Prove that the dual of a Hilbert space is itself.
- User
Prove that the dual of a Hilbert space is itself.
- Assistant
Deciding on inner product convention
I’m thinking about which inner product convention to adopt. I’ll go for linear in the first argument and conjugate-linear in the second since that seems increasingly common for complex spaces, though older texts might differ. It's essential to make a note of this for clarity. I'll skip any introductory phrases like "Great question" and dive right into it: "Let H be a Hilbert space..." So, let’s get started on the writing! Let (H) be a Hilbert space (real or complex), with inner product (\langle \cdot,\cdot\rangle).
By the dual (H^*) we mean the space of continuous linear functionals (f : H \to \mathbb{F}) ((\mathbb{F}=\mathbb{R}) or (\mathbb{C})).The statement “the dual of a Hilbert space is itself” means:
Every (f \in H^) is of the form
[ f(x) = \langle x, y\rangle \quad\text{for a unique } y\in H, ] and the correspondence (y \mapsto f_y) is an isometric isomorphism between (H) and (H^).This is the Riesz Representation Theorem. I’ll give a self-contained proof.
Step 1: A geometric lemma in Hilbert spaces
Lemma. Let (C \subset H) be nonempty, closed, and convex. Then there exists a unique element (z \in C) of minimal norm: [ |z| = \inf_{x\in C} |x|. ]
Proof.
Let (\delta = \inf_{x\in C}|x|). Choose a sequence ((x_n)\subset C) such that (|x_n|\to\delta); this is a minimizing sequence.We show ((x_n)) is Cauchy. The parallelogram law in a Hilbert space states: [ |x_n - x_m|^2 + |x_n + x_m|^2 = 2|x_n|^2 + 2|x_m|^2. ] Since (C) is convex, (\frac{x_n+x_m}{2}\in C), so [ \left|\frac{x_n + x_m}{2}\right| ;\ge; \delta. ] Thus [ |x_n + x_m|^2 ;\ge; 4\delta^2. ] Combine with the parallelogram law: [ |x_n - x_m|^2 = 2|x_n|^2 + 2|x_m|^2 - |x_n + x_m|^2 \le 2|x_n|^2 + 2|x_m|^2 - 4\delta^2. ] As (n,m\to\infty), (|x_n|,|x_m|\to\delta), so the RHS (\to 0). Hence ((x_n)) is Cauchy. Since (H) is complete, there is (z\in H) with (x_n\to z).
Because (C) is closed and (x_n\in C), we have (z\in C). By continuity of the norm, [ |z| = \lim_{n\to\infty} |x_n| = \delta. ] So (z) attains the infimum.
For uniqueness, suppose (z_1,z_2\in C) both satisfy (|z_1|=|z_2|=\delta). Then [ \left|\frac{z_1+z_2}{2}\right|^2 = \frac{1}{2}|z_1|^2 + \frac{1}{2}|z_2|^2 - \frac{1}{4}|z_1-z_2|^2 = \delta^2 - \frac{1}{4}|z_1-z_2|^2. ] But (\frac{z_1+z_2}{2}\in C), so its norm is at least (\delta). Hence [ \delta^2 \le \left|\frac{z_1+z_2}{2}\right|^2 \le \delta^2 - \frac{1}{4}|z_1-z_2|^2, ] forcing (|z_1 - z_2| = 0), i.e. (z_1 = z_2). ∎
Step 2: Apply the lemma to a nonzero functional
Let (f \in H^*). If (f=0), then clearly [ f(x) = \langle x, 0\rangle, ] so we focus on (f\neq 0).
-
Normalize a vector with nonzero functional value.
Pick any (x_0\in H) with (f(x_0)\neq 0), and rescale so (f(x_0)=1) (replace (x_0) by (x_0 / f(x_0))). -
Define an affine hyperplane.
Consider the set [ S := {x\in H : f(x)=1}. ] This is an affine translate of the kernel of (f): [ S = x_0 + \ker f. ] So (S) is nonempty, closed, and convex. -
Find the minimal-norm element of (S).
