Algebraic Quantum Channels and Localized Operations

A localized quantum operation is identified by its action on the observable net, not by the appearance of a Kraus list in one Hilbert-space representation. For a nonselective operation supported in a compact region $K$, every observable in the causal complement must remain fixed. A compact system–probe dilation can explain why that condition holds, but the existence of an abstract completely positive map alone does not guarantee such a physical realization.

Required background. Signaling and causal composition supplies the operational criterion that localization must protect.

Helpful background. Local regions and algebras supplies the net $O\mapsto\mathcal A(O)$.

For visual orientation, use the chapter’s canonical sender–field–receiver task map, protocol table, and localization failure-control map. The algebraic tests below replace neither diagram; they supply the map-level statements that the diagrams summarize.

Operations on a local net

Let $O\mapsto\mathcal A(O)$ be an isotonic net of von Neumann algebras in a fixed representation. Write $\mathfrak A$ for the norm-closed quasilocal C*-algebra and $\mathcal M=\mathfrak A''$ for its represented von Neumann closure. We use the Heisenberg picture. A nonselective channel on normal states is dual to a normal, unital, completely positive map

$$\mathcal T:\mathcal M\longrightarrow\mathcal M.$$

Normality means preservation of suprema of bounded increasing nets of positive observables; operationally, it ensures countable additivity for normal states and POVMs. Complete positivity protects positivity when an arbitrary spectator system is added. Unitality is dual to trace preservation.

The basic action criterion for localization in $K$ is

$$\mathcal T(B)=B, \qquad B\in\mathcal A(K^\perp).$$

This statement is representation independent once the algebra and its inclusion into the net are fixed: it quantifies over all observables in the causal complement. It immediately gives no-signaling, because every normal state $\omega$ obeys

$$(\omega\circ\mathcal T)(B)=\omega(B), \qquad B\in\mathcal A(K^\perp).$$

There is also useful joint control. Since $\mathcal T$ fixes the complementary algebra pointwise, each of its unitaries lies in the multiplicative domain of $\mathcal T$. Consequently,

$$\mathcal T(BA)=B\mathcal T(A), \qquad \mathcal T(AB)=\mathcal T(A)B,$$

for $A\in\mathcal M$ and $B\in\mathcal A(K^\perp)$. To see why, apply the Schwarz inequality for unital CP maps to a complementary unitary $U$: both $\mathcal T(U^*U)$ and $\mathcal T(U)^*\mathcal T(U)$ equal $\mathbf1$, so equality in Schwarz puts $U$ in the multiplicative domain. Linear combinations of unitaries then cover the complementary algebra. Identity action therefore controls cross-observables as well as isolated marginals.

The criterion is deliberately one-sided. It says that an operation supported in $K$ cannot alter the causal complement, but it does not require every observable changed by the operation to lie in $\mathcal A(K)$. Dynamics can carry the effect into the causal future, and the time-slice property can give one observable several localization regions. Nor should one silently identify $\mathcal A(K^\perp)$ with the commutant $\mathcal A(K)'$: locality gives the inclusion $\mathcal A(K^\perp)\subseteq\mathcal A(K)'$, while equality is Haag duality and is an additional hypothesis.

Identity action is therefore a necessary operational localization test, but a claim of physical implementability in $K$ normally asks for more. One may require compatibility with the net under enlarging $K$, causal factorization with independently supported operations, and a system–probe construction whose coupling really has support in $K$. These requirements separate three questions beginners often merge: whether a map is CP, whether it is no-signaling to a complement, and whether a compact apparatus can realize it. The implications depend on split, funnel, additivity, or duality assumptions; they are not automatic consequences of finite-dimensional Kraus theory.

For a selective outcome $x$, the operation $\mathcal I_x$ is normal and CP but generally subunital. The sum $\sum_x\mathcal I_x$ is the nonselective localized channel. Demanding $\mathcal I_x(B)=B$ for each branch would usually be wrong: even $\mathcal I_x(\mathbf1)$ records the outcome probability. This is the algebraic version of the postselection warning on the preceding page.

