For an operator , for finite dimensional, the definitions of Self-Adjoint Operator, Hermitian Operator and Symmetric Operator are equivalent.
Definitions
Conventions differ between authors, so we fix them. Let be a complex Hilbert Space and a linear operator with dense domain . The Adjoint Operator has domain and is defined by for .
- Symmetric: for all . Equivalently : and agrees with on .
- Self-adjoint: , including equality of domains .
- Hermitian: symmetric and defined everywhere, , so that for all . Many physics texts use “Hermitian” as a synonym of one of the two other notions; here it is the everywhere-defined version.
In general: Hermitian self-adjoint symmetric. The first implication holds since and force .
Proof
Let .
- Every subspace of is closed, so the only dense subspace is itself. Hence for any densely defined operator.
- Every linear map on a finite-dimensional normed space is bounded: writing in an orthonormal basis, by the Cauchy-Schwarz inequality. Therefore is bounded for every , so as well.
Now suppose is symmetric. By 1, , so is Hermitian. By 2, , and then gives , so is self-adjoint. Combined with the general implications above, the three notions coincide. In an orthonormal basis all three say that the matrix of equals its conjugate transpose, .
The Hellinger-Toeplitz theorem
In infinite dimensions the notions separate, and the Hellinger-Toeplitz theorem explains why the separation only concerns unbounded operators:
If is defined on all of a Hilbert space and is symmetric, for all , then is bounded.
Proof: by the closed graph theorem it suffices to show that the graph of is closed. Let and . For every ,
so , the graph is closed and is bounded.
Why it matters here: the theorem says Hermitian operators (in the convention above) are exactly the bounded self-adjoint operators. Hence an unbounded symmetric operator, such as the position, momentum or a typical Hamiltonian in quantum mechanics, can never be defined on all of . It must live on a dense proper domain, and then symmetry and self-adjointness genuinely differ, because may be strictly larger than (see Not all self-adjoint operators are hermitian for a domain mismatch of this kind). Only self-adjoint operators have a Spectral theorem with real spectrum and generate unitary time evolution by Stone’s theorem, which is why the distinction matters physically. In finite dimensions both obstacles, unbounded operators and proper dense domains, are absent, which is exactly what the proof above uses.