1,\pmb\sigma)\,,\label{Eq:CoefficientEG1a}
\eea
%
%
\bea
C(\ell;1,\pmb\sigma)\cong\left[(\epsilon_p)_{\mu}(\epsilon_q)_{\nu}-{\epsilon_p\cdot\epsilon_q\over k_p\cdot k_q}\,(k_p)_{\mu}(k_q)_{\nu}\right]X^{\mu}_p(\ell;1,\pmb\sigma)X^{\mu}_q(\ell;1,\pmb\sigma)\,,\label{Eq:YMSExp2GravitonsCF}
%
%
\eea
%
\bea
\begin{matrix}
%
C(\ell;1,\pmb\sigma)&=&\biggl[\,(\epsilon_p)_{\mu}(\epsilon_q)_{\nu}(\epsilon_s)_{\tau}-{\epsilon_p\cdot\epsilon_q\over k_p\cdot k_q}\,(k_p)_{\mu}(k_q)_{\nu}(\epsilon_s)_{\tau}&\\
&&-{\epsilon_p\cdot\epsilon_s\over k_p\cdot k_s}\,(k_p)_{\mu}(\epsilon_q)_{\nu}(k_s)_{\tau}-{\epsilon_q\cdot\epsilon_s\over k_q\cdot k_s}\,(\epsilon_p)_{\mu}(k_q)_{\nu}(k_s)_{\tau}&\biggr]X^{\mu}_p(\ell;1,\pmb\sigma)X^{\mu}_q(\ell;1,\pmb\sigma)X^{\tau}_{s}(\ell;1,\pmb\sigma)
\end{matrix},\label{Eq:YMSExp3GravitonsCF}
%
%
\eea
%
where $p$, $q$, $s$ denotes the gluons, while $X^{\mu}_{i}(\ell;\pmb\sigma)$ $(i=p,q, \text{or}, s)$ refers to the sum of $\ell^{\mu}$ and the total momentum of external particles between $\ell^{\mu}$ and the gluon $i$.
\begin{thebibliography}{10}
\bibitem{Bern:2008qj}
Z.~Bern, J.~J.~M. Carrasco, and H.~Johansson, {\it {New Relations for
Gauge-Theory Amplitudes}}, {\em Phys. Rev. D} {\bf 78} (2008) 085011,
[\href{http://arxiv.org/abs/0805.3993}{{\tt arXiv:0805.3993}}].
\bibitem{Bern:2010ue}
Z.~Bern, J.~J.~M. Carrasco, and H.~Johansson, {\it {Perturbative Quantum
Gravity as a Double Copy of Gauge Theory}}, {\em Phys. Rev. Lett.} {\bf 105}
(2010) 061602, [\href{http://arxiv.org/abs/1004.0476}{{\tt
arXiv:1004.0476}}].
\bibitem{Fu:2017uzt}
C.-H. Fu, Y.-J. Du, R.~Huang, and B.~Feng, {\it {Expansion of
Einstein-Yang-Mills Amplitude}}, {\em JHEP} {\bf 09} (2017) 021,
[\href{http://arxiv.org/abs/1702.08158}{{\tt arXiv:1702.08158}}].
\bibitem{Chiodaroli:2017ngp}
M.~Chiodaroli, M.~Gunaydin, H.~Johansson, and R.~Roiban, {\it {Explicit
Formulae for Yang-Mills-Einstein Amplitudes from the Double Copy}}, {\em
JHEP} {\bf 07} (2017) 002, [\href{http://arxiv.org/abs/1703.00421}{{\tt
arXiv:1703.00421}}].
\bibitem{Teng:2017tbo}
F.~Teng and B.~Feng, {\it {Expanding Einstein-Yang-Mills by Yang-Mills in CHY
frame}}, {\em JHEP} {\bf 05} (2017) 075,
[\href{http://arxiv.org/abs/1703.01269}{{\tt arXiv:1703.01269}}].
\bibitem{Du:2017kpo}
Y.-J. Du and F.~Teng, {\it {BCJ numerators from reduced Pfaffian}}, {\em JHEP}
{\bf 04} (2017) 033, [\href{http://arxiv.org/abs/1703.05717}{{\tt
arXiv:1703.05717}}].
\bibitem{Du:2017gnh}
Y.-J. Du, B.~Feng, and F.~Teng, {\it {Expansion of All Multitrace Tree Level
EYM Amplitudes}}, {\em JHEP} {\bf 12} (2017) 038,
[\href{http://arxiv.org/abs/1708.04514}{{\tt arXiv:1708.04514}}].
\bibitem{ref-Kiermaier} Kiermaier, M. \textquotedblright Gravity
as the square of gauge theory, talk at Amplitudes 2010.\textquotedblright{}
Queen Mary, University of London (2010).
\bibitem{Bern:2010yg}
Z.~Bern, T.~Dennen, Y.-t. Huang, and M.~Kiermaier, {\it {Gravity as the Square
of Gauge Theory}}, {\em Phys. Rev. D} {\bf 82} (2010) 065003,
[\href{http://arxiv.org/abs/1004.0693}{{\tt arXiv:1004.0693}}].
\bibitem{Bjerrum-Bohr:2012kaa}
N.~E.~J. Bjerrum-Bohr, P.~H. Damgaard, R.~Monteiro, and D.~O'Connell, {\it
{Algebras for Amplitudes}}, {\em JHEP} {\bf 06} (2012) 061,
[\href{http://arxiv.org/abs/1203.0944}{{\tt arXiv:1203.0944}}].
