Entanglement Witnesses
A Bell inequality detects entanglement only when correlations are strong enough to be nonlocal. Many entangled states are not nonlocal yet are still entangled. Entanglement witnesses are a more flexible tool: a single observable whose expectation value reveals entanglement directly.
The idea
The set of separable (unentangled) states is convex — it is closed under classical mixing. A state outside this set is entangled. Because separable states form a convex set, a separating hyperplane exists: a Hermitian operator such that
while for at least one entangled state. Measuring a negative witness value therefore certifies entanglement — no separable state could have produced it.
A witness for the Bell state
A simple witness for the Bell state is
The constant is the maximum squared overlap any separable state can have with — the maximal fidelity . Hence for every separable ,
The witness is non-negative on all separable states, as required.
Evaluating it on the Bell state
For a pure state , . Take itself. A normalized state overlaps itself perfectly, , so
The negative value flags as entangled — exactly what we expect for a maximally entangled state.
Witnesses vs Bell inequalities
A Bell inequality is a special, device-independent kind of witness: it certifies entanglement (in fact nonlocality) without trusting the measurement apparatus, but it only fires for sufficiently nonlocal states. A general witness like above assumes you know what observables you are measuring, but in return it can detect weakly entangled states that violate no Bell inequality. The two tools are complementary.
Try it
This is a numerical exercise — your code should return a number. Using with (so the squared overlap is ),
compute the witness value. You should get — a negative number certifying entanglement.
Sign in on the full site to ask questions and join the discussion.