Defining T* through pairing
Given a linear operator T on V, fix a vector w and consider v -> <Tv, w>. That is a linear functional in v, so Riesz produces a unique vector — call it T*w. The adjoint operator T* is defined by the single identity <Tv, w> = <v, T*w> for all v, w. It is the operator that lets you slide T from one side of the inner product to the other.
Computing it: the conjugate transpose
In an orthonormal basis the abstract definition becomes concrete: the matrix of T* is the conjugate transpose of the matrix of T, written A* = conj(A)^T (also A^H). Over R this is just the transpose A^T; over C you transpose and conjugate every entry. The conjugation is exactly the price the Hermitian inner product charged us in guide one.
A = [ 1+i, 2 ]
[ 3, 4-i ]
Conjugate transpose A* = conj(A)^T:
step 1 transpose: [ 1+i, 3 ]
[ 2, 4-i ]
step 2 conjugate each: A* = [ 1-i, 3 ]
[ 2, 4+i ]
Sanity check the defining identity on e1, e2 (standard inner product):
<A e1, e2> = (A)_{21} = 3 (entry row 2, col 1)
<e1, A* e2> = conj( (A*)_{12} ) = conj(3) = 3 -> they match.The algebra you must internalize
The adjoint identities turn it into a tidy calculus. The starred map reverses products and is conjugate-linear in scalars, while being an involution:
(S + T)* = S* + T* (c T)* = conj(c) T* (conjugation appears over C) (S T)* = T* S* (order reverses, like the transpose) (T*)* = T (an involution: starring twice undoes itself) (T^-1)* = (T*)^-1 (when T is invertible)
One structural gem connects the adjoint to last guide's geometry: ker(T*) = (im T)-perp, the kernel of the adjoint is the orthogonal complement of the image. This is the conceptual upgrade of Vol I's "row space and null space are orthogonal" — the four fundamental subspaces, now stated operator-to-operator without ever choosing coordinates.