Dynamically stable two-mode squeezing in cavity optomechanics

Extracted source text · arXiv 2504.03247

%% 
%% Copyright 2007-2020 Elsevier Ltd
%% 
%% This file is part of the 'Elsarticle Bundle'.
%% ---------------------------------------------
%% 
%% It may be distributed under the conditions of the LaTeX Project Public
%% License, either version 1.2 of this license or (at your option) any
%% later version.  The latest version of this license is in
%%    http://www.latex-project.org/lppl.txt
%% and version 1.2 or later is part of all distributions of LaTeX
%% version 1999/12/01 or later.
%% 
%% The list of all files belonging to the 'Elsarticle Bundle' is
%% given in the file `manifest.txt'.
%% 
%% Template article for Elsevier's document class `elsarticle'
%% with harvard style bibliographic references

%\documentclass[preprint,12pt,authoryear]{elsarticle}

%% Use the option review to obtain double line spacing
%% \documentclass[authoryear,preprint,review,12pt]{elsarticle}

%% Use the options 1p,twocolumn; 3p; 3p,twocolumn; 5p; or 5p,twocolumn
%% for a journal layout:
%% \documentclass[final,1p,times,authoryear]{elsarticle}
%% \documentclass[final,1p,times,twocolumn,authoryear]{elsarticle}
%% \documentclass[final,3p,times,authoryear]{elsarticle}
%% \documentclass[final,3p,times,twocolumn,authoryear]{elsarticle}
%% \documentclass[final,5p,times,authoryear]{elsarticle}
 \documentclass[final,5p,times,twocolumn,compress]{elsarticle}

%% For including figures, graphicx.sty has been loaded in
%% elsarticle.cls. If you prefer to use the old commands
%% please give \usepackage{epsfig}

%% The amssymb package provides various useful mathematical symbols
\usepackage{amssymb}
\usepackage{amsmath}
\usepackage{lipsum}
\usepackage{color}
%% The amsthm package provides extended theorem environments
%% \usepackage{amsthm}

%% The lineno packages adds line numbers. Start line numbering with
%% \begin{linenumbers}, end it with \end{linenumbers}. Or switch it on
%% for the whole article with \linenumbers.
%% \usepackage{lineno}

%% You might want to define your own abbreviated commands for common used terms, e.g.:
%%\newcommand{\kms}{km\,s$^{-1}$}
%%\newcommand{\msun}{$M_\odot}

\journal{Physics Letters A}

\begin{document}

\begin{frontmatter}

\title{Dynamically stable two-mode squeezing in cavity optomechanics}
\author[first]{Chen Wang}
\author[first]{Shi-fan Qi\corref{cor1}}
\cortext[cor1]{Corresponding Author}
\affiliation[first]{organization={College of Physics and Hebei Key Laboratory of Photophysics Research and Application, Hebei Normal University},%Department and Organization
         city={Shijiazhuang},
         postcode={050024}, 
         country={China}}
\ead{qishifan@hebtu.edu.cn}
            
\begin{abstract}
In this work, we propose a two-mode squeezing generation scheme in a hybrid three-mode cavity optomechanical system, where a mechanical resonator couples to two microwave (or optical) photon modes. By applying modulated strong drives, we derive an effective Hamiltonian that describes mechanically mediated two-photon squeezing, which is validated by diagonalizing the system’s transition matrix in the Heisenberg picture. Our analysis reveals that stable two-mode squeezing can be achieved by optimizing the squeezing operator even in unsteady-state dynamics, with the squeezing level exceeding the maximum achievable under system stability conditions while maintaining the anti-squeezing at a proper level within a suitable time interval. Furthermore, we show that our protocol is robust against systematic errors in both driving intensity and frequency, as well as against thermal Markovian noises. Our work presents an extendable approach for generating two-mode squeezed states between indirectly coupled bosonic modes.
\end{abstract}

\begin{keyword}
 Cavity optomechanical system\sep Two-mode squeezed state\sep The effective Hamiltonian\sep Anti-squeezing effect\sep Systematic errors.  
\end{keyword}

\end{frontmatter}

\section{Introduction}
\label{introduction}
Quantum entanglement~\cite{quanentangle} plays a key role in quantum technologies, including quantum computing~\cite{quantumcomputing}, quantum communication~\cite{quantumcommunication}, and quantum sensing~\cite{quantumsense}. Many quantum platforms~\cite{cavityopto,cqed,Schrodingerstate,superconduct,magnoncavity,magnonqubit,rydberg} as well as protocols~\cite{mppentang,ghzstate,noon,atom,High,Cvoptical,Delocalization,entangledetection,ppnvcenter,Enhancementoem,WODEDO2025108364} to prepare and measure entangled states have therefore been intensively pursued for a long time and are still under active investigation. Among diverse entangled states, two-mode squeezed states (TMSS) are crucial in quantum computation~\cite{quancomcon}, information~\cite{quaninfcon,Gauinf}, teleportation~\cite{quantelep}, and metrology~\cite{quantmetro}. Various protocols have been proposed to generate TMSS with high squeezing level (SL)~\cite{nonlocalryd,kerrmagnon}, including mixing two single-mode squeezed states at a beam splitter~\cite{convariable} and employing a spontaneous parametric down-conversion process~\cite{twocolor,twinlaser,stablethreshold}. For the optical field, a non-degenerate optical parametric amplifier is often used to generate TMSS~\cite{quanphasenpo,realEPR,observasquee}. Recently, TMSS has been well established experimentally in various platforms, such as thermal gases~\cite{macroscopic}, ultra-cold atoms~\cite{homodyne,strongobserspin,probespin,spinBES,multisqueeatomic}, atomic mechanical oscillators~\cite{Atomicoscill}, spin ensembles~\cite{phasemeasure,manipulatinggrowth,powerlawspin}, antiferromagnetic magnons~\cite{antiferromagnetic}, and superconducting circuits~\cite{travelingwave,surfacephonon}. 

The cavity optomechanical system~\cite{cavityopto} provides an alternative and promising avenue to create optical~\cite{reservoiroptome,robustopto,doubleoptosys} and mechanical~\cite{twomechansquee,thermalsquee,Optoacoustic,amptmsnanomechanical} TMSS, due to its high controllability and flexibility. Two primary strategies are commonly employed to generate TMSS in this system. One approach involves constructing a two-mode squeezing interaction between the target modes~\cite{entanglemacro,probeentangle}, while the other uses reservoir engineering to tailor the target modes dissipatively into TMSS~\cite{reservoiroptome,robustopto,Entangling,nonreciprocal,generobust,Xie2023,synergizing}. The two-mode squeezing coupling naturally leads to entanglement without reservoir engineering. However, under the constraint of system stability conditions, the SL cannot go beyond $3$ dB below the vacuum limit~\cite{stablethreshold}. In contrast, the reservoir-engineering scheme ensures the system's stability and theoretically allows the SL to exceed the threshold. Nevertheless, due to the generally lower decay rate of phonons compared to photons~\cite{cavityopto,cohermechanical,entmechanical,squeemech,Superposition,coherentcoupling,active,map}, we find that the SL remains suboptimal and it is challenging to surpass the $3$ dB upper bound, even at absolute zero temperature. Consequently, beyond steady-state restrictions, some recent studies have turned to the dynamic generation of entanglement~\cite{Chen:17,entdynhightem} and the realization of other applications~\cite{Xie:20} in optomechanical systems. Moreover, squeezed states are capable of surpassing the standard quantum limit in precision sensing~\cite{SCHNABEL20171,Frequencyest}. In time-limited metrology, unsteady dynamical regimes may facilitate a faster generation of squeezing compared with steady-state schemes.

In this work, we adopt a similar dynamical perspective to generate TMSS in a three-mode optomechanical system, consisting of two target photon modes to be entangled, each coupled to an auxiliary phonon mode. The squeezing generation process is governed by an effective Hamiltonian for two-photon squeezing (TPS) coupling, which is validated by diagonalization of the system's transition matrix in the Heisenberg picture. By analyzing the system's dynamics through the effective TPS Hamiltonian in Markovian environments, we find that an asymptotically stable TMSS, characterized by the squeezing operator variance that gradually converges to a constant over time, can be obtained in unsteady evolutions, where the system's covariance matrix (CM) elements diverge exponentially, displaying an enhanced SL exceeding the steady limit. Environmental noises modify the optimized operator of TPS while simultaneously asymptotically stabilizing TMSS, even if the TPS coupling exceeds the system stability threshold~\cite{stablethreshold}. Furthermore, numerical simulations show that our protocol can maintain the anti-two-mode-squeezing at an appropriate level within a suitable time interval and is robust against systematic errors in both driving intensity and frequency, as well as against thermal Markovian noises.

The rest of this work is organized as follows. In Sec.~\ref{Secmodel}, we present a three-mode optomechanical system and derive an effective TPS Hamiltonian assisted by the phonon. In Sec.~\ref{Sectms}, we phenomenologically analyze the generation process of the TMSS within the open-quantum-system framework. We find that the stable TPS can be obtained even beyond the system stability conditions. In Sec.~\ref{secanti}, we discuss the anti-two-mode-squeezing effect beyond the system-stability regime. Systematic errors arising from the driving-enhanced optomechanical coupling strengths and driving frequencies, as well as the effects of thermal Markovian noises, are analyzed in Secs.~\ref{secsyserror} and~\ref{secthermal}, respectively. Finally, we discuss the experimental feasibility and summarize the work in Sec.~\ref{Secexpercon}.

\section{Model and the effective Hamiltonian}\label{Secmodel}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.45\textwidth]{diagram.eps}
\caption{Schematic diagram of the hybrid three-mode cavity optomechanical system. (a) A mechanical interface acts as an intermediate mode $m$, coupling with the optical cavity $a$ and $b$. (b) A mechanical resonator $m$ is capacitively coupled to two superconducting microwave resonators, $a$ and $b$. The photon modes $a$ and $b$ are driven by strong fields $\Omega_a$ and $\Omega_b$, respectively. (c) The frequencies and linewidths of the system are adopted to generate TMSS.}\label{diagram}
\end{figure}
Consider a hybrid three-mode optomechanical system as shown in Fig.~\ref{diagram}, which is composed of a mechanical oscillator and two optical cavity modes [see Fig.~\ref{diagram} (a)], or a mechanical oscillator and two superconducting microwave resonators [see Fig.~\ref{diagram} (b)]. The full system Hamiltonian ($\hbar\equiv1$) can be described as~\cite{cavityopto,reservoiroptome}
\begin{equation}\label{Hamori}
	H=\omega_mm^\dag m+\sum_{o=a,b}\omega_oo^\dag o+g_oo^\dag o(m+m^\dag)+\Omega_o (oe^{i\epsilon_ot}+o^\dag e^{-i\epsilon_ot}),
\end{equation}
where $a(a^\dag)$, $b(b^\dag)$, and $m(m^\dag)$ are the annihilation (creation) operators of two photon modes and phonon mode, with transition frequencies $\omega_a$, $\omega_b$, and $\omega_m$, respectively. $g_a$ ($g_b)$ is the single-excitation photon-phonon coupling strength between photon mode $a$ ($b$) and phonon mode $m$, which can be compensated by a strong drive. The last term describes the external driving Hamiltonian, where $\Omega_o$ is the Rabi frequency and $\epsilon_o$ is the driving frequency of mode $o$, $o=a,b$.

