Hilbert Spaces Revision Summary

1 Pre-Hilbert Spaces and Hilbert Spaces

1.1 Pre-Hilbert Space
A pre-Hilbert space (real) is a real vector space H endowed with an inner product

⟨· | ·⟩ : H 2 → R

satisfying:
• Left-linearity:

∀ x, y, z ∈ H, ∀ λ, µ ∈ R, ⟨λx + µy | z⟩ = λ⟨x | z⟩ + µ⟨y | z⟩.

• Symmetry:
∀ x, y ∈ H, ⟨x | y⟩ = ⟨y | x⟩.

• Positive definiteness:

∀ x ∈ H, ⟨x | x⟩ ≥ 0, and ⟨x | x⟩ = 0 ⇐⇒ x = 0.

Examples:
• Rm with the usual dot product.
• Mm,k (R) with the scalar product A, B 7→ Tr(AB ⊤ ).
• L2 (R) with the inner product f, g 7→ R f (λ)g(λ) dλ.
R

1.2 Product of Pre-Hilbert Spaces

For n ∈ N∗ and pre-Hilbert spaces

(H1 , ⟨· | ·⟩1 ), . . . , (Hn , ⟨· | ·⟩n ),

define the product space

H = H1 × · · · × Hn ,
with the inner product for x = (xi )1≤i≤n and y = (yi )1≤i≤n given by:
n
X
⟨x | y⟩ = ⟨xi | yi ⟩i .
i=1

For example, this defines the usual dot product on Rn .

1.3 Norm and Metric

The norm induced by the inner product is defined as:
p
∥x∥ := ⟨x | x⟩.

The distance between two points is then:

d(x, y) := ∥x − y∥.

Thus, every pre-Hilbert space is both a normed vector space and a metric space.

1
1.4 Hilbert Space
A pre-Hilbert space is called a Hilbert space if it is complete with respect to the metric induced by its
inner product (i.e., every Cauchy sequence converges in the space). In particular, every finite-dimensional
pre-Hilbert space (Euclidean space) is a Hilbert space.

1.5 Open Ball

For x ∈ H and r > 0, the open ball centered at x with radius r is:

B(x, r) = {y ∈ H | ∥x − y∥ < r}.

2 Linear and Bilinear Mappings

2.1 Continuous Linear Applications
Let (H, ∥ · ∥H ) and (K, ∥ · ∥K ) be normed vector spaces. A linear map φ : H → K is continuous if and
only if there exists C > 0 such that:

∥φ(x)∥K ≤ C∥x∥H ∀ x ∈ H.

2.2 Continuous Bilinear Applications

For normed vector spaces (H1 , ∥ · ∥H1 ), (H2 , ∥ · ∥H2 ), and (K, ∥ · ∥K ), a bilinear map φ : H1 × H2 → K
is continuous if there exists C > 0 such that:

∥φ(x1 , x2 )∥K ≤ C ∥x1 ∥H1 ∥x2 ∥H2 ∀ (x1 , x2 ) ∈ H1 × H2 .

2.3 Bounded Linear Operators

For Hilbert spaces (H, ⟨· | ·⟩H ) and (G, ⟨· | ·⟩G ), a linear operator L : H → G is bounded if:

sup ∥L(x)∥G < +∞.

x∈H
∥x∥H =1

The space of all bounded linear operators from H to G is denoted by B(H, G) and is equipped with the
operator norm:
∥L∥B(H,G) := sup ∥L(x)∥G .
∥x∥H ≤1

Properties:

• A linear map is bounded if and only if it is continuous.

• In finite dimensions, every (bi)linear map is continuous.
• For L ∈ B(H, G), ∥L(x)∥G ≤ ∥L∥B(H,G) ∥x∥H for all x ∈ H.

• If L ∈ B(H, G) and G ∈ B(G, K), then:

∥G ◦ L∥B(H,K) ≤ ∥G∥B(G,K) ∥L∥B(H,G) .

3 Riesz Representation and Spectral Theory

3.1 Riesz Representation Theorem
For every continuous linear form φ ∈ B(H, R) on a Hilbert space H, there exists a unique y ∈ H such
that:
φ(x) = ⟨y | x⟩H ∀ x ∈ H.

2
3.2 Eigenvalues, Spectrum, and Spectral Radius
For an endomorphism u on a Banach space H:

• A scalar λ ∈ R is an eigenvalue of u if λI − u is not injective.

• λ is a spectral value if λI − u is not bijective.
• The spectrum of u is denoted by sp(u).
• The spectral radius is defined as:

ρ(u) = sup{|λ| | λ ∈ sp(u)}.

In finite dimensions, eigenvalues and spectral values coincide.

3.3 Adjoint and Self-Adjoint Operators

For Hilbert spaces H and G and L ∈ B(H, G), the adjoint L∗ is the unique operator in B(G, H)
satisfying:
⟨y | L(x)⟩G = ⟨L∗ (y) | x⟩H ∀ (x, y) ∈ H × G.
An operator L ∈ B(H, H) is self-adjoint if L = L∗ .
Additional Properties:
• (G ◦ L)∗ = L∗ ◦ G∗ and (L∗ )∗ = L.

• If L is invertible, then L∗ is invertible and (L−1 )∗ = (L∗ )−1 .

• The norms satisfy: ∥L∗ ∥ = ∥L∥ and ∥L∗ ◦ L∥ = ∥L∥2 .
• For self-adjoint L on H, ∥L∥ = ρ(L).

