Two dice lie in different rooms. If they are 'entangled,' a roll of one instantly determines the outcome of the other — even at a distance. In the quantum world, such a link is called steering: one party, by performing measurements on their side, seems to dictate the state of the partner. But what happens when the dice are slightly 'shaken'—their state is not perfectly pure but mixed with a dose of chaos? For a long time, it was suspected that steering could disappear even if entanglement persists.
The authors of the new work showed that for states described by just two 'flavors' of mixture (rank two), this does not happen. They proved that any such state is necessarily steerable in at least one direction — one particle can influence the other. If the 'sizes' of the systems match, steering works in both directions. The proof does not rely on complex inequality calculations but on the geometry of the boundary that separates admissible quantum states from impossible ones. When a state touches this boundary, a 'certificate' of nonclassicality automatically appears in its structure, like a crack in the shell that reveals the contents.
This result builds a hierarchy: the rank of the mixture determines how difficult it is to 'steal' steerability from entanglement. Rank two is the first step where mixture is already present but steerability is not yet lost. Practically, this means that even imperfect sources of entangled particles can be suitable for quantum cryptography or teleportation: you don't need to chase perfect purity. The geometric approach opens the way to quick verification: just look at the shape of the boundary, not solve mile-long equations. So two dice, even slightly shaken, remain a reliable tool: their link is not so easy to break.
as written by the authors arXiv:2606.08189
Pure entangled states are already steerable by suitable projective measurements within their Schmidt supports, whereas rank two is the first genuinely mixed rank at which entanglement and Einstein--Podolsky--Rosen steering could bifurcate. We prove that this bifurcation does not occur even under the restricted measurement class of projective measurements: every rank-two bipartite entangled state in arbitrary finite local dimensions is projectively steerable in at least one direction, and is two-way projectively steerable when the effective local dimensions are equal. The proof is boundary-geometric rather than a steering-inequality optimization or a two-qubit reduction. A dimension--rank obstruction forces a projective outcome on the larger effective party to hit a pure boundary point of the trusted state cone. At such a contact, a nonzero support--kernel tangent block is simultaneously an NPT minor and a projective-steering certificate; if the contact is degenerate, a Schur-complement peel removes one product layer and preserves the same rank-forcing mechanism on the entangled residual. This gives a directional low-rank strengthening: if an effective $$m\otimes n$$ entangled state satisfies $$\rank\rho\le 1+\lfloor(m-1)/(n-1)\rfloor$$, then it is projectively steerable from $$A$$ to $$B$$, with the symmetric statement after exchanging the parties. Thus rank two is the first complete mixed stratum of a genuine rank hierarchy for steering, and the proof identifies the boundary contact that certifies the nonclassicality.
Steering is a term introduced by Erwin Schrödinger in 1935, while discussing the famous Einstein-Podolsky-Rosen paradox. He considered this ability 'magical' and doubted it could be realized.
In Hannu Rajaniemi's novel 'The Quantum Thief,' steering is used for instantaneous communication and control over remote devices — exactly what this theoretical work promises.