Under appropriate strong driving conditions and following the standard linearization approach~\cite{cavityopto}, the full system Hamiltonian turns out to be
\begin{equation}\label{Hamline}
\begin{aligned}
	H_{\rm lin}&=H_0+V,\quad H_0=\Delta_aa^\dag a+\Delta_bb^\dag b+\omega_mm^\dag m,\\
	V&=g(e^{-i\theta_a}a+e^{i\theta_a}a^\dag)(m+m^\dag)+G(e^{-i\theta_b}b+e^{i\theta_b}b^\dag)(m+m^\dag),
\end{aligned}
\end{equation}
where $\Delta_a=\omega_a-\epsilon_a$ and $\Delta_b=\omega_b-\epsilon_b$ are the detunings of mode $a$ and $b$, respectively. $g$ and $G$ are the driving-enhanced optomechanical coupling strengths, $\theta_a$ and $\theta_b$ are the corresponding phases. The details can be found in~\ref{appaline}.

In previous works using reservoir engineering method~\cite{reservoiroptome}, the parameters were set as $\Delta_a=\omega_m$, $\Delta_b=-\omega_m$, and $g<G$, to obtain the two-mode squeezing (quantum entanglement) between modes $a$ and $b$. These conditions ensure the system's stability and theoretically enable the SL to exceed the $3$ dB. However, due to the generally lower decay rate of phonon compared to the photon modes~\cite{cohermechanical,entmechanical,squeemech,Superposition,coherentcoupling,active,map}, even at zero temperature, the SL remains suboptimal and is difficult to surpass the upper bound of $3$ dB. More details and analysis can be seen in~\ref{appbtmsssteady}. To overcome this limitation, we focus on generating the TMSS by constructing an effective TPS Hamiltonian within the framework of system instability.

At the large detuning regime, i.e., $|\Delta_a-\omega_m|, |\Delta_b-\omega_m|\gg g, G$, and under the near-resonant condition $\Delta_a=-\Delta_b+\delta$, an effective Hamiltonian describing the TPS can be extracted by a perturbation theory~\cite{james}. The effective Hamiltonian is found to be
\begin{equation}\label{Heff}
	H_{\rm eff}=g_{\rm
		eff}(e^{-i\theta}ab+e^{i\theta}a^\dag b^\dag),
\end{equation}
where $\theta=\theta_a+\theta_b$. The effective coupling strength and the energy shift are
\begin{equation}\label{geffdelta}
	g_{\rm eff}=\frac{2\omega_mgG}{\Delta^2_b-\omega^2_m},\quad \delta=\frac{2\omega_m(g^2+G^2)}{\omega^2_m-\Delta^2_b},
\end{equation}
respectively. The derivation details are presented in~\ref{appcHeff}. The effective Hamiltonian in Eq.~(\ref{Heff}) can naturally generate the TMSS without reservoir engineering~\cite{stablethreshold}.

The effective TPS Hamiltonian does not conserve excitation, making it challenging to validate the effective Hamiltonian using previous methods~\cite{intermagnon,oneexcite,nonlinear}, i.e., a standard numerical diagonalization of the system Hamiltonian in a truncated Hilbert space. To address this difficulty, we introduce an interesting approach involving diagonalization of the transition matrix of the whole system, which enables the evaluation of the effective two-mode squeezing Hamiltonian induced by the virtual process. The details can be seen in~\ref{appcHeff}. In~\ref{appcHeff}, we benchmark the validity ranges of the coupling strengths $g$ and $G$, as well as the detunings $\Delta_b$, by comparing the analytical predictions for $g_{\rm eff}$ and $\delta$ derived from the effective Hamiltonian with numerical simulations based on the full system's transition matrix. The results identify an approximate parameter regime in which the effective Hamiltonian remains both valid and well-performing, namely, $0.1\omega_m\le g\le0.2\omega_m, 8g\le(\Delta_b-\omega_m)\le10g$. Within this region, the effective coupling strength satisfies $g_{\rm eff}\gtrsim0.01\omega_m$, while both the absolute and relative errors remain small.

\section{Two-photon squeezing}\label{Sectms}
Using the effective Hamiltonian in Eq.~(\ref{Heff}), one can naturally generate the TMSS between two photon modes $a$ and $b$. Within the open-quantum-system framework, this section analyzes the system's dynamics and elucidates the generation mechanism of stable TMSS under system instability conditions. Under the standard Markovian assumptions, the dynamics of the quantum system are governed by the quantum Langevin equation, written in a matrix form
\begin{equation}\label{qleeff}
	\dot{u}^{\rm eff}(t)=A_{\rm eff}u^{\rm eff}(t)+n^{\rm eff}(t),
\end{equation}
where $u^{\rm eff}(t)=[X_a(t), Y_a(t), X_b(t), Y_b(t)]^T$ with $X_o=(e^{-i\theta_o}o+e^{i\theta_o}o^\dag)/\sqrt{2}$, $Y_o=(e^{-i\theta_o}o-e^{i\theta_o}o^\dag)/i\sqrt{2}$, $o=a,b$. The transition matrix $A_{\rm eff}$ is
\begin{equation}\label{aeff}
A_{\rm eff}=-\begin{bmatrix}\kappa_a & 0 &0 &g_{\rm eff}\\
0 &\kappa_a &g_{\rm eff} &0\\
0 & g_{\rm eff} &\kappa_b &0\\
g_{\rm eff} & 0 & 0 &\kappa_b\end{bmatrix},
\end{equation} 
where $n^{\rm eff}(t)=[\sqrt{2\kappa_a}X_a^{in}(t),\sqrt{2\kappa_a}Y_a^{in}(t),\sqrt{2\kappa_b}X_b^{in}(t),\sqrt{2\kappa_b}Y_b^{in}(t)]^T$ is the vector of Gaussian noise operators, and $X^{in}_o=(e^{-i\theta_o}o_{in}+e^{i\theta_o}o_{in}^\dag)/\sqrt{2}$,$Y^{in}_o=(e^{-i\theta_o}o_{in}-e^{i\theta_o}o_{in}^\dag)/i\sqrt{2}$. $\kappa_a$ and $\kappa_b$ are the decay rates of the modes $a$ and $b$, respectively. $o_{in}$ is the input noise operators for the mode $o$, which is characterized by the covariance functions: $\langle o^\dag_{in}(t)o_{in}(t')\rangle=N_o\delta(t-t')$ and $\langle o_{in}(t)o^\dag_{in}(t')\rangle=[N_o+1]\delta(t-t')$, under the Markovian approximation. $N_o$ is the thermal occupation number of mode $o$ at equilibrium state and $N_o=[\exp(\hbar\omega_o/k_BT)-1]^{-1}$.

At the initial time, we assume the microwave mode $a$ and optical mode $b$ are both in thermal equilibrium states. Then, the system state evolves as a Gaussian state, due to the above-linearized dynamics in Eq.~(\ref{qleeff}), which can be completely characterized by a $4\times 4$ CM $V^{\rm eff}(t)$. The dynamics of the CM $V^{\rm eff}(t)$ satisfies
\begin{equation}\label{cmeff}
	\dot{V}^{\rm eff}(t)=A_{\rm eff}V^{\rm eff}(t)+V^{\rm eff}(t)A^T_{\rm eff}+D^{\rm eff}.
\end{equation}
The elements of $V^{\rm eff}(t)$ are defined as
\begin{equation}\label{Veffdefin}
	V^{\rm eff}_{ij}(t)=\frac{\langle u^{\rm eff}_i(t)u^{\rm eff}_j(t)+u^{\rm eff}_j(t)u^{\rm eff}_i(t)\rangle}{2}-\langle u^{\rm eff}_i(t)\rangle\langle u^{\rm eff}_j(t)\rangle,
\end{equation}
where $u^{\rm eff}_i(t)$ is the $i$ term of $u^{\rm eff}(t)$. $D^{\rm eff}$ is the diffusion matrix, which is diagonal with elements given by $D^{\rm eff}_{11}=D^{\rm eff}_{22}=\kappa_a(2N_a+1)$ and $D^{\rm eff}_{33}=D^{\rm eff}_{44}=\kappa_b(2N_b+1)$.
This matrix is defined through $D^{\rm eff}_{ij}(t)=\langle n^{\rm eff}_i(t)n^{\rm eff}_j(t)+n^{\rm eff}_j(t)n^{\rm eff}_i(t)\rangle/2$. The CM elements can be analytically calculated, with explicit expressions provided in~\ref{appcm}, Eq.~(\ref{Veffelement}). In the system stability condition, the CM is invariant, i.e., $\dot{V}^{\rm eff}=0$ in Eq.~(\ref{cmeff}), which requires $g^2_{\rm eff}<\kappa_a\kappa_b$. In regimes of system dynamical instability, the CM diverges.

Beyond the system stability region, $g^2_{\rm eff}>\kappa_a\kappa_b$, we demonstrate that stable, rather than merely transient, two-mode squeezing can emerge, by optimizing the two-mode squeezing quadrature operator. The quadrature operator is given by
\begin{equation}\label{quadratureX}
	X=\cos\tilde{\phi}X_a+\sin\tilde{\phi}Y_b,
\end{equation}
and the associated variance becomes
\begin{equation}\label{Deltaxphit}
	\Delta X(t)=2C_-e^{-(\Omega+\kappa_a+\kappa_b)t}+\frac{\kappa_++\cos(2\tilde{\phi})\kappa_-}{2(\Omega+\kappa_a+\kappa_b)},
\end{equation}
where
\begin{equation}\label{Omegphi}
\begin{aligned}
&\tan(2\tilde{\phi})=\frac{2g_{\rm eff}}{\kappa_a-\kappa_b},\quad \Omega=\sqrt{4g^2_{\rm eff}+(\kappa_a-\kappa_b)^2},\\
&\kappa_\pm=\kappa_a(2N_a+1)\pm\kappa_b(2N_b+1),\quad N_\pm=N_a\pm N_b,\\
&C_-=\left[\frac{N_++1+\cos(2\tilde{\phi})N_-}{4}-\frac{\kappa_++\cos(2\tilde{\phi})\kappa_-}{4(\Omega+\kappa_a+\kappa_b)}\right].
\end{aligned}
\end{equation}
More details can be seen in~\ref{appcm}.

The condition $\Delta X(t)<0.5$ indicates the emergence of two-mode squeezing, with smaller values of $\Delta X(t)$ corresponding to higher squeezing. Equation~(\ref{Deltaxphit}) shows an asymptotically stable TPS over a long time evolution, i.e.,
\begin{equation}\label{Deltaxphiinfty}
	\Delta X(\infty)=\frac{\Omega\kappa_++(\kappa_a-\kappa_b)\kappa_-}{2\Omega(\Omega+\kappa_a+\kappa_b)}.
\end{equation}
Based on the definition of $\Omega$ in Eq.~(\ref{Omegphi}), $\Delta X(\infty)$ decreases monotonically with increasing $g_{\rm eff}$, indicating a corresponding enhancement of the TPS. Additionally, we define a sufficiently long period $\tau$ to approximate infinite time, which is given by $\tau=2\pi/(\Omega+\kappa_a+\kappa_b)$. It can be demonstrated that the difference $\Delta X(\tau)-\Delta X(\infty)=2C_-/e^{2\pi}$ is sufficiently small. We also simulate the SL to quantify the TPS, which in the decibel unit is defined by 
\begin{equation}\label{squzelevel}
	S=-10{\rm log}_{10}\left(\Delta X/\Delta X_{zp}\right),
\end{equation}
where $\Delta X_{zp}=0.5$ is the fluctuation in the zero-point level.

