The result
Consider a quantum mechanical system composed of ''n'' subsystems. The state space ''H'' of such a system is the tensor product of those of the subsystems, i.e. . For simplicity we will assume throughout that all relevant state spaces are finite-dimensional. The criterion reads as follows: If ρ is a separable mixed state acting on ''H'', then the range of ρ is spanned by a set of product vectors.Proof
In general, if a matrix ''M'' is of the form , the range of ''M'', ''Ran(M)'', is contained in the linear span of . On the other hand, we can also show lies in ''Ran(M)'', for all ''i''. Assume without loss of generality ''i = 1''. We can write , where ''T'' is Hermitian and positive semidefinite. There are two possibilities: 1) ''span''''Ker(T)''. Clearly, in this case, ''Ran(M)''. 2) Notice 1) is true if and only if ''Ker(T)'' ''span'', where denotes orthogonal complement. By Hermiticity of ''T'', this is the same as ''Ran(T)'' ''span''. So if 1) does not hold, the intersection ''Ran(T)'' ''span'' is nonempty, i.e. there exists some complex number α such that . So : Therefore lies in ''Ran(M)''. Thus ''Ran(M)'' coincides with the linear span of . The range criterion is a special case of this fact. A density matrix ρ acting on ''H'' is separable if and only if it can be written as : where is a (un-normalized) pure state on the ''j''-th subsystem. This is also : But this is exactly the same form as ''M'' from above, with the vectorial product state replacing . It then immediately follows that the range of ρ is the linear span of these product states. This proves the criterion.References
* P. Horodecki, "Separability Criterion and Inseparable Mixed States with Positive Partial Transposition", ''Physics Letters'' A 232, (1997). Quantum information science