Quantitative Entanglement Witnesses

• Quantitative entanglement witnesses are observables that provide certified lower bounds on entanglement measures using minimal expectation values.
• They employ optimized techniques such as fidelity and Frobenius distance measurements to assess multipartite and high-dimensional entanglement.
• These methods offer actionable insights for experimental entanglement certification and resource quantification without full state tomography.

Quantitative entanglement witnesses are witness-based constructions in which experimentally accessible observables do more than separate entangled from separable states. In the narrow sense, they provide certified lower bounds on entanglement monotones such as entanglement of formation, concurrence, geometric entanglement, or related distance-based quantities from one or a few expectation values. In a broader but established usage, they also quantify multipartite Schmidt structure, task-specific resource content, entanglement dimensionality, or even the information yield of a witness protocol itself. The literature therefore uses the same label for several related programs: monotone lower bounds from witness data, partition-dependent structural certification, optimized witnesses adapted to restricted data, and information-theoretic assessments of witness usefulness (Silvi et al., 2011, Shahandeh et al., 2014, Cavalcanti et al., 2023).

1. Witness theory and the meaning of “quantitative”

A standard entanglement witness is a Hermitian operator whose expectation value is nonnegative on all separable states and negative for at least one entangled state. In one common formulation, a measured observable $L$ defines a witness $W=g_s\mathbbm 1-L$, where $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$; then $\operatorname{Tr}(\rho L)>g_s$ certifies entanglement. Quantitative witness theory begins when one asks not only whether $\rho$ lies outside the separable set, but how the measured value constrains an entanglement quantity or an entanglement hierarchy (Gachechiladze et al., 2018).

One influential definition says that an observable $Q$ is a quantitative entanglement witness for an entanglement measure $E$ if there exists a nonnegative function $b$ such that

$$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$

for every bipartite state $\rho$. The optimal bound is

$W=g_s\mathbbm 1-L$0

This is the monotone-calibrated sense of quantitative witnessing (Silvi et al., 2011).

A distinct multipartite usage treats witness expectation values as lower bounds on location within a nested convex hierarchy. For a fixed partition $W=g_s\mathbbm 1-L$1, the sets $W=g_s\mathbbm 1-L$2 of states with multipartite Schmidt number at most $W=g_s\mathbbm 1-L$3 are convex and nested, and a witness of the form $W=g_s\mathbbm 1-L$4 becomes quantitative because $W=g_s\mathbbm 1-L$5 certifies multipartite Schmidt number strictly larger than $W=g_s\mathbbm 1-L$6 for that partition (Shahandeh et al., 2014).

The literature also contains broader interpretations. Some papers quantify the usefulness of a witness family in bits of mutual information about the latent variable “entangled or separable,” rather than quantifying entanglement itself (Cavalcanti et al., 2023). Others treat entangled states as higher-level witnesses and define a nonnumeric ordering by inclusion of the sets of witnesses they detect, which is a structural rather than scalar quantification (Wang, 2016).

2. From witness expectation values to entanglement measures

The most direct quantitative witness results derive explicit lower-bounding functions for standard monotones. A canonical example is the Werner- and isotropic-class construction. On $W=g_s\mathbbm 1-L$7, the swap operator

$W=g_s\mathbbm 1-L$8

and the maximally entangled projector

$W=g_s\mathbbm 1-L$9

are quantitative entanglement witnesses for entanglement of formation because their optimal bounding functions can be derived analytically from twirling. For $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$0,

$$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$1

while for $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$2,

$$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$3

with

$$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$4

The bounds are optimal because equality is attained on Werner and isotropic states with the same witness expectation value (Silvi et al., 2011).