By the lemma, there is a unique (z\in S) such that [ |z| = \inf_{x\in S} |x|. ] -
Show (z) is orthogonal to (\ker f).
Let (h\in \ker f). For any scalar (t\in\mathbb{R}) (or (\mathbb{C})), note that [ f(z + t h) = f(z) + t f(h) = 1 + t\cdot 0 = 1, ] so (z+th \in S) for all (t). By minimality of (|z|), [ |z| \le |z + th| \quad \forall t. ] Expand: [ |z + th|^2 = |z|^2 + 2,\Re\big(t \langle z, h\rangle \big) + |t|^2|h|^2 \ge |z|^2. ] This inequality for all real (t) implies (\Re \langle z,h\rangle = 0); doing the same with purely imaginary (t) gives (\Im \langle z,h\rangle = 0). Hence [ \langle z, h \rangle = 0 \quad \text{for all }h \in \ker f, ] i.e. (z \perp \ker f).In particular, (\ker f) is a closed subspace and (z\in (\ker f)^\perp).
Step 3: Represent (f) as inner product with (z)
We now derive an explicit formula for (f) in terms of (z).
Take any (x \in H). Consider the vector [ w := x - f(x), z. ] Then [ f(w) = f(x) - f(x),f(z) = f(x) - f(x)\cdot 1 = 0, ] so (w \in \ker f).
Since (w \in \ker f) and (z \perp \ker f), we have [ 0 = \langle w, z\rangle = \langle x - f(x),z,, z\rangle = \langle x, z\rangle - f(x),\langle z, z\rangle. ] Therefore [ f(x) = \frac{\langle x, z\rangle}{\langle z, z\rangle}. ] Set [ y := \frac{z}{\langle z, z\rangle} \in H. ] Then for every (x\in H), [ f(x) = \langle x, y\rangle. ]
So we have shown:
For each nonzero (f \in H^*), there exists (y\in H) such that (f(x)=\langle x,y\rangle) for all (x).
For (f=0), we just take (y=0). So this holds for all (f\in H^*).
Step 4: Uniqueness of the representing vector
Suppose (\langle x, y_1\rangle = \langle x, y_2\rangle) for all (x\in H). Then [ \langle x, y_1 - y_2\rangle = 0 \quad \forall x. ] Taking (x = y_1 - y_2) gives (|y_1 - y_2|^2 = 0), so (y_1 = y_2). Thus the representing vector is unique.
Step 5: The isometric isomorphism (H \cong H^*)
Define the map [ J : H \to H^*, \qquad J(y)(x) := \langle x, y\rangle. ]
-
Linearity.
If the inner product is linear in the first argument (common in functional analysis), then (x\mapsto \langle x, y\rangle) is linear in (x), and
[ J(\alpha y_1 + \beta y_2)(x) = \langle x, \alpha y_1 + \beta y_2\rangle = \alpha \langle x, y_1\rangle + \beta \langle x, y_2\rangle. ] So (J) is linear.
(If your convention is linear in the second argument, then (J) is conjugate-linear; in that case (H^*) is anti-linearly isomorphic to (H).) -
Isometry. For (y\in H), [ |J(y)|{H^*} = \sup{|x|=1} |J(y)(x)| = \sup_{|x|=1} |\langle x,y\rangle| \le \sup_{|x|=1} |x|,|y| = |y| ] by Cauchy–Schwarz.
Taking (x = y/|y|) (if (y\neq 0)), we get [ |J(y)(x)| = \left|\left\langle \frac{y}{|y|}, y\right\rangle\right| = |y|. ] Hence (|J(y)| = |y|); thus (J) is an isometry. -
Surjectivity.
Step 3 showed that every (f \in H^*) has the form (f(x)=\langle x,y\rangle) for some (y\in H); that is, (f = J(y)). So (J) is onto.
Therefore (J) is a bijective isometry (H \to H^*).
So:
- For real Hilbert spaces, (H^*) is linearly isometrically isomorphic to (H).
- For complex Hilbert spaces, (H^*) is conjugate-linearly isometrically isomorphic to (H).
This is what is meant by saying:
The dual of a Hilbert space is (canonically) the Hilbert space itself.
-
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.