Moreover, we numerically calculate the whole system's dynamic using the full Hamiltonian $H_{\rm lin}$ in Eq.~(\ref{Hamline}) to confirm the above analytical results. Similar to Eq.~(\ref{cmeff}), the dynamic of the whole system CM $V(t)$ satisfies
\begin{equation}\label{wholedotV}
	\dot{V}(t)=AV(t)+V(t)A^T+D.
\end{equation}
The elements of $V(t)$ are given by
\begin{equation}\label{Vt}
	V_{ij}=\frac{\langle u_i(t)u_j(t)+u_j(t)u_i(t)\rangle}{2}-\langle u_i(t)\rangle\langle u_j(t)\rangle,
\end{equation}
where $u_i(t)$ is the $i$ term of $u(t)$ and $i=1,2,\cdots,6$. $u^T(t)=[u^{\rm eff}(t),X_m(t),Y_m(t)]$ is the vector of quadrature operators, where $u^{\rm eff}$ is defined in Eq.~(\ref{qleeff}), $X_m=(m+m^\dag)/\sqrt{2}$, and $Y_m=(m-m^\dag)/i\sqrt{2}$. $A=i\mathcal{L}-\tilde{A}$, where $\mathcal{L}$ is a matrix defined in Eq.~(\ref{matrix}) in~\ref{appcHeff}, and $\tilde{A}=Diag[\kappa_a,\kappa_a,\kappa_b,\kappa_b,\kappa_m,\kappa_m]$. $D$ is the matrix of noise covariance, where the diagonal elements are 
$D_{11}=D_{22}=\kappa_a(2N_a+1)$, $D_{33}=D_{44}=\kappa_b(2N_b+1)$, and $D_{55}=D_{66}=\kappa_m(2N_m+1)$, all non-diagonal elements are zero. Then, the dynamic of $\Delta X(t)$ can be numerically obtained by calculating the CM $V(t)$,
\begin{equation}\label{deltaxfull}
	\Delta X(t)=\cos^2\tilde{\phi} V_{11}(t)+\sin^2\tilde{\phi} V_{44}(t)+\sin(2\tilde{\phi})V_{14}(t).
\end{equation}
Here, the mechanical mode is initially assumed to be in the vacuum state, and its corresponding environment noises are zero-mean Gaussian noises.

\begin{figure}[t]
	\centering
	\includegraphics[width=0.235\textwidth]{Vt.eps}
	\includegraphics[width=0.235\textwidth]{DeltaXt.eps}
	\caption{(a) Dynamics of the CM elements using the effective Hamiltonian~\eqref{Heff} or the full system Hamiltonian~\eqref{Hamline}. (b) Dynamics of the $\Delta X(t)$ and $\Delta\tilde{X}(t)$ with the effective Hamiltonian~\eqref{Heff} or the full system Hamiltonian~\eqref{Hamline}. Here, the parameters are set as $g=G=0.1\omega_m,\Delta_b=\omega_m+10g,\kappa_a=\kappa_b=10^{-3}\omega_m,\kappa_m=10^{-6}\omega_m$, and the thermal occupation numbers $N_a=N_b=0,N_m=10$. Moreover, all modes are initially assumed to be in vacuum states, i.e., $V(0)=I_6/2$, where $I_6$ is the $6\times 6$ identity matrix.}
	\label{vtsqueeze}
\end{figure}

In Fig.~\ref{vtsqueeze} (a), it can be observed that both of the CM elements $V_{11}$ (dark dotted line with circles) and $V_{14}$ (purple dotted line with diamonds) exhibit excellent agreement with the corresponding analytical results via the effective Hamiltonian within the time regime $\omega_mt\le 600$. In Fig.~\ref{vtsqueeze} (b), we plot $\Delta X(t)$ using the effective Hamiltonian in Eq.~(\ref{Heff}) by the blue solid line. It tends to stabilize a certain value of $0.066$ after prolonged evolution ($\omega_m t\ge250\approx 0.6\tau$), and the corresponding SL is approximately $8.79$ dB, which is greater than the upper bound $3$ dB in the system stability condition. However, for longer time evolution, it is found that the result about $\Delta X(t)$ using the effective Hamiltonian~(\ref{Heff}) differs obviously from that using the full Hamiltonian~(\ref{Hamline}) (red dashed line), although the CM elements remain in close agreement. In the system instability regime, the CM elements $V_{11}$ and $V_{14}$ exhibit exponential dependence on time, as shown in Fig.~\ref{vtsqueeze} (a). Over a long time, the absolute values of these elements rapidly increase. In this condition, even a slight deviation in the CM elements can significantly affect the variance $\Delta X(t)$, decreasing the SL instead of the expected enhancement. However, it is observed that the squeeze remains nearly stable within the approximate time interval $t\in[0.75\tau,1.25\tau]$.

Since the effective Hamiltonian fails to fully capture the whole system's evolution over a long time, the optimal squeezing operator in practice is not the theoretical prediction $X$ shown in Eq.~(\ref{quadratureX}). The genuine optimal operator $\tilde{X}$ is obtained by numerically minimizing the $\Delta\tilde{X}$ of the general quadrature operator
$(X_a\cos\phi_1+Y_a\sin\phi_1)\cos\phi_3+(X_b\cos\phi_2+Y_b\sin\phi_2)\sin\phi_3$, with respect to $\phi_1$, $\phi_2$, and $\phi_3$. It is demonstrated that the variance $\Delta\tilde{X}$ of the genuine optimal operator $\tilde{X}$ corresponds to the minimal eigenvalue of sub-CM $V_4=V(1:4,1:4)$, where $V$ is the full CM defined in Eq.~(\ref{wholedotV}). The variance $\Delta\tilde{X}(t)$, numerically calculated by the full system Hamiltonian in Eq.~(\ref{Hamline}), is presented as a black dash-dotted line in Fig.~\ref{vtsqueeze} (b). These results do match well with the theoretical predictions (blue solid line) and maintain consistency over time. The discrepancy between the results for $\Delta X$ and $\Delta\tilde{X}$ becomes increasingly pronounced over time. This behavior arises because the relevant CM elements are exponentially amplified in the unsteady-state regime, such that even a slight misalignment of the quadrature operator can lead to a substantial deviation in the observed SL. Consequently, there exists a favorable time window in which the squeezing remains appreciably strong while the requirement for angular precision is comparatively relaxed. We further discuss this favorable time window and the associated angular precision in Sec.~\ref{secanti}.

\begin{figure}[t]
	\centering
	\includegraphics[width=0.235\textwidth]{squeeze.eps}
	\includegraphics[width=0.235\textwidth]{squeeze2.eps}
	\includegraphics[width=0.235\textwidth]{SL05tau.eps}
	\includegraphics[width=0.235\textwidth]{Sltau.eps}
	\includegraphics[width=0.235\textwidth]{SL2tau.eps}
	\includegraphics[width=0.235\textwidth]{epsilonSL.eps}
	\caption{(a) Time evolution of the SL $S$ for operator $X$~\eqref{quadratureX} under the full system Hamiltonian~\eqref{Hamline} at varying coupling strengths. (b) Time evolution of the SL $\tilde{S}$ for numerically optimized quadrature operator $\tilde{X}$ with the full system Hamiltonian~\eqref{Hamline} at different coupling strengths. [(c), (d), and (e)] Comparison of the analytical SL $S_{\rm eff}$ with the numerically calculated results $S_{\rm lin}$ and $\tilde{S}_{\rm lin}$ under different normalized coupling strengths $g/\omega_m$. (f) The relative errors of SL, $\epsilon$ and $\tilde{\epsilon}$, as functions of $g/\omega_m$. All other parameters and initial conditions are the same as those in Fig.~\ref{vtsqueeze}.}
	\label{squeeze}
\end{figure}

In Figs.~\ref{squeeze} (a) and (b), we plot the dynamics of the SL $S$ [as defined in Eq.~(\ref{squzelevel})] and $\tilde{S}$ using the full system Hamiltonian~(\ref{Hamline}) under various coupling strengths, respectively. Here, the quantity $\tilde{S}$ represents the SL of the genuine optimal quadrature $\tilde{X}$, which is defined as $\tilde{S}=-10{\rm log}_{10}\left(\Delta \tilde{X}/\Delta X_{zp}\right)$. The numerical annotations in the figure represent the maximum SL attained during the evolution period. From Figs.~\ref{squeeze} (a) and (b), one can conclude that the SL is enhanced by increasing the coupling strength, which is consistent with the analytical results. Furthermore, as shown in Fig.~\ref{squeeze} (a), the duration of stable squeezing for $S$ decreases as the coupling strength increases. For $g=0.05\omega_m$ and $g=0.1\omega_m$, the SL has no significant degradation during the time interval ($\omega_mt\le 500$). However, when $g=0.2\omega_m$ and $g=0.25\omega_m$, a notable decline in the SL is observed at $\omega_mt\approx350$ and $\omega_mt\approx300$, respectively. In contrast, the SL $\tilde{S}$ of the numerically optimized quadrature $\tilde{X}$ maintains asymptotic stable upon attaining a plateau, independent of the coupling strength.

In Figs.~\ref{squeeze} (c), (d), and (e), the SL $S_{\rm eff}$ for quadrature operator $X$~(\ref{quadratureX}) with the effective Hamiltonian~(\ref{Heff}), the SL $S_{\rm lin}$ for $X$~(\ref{quadratureX}) with the full system Hamiltonian~(\ref{Hamline}), and the SL $\tilde{S}_{\rm lin}$ for the numerically optimized operator $\tilde{X}$ with full system Hamiltonian~(\ref{Hamline}), are shown with blue solid line, red dashed line with squares, and black dotted line with circles, respectively. Panels (c), (d), and (e) display the SL at specific moments $\tau/2$, $\tau$, and $3\tau/2$, respectively. Comparison of the results in Figs.~\ref{squeeze} (c), (d), and (e) reveals that, regardless of the coupling strength, the SL $S_{\rm eff}$ and $\tilde{S}_{\rm lin}$ at time $\tau$ exhibit only a slight difference compared to those at $3\tau/2$, while being $1-2$ units larger than the corresponding values at $\tau/2$. The two-mode squeezing has become stable at time $\tau$. In contrast, the SL $S_{\rm lin}$ at $3\tau/2$ is consistently smaller than those at $\tau$, with the magnitude of this difference growing as the coupling strength $g$ increases. This implies that, when selecting $X$ as the quadrature operator, while the SL enhances with stronger coupling, the duration of high SL decreases proportionally. These observations are consistent with the theoretical predictions and the numerical results shown in Figs.~\ref{squeeze} (a) and (b).

