---
title: Co-STEER Mechanism
url: https://www.emergentmind.com/topics/co-steer-mechanism
type: topic
---

# Co-STEER Mechanism

Searching arXiv for the cited papers to ground the article in the current record.
arXiv search query: 1912.04292
Co-STEER mechanism denotes a family of steering constructions in which a controllable auxiliary structure converts coherence, entanglement, measurement back-action, or collective coupling into directed dynamical bias. In the cited literature, the label spans repeated non-selective measurement protocols that drive a quantum system toward a dark target state, reservoir-engineering schemes that preserve maximal steered coherence, boundary-geometric criteria under which entanglement becomes projective steering, and cavity-enabled mechanisms that steer nonequilibrium molecular or optomechanical dynamics [1912.04292], [2008.10836], [2605.21245], [2412.07593], [1706.04474]. This suggests that Co-STEER is best understood as an umbrella designation for steering-by-design rather than as a single canonical formalism.

## 1. Range of meanings and common structure

Across the relevant arXiv literature, Co-STEER refers to technically distinct mechanisms.

| Usage | Physical setting | Operational core |
|---|---|---|
| Measurement-induced steering | System plus fresh detector qubits | Repeated unitary coupling, detector reset, and non-selective back-action [1912.04292] |
| Co-STEER protection | Two-qubit steering ellipsoid under decoherence | Auxiliary qubits enlarge a decoherence-free subspace and preserve MSC [2008.10836] |
| Boundary-geometric Co-STEER | Product-null boundary strata of two-qubit state space | Tangential coherence at boundary contact defeats finite-measure LHS models [2605.21245] |
| Cooperative cavity steering | Reactive molecule in an optical cavity with an auxiliary ensemble | Colored noise and always-negative feedback reshape thermalization [2412.07593] |
| Closed-loop phase steering | Three-mode optomechanical system | Relative phase transfers steering between bipartite and collective channels [1706.04474] |

The common structural motif is not a shared Hamiltonian or a shared resource monotone. Rather, the commonality is architectural: an auxiliary sector is engineered so that the induced reduced dynamics favor a selected steering outcome. In some cases the relevant auxiliary sector is a stream of detector qubits; in others it is a reservoir-coupled ensemble, a product-null boundary contact, or an interferometric loop. A recurring misconception is that steering must arise only from pre-existing entanglement or from uncontrolled open-system relaxation. The cited works instead emphasize controlled back-action, controlled geometry, or controlled collective response as the operative ingredient.

## 2. Measurement-induced steering as repeated ancilla-driven state engineering

In "Measurement-induced steering of quantum systems" [1912.04292], steering is defined in a broad Schrödinger sense: detector degrees of freedom are used to induce a chosen target state from arbitrary initial conditions. One steering event consists of three steps. First, detector qubits are prepared in a fixed state $\ket{\Phi_d}$ with density matrix $\rho_d$, independent of the system state. Second, the joint state evolves for a short time $\delta t$ under a coupling Hamiltonian $H_{s\text{-}d}$,
$$
\rho_{s\text{-}d}(t+\delta t)=e^{-iH_{s\text{-}d}\delta t}\,\rho_d\otimes\rho_s(t)\,e^{iH_{s\text{-}d}\delta t}.
$$
Third, the detectors are removed,
$$
\rho_s(t+\delta t)=\mathrm{Tr}_d\,\rho_{s\text{-}d}(t+\delta t).
$$
Because fresh detectors are prepared each cycle and the previous ones are discarded or projectively measured with outcomes averaged over, the environment is effectively Markovian.

The steering resource is explicitly the back-action generated by system-detector entanglement. The detector is driven out of its initial state when the system has amplitude in the unwanted sector, while the target is made dark under the coupling. The design principle is
$$
H_{s-d}=\sum_n\left(O_d^{(n)}\ket{\Phi_d}\bra{\Phi_d}\right)\otimes U_s^{(n)}+\mathrm{h.c.},
$$
with detector operators satisfying $\langle \Phi_d|O_d^{(n)}|\Phi_d\rangle=0$ and system operators chosen so that
$$
U_s^{(n)}\ket{\Psi}=0,\qquad U_s^{(n)}U_s^{(n)\dagger}\ket{\Psi}=\ket{\Psi},
$$
for the target $\ket{\Psi}$. The paper’s “golden inequality”
$$
\langle \Psi|\rho_s(t+\delta t)|\Psi\rangle \ge \langle \Psi|\rho_s(t)|\Psi\rangle,
$$
with equality only when $\rho_s(t)=\rho_\Psi$, establishes monotonic increase of target overlap in the ideal single-target case. For multiple detectors,
$$
\langle \Psi|\rho_s(t+\delta t)|\Psi\rangle = \langle \Psi|\rho_s(t)|\Psi\rangle + Q\,\sin^2(\delta t),
$$
where
$$
Q=\sum_n \langle \Psi|U_s^{(n)}\,\rho_s(t)\,U_s^{(n)\dagger}|\Psi\rangle\ge 0.
$$
Hence the overlap cannot decrease.

