Joint Measurements and Spacelike Composition

Two spacelike measurements admit an order-independent joint instrument when their complete localized operations causally factor. The word complete matters: commuting POVM effects guarantee compatible current probabilities, but do not determine the state updates, nonselective channels, or physical supports. This page derives the composition rule, checks it in a correlated two-mode field benchmark, and then constructs a control in which all effects commute while the updates do not.

Required background. Causal quantum channels supplies the localization and composition criteria.

Helpful background. Operational locality supplies the supported spacetime protocol and record graph.

Chapter map. Use the route from a local coupling to a field instrument, the chapter claim-validity table, and the three independent validity questions to place this construction within the full measurement protocol.

Branch maps and order independence

Let $\{\mathcal I_x^A\}$ and $\{\mathcal J_y^B\}$ be instruments induced by couplings supported in spacelike regions $K_A$ and $K_B$. Their branch maps are completely positive and trace-nonincreasing. If

$$\mathcal J_y^B\circ\mathcal I_x^A =\mathcal I_x^A\circ\mathcal J_y^B \quad\text{for every }x,y,$$

then

$$p(x,y\mid\rho) =\operatorname{tr}\!\left[ (\mathcal J_y^B\circ\mathcal I_x^A)(\rho) \right]$$

is independent of the arbitrary mathematical ordering. The common branch is itself completely positive. Summing over one result gives

$$\sum_y p(x,y\mid\rho) =\operatorname{tr}\mathcal I_x^A(\rho),$$

because $\mathcal E_B=\sum_y\mathcal J_y^B$ is trace preserving. Thus the local marginal does not depend on whether the remote, nonselective measurement was performed.

Order independence does not imply statistical independence. An initially correlated field state can have

$$p(x,y)\ne p_A(x)p_B(y)$$

while both marginals remain unchanged. Correlation becomes a usable message only if one setting changes the remote marginal without a classical record being sent.

For compactly supported system–probe couplings, the physical input is causal factorization of the scattering maps. Fewster and Verch 2020, Theorem 3.5, Eqs. (3.28)–(3.30) proves that the composition of the individual pre-instruments equals the combined pre-instrument and is independent of order when the coupling regions are causally disjoint. The support in this theorem is the interaction region, not the later spacetime point at which the two classical records are compared.

Correlated two-mode benchmark

Use a split or lattice regulator with two commuting wavepacket-mode algebras $A$ and $B$. Let

$$R=(q_A,p_A,q_B,p_B)^T, \qquad [q_j,p_k]=i\delta_{jk},$$

and prepare a two-mode squeezed vacuum with $r=1/2$. Its covariance matrix is

$$V=\frac12 \begin{pmatrix} c&0&s&0\\ 0&c&0&-s\\ s&0&c&0\\ 0&-s&0&c \end{pmatrix}, \qquad c=\cosh(2r), \quad s=\sinh(2r).$$

Measure $q_A$ and $q_B$ with identical Gaussian resolution $\Delta=1/2$,

$$M_x^A=(2\pi\Delta^2)^{-1/4} \exp\!\left[-\frac{(x-q_A)^2}{4\Delta^2}\right],$$ $$N_y^B=(2\pi\Delta^2)^{-1/4} \exp\!\left[-\frac{(y-q_B)^2}{4\Delta^2}\right].$$

Because $[q_A,q_B]=0$, $M_x^A$ commutes with $N_y^B$ for every $x,y$. Hence the selective maps, their nonselective sums, and the joint density are exactly order independent. The recorded vector $(x,y)$ is a centered bivariate Gaussian with covariance

$$\Sigma= \begin{pmatrix} v&k\\ k&v \end{pmatrix}, \qquad v=\frac12\cosh(1)+\frac14, \qquad k=\frac12\sinh(1).$$

Numerically,

Quantity	Value
Marginal variance $v$	$1.0215403174$
Outcome covariance $k$	$0.5876005968$
Correlation coefficient $k/v$	$0.5752103826$
$\det\Sigma$	$0.6982701587$
Joint density $p(0,0)=1/(2\pi\sqrt{\det\Sigma})$	$0.1904620219$

Computing $N_y^BM_x^A\rho M_x^AN_y^B$ and $M_x^AN_y^B\rho N_y^BM_x^A$ gives the same operator before normalization, not merely the same trace. The nonzero $k$ is shared-state correlation; it does not select a preferred spacelike order and does not change either marginal.