We introduce the relative errors of SL $S_{\rm lin}$ and $\tilde{S}_{\rm lin}$ to more clearly demonstrate the validity of our protocol. The relative errors are defined as
\begin{equation}\label{epsilon}
	\epsilon(t)=\left|\frac{S_{\rm lin}(t)-S_{\rm eff}(t)}{S_{\rm eff}(t)}\right|, \tilde{\epsilon}(t)=\left|\frac{\tilde{S}_{\rm lin}(t)-S_{\rm eff}(t)}{S_{\rm eff}(t)}\right|,
\end{equation}
respectively, which numerical results at characteristic moment $\tau$ are presented in Fig.~\ref{squeeze} (f). As the coupling strength $g$ increases, the relative error $\epsilon(\tau)$ associated with SL $S_{\rm lin}(\tau)$ roughly tends to increase, whereas the relative error $\tilde{\epsilon}(\tau)$ corresponding to the optimal operator SL $\tilde{S}_{\rm lin}(\tau)$ shows a gradual decrease. Furthermore, the $\tilde{\epsilon}(\tau)$ remains significantly smaller than $\epsilon(\tau)$ for all coupling strengths. Specifically, for $g\ge0.15\omega_m$, $\tilde{\epsilon}(\tau)\le 0.02$ while $\epsilon(\tau)\ge0.06$. The relative error $\tilde{\epsilon}$ closely follows the trend of the relative error of effective coupling, $\sigma$, shown in Fig.~\ref{effgdelta} (f) in~\ref{appcHeff}, while the $\epsilon$ shows a distinctly opposite trend. It indicates that, although the effective Hamiltonian can approximately capture the whole system dynamics, the squeezing operator $X$ in Eq.~(\ref{quadratureX}) derived from this framework deviates slightly from the genuine optimal operator.

\section{Performance analysis}
\subsection{Anti-two-mode-squeezing effect}\label{secanti}
\begin{figure}[ht]
	\centering
	\includegraphics[width=0.235\textwidth]{antitime.eps}
	\includegraphics[width=0.235\textwidth]{antisqueeze.eps}
	\caption{[(a), (b)] Time evolution of the variances $\Delta X(t)$ and $\Delta Y(t)$ using the system Hamiltonian~(\ref{Hamline}) under various coupling strengths. [(c), (d)] Squeezing level $S_{\rm lin}$ and anti-squeezing level $S'_{\rm lin}$ computed from the full system Hamiltonian~(\ref{Hamline}) at selected moments under various coupling strengths. All other parameters and initial conditions are identical to those in Fig.~\ref{vtsqueeze}.}
	\label{antisqueeze}
\end{figure}
Operating within the system instability regime results in a significant amplification of noise over time, causing the anti-squeezing effect to become increasingly pronounced during the time evolution. In this section, we analyze the anti-two-mode-squeezing effect inherent in our scheme. The anti-two-mode-squeezing operator orthogonal to the two-mode squeezing operator $X$~(\ref{quadratureX}) can be explicitly formulated as
\begin{equation}\label{quadratureY}
	Y=\cos\tilde{\phi}Y_a-\sin\tilde{\phi}X_b,
\end{equation}
which is utilized to assess the anti-two-mode-squeezing effect in our scheme. The corresponding variance of quadrature operator $Y$~(\ref{quadratureY}) can be derived as
\begin{equation}\label{Deltayphit}
	\Delta Y(t)=2C_+e^{(\Omega-\kappa_a-\kappa_b)t}+\frac{N_++1+\cos(2\tilde{\phi})N_-}{2}-2C_+,
\end{equation}
where the parameters $C_+$, $N_+$, and $N_-$ are defined in Eq.~(\ref{parameter}). When the system is unsteady, the exponential term satisfies $\Omega-\kappa_a-\kappa_b>0$, leading to an exponential divergence of the variance $\Delta Y$. Similarly, the anti-squeezing level $S'$ can be defined as $S'=10{\rm log}_{10}(\Delta Y/\Delta Y_{zp})$, where $\Delta Y_{zp}=0.5$ is the standard fluctuation in the zero-point level.

In Fig.~\ref{antisqueeze} (a) and (b), we plot the time evolution of the variances $\Delta X(t)$ and $\Delta Y(t)$, respectively. Obviously, $\Delta Y$ exhibits exponential divergence as $\Delta X$ approaches a steady value. Furthermore, as $g$ increases from $0.05\omega_m$ to $0.1\omega_m$, the variance $\Delta Y$ enhances significantly. More numerical results about the SL and anti-squeezing level are shown in Figs.~\ref{antisqueeze} (c) and (d), respectively. Consistent with the findings presented in Fig.~\ref{squeeze}, the SL $S_{\rm lin}$ does not follow a simple monotonic increase over time. Notably, the SL is maximal at time $\tau$, compared to the times $3\tau/4$ and $5\tau/4$. Conversely, the anti-squeezing level, as presented in Fig.~\ref{antisqueeze} (d), demonstrates a monotonic increase with time. Consequently, measuring the TMSS with the interval $[3\tau/4,\tau]$ yields a higher degree of squeezing while simultaneously minimizing the anti-squeezing effect. Within the optimal time window, the anti-squeezing level obtained by our scheme is comparable to that obtained in previous reservoir-engineering approaches under system stability conditions [see~\ref{appbtmsssteady}, Figure~\ref{Hamlinearize} (c)], while simultaneously enabling a significantly higher SL than those earlier methods. Moreover, according to the definition $\tau = 2\pi/(\Omega+\kappa_a+\kappa_b)$ and the expressions for $\Omega$ given in Eq.~\eqref{Omegphi}, this useful time window becomes narrower as the coupling strength increases.

Furthermore, the enhancement of anti-squeezing imposes a stricter requirement on the angular control of the measurement operator. Specifically, consider the actual measurement operator $O=\cos\theta X+\sin\theta Y$, which coincides with the ideal squeezing operator $X$ at $\theta=0$. For a misalignment $\theta\neq 0$, the variance becomes $\Delta O=\cos^2\theta\Delta X+\sin^2\theta\Delta Y$. Therefore, a larger anti-squeezing variance $\Delta Y$ implies a smaller allowable misalignment angle $\theta$ for maintaining $\Delta O<1/2$, leading to more stringent requirements on angle control. This further highlights the importance of selecting an appropriate time window in which the anti-squeezing remains within an acceptable range.

\subsection{Robustness to systematic errors}\label{secsyserror}
In the ideal situation, our TPS protocol assisted by the mechanical mode relies on the precise realization of the effective Hamiltonian given in Eq.~(\ref{Heff}). However, in practice, experimental imperfections and technical limitations prevent perfect control over the parameters of the full Hamiltonian~(\ref{Hamline}). In this section, we analyze the systematic errors arising from the driving-enhanced optomechanical coupling strengths and the mismatch of the photon frequency detunings. Specifically, the systematic error in the optomechanical coupling strengths originates from the fluctuations in the Rabi frequency $\Omega_o,o=a,b$ of the driving $H_d$ in Eq.~(\ref{Hamori}), while the detuning deviation is primarily due to the inaccuracies in the laser driving frequency $\epsilon_o$. The Hamiltonian implemented in experiments is assumed to be
\begin{equation}\label{Hamexpsys}
	H_{\rm exp}=H_{\rm int}+\gamma H_g+\eta H_{\Delta}
\end{equation}
where $H_{\rm int}$ is the ideal full system Hamiltonian in Eq.~(\ref{Hamline}) and $H_g$ and $H_{\Delta}$ are,
\begin{equation}
	\begin{aligned}
		H_g&=g(e^{-i\theta_a}a+e^{i\theta_a}a^\dag)(m+m^\dag)-G(e^{-i\theta_b}b+e^{i\theta_b}b^\dag)(m+m^\dag),\\
		H_\Delta&=\Delta_a a^\dag a-\Delta_b b^\dag b,
	\end{aligned}
\end{equation}
where $\gamma$ and $\eta$ are dimensionless perturbation coefficients for the coupling strength and the frequency detuning, respectively.
\begin{figure}[t]
\centering
\includegraphics[width=0.235\textwidth]{syserrorcoupling.eps}
\includegraphics[width=0.235\textwidth]{sysdelta.eps}
\includegraphics[width=0.235\textwidth]{syserrorcoupling2.eps}
\includegraphics[width=0.235\textwidth]{sysdelta2.eps}
\includegraphics[width=0.235\textwidth]{syscouplingspace.eps}
\includegraphics[width=0.235\textwidth]{sysdeltaspace.eps}
\caption{[(a),(b)] The SL $S_{\rm lin}$ and $\tilde{S}_{\rm lin}$ at time $\tau$ evaluated under the system Hamiltonian~\eqref{Hamexpsys} as a function of the systematic error associated with the optomechanical coupling strength, respectively. (c) The numerical optical SL $\tilde{S}_{\rm lin}(\tau)$ under the space spanned by the detuning $\Delta_b$ and coupling error magnitude $\gamma$. [(d), (e)] The SL $S_{\rm lin}(\tau)$ and $\tilde{S}_{\rm lin}(\tau)$ under the system Hamiltonian~\eqref{Hamexpsys} as a function of the systematic error associated with the detuning frequency, respectively. (f) The SL $\tilde{S}_{\rm lin}(\tau)$ under the space spanned by the detuning $\Delta_b$ and detuning error magnitude $\eta$. For panels (a),(b),(d), and (e), the detuning is fixed at $\Delta_b=\omega_m+10g$. For panels (c) and (f), the coupling strengths are set as $g=G=0.1\omega_m$. All other parameters and initial system states are identical to those in Fig.~\ref{vtsqueeze}.}
\label{syserror}
\end{figure}

Under the Hamiltonian in Eq.~(\ref{Hamexpsys}) with systematic errors, the SL $S_{\rm lin}$ corresponding to the quadrature operator $X$~(\ref{quadratureX}), as well as the numerically obtained optimal $\tilde{S}_{\rm lin}$, are presented in Fig.~\ref{syserror}. It is found in Fig.~\ref{syserror} (a) that the SL $S_{\rm lin}$ sensitivity to the coupling strength error $\gamma$ is amplified with increasing strength $g$. However, when $\gamma\le 0.1$, all cases exhibit strong robustness against such errors. In contrast, as shown in Fig.~\ref{syserror} (b), the optimized SL $\tilde{S}_{\rm lin}$ are roughly invariant in the presence of errors $\gamma$, even when the error magnitude reaches $20\%$. Fig.~\ref{syserror} (c) presents the SL $\tilde{S}_{\rm lin}$ under various detuning $\Delta_b$ and coupling error magnitude $\gamma$. It can be observed that smaller detuning lead to higher SL. This behavior can be understood as a consequence of smaller detuning resulting in larger effective coupling $g_{\rm eff}$~(\ref{geffdelta}), regardless of the error magnitude. Therefore, we conclude that systematic errors in the optomechanical coupling have a slight influence on the selection of the optimal quadrature operator, causing a slight deviation from the theoretical prediction $X$~(\ref{quadratureX}) while leaving the two-mode squeezing phenomenon largely unaffected.

Similarly, the SL robustness against the frequency deviation of the two photonic modes is presented in Figs.~\ref{syserror} (d), (e), and (f). As shown in Fig.~\ref{syserror} (d), the SL $S_{\rm lin}$ decreases significantly with increasing error magnitude $\eta$, and the sensitivity becomes more pronounced for larger values of $g$. For instance, when $g=0.2\omega_m$, the $S_{\rm lin}$ declines into $0$ as $\eta$ approaches $10^{-3}$. This indicates that detuning errors have a substantial impact on the achievable $S_{\rm lin}$, which is more detrimental than errors in the coupling strength. However, as seen in Figs.~\ref{syserror} (e) and (f), this deviation has no significant effect on the optimized SL $\tilde{S}_{\rm lin}$. Theoretically, the quadrature operator $X$~(\ref{quadratureX}), derived using the effective Hamiltonian, is obtained under a strict condition, i.e., $\Delta_a=-\Delta_b+\delta$, where $\delta\sim0.01-0.1\Delta_b$ (see~\ref{appcHeff}). As a result, frequency deviations between the two photonic modes can significantly affect the selection of the optimal quadrature operator. Nevertheless, such deviations do not prevent the emergence of a stable TPS, as the effective TPS coupling $g_{\rm eff}$ can still be successfully constructed when $\Delta_a\approx-\Delta_b$. In fact, under this condition, the system is governed by the effective Hamiltonian given in Eq.~(\ref{Heffinteract}), rather than the $H_{\rm eff}$ shown in Eq.~(\ref{Heff}). 