A more recent general route starts from a normalized witness. If $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$5 and $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$6 are the largest and smallest eigenvalues of $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$7, then shifting and rescaling $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$8 so that $$g_s=\sup_{\sigma\in\mathcal S_{\mathrm{sep}}}\operatorname{Tr}(\sigma L)$$9 yields a lower bound on the minimum trace distance from $\operatorname{Tr}(\rho L)>g_s$0 to the separable set: $\operatorname{Tr}(\rho L)>g_s$1 From that single quantity one obtains

$\operatorname{Tr}(\rho L)>g_s$2

$\operatorname{Tr}(\rho L)>g_s$3

Here the witness value becomes a certified bound on operational distinguishability from separable states and, through standard inequalities, on relative entropy of entanglement, entanglement of formation, geometric measure, concurrence, and robustness-type quantities (Sun et al., 2023).

Another general construction uses the Frobenius distance to the separable set,

$\operatorname{Tr}(\rho L)>g_s$4

Given a witness $\operatorname{Tr}(\rho L)>g_s$5, one obtains

$\operatorname{Tr}(\rho L)>g_s$6

with

$\operatorname{Tr}(\rho L)>g_s$7

This implies

$\operatorname{Tr}(\rho L)>g_s$8

$\operatorname{Tr}(\rho L)>g_s$9

thereby turning witness violations and witness-like criteria into lower bounds on concurrence, entanglement of formation, and geometric entanglement without full tomography (Shi, 2023).

3. Optimized witnesses: subspaces, constrained observables, and GHZ-type quantification

A major development is the replacement of a fixed target-state witness by an optimized family. For a pure target $\rho$0, the usual fidelity witness is $\rho$1. The subspace witness introduced for Bell-subspace targets instead optimizes over locally phase-shifted target states and uses

$\rho$2

In the Bell-subspace case,

$\rho$3

so the witness depends on the coherence magnitude rather than on one phase-dependent projection. This always satisfies $\rho$4, with strict inequality when the fixed witness is not phase matched. Quantitatively, for two qubits,

$\rho$5

The same measurements also expose the many-body coherence terms needed for lower bounds on genuine multipartite concurrence once additional outside-subspace populations are measured (Sun et al., 2019).

The same optimization logic appears in two-observable witness theory. Given $\rho$6 and $\rho$7, the natural quantitative object is

$\rho$8

For convex $\rho$9, Legendre duality yields

$Q$0

For the geometric measure with $Q$1 and $Q$2, the explicit bound is

$Q$3

under the assumptions $Q$4 and $Q$5. In this framework, knowing two expectation values does not merely tighten a separability threshold; it yields the optimal lower bound on the compatible entanglement measure (Gachechiladze et al., 2018).

For GHZ-type multipartite entanglement, a related program combines SLOCC normal forms, GHZ symmetrization, and exact knowledge of the three-tangle on GHZ-symmetric states. For arbitrary three-qubit mixed states this gives an optimized lower bound on $Q$6, while on GHZ-symmetric states and locally equivalent states it is exact. The standard projective GHZ witness

$Q$7

also yields the quantitative but non-optimal bound

$Q$8

The two-qubit specialization recovers Wootters’ concurrence formula (Eltschka et al., 2013).

4. Multipartite structural quantification

Quantitative witnessing in multipartite systems is not restricted to monotone lower bounds. In the structural quantification of entanglement framework, one fixes a partition

$Q$9

and considers the closed convex sets $E$0 of states with multipartite Schmidt number at most $E$1. These sets obey

$E$2

and refinement of partitions gives additional nesting. An optimal witness has the form $E$3, with

$E$4

so the experimentally checkable condition is

$E$5

The resulting certification is quantitative because it lower-bounds multipartite Schmidt number for a specified partition, rather than merely detecting nonseparability (Shahandeh et al., 2014).

A different multipartite route uses entropic correlations. For tripartite systems, the tripartite entanglement of formation is

$E$6

and it is nonzero iff the state is genuinely tripartite entangled. The entropic witness constructed for three parties gives

$E$7

where $E$8. Here the degree of witness violation is itself a lower bound on a genuine-tripartite entanglement measure. The same paper then replaces the quantum conditional entropies by lower bounds derived from experimentally accessible Shannon entropies through entropic uncertainty relations with side information (Schneeloch et al., 2020).