Commuting effects with noncommuting updates

Now deliberately remove the localization guarantee. On one qubit define two binary instruments

$$\mathcal I_x(\rho)=\frac12U_x\rho U_x^\dagger, \qquad \mathcal J_y(\rho)=\frac12V_y\rho V_y^\dagger,$$

with

$$U_0=V_0=\mathbf1, \qquad U_1=e^{-i\pi\sigma_x/4}, \qquad V_1=e^{-i\pi\sigma_z/4}.$$

Every effect is $\mathbf1/2$, so all effects commute and every joint outcome has probability $1/4$ in either order. Nevertheless the $x=y=1$ branches do not commute. Starting from $|0\rangle$,

$$V_1U_1|0\rangle\sim|{+x}\rangle, \qquad U_1V_1|0\rangle\sim|{-y}\rangle.$$

The normalized outputs have overlap squared $1/2$, hence trace distance

$$D=\sqrt{1-|\langle +x|-y\rangle|^2} =\frac{1}{\sqrt2}.$$

This control is not a model of a legitimate spacelike apparatus: both updates act on the same algebra. That is precisely why it is useful. It proves that commuting effects, identical current probabilities, and even no visible discrepancy in a one-shot record do not certify order-independent state updates. The distinction between causal and physically localizable operations is developed in Beckman et al. 2001, § II, Eqs. (1)–(4) and (15)–(16).

What a spacelike-composition test must verify

For an actual detector pair, compute both orderings with the same initial state, switching and smearing functions, cutoffs, integration grids, and perturbative order. Report

$$\delta_{xy} =p_{A\prec B}(x,y)-p_{B\prec A}(x,y)$$

together with a numerical and truncation uncertainty. Also compare the branch outputs or a spanning set of later observables; $\delta_{xy}=0$ alone can miss the qubit failure above. The test should include:

• disjointness of the full coupling supports, including switching and smearing tails;
• equality of every selective branch in both orders, not only commutation of effects;
• equality of the nonselective channels and invariance of remote marginals;
• formation of a joint classical record only in the common causal future;
• a seeded support overlap or nonlocal update that the test demonstrably detects.

The causal analysis of Bostelmann, Fewster, and Ruep 2021, Theorem 2, Eqs. (11)–(12) shows why locality of the system–probe coupling, rather than an ideal measurement label, protects later spacelike observables from signaling.

Limitations

The Gaussian calculation is an exact benchmark of two regulated commuting mode algebras. It does not prove a split property for arbitrary continuum regions, nor that every abstract Gaussian instrument has a compactly supported field-theoretic dilation. In a continuum application, causal factorization must be established for the actual coupling. Gaussian switching or smearing has tails, so it supports only a quantitative approximate-separation statement unless those tails are bounded.

Common pitfalls

Confusing correlation with signaling. Vacuum or squeezed-state correlations can make the records strongly dependent. Signaling concerns a change in one marginal under a remote choice.

Checking only the POVM. The qubit control has commuting effects and identical joint probabilities but order-dependent posterior states. Test the complete branch maps.

Locating the record instead of the coupling. Moving a data file does not move the physical intervention. Causal support is determined by the entire system–probe interaction.

Exercises

Assume every branch map commutes across $A$ and $B$. Show that the nonselective maps commute and that both orderings give the same marginals.

Solution

Sum $\mathcal I_x^A\mathcal J_y^B=\mathcal J_y^B\mathcal I_x^A$ over $x$ and $y$ to obtain $\mathcal E_A\mathcal E_B=\mathcal E_B\mathcal E_A$. Summing the common joint distribution over $y$ gives

$$\sum_y\operatorname{tr}\mathcal J_y^B\mathcal I_x^A(\rho) =\operatorname{tr}\mathcal E_B\mathcal I_x^A(\rho) =\operatorname{tr}\mathcal I_x^A(\rho),$$

because $\mathcal E_B$ is trace preserving. The other marginal follows symmetrically.

Reproduce the numerical covariance table for $r=\Delta=1/2$ and verify the quoted density at the origin.

Solution

Using $\cosh 1=1.5430806348$ and $\sinh 1=1.1752011936$ gives

$$v=\frac{1.5430806348}{2}+0.25=1.0215403174,$$ $$k=\frac{1.1752011936}{2}=0.5876005968.$$

Then $v^2-k^2=0.6982701587$, $k/v=0.5752103826$, and $1/(2\pi\sqrt{v^2-k^2})=0.1904620219$.

Verify the adversarial trace distance for the $x=y=1$ qubit branches.

Solution

$U_1|0\rangle=(|0\rangle-i|1\rangle)/\sqrt2=|-y\rangle$. Applying $V_1$ changes the relative phase and gives $|+x\rangle$ up to a global phase, whereas applying $V_1$ first changes only the phase of $|0\rangle$, so $U_1V_1|0\rangle\sim|-y\rangle$. The Bloch vectors $+x$ and $-y$ are orthogonal, so their state overlap has squared magnitude $1/2$. The pure-state trace distance is therefore $\sqrt{1-1/2}=1/\sqrt2$.

References

• Beckman, D., Gottesman, D., Nielsen, M. A., and Preskill, J. (2001). “Causal and Localizable Quantum Operations.” Physical Review A 64, 052309. DOI. Open PDF.
• Bostelmann, H., Fewster, C. J., and Ruep, M. H. (2021). “Impossible Measurements Require Impossible Apparatus.” Physical Review D 103, 025017. DOI. Open PDF.
• Fewster, C. J., and Verch, R. (2020). “Quantum Fields and Local Measurements.” Communications in Mathematical Physics 378, 851–889. DOI. Open PDF.