\subsection{Feasibility in thermal environment noises}\label{secthermal}
\begin{figure}[ht]
	\centering
	\includegraphics[width=0.235\textwidth]{thermal1.eps}
	\includegraphics[width=0.235\textwidth]{thermal2.eps}
	\caption{(a) Time evolution of $\Delta X(t)$ obtained using either the effective Hamiltonian~(\ref{Heff}) or the full system Hamiltonian~(\ref{Hamline}) at various temperatures. (b) The SL computed from the full system Hamiltonian~(\ref{Hamline}) at selected moments under different temperatures. All other parameters are identical to those in Fig.~\ref{vtsqueeze}, except for the thermal occupation numbers. Specifically, the photon modes are initially prepared in thermal equilibrium states, while the phonon mode $m$ is initially in the vacuum state.}
	\label{thermal}
\end{figure}
Our protocol is primarily focused on generating phonon-assisted two-photon-mode squeezing in the specific cavity optomechanical system. In the preceding numerical simulations, we assume the thermal occupations of photon and phonon modes to be $N_a=N_b=0$ and $N_m=10$, respectively, approximate values corresponding to a low temperature of $T\sim 10$ mK. In this section, we analyze the performance of two-mode squeezing across a range of temperatures, employing a representative set of experimentally accessible parameters. Specifically, we consider microwave resonator transition frequencies of $\omega_a/2\pi=\omega_b/2\pi=10$ GHz and a mechanical mode frequency of $\omega_m/2\pi=10$ MHz in our simulations. The corresponding numerical results are presented in Fig.~\ref{thermal}.

Figure~\ref{thermal} (a) compares the variance dynamics $\Delta X(t)$ with the effective Hamiltonian~(\ref{Heff}) and the full system Hamiltonian~(\ref{Hamline}) under varying thermal conditions. Within the time interval, $\omega_mt\le 500$, the analytical results from Eq.~(\ref{Deltaxphit}) show excellent agreement with the corresponding numerical results derived from Eq.~(\ref{wholedotV}). As the temperature $T$ increases from $20$ mK to $200$ mK, $\Delta X(t)$ shows only a slight increase, indicating that the high-quality TPS is well preserved even at elevated temperatures. In Fig.~\ref{thermal} (b), we plot the SL $S_{\rm lin}(\tau/2)$, $S_{\rm lin}(\tau)$, and $S_{\rm lin}(3\tau/2)$ as a function of temperature, using the full system Hamiltonian. The results demonstrate that the protocol sustains a high SL, with $S_{\rm lin}(\tau)\ge 8$ dB, at low temperatures $T\le 100$ mK. As the temperature gradually increases, the SL $S_{\rm lin}(\tau)$ gradually declines. However, a substantial squeezing of $5.5$ dB is still achieved at $T=100$ mK. Besides, consistent with the results in Fig.~\ref{squeeze}, both $S_{\rm lin}(\tau/2)$ and $S_{\rm lin}(3\tau/2)$ are approximately $1$ dB lower than $S_{\rm lin}(\tau)$ across the considered temperature range. 

\section{Discussion and Conclusion}\label{Secexpercon}
Our protocol is designed for hybrid cavity optomechanical systems. In recent experiments~\cite{cohermechanical,entmechanical,squeemech,Superposition,coherentcoupling,active,map}, the phonon frequency $\omega_m/2\pi\sim 10-100$ MHz, with a low decay rate $\kappa_m/2\pi\sim 10-100$ Hz, corresponding to $\kappa_m/\omega_m\sim 10^{-6}$. The cavity quality factor $Q=\omega/\kappa$, where $\omega$ is the transition frequency and $\kappa$ is the decay rate, depends on the specific cavity structure. Fabry-P\'erot~\cite{Rempe:92} and whispering-gallery-mode~\cite{Gorodetsky:96} (micro) cavities exhibit ultrahigh quality factors on the order of $10^{10}$, with decay rates $\kappa/2\pi \sim 10$ KHz. For microwave resonators, the transition frequency of the photon mode is typically around $10$ GHz, with a high-quality factor $Q\sim 10^{4}- 10^{7}$. The corresponding photon loss rates are $\kappa/2\pi\sim 10^{-3}\omega_m$. The Rabi frequency of external driving is defined as $\Omega\equiv \sqrt{\kappa P_d/(\hbar\omega_d)}$~\cite{synergizing}, where $P_d$ and $\omega_d$ represent the power and frequency of the drive, respectively. In recent experiments~\cite{MechanicalBis}, $\kappa/2\pi\sim 0.01-0.1$ MHz and $P_d\sim 20-30$ dBm ($100-1000$ mW), yielding a Rabi frequency $\Omega\sim 10^{14}-10^{15}$ Hz, and consequently, the mean excitation value shown in Eq.~(\ref{steadyope}) is about $|\alpha|\approx 10^{7}\sim 10^{8}$. The driving-enhanced photon-phonon coupling is given by $g\equiv g_a\alpha \sim 0.1\omega_m$, where the single excitation coupling $g_a$ is about $1-10$ Hz. As shown in Sec.~\ref {secsyserror}, the coupling strength $g$ can tolerate systematic errors of up to $\sim 10\%$ ($\sim 0.1-1$ MHz). Owing to the dependence of $g$ on the driving power $P_d$, the scheme is robust against power fluctuations at the $1\%$ level. Moreover, it remains valid for driving-frequency deviations of $10-100$ kHz. The thermal occupations of photon modes and phonons are respectively $N_a\approx N_b\approx 0$ and $N_m\approx 10$ at $T\sim 10$ mK. All of the values are consistent with the parameters used in Secs.~\ref {Sectms},~\ref{secanti}, and~\ref{secsyserror}. In addition, beyond the cavity optomechanical system, this scheme is extendable to other hybrid platforms involving three bosonic modes. For example, a magnonic interface could be employed to realize microwave–optical squeezing~\cite{Nonrecisteer}.

In summary, we have proposed a protocol for generating TMSS in a three-mode cavity optomechanical system, wherein a mechanical resonator is simultaneously coupled to two microwave (or optical) modes. The protocol relies on an effective two-mode squeezing Hamiltonian mediated by the mechanical mode. To numerically validate the effective Hamiltonian in the presence of nonconservative excitations, we employ a method based on diagonalizing the full system’s transition matrix. Within the open quantum systems framework, we derive the dynamical evolution of TMSS generation governed by the effective Hamiltonian. Our analysis demonstrates that stable TMSS with a high SL can be achieved even beyond the system stability regime, with high-SL TMSS dynamically generated within an appropriate time window while maintaining the anti-squeezing at a controlled level. Moreover, the protocol exhibits robustness against systematic errors and thermal environmental noises. It offers notable advantages in terms of system controllability and provides a practical route to implementing TMSS generation in the presence of environmental noises and technical imperfections. Overall, our protocol establishes a versatile and scalable framework for TMSS generation, with broad potential applications in quantum information processing and quantum precision measurement using bosonic systems.

\section{Data availability statement}
Data will be made available on request.

\section{Acknowledgements} 
We acknowledge financial support from the National Science Foundation of China (12404405) and the Science Foundation of Hebei Normal University of China (L2024B10).

\appendix
\section{System linearized Hamiltonian}\label{appaline}
This appendix contributes to deriving the linearized Hamiltonian in Eq.~(\ref{Hamline}). With respect to the transformation $U(t)=\exp{(i\epsilon_ata^\dag a+i\epsilon_btb^\dag b)}$, the original Hamiltonian in Eq.~(\ref{Hamori}) turns out to be 
\begin{equation}\label{Ham1}
H_s=\omega_mm^\dag m+\sum_{o=a,b}\Delta_oo^\dag o+g_oo^\dag o(m+m^\dag)+\Omega_o(o+o^\dag),
\end{equation}
where $\Delta_o=\omega_o-\epsilon_o$. Using the quantum Langevin equation, the time evolution of the system operators can be written as
\begin{equation}\label{qleoperator}
\begin{aligned}
	\dot{a}&=-(i\Delta_a+\kappa_a)a-ig_aa(m+m^\dag)-i\Omega_a+\sqrt{2\kappa_a}a_{in},\\
	\dot{b}&=-(i\Delta_b+\kappa_b)b-ig_bb(m+m^\dag)-i\Omega_b+\sqrt{2\kappa_b}b_{in},\\
	\dot{m}&=-(i\omega_m+\kappa_m)m-ig_aa^\dag a-ig_bb^\dag b+\sqrt{2\kappa_m}m_{in},\\
\end{aligned}
\end{equation}
where $\kappa_a$, $\kappa_b$, and $\kappa_m$ are the decay rates of the modes $a$, $b$, and $m$, respectively. $o_{in}, o=a,b,m$ is the input noise operator for the mode $o$, which is characterized by the covariance functions: $\langle o_{in}(t)o^\dag_{in}(t')\rangle=[N_o+1]\delta(t-t')$ and $\langle o^\dag_{in}(t)o_{in}(t')\rangle=N_o\delta(t-t')$, under the Markovian approximation. $N_o=[\exp(\hbar\omega_o/k_BT)-1]^{-1}$ is the thermal occupation number of mode $o$.

Under proper strong driving pulses, by performing the standard linearization approach, we write the operators as their classical values plus small fluctuations, i.e., $a=\alpha+\delta a$, $b=\beta+\delta b$, and $m=M+\delta m$. Here, $\alpha,\beta, M$ are the complex numbers and $\delta o, o=a,b,m$ are the fluctuation operators. The classical values are determined by
\begin{equation}\label{steadyoperator}
\begin{aligned}
	\dot{\alpha}&=-(i\Delta_a+\kappa_a)\alpha-ig_a\alpha(M+M^*)-i\Omega_a,\\
	\dot{\beta}&=-(i\Delta_b+\kappa_b)\beta-ig_b\beta(M+M^*)-i\Omega_b,\\
	\dot{M}&=-(i\omega_m+\kappa_m)M-ig_a|\alpha|^2-ig_b|\beta|^2,\\
\end{aligned}
\end{equation}
It is important to note that $\alpha, \beta, M$ can, in principle, achieve any desirable values by time-dependent modulation of the corresponding driving fields $\Omega_a$ and $\Omega_b$. These quantities rapidly converge to their respective constant magnitudes through appropriate tuning of the Rabi frequencies, which occurs on a timescale much faster than that of the fluctuation dynamics. The classical constants are obtained by setting the time derivatives in Eq.~(\ref{steadyoperator}) to zero, which satisfy
\begin{equation}\label{steadyope}
\begin{aligned}
\alpha&=-\frac{\Omega_a}{\Delta_a-i\kappa_a-2g_a{\rm Re}(M)},\\
\beta&=-\frac{\Omega_b}{\Delta_b-i\kappa_b-2g_b{\rm Re}(M)},\quad M=-\frac{g_a|\alpha|^2+g_b|\beta|^2}{\omega_m-i\kappa_m},
\end{aligned}
\end{equation}
where ${\rm Re}(M)$ is the real part of $M$. When $\kappa_a,\kappa_b,\kappa_m\ll|\Delta_a|,|\Delta_b|,\omega_m$ and both $g_a$ and $g_b$ are significantly small, the classical constants of photon modes $a$ and $b$ approximately equal to $|\alpha|\approx\Omega_a/\Delta_a$ and $|\beta|\approx \Omega_b/\Delta_b$, respectively.