The continuous-time limit yields a Lindblad equation rather than a merely heuristic dissipator. After expanding to second order in $\delta t$ and rescaling $U\to \tilde U=U\sqrt{\delta t}$,
$$
\partial_t\rho_s=-i[V,\rho_s]+\left(\tilde U\rho_s\tilde U^\dagger-\tfrac12\{\tilde U^\dagger \tilde U,\rho_s\}\right),
$$
and for multiple detectors,
$$
\partial_t\rho_s=-i[V,\rho_s]+\sum_n\left(\tilde U_n\rho_s\tilde U_n^\dagger-\tfrac12\{\tilde U_n^\dagger \tilde U_n,\rho_s\}\right).
$$
The jump operators are thus fixed microscopically by the measurement couplings. The steady-state manifold is the zero-eigenvalue sector of the Lindbladian, and the convergence rate is governed by the Lindblad gap $\Delta$.

The paper gives both few-body and many-body realizations. For two spin-$\tfrac12$ particles, the target is the singlet
$$
\ket{S_0}=\frac{\ket{\uparrow\downarrow}-\ket{\downarrow\uparrow}}{\sqrt2},
$$
with three detector qubits removing weight from the triplet sector. The singlet population approaches $1$, triplet components decay exponentially, and the Lindbladian has a unique zero mode at $\ket{S_0}$. For a spin-1 chain, the target is the AKLT state, with bond-local couplings that remove weight from the $S^{\mathrm{tot}}=2$ sector. On the periodic chain, the AKLT ground state is the unique steady state, the many-body state approaches $\rho_{\mathrm{AKLT}}$ exponentially, and the Lindbladian gap remains finite in the thermodynamic limit, so the steering time does not diverge with system size. A subtlety is that local bond steering need not make the AKLT energy or the trace distance monotone at every step; convergence to the target and monotonicity of a chosen distance measure are distinct properties.

## 3. Reservoir-engineered protection of steered coherence

In "Maximal Steered Coherence Protection by Quantum Reservoir Engineering" [2008.10836], Co-STEER refers to preservation of the coherence content of Bob’s quantum steering ellipsoid by coupling auxiliary qubits to the same reservoir. The relevant geometric object is the quantum steering ellipsoid $\varepsilon_B$ for a general two-qubit state
$$
\rho_{AB}=\frac{1}{4} \Big[I\otimes I+\mathbf{a}\cdot\boldsymbol{\sigma}\otimes I+I\otimes \mathbf{b}\cdot\boldsymbol{\sigma}+\sum_{m,n=1}^{3}T_{nm}\sigma_{n}\otimes \sigma_{m}\Big].
$$
Its center is
$$
C_{B}=\frac{\mathbf{b}-T^{T}\mathbf{a}}{1-\mathbf{a}^2},
$$
and its shape is encoded by
$$
Q_{B}=\frac{(T^{T}-\mathbf{b}\mathbf{a}^{T})}{1-\mathbf{a}^2} \left(I+\frac{\mathbf{a}\mathbf{a}^{T}}{1-\mathbf{a}^2}\right) (T-\mathbf{a}\mathbf{b}^{T}).
$$
The eigenvalues of $Q_B$ give the squared semiaxes.

The coherence quantity of interest is the maximal steered coherence (MSC). If Alice performs a POVM element $M$, Bob’s post-measurement state is
$$
\rho^M_B=\frac{\mathrm{tr}_A(M\otimes I\,\rho_{AB})}{p_M},\qquad p_M=\mathrm{tr}(M\otimes I\,\rho_{AB}),
$$
and the coherence in the eigenbasis $\Xi=\{|\chi_i\rangle\}$ of $\rho_B$ is
$$
C(\rho^M_B,|\chi_i\rangle) =\frac{1}{p_M}\sum_{i\neq j} \left|\langle \chi_i|\mathrm{tr}_A(M\otimes I\,\rho_{AB})|\chi_j\rangle\right|.
$$
The MSC is
$$
MSC(\rho^M_B):= \inf_{\Xi}\left\{ \max_{M\in POVM} \left[ \frac{1}{p_M}\sum_{i\neq j} \left|\langle \chi_i|\mathrm{tr}_A(M\otimes I\,\rho_{AB})|\chi_j\rangle\right| \right] \right\}.
$$
For the states analyzed in the paper, this reduces to
$$
MSC(\rho^M_B)=\max\{s_1,s_2,s_3\},
$$
the length of the largest semiaxis of Bob’s steering ellipsoid.

