--- title: Centered Trace-Dobrushin Coefficient in Quantum Channels url: https://www.emergentmind.com/topics/centered-trace-dobrushin-coefficient type: topic --- # Centered Trace-Dobrushin Coefficient in Quantum Channels The **centered trace-Dobrushin coefficient** is a Dobrushin-type contraction coefficient defined for positive trace-preserving maps on finite-dimensional matrix algebras by restricting the channel to the real trace-zero self-adjoint subspace and measuring its operator norm in trace norm. In the explicit terminology of the literature, this object is defined for products of quantum channels and used to quantify the residual dependence of the output on the input state; its decay is the criterion for trace-norm forgetting and for asymptotic replacement by a moving replacement channel [2605.00157]. Closely related antecedents appear in generalized Dobrushin coefficients relative to a projection on abstract state spaces and in cone-based noncommutative analogues of Dobrushin’s ergodicity coefficient, where the relevant norm becomes the trace norm in the quantum setting [2001.06266] [1302.5226]. ## 1. Formal definition and equivalent characterizations For a positive and trace-preserving map \(T:M_d(\mathbb C)\to M_d(\mathbb C)\), the centered trace-Dobrushin coefficient is defined on the real trace-zero self-adjoint subspace \[ H:=\{X=X^*\in M_d(\mathbb C):\operatorname{Tr}X=0\} \] by \[ \kappa_{\rm tr}(T):=\sup_{X\in H\setminus\{0\}}\frac{\|T(X)\|_1}{\|X\|_1}. \] Equivalently, it is the operator norm of \(T\) restricted to \((H,\|\cdot\|_1)\). For a product of channels \(\Phi_{t:s}:=\Phi_{t-1}\circ\cdots\circ\Phi_s\), the product coefficient is \[ \kappa_{t:s}:=\kappa_{\rm tr}(\Phi_{t:s}) . \] This formulation makes the adjective **centered** literal: the coefficient ignores multiples of the identity by working on the trace-zero sector, which is precisely the space of deviations from a state-independent output [2605.00157]. The coefficient admits several exact reformulations. For any positive trace-preserving \(T\), \[ \kappa_{\rm tr}(T)=\sup_{\substack{\rho,\sigma\in\mathcal S_d\\ \rho\neq\sigma}} \frac{\|T(\rho)-T(\sigma)\|_1}{\|\rho-\sigma\|_1} =\frac12\sup_{\rho,\sigma\in\mathcal S_d}\|T(\rho)-T(\sigma)\|_1 . \] It also satisfies \[ \kappa_{\rm tr}(T)=\frac12\sup\{\|T(E-F)\|_1:\ E,F\text{ orthogonal rank-one projections}\}. \] Hence \(0\le \kappa_{\rm tr}(T)\le 1\), and \[ \kappa_{\rm tr}(T)=0 \iff T(X)=\operatorname{Tr}(X)\,\tau \text{ for some state }\tau . \] Thus vanishing of the coefficient is equivalent to \(T\) being a **replacement channel**. For products, \[ \sup_{\rho,\sigma\in\mathcal S_d}\|\Phi_{t:s}(\rho)-\Phi_{t:s}(\sigma)\|_1 = 2\kappa_{t:s}, \] so \(2\kappa_{t:s}\) is exactly the trace-norm diameter of the evolved state set \(\Phi_{t:s}(\mathcal S_d)\) [2605.00157]. These identities place the coefficient simultaneously in operator-theoretic, probabilistic, and distinguishability-based frameworks. As an operator norm, it measures contraction on centered observables or centered states; as a state-space diameter, it measures the worst residual input dependence after evolution. ## 2. Deterministic channel products and asymptotic replacement For deterministic products of channels, the centered trace-Dobrushin coefficient is the exact product-level measure of residual