By substituting these classical constants in Eq.~(\ref{steadyope}) into the Eq.~(\ref{qleoperator}) and ignoring all the high-order terms of fluctuations, the quantum Langevin equations describing the fluctuation operator $\delta o$ can be written as
\begin{equation}\label{flucope}
\begin{aligned}
\dot{\delta a}&=-[i\Delta_a+2ig_a{\rm Re}(M)+\kappa_a]\delta a-ig_a\alpha(\delta m+\delta m^\dag)+\sqrt{2\kappa_a}a_{in},\\
\dot{\delta b}&=-[i\Delta_b+2ig_b{\rm Re}(M)+\kappa_b]\delta b-ig_b\beta(\delta m+\delta m^\dag)+\sqrt{2\kappa_b}b_{in},\\
\dot{\delta m}&=-(i\omega_m+\kappa_m)\delta m-ig_a(\alpha a^\dag+\alpha^*a)-ig_b(\beta b^\dag+\beta^*b)\\
&+\sqrt{2\kappa_m}m_{in}.\\
\end{aligned}
\end{equation}
The corresponding effective linearized Hamiltonian can be described as
\begin{equation}\label{hamlineapp}
\begin{aligned}
	H_{\rm lin}&=\tilde{\Delta}_a\delta a^\dag\delta a+\tilde{\Delta}_b\delta b^\dag\delta b+\omega_m\delta m^\dag\delta m\\
	&+(g^*\delta a+g\delta a^\dag)(m+m^\dag)+(G^*\delta b+G\delta b^\dag)(m+m^\dag),
\end{aligned}
\end{equation}
where $g=g_a\alpha\approx g_a\Omega_a/\Delta_a$ and $G=g_b\beta\approx g_b\Omega_b/\Delta_b$ are the driving-enhanced coupling strengths. The modified detunings $\tilde{\Delta}_a=\Delta_a-2g_a{\rm Re}(M)\approx\Delta_a$ and $\tilde{\Delta}_b=\Delta_b-2g_b{\rm Re}(M)\approx\Delta_b$ provided by $g_o{\rm Re}(M)\ll\Delta_o, o=a,b$. It is the linearized Hamiltonian in Eq.~(\ref{Hamline}) in the main text. For simplicity and with no loss of generality, we apply the convention $\tilde{\Delta}_o\to\Delta_o$, $\delta o\to o, o=a,m,b$, and $g\to ge^{-i\theta_a}$, $G\to Ge^{-i\theta_b}$ in the main manuscript and following content.

\section{Two-mode squeezing via reservoir engineering method}\label{appbtmsssteady}
Under the parameter conditions $\Delta_a=\omega_m, \Delta_b=-\omega_m,$ and $g, G\ll \Delta_a,\Delta_b,\omega_m$, after the rotating wave approximation, the linearized system Hamiltonian $H_{\rm lin}$ turns into
\begin{equation}\label{Hintappb}
	H_{\rm int}=g(e^{i\theta_a}a^\dag m+e^{-i\theta_a}am^\dag)+G(e^{-i\theta_b}bm+e^{i\theta_b}b^\dag m^\dag).
\end{equation} 
In the regime $g>G$, one can introduce a Bogoliubov mode $\tilde{a}$ to transform the above Hamiltonian, defined as
\begin{equation}
	\tilde{a}=e^{-i\theta_a}a\cosh r+e^{i\theta_b}b^\dag\sinh r=S(r)aS^\dag (r),
\end{equation}
where $S(r)\equiv\exp(r^*ab-ra^\dag b^\dag)$, $r=|r|e^{i(\theta_a+\theta_b)}$, and $\tanh |r|=G/g$. The joint vacuum of $\tilde{a}$ is the TMSS $|\tilde{0}\rangle=S(r)|00\rangle$, where $|00\rangle\equiv|0\rangle_a|0\rangle_b$. Consequently, reservoir engineering that cools $\tilde{a}$ into its vacuum state deterministically generates a TMSS of the original modes $a$ and $b$.

Using the quantum Langevin equation, the dynamics of the quantum system under the interaction Hamiltonian~(\ref{Hintappb}) can be written in a matrix form
\begin{equation}\label{uat}
	\dot{u}(t)=A_{\rm int}u(t)+n(t),
\end{equation}
where $u^T(t)=[X_a(t),Y_a(t),X_b(t),Y_b(t),X_m(t),Y_m(t)]$ is the vector of quadrature fluctuation operators, and $X_o=(e^{-i\theta_o}o+e^{i\theta_o}o^\dag)/\sqrt{2}$, $Y_o=(e^{-i\theta_o}o-e^{i\theta_o}o^\dag)/i\sqrt{2}$, $o=a,m,b$, $\theta_m=0$. $n^T(t)=[X^{in}_a(t),Y^{in}_a(t),X^{in}_b(t),Y^{in}_b(t),X^{in}_m(t),Y^{in}_m(t)]$ is the vector of corresponding noise operators, and $X^{in}_o=(e^{-i\theta_o}o_{in}+e^{i\theta_o}o_{in}^\dag)/\sqrt{2}$, $Y^{in}_o=(e^{-i\theta_o}o_{in}-e^{i\theta_o}o_{in}^\dag)/i\sqrt{2}$. The transition matrix $A_{\rm int}$ is
\begin{equation}\label{alinmatrix}
\begin{aligned}
	A_{\rm int}=\begin{bmatrix}
		-\kappa_a & 0 & 0 & 0 &0 & g\\
		0 &-\kappa_a &0 & 0 &-g & 0\\
		0 &0 &-\kappa_b &0 &0 & -G\\
		0 &0 &0 &-\kappa_b &-G & 0\\
		0 &g &0 &-G &-\kappa_m & 0\\
		-g &0 &-G &0 &0& -\kappa_m\\
	\end{bmatrix}
\end{aligned}
\end{equation}

Due to the above dynamics in Eq.~(\ref{uat}) and the zero-mean Gaussian nature of the quantum noises, the hybrid system evolves as a Gaussian state, which can be completely characterized by a $6\times 6$ CM $V(t)$. The dynamics of CM $V(t)$ satisfies 
\begin{equation}
	\dot{V}(t)=A_{\rm int}V(t)+V(t)A_{\rm int}^T+D.
\end{equation}
The element of $V(t)$ is given by $V_{ij}=\langle u_i(t)u_j(t)+u_j(t)u_i(t)\rangle/2$, where $u_i(t)$ is the $i$ term of $u(t)$ and $i=1,2,\cdots,6$. $D$ is the noise covariance matrix, where the diagonal elements are 
$D_{11}=D_{22}=\kappa_a(2N_a+1)$, $D_{33}=D_{44}=\kappa_b(2N_b+1)$, and $D_{55}=D_{66}=\kappa_m(2N_m+1)$, all non-diagonal elements are zero.

The steady-state CM can be achieved by setting $\dot{V}(t)=0$ and its elements are
\begin{equation}\label{cmsteadyele}
\begin{aligned}
V_{11}&=\frac{1}{2}+\frac{G^2g^2(\kappa_m+2\kappa_a)}{\tilde{v}},\quad\\
V_{33}&=\frac{1}{2}+\frac{G^2[2(\kappa_a+\kappa_m)(g^2+\kappa_a\kappa_m)-G^2\kappa_m)]}{\tilde{v}},\\
V_{66}&=\frac{1}{2}+\frac{G^2\kappa_a[g^2-G^2+2\kappa_a(\kappa_a+\kappa_m)]}{\tilde{v}},\quad\\
V_{13}&=-\frac{Gg[G^2\kappa_a+(g^2+\kappa_a\kappa_m)(\kappa_a+\kappa_m)]}{\tilde{v}},\\
V_{16}&=-\frac{gG^2\kappa_a(2\kappa_a+\kappa_m)}{\tilde{v}},\quad\\
V_{36}&=-\frac{G[2\kappa_a(\kappa_a+\kappa_m)(g^2+\kappa_a\kappa_m)-G^2\kappa_a\kappa_m]}{\tilde{v}},\\\end{aligned}
\end{equation}
where 
\begin{equation}
	\tilde{v}=(\kappa_a+\kappa_m)(G^2-g^2-\kappa_a\kappa_m)[G^2-g^2-2\kappa_a(\kappa_a+\kappa_m)],
\end{equation}
and non-zero matrix elements are $V_{22}=V_{11}$, $V_{44}=V_{33}$, $V_{31}=V_{13}$, $V_{24}=V_{42}=-V_{13}$, $V_{61}=V_{16}$, $V_{25}=V_{52}=-V_{26}$, and $V_{45}=V_{54}=V_{63}=V_{36}$. For simplicity and with no loss of generality, here we assume $\kappa_b=\kappa_a$ and $N_a=N_b=N_m=0$. 

\begin{figure}[b]
	\centering
	\includegraphics[width=0.235\textwidth]{squapp.eps}
	\includegraphics[width=0.235\textwidth]{antisquapp.eps}
	\caption{[(a), (b)] The SL $S$ and anti-squeezing level $S'$ as a function of the coupling strength $G/g$ using the whole system Hamiltonian in Eq.~(\ref{Hamline}), respectively. Here, the parameters are set as $\Delta_a=\omega_m$, $\Delta_b=-\omega_m$, $\kappa_a=\kappa_b=10^{-3}\omega_m$, $\kappa_m=10^{-6}\omega_m$, and the thermal occupation numbers $N_a=N_b$ and $N_m=10$.}\label{Hamlinearize}
\end{figure}

The two-mode squeezing operator can be written as $X=(X_a+X_b)/\sqrt{2}$. Then, with the CM elements shown in Eq.~(\ref{cmsteadyele}), its variance 
$\Delta X=\langle X\rangle^2-\langle X\rangle^2$ can be described as
\begin{equation}
\begin{aligned}
\Delta X&=\frac{1}{2}+\frac{G^2(g^2-G^2)\kappa_m}{2\tilde{v}}+\frac{G^2(g^2-Gg)\kappa_a}{\tilde{v}}\\
&+\frac{(G^2-Gg)(g^2+\kappa_a\kappa_m)(\kappa_a+\kappa_m)}{\tilde{v}}.
\end{aligned}
\end{equation}
In the strong coupling regime and the decay rate of photon is larger than the one of phonon~\cite{cohermechanical,entmechanical,squeemech,Superposition,coherentcoupling,active,map}, i.e., $g, G>\kappa_a\gg\kappa_m$, the variance $\Delta X$ can be approximated as
\begin{equation}\label{Deltaxapprox}
\Delta X\approx \frac{1}{2}-\frac{Gg}{(G+g)^2}.
\end{equation}
The variance reaches its minimal value, $\Delta X_{\rm min}=1/4$, when $G\to g$, which corresponds to a SL of $3$ dB. The SL $S$ in the decibel unit is defined by $S=-10\log_{10}(\Delta X/\Delta X_{zp})$, where $\Delta X_{zp}=0.5$ is the standard fluctuation in the zero-point level. Obviously, a smaller $\Delta X$ yields a higher SL $S$. In practical implementations, cooling the phonon mode to its ground state remains a challenging task. Moreover, a higher phonon occupation number $N_m$ degrades the squeezing performance. Consequently, under realistic conditions, the achievable SL of this method is difficult to exceed the $3$ dB limit.