Local dilations and their hypotheses

Stinespring’s theorem represents a normal CP map as

$$\mathcal T(A)=V^*\pi(A)V$$

for a representation $\pi$ and an isometry $V$; see Stinespring 1955, Theorem 1, pp. 212–214. This is a structural theorem, not a spacetime-locality theorem. It does not say that the ancillary system is a realizable probe, that the interaction is compactly supported in $K$, or that the dilation respects causal factorization.

A stronger physical construction specifies a probe algebra, an initial probe state, and a system–probe coupling supported in $K$. The associated scattering morphism induces a normal CP operation and yields causal composition when the dynamics obeys the required factorization. Fewster and Verch 2020, §§ 3–5 formulate this construction for local covariant QFT. For general von Neumann-algebra instruments, realizability needs an extension property; Okamura and Ozawa 2016, §§ III–VI show why not every abstract instrument automatically comes from a measuring process and how the split property enters local extensions.

Likewise, equivalence with a local Kraus approximation requires additional net hypotheses. Kitajima 2017, Theorem 15 and Corollary 16 uses the funnel property to obtain the relevant approximation. These qualifications matter in QFT because local algebras are generally not finite-dimensional matrix factors.

First QFT benchmark: compact Gaussian noise on an interval

Consider a free scalar field in a regular representation. Let $W(f)$ be the Weyl observable associated with a real test function $f$, with convention

$$W(f)W(g) =e^{-i\sigma(f,g)/2}W(f+g), \qquad W(f)^*=W(-f).$$

Here $\sigma$ is the Klein–Gordon symplectic form. Choose a smooth real $h$ with compact support in a spacetime region $K$ whose Cauchy-surface base is an interval $I$. For $\nu\geq0$, average local Weyl displacements with a centered Gaussian of variance $\nu$:

$$\mathcal T_\nu(A) =\int_{-\infty}^{\infty} \frac{d\xi}{\sqrt{2\pi\nu}} e^{-\xi^2/(2\nu)} W(\xi h)^* A W(\xi h).$$

At $\nu=0$ the integral means the identity channel. For $\nu>0$ it is a weak-operator integral of unitary conjugations, hence normal, unital, and CP. Weyl multiplication gives

$$W(\xi h)^*W(f)W(\xi h) =e^{i\xi\sigma(h,f)}W(f),$$

and the elementary Gaussian characteristic function therefore yields

$$\mathcal T_\nu(W(f)) =\exp\!\left[-\frac{\nu}{2}\sigma(h,f)^2\right]W(f).$$

This is a Gaussian channel defined directly on the field algebra. If $\operatorname{supp}f\subset K^\perp$, causal propagation gives $\sigma(h,f)=0$, so $\mathcal T_\nu(W(f))=W(f)$ exactly. Since the spacelike Weyl operators generate the complementary algebra, the channel passes the identity-action test. Compact support is doing real work: replacing $h$ by a Gaussian profile would provide only approximate localization controlled by its tails.

For a one-mode regulator with $[Q,P]=i$, choose $W(\xi h)=e^{i\xi Q}$. Then $Q$ is unchanged while $P$ receives a classical random shift of variance $\nu$. In the ordering $R=(Q,P)^T$,

$$m\longmapsto m, \qquad V\longmapsto V+ \begin{pmatrix} 0&0\\ 0&\nu \end{pmatrix}.$$

The Gaussian CP condition is immediate because $X=\mathbf1$ and the added-noise matrix is positive semidefinite.

The benchmark is reproducible with $\nu=0.36$. Evaluate the attenuation for three test functions satisfying $c=\sigma(h,f)$:

$$\begin{array}{c|ccc} c&0&1&2\\ \hline e^{-\nu c^2/2}&1&0.835270&0.486752 \end{array}$$

The $c=0$ column is the spacelike-complement control. The other columns quantify overlap with the affected canonical direction. A numerical implementation should reproduce these six-digit values, preserve $V+i\Omega/2\geq0$, and verify exact identity on compactly spacelike test functions before any continuum claim is made.

Kraus descriptions do not define localization

In a type-I regulator, a channel may be written as $\mathcal T(A)=\sum_jK_j^*AK_j$. A second Kraus family related by an isometry, $L_\alpha=\sum_j u_{\alpha j}K_j$, gives the same map. Within one faithful tensor-product representation, isometric mixing of operators in $\mathcal B(\mathcal H_K)\otimes\mathbf1$ keeps them in that same local operator space. Thus a purported “nonlocal Kraus rewrite” of the same full matrix-algebra channel must have changed either the representation or the algebra being tested.