5. High-dimensional entanglement and dimension-sensitive witnesses

High-dimensional settings have motivated witness schemes that are simultaneously scalable and quantitative. One route uses EPR-type entropic correlations. For two pairs of complementary observables, the uncertainty relation with quantum memory leads to

$E$9

in discrete variables, and in the continuous Fourier-conjugate case

$b$0

The same witness structure lower-bounds $b$1, $b$2, and $b$3. The method requires only the joint distributions for two complementary measurement settings and can certify many ebits in common high-dimensional EPR-like systems such as spatially or temporally entangled photons from parametric down-conversion (Schneeloch et al., 2017).

A different high-dimensional proposal is the $b$4-concurrence,

$b$5

defined as the product of ordinary two-qubit concurrences over all relevant qubit subspaces of a $b$6-dimensional bipartite state. The construction is intended to be both a quantitative entanglement measure on $b$7 and a dimension witness: $b$8 whenever entanglement fails to occupy the full target dimension. Its experimental attraction is that it uses qubit-subspace tomography rather than full $b$9 tomography. The paper is explicit, however, that it does not provide a formal proof of LOCC monotonicity and only anticipates applicability to mixed high-dimensional states, so its status is more operational than axiomatic (Ndagano et al., 2017).

6. Operational, informational, and device-independent extensions

Some witness families are quantitative because they exactly characterize an operational task rather than a general monotone. Zhao, Fei, and Li-Jost construct a complete family of teleportation witnesses such that

$$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$0

The optimized witness value satisfies

$$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$1

with $$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$2 the fully entangled fraction, and therefore

$$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$3

This is quantitative in an operational sense: the witness family determines the exact optimal teleportation fidelity after optimization, but it does not quantify entanglement in a resource-independent way (Zhao et al., 2012).

Measurement-device-independent witness theory addresses a different limitation. MDI-EWs convert standard witnesses into inequalities based on trusted quantum inputs and untrusted measurements, so that

$$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$4

for all separable states, while some entangled states yield $$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$5. This construction certifies all entangled states with untrusted measurement apparatuses and is loss tolerant, but it is fundamentally a detection framework; the witness value is not converted into a bound on a standard monotone in the original MDI papers (Branciard et al., 2012, Xu et al., 2014).

Another broadening of the term “quantitative” is explicitly information theoretic. Entanglement witnessing can be modeled as a channel

$$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$6

where $$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$7 is the latent entanglement-status variable, $$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$8 is the full witness expectation value, and $$E(\rho)\ge b\!\left(\operatorname{Tr}[Q\rho]\right)$$9. The quantities

$\rho$0

measure how many bits about entanglement status are learned from the witness value or its sign. The central conclusion is that there is more information in $\rho$1 than in $\rho$2; standard sign-based post-processing discards information present in the full expectation value (Cavalcanti et al., 2023).

A further development is data-driven optimization from incomplete measurement records. Given a finite set of observables $\rho$3, one can map separability compatibility onto a convex inverse statistical problem and construct the optimal witness in the span of the measured observables. The resulting witness is quantitative in the sense of maximal violation margin in the accessible data space and maximal robustness to uncertainty in those observables, but it is not, in that paper, converted into a lower bound on a standard entanglement measure (Frérot et al., 2021).

Taken together, these lines of work show that quantitative entanglement witnesses do not denote a single formalism. In the strictest usage they are witnesses with explicit lower-bounding functions for entanglement monotones. In wider usage they include witness-based certification of Schmidt-number structure, genuine multipartite formation cost, entanglement dimensionality, task-specific figures of merit, and the information content of witness data. The unifying theme is that the witness value is treated as more than a binary flag.