# Entanglement witness

An **entanglement witness** is a Hermitian observable used to certify that a quantum state is entangled: its expectation value is nonnegative on every separable state, while at least one entangled state gives a negative value, so a negative expectation value of the witness proves that the measured state is entangled.<sup>[1](https://chaos.fmf.uni-lj.si/wp-content/uploads/2017/06/a7_45_017.pdf)</sup><sup> • </sup><sup>[12](https://www.nature.com/articles/s41534-024-00810-3)</sup> Witnesses are a practical alternative to full quantum state tomography, which requires far more measurements, and they underpin experimental entanglement certification in quantum communication, sensing, and quantum computing.<sup>[2](https://www.nature.com/articles/s42254-018-0003-5)</sup>

| Key fact | Detail |
|---|---|
| Definition | Hermitian operator W with Tr(Wρ) ≥ 0 for all separable states ρ and Tr(Wρ_ent) < 0 for at least one entangled state<sup>[1](https://chaos.fmf.uni-lj.si/wp-content/uploads/2017/06/a7_45_017.pdf)</sup> |
| Completeness | A state is separable if and only if Tr(ρW) ≥ 0 for all entanglement witnesses W<sup>[3](https://link.springer.com/article/10.1007/s10701-021-00499-w)</sup> |
| Existence | For any particular entangled state, one can always find an observable that serves as an entanglement witness<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> |
| Fidelity witness | Built from an entangled reference state and a positive multiple of the identity; it detects exactly the states whose fidelity with the reference exceeds a threshold<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> |
| Threshold | The threshold equals the square of the largest Schmidt coefficient of the reference state; for a two-qubit maximally entangled state it is 1/2<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> |
| Measurement cost | A witness for the Bose et al. gravitational-entanglement proposal needs five nontrivial local spin measurements, against 15 for full-state tomography<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> |
| Trust assumptions | Measurement-device-independent witnesses certify entanglement of all entangled states with untrusted measurement apparatus<sup>[5](https://arxiv.org/abs/1210.8037)</sup> |

## Definition and geometric picture

Formally, an observable W is an entanglement witness if it is not a positive operator, but \( (x_1 \otimes x_2)^{\dagger} W (x_1 \otimes x_2) \ge 0 \) for all product vectors, equivalently Tr(Wρ) ≥ 0 for all separable states ρ.<sup>[3](https://link.springer.com/article/10.1007/s10701-021-00499-w)</sup> The set of separable states is convex, so any entangled state lying outside it can be separated from it by a hyperplane, and the operator defining that hyperplane is a witness.<sup>[6](https://pmc.ncbi.nlm.nih.gov/articles/PMC4870702/)</sup> A negative expectation value of W on the measured state therefore certifies entanglement directly, without reconstructing the density matrix.<sup>[7](https://arxiv.org/pdf/1303.2433)</sup>

The criterion is complete in a precise sense: a state is separable if and only if its expectation value is nonnegative on every entanglement witness.<sup>[3](https://link.springer.com/article/10.1007/s10701-021-00499-w)</sup> [Existence](https://www.edgepedia.org/existence), however, does not mean convenience. Many witnesses detect entanglement for only a small set of states, and in experiments the witness must be measured through local measurements on the subsystems, so observers prefer observables that need fewer measurement settings than full-state tomography.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup>

## Fidelity witnesses

A widely used class is the **fidelity witness**, constructed from an entangled [pure state](https://www.edgepedia.org/pure-state) \( |\phi\rangle \) and the identity: \( W = \alpha \cdot \mathbb{1} - |\phi\rangle\langle\phi| \), where \( \alpha = \max_{|\varphi\rangle \in S} |\langle\varphi|\phi\rangle|^{2} \) over the set S of separable states.<sup>[8](https://www.nature.com/articles/srep08048)</sup> This operator is nonnegative on all separable states and negative on the reference state itself, so it is a witness.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup>

The coefficient α measures the "distance" of the reference state to the nearest product state and sets the size of the detection region: the witness flags entanglement for exactly those states whose fidelity with \( |\phi\rangle \) exceeds α.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> Using the [Schmidt decomposition](https://www.edgepedia.org/schmidt-decomposition), α is bounded by the square of the largest Schmidt coefficient, and this bound is attained; the smallest possible value occurs for maximally entangled states, giving \( \alpha = 1/2 \) for two qubits.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> Fidelity witnesses built from maximally entangled states therefore yield the widest detection areas.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup>

**Choosing the reference state.** Building the witness from the target state itself can be a poor choice when that state is only weakly entangled: α is then close to 1, the detection region is small, and small experimental errors that push the state away from the target make the witness fail. The recommended construction instead uses a fidelity witness built from a maximally entangled state closest to the state of interest, which maximizes the detection area.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup>