\bibitem{Fu:2012uy}
C.-H. Fu, Y.-J. Du, and B.~Feng, {\it {An algebraic approach to BCJ
numerators}}, {\em JHEP} {\bf 03} (2013) 050,
[\href{http://arxiv.org/abs/1212.6168}{{\tt arXiv:1212.6168}}].
\bibitem{Fu:2014pya}
C.-H. Fu, Y.-J. Du, and B.~Feng, {\it {Note on symmetric BCJ numerator}}, {\em
JHEP} {\bf 08} (2014) 098, [\href{http://arxiv.org/abs/1403.6262}{{\tt
arXiv:1403.6262}}].
\bibitem{Mason:2013sva}
L.~Mason and D.~Skinner, {\it {Ambitwistor strings and the scattering
equations}}, {\em JHEP} {\bf 07} (2014) 048,
[\href{http://arxiv.org/abs/1311.2564}{{\tt arXiv:1311.2564}}].
\bibitem{He:2015yua}
S.~He and E.~Y. Yuan, {\it {One-loop Scattering Equations and Amplitudes from
Forward Limit}}, {\em Phys. Rev. D} {\bf 92} (2015), no.~10 105004,
[\href{http://arxiv.org/abs/1508.06027}{{\tt arXiv:1508.06027}}].
\bibitem{Cachazo:2015aol}
F.~Cachazo, S.~He, and E.~Y. Yuan, {\it {One-Loop Corrections from Higher
Dimensional Tree Amplitudes}}, {\em JHEP} {\bf 08} (2016) 008,
[\href{http://arxiv.org/abs/1512.05001}{{\tt arXiv:1512.05001}}].
\bibitem{He:2016mzd}
S.~He and O.~Schlotterer, {\it {New Relations for Gauge-Theory and Gravity
Amplitudes at Loop Level}}, {\em Phys. Rev. Lett.} {\bf 118} (2017), no.~16
161601, [\href{http://arxiv.org/abs/1612.00417}{{\tt arXiv:1612.00417}}].
\bibitem{He:2017spx}
S.~He, O.~Schlotterer, and Y.~Zhang, {\it {New BCJ representations for one-loop
amplitudes in gauge theories and gravity}}, {\em Nucl. Phys. B} {\bf 930}
(2018) 328--383, [\href{http://arxiv.org/abs/1706.00640}{{\tt
arXiv:1706.00640}}].
\bibitem{Geyer:2017ela}
Y.~Geyer and R.~Monteiro, {\it {Gluons and gravitons at one loop from
ambitwistor strings}}, {\em JHEP} {\bf 03} (2018) 068,
[\href{http://arxiv.org/abs/1711.09923}{{\tt arXiv:1711.09923}}].
\bibitem{Porkert:2022efy}
F.~Porkert and O.~Schlotterer, {\it {One-loop amplitudes in Einstein-Yang-Mills
from forward limits}}, {\em JHEP} {\bf 02} (2023) 122,
[\href{http://arxiv.org/abs/2201.12072}{{\tt arXiv:2201.12072}}].
\bibitem{Cao:2025ygu}
Q.~Cao, S.~He, Y.~Zhang, and F.~Zhu, {\it {Loop-Level Double Copy Relations
from Forward Limits}}, \href{http://arxiv.org/abs/2509.25129}{{\tt
arXiv:2509.25129}}.
\bibitem{Du:2025yxz}
Y.-J. Du, C.-H. Fu, Y.~Wang, and C.~Xie, {\it {Algebraic Consistency and
Explicit Construction of One-Loop BCJ Numerators of Yang-Mills and Related
Theories}}, \href{http://arxiv.org/abs/2511.10963}{{\tt arXiv:2511.10963}}.
\bibitem{Xie:2024pro}
C.~Xie and Y.-J. Du, {\it {Extracting quadratic propagators by refined graphic
rule}}, {\em JHEP} {\bf 02} (2025) 068,
[\href{http://arxiv.org/abs/2403.03547}{{\tt arXiv:2403.03547}}].
\bibitem{Xie:2025utp}
C.~Xie and Y.-J. Du, {\it {Local vertices, quadratic propagators and
double-copy structure of one-loop integrands from forward limits}},
\href{http://arxiv.org/abs/2509.25632}{{\tt arXiv:2509.25632}}.
\bibitem{Kleiss:1988ne}
R.~Kleiss and H.~Kuijf, {\it {Multi - Gluon Cross-sections and Five Jet
Production at Hadron Colliders}}, {\em Nucl. Phys. B} {\bf 312} (1989)
616--644.
\bibitem{DelDuca:1999rs}
V.~Del~Duca, L.~J. Dixon, and F.~Maltoni, {\it {New color decompositions for
gauge amplitudes at tree and loop level}}, {\em Nucl. Phys. B} {\bf 571}
(2000) 51--70, [\href{http://arxiv.org/abs/hep-ph/9910563}{{\tt
hep-ph/9910563}}].
\bibitem{Mafra:2016ltu}
C.~R. Mafra, {\it {Berends-Giele recursion for double-color-ordered
amplitudes}}, {\em JHEP} {\bf 07} (2016) 080,
[\href{http://arxiv.org/abs/1603.09731}{{\tt arXiv:1603.09731}}].
\end{thebibliography}
%\bibliographystyle{JHEP}
%\bibliography{reference.bib}
\end{document}