Local Measurement Instruments in QFT

A POVM answers only “how likely is each outcome?” A measurement instrument also answers “what operation occurred on that branch?” In QFT the distinction is decisive: two devices can have identical readout statistics but different disturbance, energy injection, or causal support. This page constructs a Gaussian unsharp instrument for a compactly smeared field observable and checks complete positivity, normalization, and locality independently.

Required background. Influence functionals supplies the system–environment reduction viewpoint. System–probe scattering derives an instrument from a localized coupling.

Helpful background. Symmetry-constrained operations explains when charge or reference-frame restrictions narrow the admissible instrument set.

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 distinguish probability, update, and localization claims.

Effects and operations are different data

In a type-I or regulated Schrödinger picture, an instrument is a countably additive family $X\mapsto\mathcal I_X$ of completely positive, trace-nonincreasing maps. For an input state $\rho$,

$$p_\rho(X)=\operatorname{tr}\mathcal I_X(\rho), \qquad \rho_X=\frac{\mathcal I_X(\rho)}{p_\rho(X)}$$

when $p_\rho(X)>0$. The total operation $\mathcal I_\Omega$ is trace preserving. Its Heisenberg dual satisfies

$$E(X)=\mathcal I_X^*(\mathbf1), \qquad p_\rho(X)=\operatorname{tr}[\rho E(X)], \qquad \mathcal I_\Omega^*(\mathbf1)=\mathbf1.$$

The effects $E(X)$ form the POVM. They determine the current outcome probabilities, but they do not determine $\mathcal I_X^*(A)$ for $A\ne\mathbf1$. This measure-valued operation is the instrument introduced in Davies and Lewis 1970, pp. 243–246.

In an algebraic QFT representation one normally works directly with normal CP maps on a local von Neumann algebra and its normal states. A density matrix for the entire continuum theory need not exist; the regulated formulas below are a computational realization, not a claim that a type-III local algebra factorizes.

Gaussian unsharp measurement of a smeared field

Let $f\in C_0^\infty(M)$ be real with $\operatorname{supp}f\subset K$, and let

$$Q=\Phi(f)$$

be self-adjoint in the chosen regulated representation. For resolution $\delta>0$, define the outcome-density Kraus operator

$$M_x=(2\pi\delta^2)^{-1/4} \exp\!\left[-\frac{(x-Q)^2}{4\delta^2}\right], \qquad x\in\mathbb R,$$

and the operation density

$$\mathfrak i_x(\rho)=M_x\rho M_x.$$

The instrument itself is indexed by Borel events $X\subset\mathbb R$:

$$\mathcal I_X(\rho) =\int_X dx\,\mathfrak i_x(\rho), \qquad E(X)=\int_X dx\,M_x^2.$$

The first integral is taken in trace norm in the regulated Schrödinger realization; equivalently, the dual integral is ultraweak. Each event map is CP and trace nonincreasing. The pointwise quantity $p_\rho(x)=\operatorname{tr}\mathfrak i_x(\rho)$ is a probability density, not the probability of the singleton $\{x\}$. A point-conditioned posterior $\mathfrak i_x(\rho)/p_\rho(x)$ is consequently a regular conditional state defined only almost everywhere; finite experimental records condition on bins $X$ and use $\mathcal I_X(\rho)/p_\rho(X)$.

To check normalization, use the spectral measure $P_Q(dq)$:

$$\begin{aligned} \int_{-\infty}^{\infty}dx\,M_x^2 &=\int P_Q(dq)\, \frac{1}{\sqrt{2\pi\delta^2}} \int_{-\infty}^{\infty}dx\, e^{-(x-q)^2/(2\delta^2)}\\ &=\int P_Q(dq)=\mathbf1. \end{aligned}$$

This is an operator identity, so it proves $\mathcal I_{\mathbb R}$ is trace preserving and $0\le E(X)\le\mathbf1$ for every event. It is the Gaussian specialization of the event-valued strictly local instrument written in Okamura and Ozawa 2016, Eq. (101), PDF.

The effect density is $E_x=M_x^2$. If the sharp distribution of $Q$ is Gaussian with mean $\mu$ and variance $v$, the detector distribution is the convolution

$$p(x)=\frac{1}{\sqrt{2\pi(v+\delta^2)}} \exp\!\left[-\frac{(x-\mu)^2}{2(v+\delta^2)}\right].$$

Thus the apparatus adds the declared noise variance $\delta^2$ rather than silently replacing the field variance. For the reproducible benchmark

$$\mu=0, \qquad v=1, \qquad \delta=\frac12,$$

the output variance is $5/4$ and

$$\Pr(|x|\le1) =\operatorname{erf}\!\sqrt{\frac25} =0.628906\ldots.$$

The error-function value is analytic; the decimal is rounded rather than statistically estimated. In a numerical field calculation, disagreement with it separates quadrature or cutoff error from the physical detector resolution $\delta$.

The exact nonselective disturbance is visible in the $Q$ spectral representation:

$$\langle q|\mathcal I_\Omega(\rho)|q'\rangle =\exp\!\left[-\frac{(q-q')^2}{8\delta^2}\right] \langle q|\rho|q'\rangle.$$

Diagonal $Q$ statistics are unchanged, while coherences between widely separated $Q$ values are suppressed. The sharp limit $\delta\to0$ is therefore not harmless: it produces strong conjugate disturbance and can have unbounded domain or energy cost.