memory. The defining asymptotic property is **trace-norm forgetting**: \[ \lim_{t\to\infty}\sup_{\rho,\sigma\in\mathcal S_d} \|\Phi_{t:s}(\rho)-\Phi_{t:s}(\sigma)\|_1=0, \] which is equivalent to \[ \kappa_{t:s}\to 0 . \] Small \(\kappa_{t:s}\) therefore means that all input states are sent to nearly the same output state [2605.00157]. A replacement channel with center \(\eta\in\mathcal S_d\) is \[ R_\eta(X):=\operatorname{Tr}(X)\eta . \] For a fixed reference state \(\tau\), the moving reference replacement channel is \[ R_{t:s}^{(\tau)}(X):=\operatorname{Tr}(X)\,\Phi_{t:s}(\tau). \] The key quantitative estimate is \[ \kappa_{t:s}\le \|\Phi_{t:s}-R_{t:s}^{(\tau)}\|_{1\to1}\le 4\,\kappa_{t:s}. \] Accordingly, decay of \(\kappa_{t:s}\) is equivalent, up to universal constants, to convergence of the product channel to a replacement channel [2605.00157]. The deterministic theory identifies four equivalent formulations for fixed \(s\): 1. \(\kappa_{t:s}\to 0\) as \(t\to\infty\). 2. \(\Phi_{t:s}\) is strongly asymptotically replacing from \(s\): there exist states \(\eta_{t:s}\) such that \[ \|\Phi_{t:s}-R_{t:s}\|_{1\to1}\to0,\qquad R_{t:s}(X)=\operatorname{Tr}(X)\eta_{t:s}. \] 3. For every reference state \(\tau\), \[ \|\Phi_{t:s}-R_{t:s}^{(\tau)}\|_{1\to1}\to0. \] 4. The same convergence holds for some reference state \(\tau\). The significance of these equivalences is structural rather than merely quantitative. They show that the coefficient is not only a bound on distinguishability contraction; it is the intrinsic criterion for whether a non-homogeneous channel product eventually behaves like a rank-one, state-forgetting evolution. ## 3. Pullback forgetting and boundary states In the two-sided setting indexed by \(\mathbb Z\), the relevant limit is pullback rather than forward evolution. The pullback memory-loss condition is \[ \lim_{s\to-\infty}\kappa_{t:s}=0 \qquad \text{for each fixed } t . \] Under this condition there exists a unique **pullback boundary state** \[ (\rho_t)_{t\in\mathbb Z}\subset\mathcal S_d \] satisfying \[ \rho_{t+1}=\Phi_t(\rho_t),\qquad \rho_t=\Phi_{t:s}(\rho_s). \] Moreover, \[ \rho_t=\lim_{s\to-\infty}\Phi_{t:s}(\tau) \] for every reference state \(\tau\), and the limit is independent of \(\tau\). The associated canonical replacement family is \[ R_t(X):=\operatorname{Tr}(X)\rho_t . \] Quantitatively, \[ \kappa_{t:s}\le \sup_{\sigma\in\mathcal S_d}\|\Phi_{t:s}(\sigma)-\rho_t\|_1 \le 2\kappa_{t:s}, \] and \[ \kappa_{t:s}\le \|\Phi_{t:s}-R_t\|_{1\to1}\le 4\kappa_{t:s}. \] Thus the remote-past limit canonically selects a moving replacement center and upgrades mere forgetting to a boundary-state theory [2605.00157]. This boundary-state interpretation is one of the main reasons the coefficient is useful beyond one-step contraction analysis. It identifies a dynamically consistent family of effective equilibria even when the channel sequence is time-dependent and non-stationary. ## 4. Random cocycles, Lyapunov exponents, and convergence rates For a measurable family of CPTP maps \(\Phi_\omega\) over an invertible probability-preserving system \((\Omega,\mathcal F,\mathbb P,\theta)\), define the cocycle \[ \Phi_\omega^{(n)}=\Phi_{\theta^{n-1}\omega}\circ\cdots\circ\Phi_\omega, \qquad \kappa_n(\omega):=\kappa_{\rm tr}(\Phi_\omega^{(n)}). \] The basic structural property is submultiplicativity: \[ \kappa_{n+m}(\omega)\le \kappa_n(\theta^m\omega)\,\kappa_m(\omega). \] Equivalently, \[ \log\kappa_{n+m}(\omega)\le \log\kappa_m(\omega)+\log\kappa_n(\theta^m\omega). \] By Kingman’s theorem, the **trace-Dobrushin Lyapunov exponent** \[ \lambda_{\rm tr}(\omega):=\lim_{n\to\infty}\frac1n\log\kappa_n(\omega)\in[-\infty,0] \] exists almost surely; it is \(\theta\)-invariant, and under ergodicity of \(\theta\) it is almost surely constant with \[ \lambda_{\rm tr}=\inf_{n\ge1}\frac1n\mathbb E[\log\kappa_n]. \] The paper proves the equivalence \[ \lambda_{\rm tr}(\omega)<0\ \text{a.s.} \iff \kappa_{\omega;n:0}\to0\ \text{a.s.} \iff \kappa_{\omega;0:-n}\to0\ \text{a.s.} \] This identifies negativity of the exponent with quenched trace-memory loss in both forward and pullback directions [2605.00157]. Assuming \(\lambda_{\rm tr}<0\) almost surely, there is a unique dynamically stationary random state \[ \rho:\Omega\to\mathcal S_d,\qquad \Phi_\omega(\rho_\omega)=\rho_{\theta\omega}, \] and there exist measurable \(\theta\)-invariant \(\beta(\omega)\in(-\infty,0)\) and almost surely finite random variables \(C_\beta^\pm\) such that \[ \|\Phi_{\omega;n:0}-R_{\theta^n\omega}\|_{1\to1} \le 4C_\beta^+(\omega)e^{\beta(\omega)n}, \] \[ \|\Phi_{\omega;0:-n}-R_\omega\|_{1\to1} \le 4C_\beta^-(\omega)e^{\beta(\omega)n}. \] The state-level estimates have constant \(2\) instead of \(4\). The coefficient therefore controls both existence and quantitative rate of replacement mixing [2605.00157]. A further refinement concerns annealed decay. If the stationary one-step channel process is \(\varrho\)-mixing, in the sense that its maximal-correlation profile \(\varrho_m\) tends to zero, then for every \(p\in\mathbb N\), \[ \mathbb E[\Delta_n^{\rm op}] \le C_p n^{-p}. \] If the one-step channels are independent, then \[ \mathbb E[\Delta_n^{\rm op}] \le Ce^{-\gamma n}. \] This shows that the same coefficient supports both quenched and annealed asymptotic regimes. ## 5. Relation to earlier Dobrushin frameworks Before the explicit terminology of centered trace-Dobrushin coefficients for channel products, several closely related Dobrushin constructions had already isolated the same contraction mechanism in different languages. The following comparison captures the main line of descent. | Setting | Quantity | Relation | |---|---|---| | Product of CPTP maps | \(\kappa_{\rm tr}(T)\) on \(H=\{X=X^*:\operatorname{Tr}X=0\}\) | Explicit named centered trace-Dobrushin coefficient [2605.00157] | | Abstract state spaces | \(\delta_P(T)=\sup_{x\in N_P,\ x\neq0}\|Tx\|/\|x\|\), \(N_P=\ker P\) | Projection-centered Dobrushin coefficient; trace-class settings included [2001.06266] | | Cones and quantum channels | Hopf/Hilbert contraction coefficient with dual trace norm in the matrix case | Noncommutative analogue of Dobrushin’s ergodicity coefficient [1302.5226] | | Energy-constrained quantum channels | \(F_E(\delta)\) | Trace-norm constrained refinement of the ordinary quantum Dobrushin coefficient [1808.03182] | | Measures on countable products | \(D(\mu,\nu)=d(\mu,\nu)\) | Oscillation and coupling formulations coincide exactly [2104.08365] | In the abstract-state-space setting, the generalized Dobrushin coefficient relative to a projection \(P\) is \[ \delta_P(T)=\sup_{x\in N_P,\;x\neq 0}\frac{\|Tx\|}{\|x\|}, \qquad N_P:=\{x\in