## Experimental implementation

To measure a witness, it is decomposed into a sum of local observables that can be evaluated with ordinary local measurements. For a two-qubit witness given by a partially transposed projector, such a decomposition requires at least three local measurement settings.<sup>[9](https://arxiv.org/abs/quant-ph/0210134)</sup> Optimization of these decompositions, and of witnesses themselves, is an active area: necessary and sufficient conditions for a witness to be optimal were established early in the field's development.<sup>[10](https://arxiv.org/abs/quant-ph/0005014)</sup>

A concrete application comes from proposals to detect gravitationally induced entanglement between two optomechanical systems, following S. Bose and colleagues (Phys. Rev. Lett. 119, 240401 (2017)). When the optimal fidelity-witness framework is applied to that setup, the resulting witness needs just five nontrivial local spin measurements, versus the 15 measurements required by full-state tomography; after postprocessing, the witness reveals entanglement regardless of which phases are chosen in the setup—apart from a measure-zero set—when the system is closed.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup> The ability of these witnesses to detect entanglement decreases as dephasing noise grows.<sup>[4](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)</sup>

## Relation to other certification tools

Bell inequalities can sometimes discriminate entangled from separable states, but they may fail to detect genuinely entangled states, which is one motivation for witnesses.<sup>[11](https://academic.oup.com/book/8602/chapter-abstract/154566500)</sup> Conversely, standard witnesses assume that the measurements are implemented exactly as modeled; a mismatch between prescribed and performed measurements can produce false-positive entanglement detection, and an artificial truncation of the [Hilbert space](https://www.edgepedia.org/hilbert-space) can likewise lead to wrong conclusions.<sup>[7](https://arxiv.org/pdf/1303.2433)</sup>

Two extensions relax these assumptions. **Measurement-device-independent** entanglement witnesses, introduced by Cyril Branciard (CNRS researcher in quantum information theory), Denis Rosset, Yeong-Cherng Liang, and [Nicolas Gisin](https://www.edgepedia.org/nicolas-gisin) (professor at the University of Geneva), certify entanglement of all entangled states even when the measurement apparatus is untrusted, are loss-tolerant, and can be implemented with current technology.<sup>[5](https://arxiv.org/abs/1210.8037)</sup> **Device-independent** witnesses go further, assuming only that measurements are performed locally on each subsystem; they were experimentally tested on up to six entangled calcium-40 ions.<sup>[7](https://arxiv.org/pdf/1303.2433)</sup> The witness concept has also been transferred beyond quantum states: witnesses of causal nonseparability are Hermitian operators that certify that a process has no definite causal order, defined in direct analogy with entanglement witnesses.<sup>[6](https://pmc.ncbi.nlm.nih.gov/articles/PMC4870702/)</sup>

## References

1. [Detecting entanglement of random states with an entanglement witness](https://chaos.fmf.uni-lj.si/wp-content/uploads/2017/06/a7_45_017.pdf)
2. [Entanglement certification from theory to experiment, Nature Reviews Physics](https://www.nature.com/articles/s42254-018-0003-5)
3. [The Weirdness Theorem and the Origin of Quantum Paradoxes, Foundations of Physics](https://link.springer.com/article/10.1007/s10701-021-00499-w)
4. [Optimal fidelity witnesses for gravitational entanglement, Phys. Rev. A 105, 022444 (2022)](https://journals.aps.org/pra/abstract/10.1103/PhysRevA.105.022444)
5. [Measurement-Device-Independent Entanglement Witnesses for All Entangled Quantum States, arXiv:1210.8037](https://arxiv.org/abs/1210.8037)
6. [Witnesses of causal nonseparability: an introduction and a few case studies](https://pmc.ncbi.nlm.nih.gov/articles/PMC4870702/)
7. [Device-independent demonstration of genuine multipartite entanglement, arXiv:1303.2433](https://arxiv.org/pdf/1303.2433)
8. [Experimental Measurement-Device-Independent Entanglement Detection, Scientific Reports](https://www.nature.com/articles/srep08048)
9. [Experimental detection of entanglement via witness operators and local measurements, arXiv:quant-ph/0210134](https://arxiv.org/abs/quant-ph/0210134)
10. [Optimization of entanglement witnesses, arXiv:quant-ph/0005014](https://arxiv.org/abs/quant-ph/0005014)
11. [Witnessing quantum entanglement, in V. Vedral, Introduction to Quantum Information Science, Oxford](https://academic.oup.com/book/8602/chapter-abstract/154566500)
12. [Valid and efficient entanglement verification with finite copies of a quantum state, npj Quantum Information](https://www.nature.com/articles/s41534-024-00810-3)

---
*Topic: Encyclopedia › Physical world and mathematics › Physics › Quantum physics › Quantum information science › Quantum communication and information theory › Quantum information theory › Entanglement theory › Separability criteria and entanglement detection*

*Initially written Sep 17, 2026 · Reviewed: Oct 15, 2026 · Edited: Oct 15, 2026 · Last review: Oct 15, 2026*

*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*

License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license