where is a Positive operator.
Here means for all . We prove this for acting on a finite-dimensional complex Inner Product Space , and sketch the bounded infinite-dimensional case at the end.
Proof
Step 0, positive operators are self-adjoint with non-negative eigenvalues. Since is real, for all . The polarisation identity expresses through such diagonal values, and the same identity applied to shows , so is self-adjoint. If with , then .
Existence. By the Spectral theorem, , where are the distinct eigenvalues of and is the orthogonal projection onto the eigenspace , with and . Define . Then
and , so .
Uniqueness. Let with . By step 0 applied to , with distinct and the orthogonal projections onto the eigenspaces , which are mutually orthogonal and sum to . Then:
- , so every satisfies , giving .
- Since is injective on , the values are distinct. Each is therefore one of the , with different giving different , so for an injective map .
- The span and the are mutually orthogonal, so . This forces to be a bijection and for every .
Hence and , so .
Remark: positivity of is what gives uniqueness. Without it, are all square roots of , and if some eigenvalue is degenerate there are infinitely many self-adjoint square roots (for on , every reflection with squares to ).
Bounded operators on a Hilbert space
The statement holds for any bounded positive on a complex Hilbert Space. Existence follows from the continuous functional calculus, , or directly by approximating uniformly by polynomials on (Weierstrass approximation). For uniqueness, if and , then commutes with , hence with every polynomial in , hence with their norm limit . Then for any , with ,
using . Both terms are non-negative, so both vanish, which for a positive operator gives , hence . Therefore and . See Reed and Simon, Methods of Modern Mathematical Physics I, Section VI.4, for this square root lemma and its use in the Polar decomposition.