X:\ Px=0\}. \] When \(P\) is a rank-one Markov projection, this recovers the classical coefficient. In noncommutative \(L^1\)-spaces and trace-class settings, the same theory applies, so \(N_P\) plays the role of the centered subspace of deviations from equilibrium [2001.06266]. This suggests that the later centered trace-Dobrushin coefficient is a concrete finite-dimensional quantum realization of a more abstract projection-centered paradigm. The cone-based formulation sharpens the same point from a dual viewpoint. There the Dobrushin-type quantity is the contraction rate with respect to Hopf’s oscillation seminorm, which is a quotient norm modulo the line \(\mathbb Re\). In the matrix setting, the dual norm becomes the trace norm, the dual simplex becomes the set of density matrices, and the resulting noncommutative coefficient is explicitly presented as the quantum analogue of Dobrushin’s ergodicity coefficient [1302.5226]. The centering mechanism is therefore quotienting out the trivial direction generated by the unit, while the trace aspect enters through the dual norm. For a single quantum channel, the ordinary trace-norm Dobrushin coefficient is \[ \eta(\Phi)=\sup_{\rho_0,\rho_1} \frac{\|\Phi(\rho_0)-\Phi(\rho_1)\|_1}{\|\rho_0-\rho_1\|_1}. \] The corresponding unconstrained Dobrushin curve is linear: \[ f(\delta)=\frac{\delta}{2}\eta(\Phi). \] Because this unconstrained curve is trivial, the energy-constrained Dobrushin curve \[ F_E(\delta)=\sup_{\substack{\rho_0,\rho_1\in\mathcal G_E\\ \|\rho_0-\rho_1\|_1\le \delta}} \|\Phi(\rho_0)-\Phi(\rho_1)\|_1 \] becomes the nontrivial refinement [1808.03182]. This is not called centered, but it belongs to the same family of trace-distance contraction quantities. The measure-theoretic equality \[ D(\mu,\nu)=d(\mu,\nu) \] on countable products of Polish spaces provides another instructive analogue: a Dobrushin metric defined through coordinatewise oscillation/Lipschitz control coincides exactly with a coupling metric defined through optimal joinings [2104.08365]. A plausible implication is that “oscillation” and “trace-like” formulations are often two representations of the same contraction phenomenon rather than genuinely different invariants. ## 6. Applications to inhomogeneous matrix product states and terminological scope The coefficient transfers directly to inhomogeneous matrix product states in left-canonical CPTP gauge, where the auxiliary transfer maps are channels. Writing the right-tail transfer product as \[ \Theta_{m,n}:=\Phi_{m+1}\circ\cdots\circ\Phi_n, \] the condition \[ \kappa_{\rm tr}(\Theta_{q,n})\to0 \qquad (n\to\infty) \] for every fixed \(q\) implies the existence of a unique right boundary sequence \((\rho_r)_{r\ge1}\) satisfying \[ \rho_r=\Phi_r(\rho_{r+1}), \] and a unique infinite-volume state \(\varphi_\infty\) on the quasi-local algebra given locally by \[ \varphi_\infty(X)=\operatorname{Tr}\bigl(\widehat X_{[1,m]}(\rho_{m+1})\bigr), \qquad X\in\mathcal A_{[1,m]}. \] For trace-closed finite-volume MPS vectors \(\Psi_n\), the normalized finite-volume expectations converge to \(\varphi_\infty\), with quantitative estimate \[ |\varphi_n(X)-\varphi_\infty(X)| \le 16\,D_{\mathcal H}^2\,\|X\|_\infty\,\kappa_{\rm tr}(\Theta_{m,n}). \] For separated local observables \(A\in\mathcal A_{[p,q]}\) and \(B\in\mathcal A_{[r,s]}\) with \(q+1