In Figs.~\ref{Hamlinearize}(a) and (b), we present the numerical results for the SL $S$ and the anti-squeezing level $S'$ under different coupling strengths, respectively. The anti-squeezing level is defined as $S'=10\log_{10}(\Delta Y/\Delta Y_{zp})$, where $\Delta Y$ is the variance of the anti-squeezing operator $Y=(Y_a+Y_b)/\sqrt{2}$ and $\Delta Y_{zp}=0.5$ denotes the zero-point fluctuation. As shown in Fig.~\ref{Hamlinearize}(a), the SL $S$ remains below the $3$ dB limit for the chosen parameters. With increasing coupling strength $g$, the optimal ratio $G/g$ (corresponding to the inflection point where maximal squeezing occurs) shifts to larger values, accompanied by an enhanced achievable SL. However, for values of $G/g$ far from this optimum, further increasing $g$ instead suppresses the squeezing. Figure~\ref{Hamlinearize}(b) shows the corresponding anti-squeezing level. The three curves nearly overlap and lie within the range of $22-45$ dB, indicating that the anti-squeezing is largely insensitive to variations in $g$. In addition, the anti-squeezing level increases monotonically with $G/g$.

\section{The effective Hamiltonian}\label{appcHeff}
This appendix contributes to deriving the effective Hamiltonian in Eq.~(\ref{Heff}) and evaluates its validity. When the detuning frequency of mode $a$ is almost opposite the detuning of mode $b$, and both of them are far resonant from the frequency of phonon, i.e., $\Delta_a+\Delta_b\approx 0$ and $|\Delta_a-\omega_m|, |\Delta_b-\omega_m|\gg g, G$, the interaction Hamiltonian $V$ in Eq.~\eqref{Hamline} can be regarded as a perturbation to the free Hamiltonian $H_0$. Under these conditions, the effective TPS coupling between photons $a$ and $b$, mediated by phonon mode $m$, can be successfully constructed. The detailed derivations are provided below.

With respect to $U(t)=\exp(-iH_0t)$, the linearized Hamiltonian $H_{\rm lin}$ in Eq.~\eqref{Hamline} can be transformed to the interaction picture
\begin{equation}\label{Hamint}
\begin{aligned}
	H_I&=ge^{-i\theta_a}ame^{-i(\Delta_a+\omega_m)t}+ge^{-i\theta_a}am^\dag e^{-i(\Delta_a-\omega_m)t}\\
	&\quad+Ge^{-i\theta_b}bme^{-i(\Delta_b+\omega_m)t}+Ge^{-i\theta_b}bm^\dag e^{-i(\Delta_b-\omega_m)t}+{\rm H.c.}.
\end{aligned}
\end{equation}
In the interaction picture, if the system is governed by the Hamiltonian $H_I$ of the form~\cite{james},
\begin{equation}\label{Hintform}
	H_I=\sum_s g_s[h_s\exp(i\omega_st)+h^\dag_s\exp(-i\tilde{\omega}_st)].
\end{equation}
When $|\omega_s-\tilde{\omega}_s|\ll g_s\ll|\omega_s|,|\tilde{\omega}_s|$, the effective Hamiltonian up to the second order can be derived as
\begin{equation}\label{Heffjames}
	H_{\rm eff}=\sum_s\frac{g_s^2}{2}\left(\frac{1}{\omega_s}+\frac{1}{\tilde{\omega}_s}\right)[h_s,h^\dag_s].
\end{equation}
Then, the effective Hamiltonian corresponding to the system's Hamiltonian~\eqref{Hamint} in the interaction picture can be written as 
\begin{equation}\label{Heffinteract}
\begin{aligned}
H_{\rm eff}&=\frac{2g^2\omega_m}{\Delta^2_a-\omega^2_m}a^\dag a+\frac{2G^2\omega_m}{\Delta^2_b-\omega^2_m}b^\dag b\\
&-\left(\frac{2g^2\Delta_a}{\Delta^2_a-\omega^2_m}+\frac{2G^2\Delta_b}{\Delta^2_b-\omega^2_m}\right)m^\dag m\\
&+\left(\frac{gG\omega_m}{\Delta^2_a-\omega^2_m}+\frac{gG\omega_m}{\Delta^2_b-\omega^2_m}\right)(e^{-i\theta}ab+e^{i\theta}a^\dag b^\dag),
\end{aligned}
\end{equation}
where $\theta=\theta_a+\theta_b$. Rotating it into the lab frame and discard the non-interacting phonon mode $m$, the effective Hamiltonian in Eq.~\eqref{Heffinteract} turns into
\begin{equation}\label{Heffint}
\begin{aligned}
H_{\rm eff}&=\left(\Delta_a+\frac{2g^2\omega_m}{\Delta^2_a-\omega^2_m}\right)a^\dag a+\left(\Delta_b+\frac{2G^2\omega_m}{\Delta^2_b-\omega^2_m}\right)b^\dag b\\
&\quad+\left(\frac{gG\omega_m}{\Delta^2_a-\omega^2_m}+\frac{gG\omega_m}{\Delta^2_b-\omega^2_m}\right)(e^{-i\theta}ab+e^{i\theta}a^\dag b^\dag).
\end{aligned}
\end{equation}

A purely two-mode squeezing coupling between modes $a$ and $b$ requires that the first two terms in Eq.~\eqref{Heffint} constitute an identity operator. Assuming the distance between $\Delta_a$ and $-\Delta_b$ is $\delta$, one can have 
\begin{equation}\label{delta}
	\begin{aligned}
		\delta&\equiv\Delta_a+\Delta_b=\frac{2(g^2+G^2)\omega_m}{\omega^2_m-\Delta^2_b},
	\end{aligned}
\end{equation}
up to the second order of the coupling strengths $g$ and $G$. At the condition $\Delta_a=-\Delta_b+\delta$, the effective Hamiltonian in Eq.~\eqref{Heffint} eventually turns into
\begin{equation}\label{Heffint2}
	H_{\rm eff}=\frac{2gG\omega_m}{\Delta^2_b-\omega^2_m}(e^{-i\theta}ab+e^{i\theta}a^\dag b^\dag).
\end{equation}
That is exactly the effective Hamiltonian in Eq.~\eqref{Heff} describing the coupling between photon modes $a$ and $b$. It is worth noting that choosing the effective Hamiltonian in Eq.~\eqref{Heffint2}, rather than the more general form in Eq.~\eqref{Heffint}, enables the theoretical derivation of an optimal squeezing operator in Sec.~\ref{Sectms}. The effective Hamiltonian given in Eq.~\eqref{Heffint} is already capable of generating the stable TMSS with a high SL.

Now, we evaluate the applicability range of the effective Hamiltonian in Eq.~\eqref{Heffint2} regarding the coupling strengths by diagonalizing the transition matrix of the whole system~\cite{Delocalization}. This approach differs from previous works~\cite{intermagnon,oneexcite,nonlinear} that diagonalize the full Hamiltonian in a truncated finite-dimensional Hilbert space. 

\begin{figure}[t]
	\centering
	\includegraphics[width=0.235\textwidth]{realeigen.eps}
	\includegraphics[width=0.235\textwidth]{imageigen.eps}
	\caption{(a) All six real parts of the normalized eigenvalues of the transition matrix $\mathcal{L}$ are plotted as a function of the detuning frequency $\Delta_a/\omega_m$. (b) Two relevant imaginary parts of the normalized eigenvalues are depicted as a function of the detuning frequency $\Delta_a/\omega_m$. The parameters used are $\Delta_b=3\omega_m$ and $g=G=0.1\omega_m$.}\label{eigendiag}
\end{figure}

Dynamics of the system quadrature operators in the Heisenberg picture under the full Hamiltonian in Eq.~\eqref{Hamline} satisfy
\begin{equation}\label{utHeisenberg}
	\dot{u}(t)=i[H,u(t)]=i\mathcal{L}u(t),
\end{equation}
where $u(t)=[X_a(t), Y_a(t), X_b(t), Y_b(t), X_m(t), Y_m(t)]^T$ is the vector of quadrature operators, and $X_m=(m+m^\dag)/\sqrt{2}$, $Y_m=(m-m^\dag)/i\sqrt{2}$. $\mathcal{L}$ represents the whole system's transition matrix, 
\begin{equation}\label{matrix}
\begin{aligned}
\mathcal{L}=i\begin{bmatrix}
	0 &-\Delta_a& 0 & 0 & 0 & 0\\
	\Delta_a& 0 & 0 & 0 & 2g &0\\
	0& 0 & 0 &-\Delta_b & 0 & 0\\
	0& 0 & \Delta_b & 0 & 2G & 0\\
	0& 0 & 0 & 0 & 0 &-\omega_m\\
	2g& 0 & 2G & 0 &\omega_m &0			
\end{bmatrix}.
\end{aligned}
\end{equation}
The Heisenberg equation in Eq.~\eqref{utHeisenberg} can be regarded as a discrete Schr\"{o}dinger equation, where $u(t)$ is conceptualized as an effective operator wave function~\cite{Delocalization}. In the energy-level diagram of the matrix $\mathcal{L}$ as a function of $\Delta_a$, the two-mode squeezing interaction can be illustrated via the level attractions of the real parts and the maximal splittings of the imaginary parts~\cite{kerrmagnon}. 

\begin{figure}[t]
	\centering
	\includegraphics[width=0.235\textwidth]{geff.eps}
	\includegraphics[width=0.235\textwidth]{delta.eps}
	\includegraphics[width=0.235\textwidth]{geffdetuning.eps}
	\includegraphics[width=0.235\textwidth]{deltadetuning.eps}
	\includegraphics[width=0.235\textwidth]{geffspace.eps}
	\includegraphics[width=0.235\textwidth]{sigmarelative.eps}
	\caption{(a) Comparison between the numerically calculated normalized effective coupling strength $|g_{\rm eff}|/\omega_m$ (points) and the corresponding analytical results (lines) in Eq.~\eqref{geffdelta} as a function of $g/\omega_m$. (b) Comparison between the numerically calculated normalized energy shift $|\delta|/\omega_m$ (points) and the corresponding analytical results (lines) in Eq.~\eqref{geffdelta} as a function of $g/\omega_m$. (c) Comparison between the numerically calculated $|g_{\rm eff}|/\omega_m$ (points) and the corresponding analytical results (lines) in Eq.~\eqref{geffdelta} as a function of $(\Delta_b-\omega_m)/g$. (d) Comparison between the numerically calculated $|\delta|/\omega_m$ (points) and the corresponding analytical results (lines) in Eq.~\eqref{geffdelta} as a function of $(\Delta_b-\omega_m)/g$. (e) The numerical effective coupling $|g_{\rm eff}|/\omega_m$ in parameter space spanned by $g$ and $\Delta_b$. (f) The relative deviation $\sigma$ in the parameter space spanned by $g$ and $\Delta_b$. The parameter is fixed as $G=g$.}\label{effgdelta}
\end{figure}

The energy levels of $\mathcal{L}$ in Eq.~\eqref{matrix} are plotted in Figs.~\ref{eigendiag} (a) and (b). The real parts of all six eigenvalues are displayed in Fig.~\ref{eigendiag} (a). The orange and purple lines correspond to the energies of the mechanical mode and are decoupled from the two-mode squeezing dynamics. In contrast, for the remaining four eigenvalues, two distinct energy-level attractions emerge simultaneously as the detuning $\Delta_a$ approaches (but does not exactly equal) $-\Delta_b$. One of the level attractions is highlighted by a dark circle, and the inset further emphasizes it. The imaginary parts of the two relevant eigenvalues (shown in blue and red lines) are presented in Fig.~\ref{eigendiag} (b). As the real parts of the two eigenvalues gradually converge, their imaginary parts grow monotonically, reaching a maximum magnitude of $|g_{\rm eff}|$ at $\Delta_a=-\Delta_b+\delta$. The shift $\delta$ arises from the mutual interaction between the photon and phonon modes.