The reservoir model contains $N$ identical two-level systems coupled to a common zero-temperature reservoir. Under the symmetric coupling assumption $g_k^j=\frac{1}{\sqrt{N}}G_k$, the collective basis diagonalizes the qubit sector so that only $\{|0\rangle,|\varphi_0\rangle\}$ couple to the reservoir, while $\{|\varphi_1\rangle,\dots,|\varphi_{N-1}\rangle\}$ are uncoupled and form a decoherence-free subspace. This is the protection mechanism: auxiliary qubits enlarge the decoherence-free sector and trap excitation away from the noisy collective mode.

For a Lorentzian spectral density
$$
J(\omega)=\frac{1}{2\pi}\frac{\gamma_0\lambda}{(\omega-\Omega_0)^2+\lambda^2},
$$
the effective spectral density becomes $\mathfrak{J}(\omega)=N J(\omega)$, and the exact amplitude is
$$
\tilde{C}_{0}(t)=e^{-\lambda t/2} \left[ \cosh\left(\frac{Dt}{2}\right) +\frac{\lambda}{D}\sinh\left(\frac{Dt}{2}\right) \right]\tilde{C}_{0}(0),
\qquad
D=\sqrt{\lambda^{2}-2N\gamma_{0}\lambda}.
$$
For Bob’s reduced dynamics, the Kraus operators contain
$$
p(t)=|\mathcal{G}(t)|^2,
$$
with
$$
\mathcal{G}(t)=e^{-i\Omega_0 t}\left[ \frac{N-1}{N} +\frac{e^{-\lambda t/2}}{N} \left( \cosh\left(\frac{Dt}{2}\right) +\frac{\lambda}{D}\sinh\left(\frac{Dt}{2}\right) \right) \right].
$$
As $N$ increases, the protected contribution $(N-1)/N$ becomes dominant.

The QSE semiaxes for the studied family are
$$
s_1=s_2=\frac{q\sqrt{p}\sin\theta}{\sqrt{1-q^2\cos^2\theta}},
\qquad
s_3=\frac{qp(1-q\cos^2\theta)}{1-q^2\cos^2\theta},
$$
so preserving $p(t)$ preserves the ellipsoid size and therefore the MSC. The mechanism operates in both Markovian and non-Markovian regimes. In the non-Markovian case, coherence and ellipsoid size can revive through information backflow; in the Markovian case, the ellipsoid contracts monotonically but more slowly under reservoir engineering.

## 4. Boundary geometry, tangential coherence, and the conversion of entanglement into steering

In "Boundary Geometry Turns Entanglement into Steering" [2605.21245], Co-STEER is a boundary-geometric mechanism: coherence at a product-null boundary contact turns entanglement into projective steering. For Alice-to-Bob projective steering, Alice measures
$$
P_t = |\xi_t\rangle\langle \xi_t|,\qquad  |\xi_t\rangle = \frac{|0\rangle+t|1\rangle}{\sqrt{1+t^2}},
$$
and Bob receives the unnormalized conditional state
$$
\sigma_t=\operatorname{tr}_A[(P_t\otimes I_B)\rho].
$$
In the Bob basis where the limiting boundary point is $|0\rangle\langle 0|$,
$$
\sigma_t= \begin{pmatrix} a_t & b_t\\ \bar b_t & d_t \end{pmatrix},
$$
with transverse displacement controlled by $b_t$ and inward defect by $d_t$:
$$
|\vec R_{t,\perp}|=2|b_t|,\qquad m_t-R_{t,z}=2d_t.
$$
The crucial scaling is
$$
|b_t|=O(t),\qquad d_t=O(t^2).
$$
The conditional state moves linearly along the tangent direction but only quadratically inward from the Bloch-sphere boundary.