3.4 Positivity of Operators

For a self-adjoint operator L ∈ B(H, H):
• Positive: ⟨x | Lx⟩ ≥ 0 for all x ∈ H.

• Positive Definite: ⟨x | Lx⟩ > 0 for all x ∈ H \ {0}.

• Strongly Positive: There exists η > 0 such that:

⟨x | Lx⟩ ≥ η∥x∥2 ∀ x ∈ H.

Remark: In finite dimensions, a self-adjoint operator is strongly positive if and only if it is positive
definite. For symmetric matrices A ∈ Sn (R), the mapping x 7→ Ax is positive (or positive definite) if
and only if A is positive semidefinite (or positive definite).

4 Functional Analysis Notions

4.1 Domains and Proper Functions
For a function f : S → [−∞, +∞] defined on a nonempty subset S ⊂ H:
• The domain of f is:
dom f = {x ∈ S | f (x) < +∞}.

• f is proper if:
1. f (x) ̸= −∞ for all x ∈ S, and
2. dom f ̸= ∅.

3
4.2 Indicator and Characteristic Functions
• The indicator function ιC of a set C ⊂ H is defined by:
(
0 if x ∈ C,
ιC (x) =
+∞ otherwise.

• The function 1C (which equals 1 on C and 0 elsewhere) is called the characteristic function in
optimization.
• Minimizing a function f on a set C can be reformulated as minimizing f + ιC on H.

4.3 Convergence
Strong Convergence: A sequence (xn )n in H converges strongly to x̂ if:

∥xn − x̂∥ → 0.

Weak Convergence: A sequence (xn )n in H converges weakly to x̂ (denoted xn ⇀ x̂) if:

∀ y ∈ H, ⟨y, xn − x̂⟩ → 0.

Properties:
• Strong convergence implies weak convergence.
• In finite dimensions, strong and weak convergence are equivalent.
• If xn ⇀ x̂ and ∥xn ∥ → ∥x̂∥, then strong convergence holds.

4.4 Boundedness, Closure, Openness, and Compactness

For a subset S ⊂ H:
• Bounded: There exist x ∈ H and r > 0 such that S ⊂ B(x, r).
• Closed: Every convergent sequence in S has its limit in S.
• Open: For every x ∈ S, there exists r > 0 such that B(x, r) ⊂ S.
• Compact: Every sequence in S has a convergent subsequence whose limit is in S.
Theorem (Finite Dimensions): In finite-dimensional spaces, every closed and bounded set is
compact.

• The closure of S is the smallest closed set containing S (or the set of limits of convergent sequences
from S).
• The interior of S is the largest open set contained within S.

4.5 Limit Inferior and Superior

For a sequence (ξn )n in R:
• Limit inferior (liminf ): lim inf n→∞ ξn is defined as the limit of the sequence of infimums.
• Limit superior (limsup): lim supn→∞ ξn is defined as the limit of the sequence of supremums.
• Properties:
– lim inf ξn = − lim sup(−ξn ).
– lim inf ξn ≤ lim sup ξn .
– The sequence converges if and only if the liminf and limsup are equal.
– There exists a subsequence converging to the liminf.

4
4.6 Epigraph and Lower Semicontinuity
• The epigraph of f : H → [−∞, +∞] is:

epi f = {(x, ζ) ∈ dom f × R | f (x) ≤ ζ}.

• A function f is lower semicontinuous (l.s.c.) if:

– For every x ∈ H and every sequence (xn )n converging to x, lim inf n→∞ f (xn ) ≥ f (x).
– Equivalently, the epigraph epi f is a closed subset of H × R.

4.7 Coercivity
A function f : H → [−∞, +∞] is coercive if:

lim f (x) = +∞.

∥x∥→+∞

5 Minimizers and the Weierstrass Theorem

5.1 Definitions of Minimizers
For a proper function f : S →] − ∞, +∞] defined on a nonempty subset S ⊂ H:

• A point x̂ ∈ S is a local minimizer if there exists an open neighborhood O of x̂ such that:

∀ x ∈ O ∩ S, f (x̂) ≤ f (x).

• It is a global minimizer if:

∀ x ∈ S, f (x̂) ≤ f (x).

• Strict minimizers require that the inequality is strict for all other points.

5.2 Weierstrass Theorem

For Compact Sets: If S is a nonempty compact subset of H and f : S →] − ∞, +∞] is proper and
l.s.c., then there exists x̂ ∈ S such that:

f (x̂) = inf f (x).

x∈S

Moreover, the set of minimizers is compact.

In Finite Dimensions: For a finite-dimensional Hilbert space H and f : H →] − ∞, +∞] that is
proper, coercive, and l.s.c., the set of minimizers is nonempty and compact.

5.3 Example: Minimization on O(n)(R)

Let
O(n)(R) = {A ∈ Mn (R) | AA⊤ = In },
i.e., the set of orthogonal matrices. Consider minimizing the trace function Tr on O(n)(R).

• The trace is real-valued, proper, and continuous (being linear in finite dimensions).
• O(n)(R) is nonempty (since In ∈ O(n)(R)),
p closed√(as A 7→ AA⊤ is continuous), and bounded
(since for every A ∈ O(n)(R), ∥A∥ = Tr(AA⊤ ) = n).
• Thus, O(n)(R) is compact, and by the Weierstrass theorem, a minimizer of Tr exists on O(n)(R).