Entanglement Witnesses and Local Tomography Limits

Local field measurements can certify entanglement without reconstructing an entire continuum state. The defensible claim is determined by the calibrated observable set: a violated witness proves entanglement under its support and confidence assumptions, while finite moments yield full tomography only inside a declared model class such as Gaussian states.

Required background. Local measurement instruments supplies the effects, updates, and calibration data behind measured correlators.

Helpful background. Choosing an entanglement measure separates the operational question from a convenient numerical diagnostic.

Witnesses are one-sided tests

An entanglement witness $W$ satisfies

$$\operatorname{tr}(W\sigma)\ge0$$

for every separable state $\sigma$ in the declared bipartition, while some entangled state $\rho$ gives $\operatorname{tr}(W\rho)<0$. An experimental estimate $\widehat w$ certifies entanglement only if its confidence interval lies below zero after calibration and model uncertainties are included. Failure to violate the witness is inconclusive because $W$ detects only part of the entangled set.

In QFT, the bipartition must be operationally specified by local algebras or by regulated wavepacket modes. The measured operators need bounded approximants or domain control, and the instrument supports must match the advertised regions.

A witness inherits the switching, smearing, noise, and support of the instruments used to estimate it. The field claim is no more local or precise than those calibrated operations. The diagram is schematic.

Two-mode covariance witness

Choose normalized wavepacket modes with quadratures

$$R=(q_A,p_A,q_B,p_B)^T, \qquad [R_j,R_k]=i\Omega_{jk}.$$

The covariance matrix is

$$V_{jk}=\frac12\langle R_jR_k+R_kR_j\rangle -\langle R_j\rangle\langle R_k\rangle.$$

For a two-mode Gaussian state, partial transposition reverses $p_B$. The state is entangled exactly when the partially transposed covariance violates the uncertainty condition,

$$V^{T_B}+\frac{i}{2}\Omega\not\ge0,$$

equivalently when its smallest symplectic eigenvalue satisfies $\widetilde\nu_-<1/2$. This is the continuous-variable positive-partial-transpose criterion established by Simon 2000, Eqs. (17)–(19), pp. 2728–2729.

To compare with a regulated exact state, compute $V$ analytically or from the lattice covariance, simulate the actual unsharp quadrature instruments, subtract only calibrated additive noise, and propagate the uncertainty to $\widetilde\nu_-$. Vary wavepacket resolution and cutoff to show that the conclusion is stable.

Why second moments are not full tomography

For a non-Gaussian state, the same covariance matrix can be shared by states with different higher moments and different entanglement. Thus covariance data can support a valid second-moment witness while failing to identify the density operator. A decisive adversarial test is to construct or optimize over non-Gaussian states consistent with the measured moments. If both separable and entangled compatible states exist, the correct output is a bound or an inconclusive result, not full tomography.

Finite detector settings create the same limitation more generally. Let $\mathcal S=\{E_1,\ldots,E_n\}$ be the calibrated effect set. Two states that agree on every $E_j$ are operationally indistinguishable in this experiment. Model-based reconstruction selects one representative; it does not prove uniqueness in the continuum state space. For the distinction between witness detection, quantitative bounds, and tomography assumptions, see Gühne and Tóth 2009, §§ 2–4, pp. 5–32.

Incomplete observables and postselection enter the inference branch. A robust negative witness can certify entanglement, but finite second moments do not certify a general non-Gaussian state. The map is schematic.

Reporting a certification

State the local algebras or mode functions, all measured settings, the instrument calibration, sample and independence assumptions, estimator, confidence construction, multiple-testing correction if used, regulator convergence, and the precise conclusion. Separate three levels: witness violation, entanglement quantification within a model, and full state reconstruction. They demand increasingly strong assumptions.

References

• Gühne, O., and Tóth, G. (2009). “Entanglement Detection.” Physics Reports 474, 1–75. DOI. Open PDF.
• Simon, R. (2000). “Peres–Horodecki Separability Criterion for Continuous Variable Systems.” Physical Review Letters 84, 2726–2729. DOI. Open PDF.