The geometric obstruction targets finite-measure local-hidden-state models. If
$$
|\vec R_{t,\perp}|\ge Lt,\qquad m_t-R_{t,z}\le Ct^2,
$$
the paper proves that no finite hidden-state measure can reproduce the assemblage. The proof splits the Bloch ball into shrinking caps
$$
\Gamma_t=\Bigl\{\vec r\in\mathbb B^3:1-z\le \frac{K^2t^2}{2}\Bigr\}
$$
and uses
$$
|\vec r_\perp|^2\le 2(1-z)
$$
to show that the required first-order transverse weight cannot be supported by any finite measure as the caps shrink to the boundary point.

The boundary contact is locally equivalent to a product vector in the kernel:
$$
\rho(|\alpha\rangle\otimes|\beta\rangle)=0.
$$
This is also the geometric condition that Bob’s steering ellipsoid touches the Bloch sphere. In the standard product-null form,
$$
\rho=\frac{1}{\operatorname{tr}H} \begin{pmatrix}
h_{00}&0&h_{01}&h_{02}\\
0&0&0&0\\
\bar h_{01}&0&h_{11}&h_{12}\\
\bar h_{02}&0&\bar h_{12}&h_{22}
\end{pmatrix},
$$
the decisive quantity is the tangential coherence
$$
h_{02}=\langle 00|\rho|11\rangle.
$$
For the one-parameter family above,
$$
b_t=\frac{th_{02}+t^2h_{12}}{(\operatorname{tr}H)(1+t^2)},\qquad
d_t=\frac{t^2h_{22}}{(\operatorname{tr}H)(1+t^2)}.
$$
Thus $h_{02}\neq 0$ enforces the boundary-contact scaling obstruction.

The same $h_{02}$ controls NPT entanglement. In the partial transpose,
$$
\rho^{\Gamma_B} = \frac{1}{\operatorname{tr}H} \begin{pmatrix}
h_{00}&0&h_{01}&0\\
0&0&h_{02}&0\\
\bar h_{01}&\bar h_{02}&h_{11}&\bar h_{12}\\
0&0&h_{12}&h_{22}
\end{pmatrix},
$$
the $2\times 2$ principal minor on $\{|01\rangle,|10\rangle\}$ has determinant
$$
\det \begin{pmatrix} 0 & h_{02}\\ \bar h_{02} & h_{11} \end{pmatrix}=-|h_{02}|^2<0
$$
whenever $h_{02}\neq 0$. The paper proves, for the standard product-null class,
$$
\rho\text{ is entangled} \iff \rho\text{ is NPT} \iff h_{02}\neq0,
$$
and consequently obtains two principal classification results:
$$
\boxed{\text{Every entangled two-qubit rank-two state is two-way projectively steerable}}
$$
and
$$
\boxed{\text{Every entangled rank-three two-qubit state with product-null kernel vector is two-way projectively steerable}.}
$$

The compact witness proposed in the paper is correspondingly local in structure: verify a product-null vector $|\alpha\rangle\otimes|\beta\rangle\in\ker\rho$, then test
$$
M_{\alpha,\alpha_\perp}=(\langle\alpha|\otimes I)\rho(|\alpha_\perp\rangle\otimes I)
$$
and check whether
$$
M_{\alpha,\alpha_\perp}|\beta\rangle\neq0.
$$
The same support-kernel logic is extended to arbitrary steering cuts by replacing the pure-contact condition with a rank-deficient trusted conditional state and the scalar tangential coherence with the support-kernel coupling $P_{\mathrm{supp}A}BP_{\ker A}\neq0$.

## 5. Cooperative vibrational strong coupling and steering of nonequilibrium molecular dynamics

In "Steering Non-Equilibrium Molecular Dynamics in Optical Cavities" [2412.07593], Co-STEER is a cooperative vibrational strong-coupling mechanism in an open quantum system. The system contains a reactive molecular subsystem, a single optical cavity mode, an auxiliary molecular ensemble, and external environments. The cavity plus auxiliary ensemble form a bosonic Tavis-Cummings (BTC) subsystem, while the reactive molecules couple more weakly to that BTC background. The full Hamiltonian is
$$
H = H_{\mathrm{reac}} + H_{\mathrm{BTC}} + H_{\mathrm{coupl}},
$$
with
$$
H_{\mathrm{reac}}=\sum_j \frac{P_j^2}{2M_j}+E(R),
$$
$$
H_{\mathrm{BTC}}=\hbar \omega_{\mathrm{c}} a^\dagger a +\hbar \sum_{k=1}^{N}\left[\omega_{\mathrm{a}} b_k^\dagger b_k + g_k(a^\dagger b_k + a b_k^\dagger)\right],
$$
and
$$
H_{\mathrm{coupl}} =\sqrt{\frac{\hbar \omega_{\mathrm{c}}}{2\epsilon_0 V}}\, \hat{e}\cdot \mathbf{\mu}(R)\,(a^\dagger+a).
$$