Locality is checked on complementary observables

Let $B$ belong to an algebra causally disjoint from $K$. Microcausality gives $[B,Q]=0$, hence $[B,M_x]=0$. The operation density obeys

$$\mathfrak i_x^*(B)=M_xBM_x=BE_x, \qquad E_x=M_x^2.$$

Integrating gives $\mathcal I_X^*(B)=BE(X)$ for every event $X$. This is not generally $p_\rho(X)B$ as an operator, so conditioning can change the expectation of $B$ when the initial state contains spacelike correlations. After outcomes are ignored, however,

$$\mathcal I_\Omega^*(B) =B\int dx\,E_x=B.$$

The nonselective intervention therefore has identity action on the causal complement. Probe-induced selective and nonselective updates, including precisely this correlation-versus-signaling distinction, are derived in Fewster and Verch 2020, Eqs. (3.19)–(3.26) and Theorem 3.4, PDF.

This action criterion is stronger than saying that an effect “looks local.” In a continuum net, a proposed instrument must extend normally and compose with the rest of the observable algebra. Under the normal extension and split hypotheses, Okamura and Ozawa 2016, Theorems III.4 and VI.2, PDF relate CP instruments to measuring processes and local extensions.

Same POVM, different update

Keep the Gaussian effects fixed and append measurable outcome-dependent feedback:

$$\widetilde M_x=U_xM_x, \qquad \widetilde{\mathfrak i}_x(\rho) =U_xM_x\rho M_xU_x^\dagger.$$

The corresponding event map is $\widetilde{\mathcal I}_X(\rho)=\int_X dx\,\widetilde{\mathfrak i}_x(\rho)$.

Because $U_x$ is unitary,

$$\widetilde M_x^\dagger\widetilde M_x=M_x^2=E_x.$$

The readout density is identical, but the posterior state need not be. The following point-conditioned formulas hold for regular conditional states at almost every $x$. For an explicit regulated canonical mode with $[Q,P]=i$, choose

$$U_x=e^{-i\lambda xP}.$$

Retaining the centered Gaussian input used in the benchmark, the unmodified branch has conditional mean

$$\mathbb E(Q\mid x) =\frac{v}{v+\delta^2}x,$$

whereas $U_x^\dagger Q U_x=Q+\lambda x$ gives

$$\widetilde{\mathbb E}(Q\mid x) =\left(\frac{v}{v+\delta^2}+\lambda\right)x.$$

With $v=1$, $\delta=1/2$, and $\lambda=1$, the two conditional means are $0.8x$ and $1.8x$ even though every outcome probability is the same. This is the manifest adversarial control: conclusions about postmeasurement observables or energy cannot be inferred from the POVM alone.

If the feedback generator $P$ is not localized in $K$, or if implementing $U_x$ requires an outcome record that has not arrived, the modified instrument does not inherit the localization of $E_x$. A local effect plus a nonlocal update is still a nonlocal instrument.

Scope and limitations

The Gaussian formula assumes a self-adjoint smeared field in a fixed regulated representation and a declared resolution $\delta$. It does not construct the hardware that realizes the map; a supported probe dilation must supply that evidence. It also does not justify an ideal measurement of a point field, the limit $f\to\delta_x$, or the sharp limit $\delta\to0$. Each limit needs separate ultraviolet, domain, energy, and causal-support control.

Exercises

1. Normalize the Gaussian instrument

Use the spectral theorem to show that $\int dx\,M_x^2=\mathbf1$, and derive the output variance $v+\delta^2$ for a Gaussian input distribution of $Q$.

Solution

On a spectral value $q$, $M_x^2$ is the normalized Gaussian density of $x$ centered at $q$; integrating it over $x$ gives one. Integrating this scalar identity against $P_Q(dq)$ yields the operator identity. The measured variable is distributed as $x=q+n$, where $q$ has variance $v$ and the independent apparatus noise $n$ has variance $\delta^2$, so $\operatorname{Var}x=v+\delta^2$.

2. Derive the coherence factor

Compute $\langle q|\mathcal I_\Omega(\rho)|q'\rangle$ by carrying out the $x$ integral.

Solution

The kernel is multiplied by

$$\frac{1}{\sqrt{2\pi\delta^2}} \int dx\, \exp\!\left[-\frac{(x-q)^2+(x-q')^2}{4\delta^2}\right].$$

Complete the square:

$$(x-q)^2+(x-q')^2 =2\left(x-\frac{q+q'}2\right)^2+\frac{(q-q')^2}{2}.$$

The normalized Gaussian integral is one, leaving $\exp[-(q-q')^2/(8\delta^2)]$.

3. Hold the effect fixed

For $U_x=e^{-i\lambda xP}$, show that the original and feedback instruments have the same effect density but different conditional $Q$ means. Which statement remains valid if $U_x$ is nonlocal?

Solution

Unitarity gives

$$(U_xM_x)^\dagger(U_xM_x)=M_x^2,$$

so all current readout probabilities agree. Because $U_x^\dagger Q U_x=Q+\lambda x$, the feedback shifts the conditional mean by $\lambda x$. If $U_x$ is nonlocal, only the POVM statement survives: the effect and readout probabilities remain the same, but neither the update localization nor any later-statistics claim can be inherited from the original instrument.

References

• Davies, E. B., and Lewis, J. T. (1970). “An Operational Approach to Quantum Probability.” Communications in Mathematical Physics 17, 239–260. DOI.
• Fewster, C. J., and Verch, R. (2020). “Quantum Fields and Local Measurements.” Communications in Mathematical Physics 378, 851–889. DOI. Open PDF.
• Okamura, K., and Ozawa, M. (2016). “Measurement Theory in Local Quantum Physics.” Journal of Mathematical Physics 57, 015209. DOI. Open PDF.