That observation supplies a useful adversarial example. Amplify a representation by an auxiliary space $\mathcal H_R$:

$$\pi(A)=A\otimes\mathbf1_R.$$

If $\{K_j\}$ represents $\mathcal T$ and $V_R$ is any unitary, define $L_j=K_j\otimes V_R$. Then

$$\sum_j L_j^*\pi(A)L_j =\mathcal T(A)\otimes\mathbf1_R.$$

The $L_j$ act on the extra factor and may look delocalized in the enlarged Hilbert-space diagram, yet they define exactly the same map on the represented observable algebra. They define a different extension to all of $\mathcal B(\mathcal H\otimes\mathcal H_R)$. Localization belongs to the algebraic map and declared net, not to unsupported labels attached to its implementers.

Now make the opposite failure. In a two-cell regulator, append a remote unitary $X_B$ to every otherwise local Kraus operator. The resulting map still passes complete positivity and normalization, but

$$\widetilde{\mathcal T}(Z_B)=X_BZ_BX_B=-Z_B.$$

It therefore fails localization immediately. A Kraus normalization check cannot replace the causal-complement action test.

Common pitfalls

Fixing one remote observable. Accidental commutation with one $B$ says nothing about the rest of $\mathcal A(K^\perp)$. Test a generating family or prove pointwise identity algebraically.

Treating every Stinespring space as apparatus. Stinespring dilation proves complete positivity has a Hilbert-space representation. A compact physical realization needs additional localization, covariance, and causal-factorization data.

Hiding a regulator. A mode partition can realize an interval channel at finite cutoff, but it is not the continuum definition. State the cutoff, support or tail bound, representation, and topology used to remove it.

Exercises

1. Derive the Weyl attenuation

Starting from the Weyl relations, derive $\mathcal T_\nu(W(f))$.

Solution

Twice applying the Weyl product gives

$$W(\xi h)^*W(f)W(\xi h) =e^{i\xi\sigma(h,f)}W(f).$$

The centered Gaussian characteristic function is $\mathbb E(e^{i\xi c})=e^{-\nu c^2/2}$. Taking $c=\sigma(h,f)$ proves the stated formula.

2. Check the one-mode covariance

Let $U_\xi=e^{i\xi Q}$ with $[Q,P]=i$. Show that averaging $U_\xi^*(\cdot)U_\xi$ over variance $\nu$ adds $\nu$ only to $V_{PP}$.

Solution

The Baker–Campbell–Hausdorff series terminates:

$$U_\xi^*QU_\xi=Q, \qquad U_\xi^*PU_\xi=P+\xi.$$

The Gaussian has zero mean, so first moments are unchanged. Its independent classical variance adds $\mathbb E(\xi^2)=\nu$ to $V_{PP}$ and nothing to the other covariance entries.

3. Distinguish an amplified representation from a remote action

Why do $L_j=K_j\otimes V_R$ leave the abstract map unchanged, while multiplying local Kraus operators by a physical $X_B$ fails localization?

Solution

In the amplified representation the declared observable algebra contains only $A\otimes\mathbf1_R$, and $V_R$ commutes with every represented observable. Hence it cancels in $L_j^*\pi(A)L_j$. A physical receiver algebra contains $Z_B$, and $X_B$ does not commute with it: $X_BZ_BX_B=-Z_B$. The first construction changes an implementation outside the declared algebra; the second changes the algebraic map itself.

References

• Fewster, C. J., and Verch, R. (2020). “Quantum Fields and Local Measurements.” Communications in Mathematical Physics 378, 851–889. DOI. Open PDF.
• Kitajima, Y. (2017). “Local Operations and Completely Positive Maps in Algebraic Quantum Field Theory.” arXiv:1704.01229.
• Okamura, K., and Ozawa, M. (2016). “Measurement Theory in Local Quantum Physics.” Journal of Mathematical Physics 57, 015209. DOI. Open PDF.
• Stinespring, W. F. (1955). “Positive Functions on C*-Algebras.” Proceedings of the American Mathematical Society 6, 211–216. DOI. Open PDF.