The collective strong-coupling condition is
$$
\sqrt{N}\,g_{\mathrm{a}} > \kappa,\gamma_{\mathrm{a}},
$$
where $g_{\mathrm{a}}$ is the RMS auxiliary coupling strength. In this regime the BTC subsystem has two bright polariton branches and $N-1$ dark states, so the reactive subsystem sees a structured bath rather than a bare cavity mode.

After including noise and decay, elimination of the BTC variables yields an effective Langevin-type equation for the reactive coordinate,
$$
M\frac{d^2 \mathbf R}{dt^2} = -\nabla E(\mathbf R) -M\alpha\frac{d\mathbf R}{dt} +F_{\mathrm{FB}} +F_{\mathrm{ST}}.
$$
This is the central reduced description. The cavity and auxiliary ensemble generate two distinct emergent terms: an extra stochastic force with memory and a coherent feedback force. The stochastic contribution has a power spectral density
$$
S_{F_{\mathrm{ST}}F_{\mathrm{ST}}} = 2\alpha M k_B T + \chi^2 \sum_{k=l,u} \frac{|\xi_k|^2 \beta_k \bar n_k^{\mathrm{th}} \Gamma_k} {\Gamma_k^2/4 + (\omega-\omega_k)^2},
$$
which is explicitly colored. Without the auxiliary ensemble there is a single cavity peak; resonance with the ensemble splits it into two peaks; detuning shifts the effective noise peak. The cavity therefore converts the bare Markovian thermal bath into a non-Markovian bath whose spectral structure is tunable by the auxiliary ensemble.

If memory effects are neglected, the same cavity contribution may be summarized by an effective temperature
$$
T_{\mathrm{eff}} = T + \frac{2\chi^2}{\alpha M k_B} \sum_{k=l,u} |\xi_k|^2 \bar n_k^{\mathrm{th}} \Gamma_k,
$$
which the paper reports to grow nonlinearly with the number of auxiliary molecules. This reframes the cooperative bath as a frequency-selective heating channel rather than as a featureless temperature shift.

The coherent backaction is the second half of the mechanism. For a single-frequency molecular motion
$$
R^{(0)}(t)=\sqrt{\frac{\hbar}{2M\omega_0}}\sin(\omega_0 t),
$$
the feedback force becomes
$$
F_{\mathrm{FB}}^{(0)}(t) = -\chi^2 \sqrt{\frac{\hbar}{2M\omega_0}} \frac{1}{\hbar\alpha} A(\omega_0)\, \sin[\omega_0 t+\phi(\omega_0)].
$$
The phase shift is such that the work done over a cycle is always negative. The backaction therefore opposes molecular motion, suppresses high-energy vibrations, and shortens excitation lifetimes. In the nonequilibrium simulations reported in the paper, stronger single-molecule coupling generally accelerates thermalization, adding more auxiliary molecules can cancel that acceleration, and the crossover becomes sharper as the auxiliary-ensemble size grows. The chemical interpretation advanced by the paper is that Co-STEER modifies bond stability and reactivity by steering the thermalization pathway rather than by merely shifting equilibrium spectra.

## 6. Closed-loop phase control and transfer between bipartite and collective steering

In "Phase control of entanglement and quantum steering in a three-mode optomechanical system" [1706.04474], Co-STEER is a phase-controlled closed-loop mechanism. The system is a cavity containing a partially transmitting dielectric membrane, with optical modes $a_1$ and $a_2$ in the two subcavities and a mechanical mode $c$ for the membrane. The coupling graph is
$$
a_1 \leftrightarrow a_2,\qquad a_1 \leftrightarrow c,\qquad a_2 \leftrightarrow c,
$$
and the cavity is driven by short laser pulses with a controlled relative phase
$$
\delta\varphi_0=\varphi_{01}-\varphi_{02}.
$$
The Hamiltonian is
$$
\begin{aligned}
H &= \hbar\omega_{1}a^{\dag}_{1}a_{1}+\hbar\omega_{2}a^{\dag}_{2}a_{2}+\hbar\omega_{m}c^{\dag}c+\hbar J(a_{1}^{\dag}a_{2}+ a_{2}^{\dag}a_{1}) \\
&\quad +\hbar (g_{0,1}a^{\dag}_{1}a_{1} + g_{0,2}a^{\dag}_{2}a_{2})(c^{\dag}+c)
+i\hbar[E_{1}(t)a_{1}^{\dag} +E_{2}(t)a_{2}^{\dag} - {\rm H.c.}] .
\end{aligned}
$$