The maximal splitting $|g_{\rm eff}|$ of the imaginary parts of the two eigenvalues [see Fig.~\ref{eigendiag} (b)] is presented in Fig.~\ref{effgdelta} (a) as a function of the photon-phonon coupling strength $g$, and compared with the analytical prediction given by Eq.~\eqref{geffdelta}. Blue dots and orange squares represent the numerical results at $\Delta_b=2\omega_m$ and $\Delta_b=3\omega_m$, respectively, while the red solid and purple dashed lines correspond to the respective analytical results. One can observe that the analytical $g_{\rm eff}$ do match well with their numerical results for the coupling strength $g\le0.25\omega_m$ at large detuning $\Delta_b=3\omega_m$. At a small detuning $\Delta_b=2\omega_m$, the valid range decreases to $g\le0.18\omega_m$. Notably, both parameter sets enter the ultrastrong coupling regime, defined by $g/\omega_m\ge0.1$. The value indicated by the black box corresponds to those in Fig.~\ref{eigendiag} (b). The energy shift $\delta$ in Eq.~\eqref{geffdelta} is validated in Figure~\ref{effgdelta} (b). It is found that the energy shift $\delta$ is valid when $g\le 0.25\omega_m$ at $\Delta_b=3\omega_m$. As $\Delta_b$ decreases into $2\omega_m$, the valid range turns into $g\le 0.18\omega_m$. Similar results for $g_{\rm eff}$ and $\delta$ are shown in Figs.~\ref{effgdelta} (c) and (d) as functions of the detuning-to-coupling ratio $(\Delta_b-\omega_m)/g$. When $(\Delta_b-\omega_m)/g \ge 8$, the analytical predictions agree well with the numerical results, demonstrating the validity of the effective Hamiltonian in this regime. As the detuning increases, the effective coupling strength $g_{\rm eff}$ gradually decreases. By contrast, reducing the detuning leads to growing deviations between analytical and numerical results, indicating that the large-detuning condition required for the effective Hamiltonian is no longer well satisfied and its validity is correspondingly reduced.

More numerical results of the effective coupling strength $g_{\rm eff}$ are presented in Fig.~\ref{effgdelta} (e). It can be observed that a larger coupling $g$ and a smaller difference $(\Delta_b-\omega_m)$ yield a stronger effective coupling strength $g_{\rm eff}$, which is approximately consistent with the analytical result given by Eq.~\eqref{geffdelta}. To more clearly delineate the valid range of the effective Hamiltonian, we introduce the relative errors of the effective coupling strength, as illustrated in Fig.~\ref{effgdelta} (f). The relative error is defined as
\begin{equation}
	\sigma=\left|\frac{{\rm num}(g_{\rm eff})-{\rm ana}(g_{\rm eff})}{{\rm ana}(g_{\rm eff})}\right|,
\end{equation}
where ${\rm num}(g_{\rm eff})$ describes the numerical result of $g_{\rm eff}$ and ${\rm ana}(g_{\rm eff})$ presents the analytical one. From Fig.~\ref{effgdelta} (f), it can be observed that the relative error $\sigma$ decreases approximately as $\Delta_b-\omega_m$ increases, which is consistent with the conditions required by the perturbation theory. For a fixed and small $\Delta_b$, the $\sigma$ exhibits a significant reduction with increasing $g$, despite this trend seeming to diverge from the results in Fig.~\ref{effgdelta} (c). Indeed, as $g$ increases, the absolute error $|{\rm num}(g_{\rm eff})-{\rm ana}(g_{\rm eff})|$ increases, while the relative error $\sigma$ decreases. By synthesizing the results presented in Figs.~\ref{effgdelta} (a)-(f), a valid and well-performing region is approximately identified where $8g\le(\Delta_b-\omega_m)\le10g, 0.1\omega_m\le g\le 0.2\omega_m$. Within this region, the effective coupling strength $g_{\rm eff}\gtrsim0.01\omega_m$ and the relative error $\sigma\le0.01$, and the absolute error remains small.

\section{Derivation of the optimal quadrature operator}\label{appcm}
This appendix presents the derivation of the quadrature operator $X$ in Eq.~(\ref{quadratureX}). Based on Eq.~(\ref{cmeff}) and under the initial condition $V^{\rm eff}(0)=Diag[N_a+1/2,N_a+1/2,N_b+1/2,N_b+1/2]$, the nonzero matrix elements of the time-dependent CM $V^{\rm eff}(t)$ can be solved as
\begin{equation}\label{Veffelement}
\begin{aligned}
V^{\rm eff}_{11}(t)&=C_+(1-\sin\varphi)e^{(\Omega-\kappa_a-\kappa_b)t}-C_0\cos\varphi e^{-(\kappa_a+\kappa_b)t}\\
&+C_-(1+\sin\varphi)e^{-(\Omega+\kappa_a+\kappa_b)t}+c_a,\\
V^{\rm eff}_{44}(t)&=C_+(1+\sin\varphi)e^{(\Omega-\kappa_a-\kappa_b)t}+C_0\cos\varphi e^{-(\kappa_a+\kappa_b)t}\\
&+C_-(1-\sin\varphi)e^{-(\Omega+\kappa_a+\kappa_b)t}+c_b,\\
V^{\rm eff}_{14}(t)&=-C_+\cos\varphi e^{(\Omega-\kappa_a-\kappa_b)t}+C_0\sin\varphi e^{-(\kappa_a+\kappa_b)t}\\
&+C_-\cos\varphi e^{-(\Omega+\kappa_a+\kappa_b)t}+c,\\
\end{aligned}
\end{equation}
and $V_{22}^{\rm eff}(t)=V^{\rm eff}_{11}(t)$, $V^{\rm eff}_{33}(t)=V^{\rm eff}_{44}(t)$, $V^{\rm eff}_{23}(t)=V^{\rm eff}_{32}(t)=V^{\rm eff}_{41}(t)=V^{\rm eff}_{14}(t)$. The parameters are defined as
\begin{equation}\label{parameter}
\begin{aligned}
\Omega&=\sqrt{4g^2_{\rm eff}+(\kappa_a-\kappa_b)^2},\quad \tan\varphi=\frac{\kappa_a-\kappa_b}{2g_{\rm eff}},\quad\\ C_0&=\frac{\cos\varphi}{2}\left[\frac{\kappa_-}{(\kappa_a+\kappa_b)}-N_-\right],\\
C_{\pm}&=\pm \frac{\kappa_+\mp\sin\varphi\kappa_-}{4[\Omega\mp(\kappa_a+\kappa_b)]}+\frac{N_+\mp\sin\varphi N_-+1}{4},\\ 
N_\pm&=N_a\pm N_b,\quad \kappa_\pm=\kappa_a(2N_a+1)\pm\kappa_b(2N_b+1).\	
\end{aligned}
\end{equation}
And
\begin{equation}\label{steadyV}
	c=\frac{g_{\rm eff}\kappa_a\kappa_b(N_a+N_b+1)}{(g^2_{\rm eff}-\kappa_a\kappa_b)(\kappa_a+\kappa_b)},~c_o=N_o+\frac{1}{2}-\frac{g_{\rm eff}}{\kappa_o}c,~o=a,b
\end{equation}
which are also the solutions of the matrix elements $V^{\rm eff}_{11}$, $V^{\rm eff}_{44}$, and $V^{\rm eff}_{14}$ in the system stability condition, respectively, obtained by setting $\dot{V}^{\rm eff}=0$ in Eq.~(\ref{cmeff}). These steady CM elements are asymptotic values as $t\to\infty$. The system stability regime requires $g^2_{\rm eff}<\kappa_a\kappa_b$. 

Beyond the system stability region, $g^2_{\rm eff}>\kappa_a\kappa_b$, all the CM elements $V^{\rm eff}_{11}(t)$, $V^{\rm eff}_{44}(t)$, and $V^{\rm eff}_{14}(t)$ in Eq.~(\ref{Veffelement}) exhibit exponential divergence due to the exponential factor $\Omega-\kappa_a-\kappa_b>0$. In the specifical situation when $\kappa_a=\kappa_b$, one can demonstrate the elements $V^{\rm eff}_{11}(t)=V^{\rm eff}_{44}(t)$ by Eq.~(\ref{Veffelement}). However, the CM instability does not imply unstable two-mode squeezing. To find a stable TMSS with a higher SL, we define a general two-mode squeezing operator 
\begin{equation}
	X_\phi=\cos\phi X_a+\sin\phi Y_b,
\end{equation} 
where $\phi$ is an angle to optimize. With the CM definition in Eq.~(\ref{Veffdefin}) and its solution in Eq.~(\ref{Veffelement}), the variance of the general quadrature operator $\Delta X_\phi=\langle X_\phi^2\rangle-\langle X_\phi\rangle^2$ can be described as
\begin{equation}\label{Deltaxeff}
\begin{aligned}
\Delta X_\phi(t)&=\cos^2\phi V^{\rm eff}_{11}(t)+\sin^2\phi V^{\rm eff}_{44}(t)+\sin(2\phi)V^{\rm eff}_{14}(t)\\
&=C_+(1-\sin\tilde{\varphi})e^{(\Omega-\kappa_a-\kappa_b)t}-C_0\cos\tilde{\varphi}e^{-(\kappa_a+\kappa_b)t}\\
&+C_-(1+\sin\tilde{\varphi})e^{-(\Omega+\kappa_a+\kappa_b)t}+C_\phi,\\
\end{aligned}
\end{equation}
where $\tilde{\varphi}=\varphi+2\phi$ and $C_\phi=\cos^2\phi c_a+\sin^2\phi c_b+\sin(2\phi)c$, $c_a, c_b, c$ are constants in Eq.~(\ref{steadyV}).

From Eq.~(\ref{Deltaxeff}), one can find that the exponential divergence term in $\Delta X_\phi(t)$ can be canceled when $\tilde{\varphi}=\pi/2$. This implies the existence of an optimized angle $\tilde{\phi}$, which satisfies
\begin{equation}\label{phi}
	\tan(2\tilde{\phi})=\cot(\varphi)=\frac{2g_{\rm eff}}{\kappa_a-\kappa_b}.
\end{equation}
Specifically, $\tilde{\phi}=\pi/4$ at $\kappa_a=\kappa_b$. At this optimized angle $\tilde{\phi}$, the corresponding quadrature operator is given by 
\begin{equation}\label{quadratureXapp}
	X=\cos\tilde{\phi}X_a+\sin\tilde{\phi}Y_b,
\end{equation}
and the associated variance becomes
\begin{equation}\label{Deltaxphitapp}
	\Delta X(t)=2C_-e^{-(\Omega+\kappa_a+\kappa_b)t}+\frac{N_++1+\cos(2\tilde{\phi})N_-}{2}-2C_-,
\end{equation}
which is the same as Eq.~(\ref{Deltaxphit}).


\bibliographystyle{elsarticle-num}  % 指定参考文献样式
\bibliography{reference}
\end{document}
\endinput