After linearization and diagonalization of the direct optical coupling, the relevant superposition modes are
$$
a_{w} = a_{1}\cos\theta +a_{2}\sin\theta,\qquad a_{u} = a_{1}\sin\theta  -a_{2}\cos\theta,
$$
with
$$
\cos^2\theta = \frac12+\frac{\Delta}{2w},\qquad w=\sqrt{J^{2}+\Delta^{2}}.
$$
Their effective couplings to the mechanics are
$$
g_{w} = g_{1}\cos\theta + g_{2}\sin\theta,\qquad g_{u} = g_{1}\sin\theta - g_{2}\cos\theta,
$$
and the relative phase $2\psi=\phi_{g_1}-\phi_{g_2}$ controls
$$
|g_{w}|^{2}=|g_{1}|^{2}\cos^{2}\theta +|g_{2}|^{2}\sin^{2}\theta + |g_{1}||g_{2}|\sin2\theta\cos2\psi,
$$
$$
|g_{u}|^{2}=|g_{1}|^{2}\sin^{2}\theta +|g_{2}|^{2}\cos^{2}\theta - |g_{1}||g_{2}|\sin2\theta\cos2\psi.
$$
If $J=0$, then $\sin2\theta=0$ and the phase sensitivity disappears. With $J\neq0$, interference between the closed-loop channels is unavoidable.

The same phase that redistributes optical population determines the steering channel. The field-mode populations satisfy
$$
\langle (A_j^{\rm out})^{\dag}A_j^{\rm out}\rangle =
\frac{\kappa \Upsilon(r)}{G(\kappa^{2} +w^{2})}|{\cal A}_{j}(\psi)|^{2},
$$
while the total population is conserved:
$$
\langle (A_1^{\rm out})^{\dag}A_1^{\rm out}\rangle+\langle (A_2^{\rm out})^{\dag}A_2^{\rm out}\rangle =
\frac{\kappa\left(|g_{1}|^{2}+|g_{2}|^{2}\right)}{G(\kappa^{2} +w^{2})}\Upsilon(r).
$$
Thus the phase does not change the total emitted population; it transfers it between channels. In the symmetric case, a suitable phase can make one collective mode bright and the other dark, or make one original cavity mode dominate the interaction with the mechanics. This is the basis for switching between collective and bipartite steering.

The first-order mutual coherence of the two optical outputs is perfect,
$$
\gamma_{12}^{(1)}=1,
$$
and likewise $\gamma_{wu}^{(1)}=1$. The paper emphasizes that this is induced coherence without induced emission: interference survives even when one output mode is unpopulated. Entanglement, however, is not between the two field modes; it is between the mechanics and either an individual optical mode or a collective superposition. Using the steering criterion
$$
E_{i|j}=\Delta_{inf,j}X_i\,\Delta_{inf,j}P_i<\frac12,
$$
the mechanics can be steered either by a single optical mode or by a collective mode. Only one of $E_{m|1}$ and $E_{m|2}$ can be below $\tfrac12$ at fixed phase, reflecting monogamy, but collective steering through $u$ or $w$ can occur while both bipartite criteria fail.

The proposed experimental signature is the coincidence rate
$$
R_{12}\sim \langle (A_1^{\rm out})^{\dagger}(A_2^{\rm out})^{\dagger}A_1^{\rm out}A_2^{\rm out}\rangle.
$$
For $\Delta=0$ and $|g_1|=|g_2|=g$,
$$
R_{12}=2\left[\frac{\kappa g^{2}\Upsilon(r)}{G(\kappa^{2} +J^{2})}\right]^{2}\left(1-\sin^{2}2\phi\sin^{2}2\psi\right).
$$
The interpretation is direct: minima of $R_{12}$ signal the bipartite steering regime, while maxima signal the collective steering regime. Co-STEER here is therefore a deterministic phase transfer of steering between individual and collective channels in a closed-loop interferometric optomechanical system.

Source: https://www.emergentmind.com/topics/co-steer-mechanism