1 Statement of the Riesz representation theorem

The Riesz representation theorem states that continuous (equivalently, bounded) linear functionals on a Hilbert space can be expressed in terms of the inner product with a unique vector of the space. This result turns questions about the dual space into questions about geometry inside the original Hilbert space.

1.1 Linear functionals and boundedness

Let \(H\) be a Hilbert space over \(\mathbb{R}\) or \(\mathbb{C}\). A linear functional is a map \(f:H\to \mathbb{F}\) (with \(\mathbb{F}=\mathbb{R}\) or \(\mathbb{C}\)) satisfying \(f(ax+by)=af(x)+bf(y)\). The functional is continuous precisely when it is bounded, meaning there exists \(C\ge 0\) such that \[

f(x)\le C\|x\|\quad\text{for all }x\in H.

\] In a Hilbert space, continuity and boundedness coincide for linear functionals.

1.2 Hilbert spaces and inner product conventions

A Hilbert space is a complete inner product space. The inner product \(\langle \cdot,\cdot\rangle\) determines the norm by \(\|x\|=\sqrt{\langle x,x\rangle}\).

Different authors adopt different linearity conventions in the first or second argument. The theorem can be written compatibly with either convention; the key point is that the representing vector appears inside the inner product in a consistent manner.

A common convention (complex case) is: the inner product is linear in the first argument and conjugate-linear in the second. Under this convention, the representation has the form \[ f(x)=\langle x, y\rangle \] for a unique \(y\in H\). With the opposite convention, one obtains \(f(x)=\langle y,x\rangle\).

1.3 Real vs. complex Hilbert spaces

The theorem holds in both real and complex Hilbert spaces. The underlying geometry is similar, but the complex case involves conjugation due to conjugate-linearity in one slot of the inner product.

Formally, for a bounded linear functional \(f\in H^*\), there exists a unique vector \(y\in H\) such that \[ f(x)=\langle x,y\rangle \quad (\text{or } \langle y,x\rangle,\text{ depending on convention}). \]

1.3.1 Uniqueness of the representing vector

If \(f(x)=\langle x,y\rangle=\langle x,z\rangle\) for all \(x\in H\), then \(\langle x,y-z\rangle=0\) for every \(x\). Taking \(x=y-z\) yields \(\|y-z\|^2=0\), so \(y=z\). Hence the representing vector is unique.

2 Constructive interpretation via inner products

Although the statement is existential, the theorem also provides a method to understand a functional through a specific vector in \(H\). This vector is often described informally as the “gradient-like” representative of the functional.

2.1 From a functional to a representing vector

Given \(f\in H^*\), one seeks \(y\in H\) such that \(f(x)=\langle x,y\rangle\) for all \(x\). Conceptually, the representing vector is characterized by the rule that it reproduces the functional’s values on every direction in \(H\).

In proofs, one typically constructs \(y\) from the orthogonal complement of \(\ker f\) and then uses density/completeness to show the constructed vector satisfies the required identity for all \(x\).

2.2 Orthogonality and kernel structure

The functional \(f\) has a kernel \(\ker f=\{x\in H: f(x)=0\}\), which is a closed subspace when \(f\) is bounded. Orthogonality becomes central because any vector representing \(f\) must be orthogonal to \(\ker f\).

Indeed, if \(f(x)=\langle x,y\rangle\), then for any \(x\in \ker f\), \[ 0=f(x)=\langle x,y\rangle, \] so \(y\in (\ker f)^\perp\). Conversely, a vector orthogonal to \(\ker f\) carries the information needed to recover \(f\).

2.3 Norm equality and estimates

A striking aspect of the theorem is that the norm of the functional equals the norm of the representing vector: \[

\|f\|=\|y\|.

\] Moreover, the identity \(f(x)=\langle x,y\rangle\) yields the estimate \[

f(x)=\langle x,y\rangle\le \|x\|\|y\|

\] by Cauchy–Schwarz, showing boundedness directly.

2.3.1 Operator norm in terms of the representing element

The operator norm of \(f\) can be expressed as \[

\|f\|=\sup_{\|x\|=1}f(x)=\sup_{\|x\|=1}\langle x,y\rangle.

\]

In a Hilbert space, this supremum is attained when \(x\) is proportional to \(y\) (or when \(y=0\)). Consequently, \(\|f\|=\|y\|\), aligning the functional’s size with the geometric magnitude of its representative.

3 Geometric consequences in Hilbert spaces

The theorem’s power lies in converting analytic objects (functionals on \(H\)) into geometric data (inner products with vectors). This supports intuitive reasoning about orthogonality, approximation, and projections.

3.1 Orthogonal decomposition framework

Given a bounded functional \(f\), the kernel \(\ker f\) admits an orthogonal decomposition \[ H=\ker f \oplus (\ker f)^\perp. \] Since the representing vector \(y\) lies in \((\ker f)^\perp\), the functional depends only on the component of \(x\) in that orthogonal complement. Vectors in \(\ker f\) contribute nothing to \(f(x)\).

3.2 The representer of a functional and best approximation

The representation interacts naturally with approximation questions. For instance, if one wants to find a vector that reproduces values of a functional as closely as possible on a subspace, the representing vector determines the optimal direction. The orthogonality properties imply that the error term lies in \(\ker f\) or its complement, yielding “best approximation” behavior typical of projection arguments in Hilbert spaces.

3.3 Projection theorem connections

Projections in Hilbert spaces are characterized by orthogonality. Since \(f\) is encoded by an inner product, properties of \(f\) translate into properties of corresponding projections.

3.3.1 Characterizing projections using represented functionals

Let \(M\subset H\) be a closed subspace and \(P_M\) the orthogonal projection onto \(M\). For any \(x\in H\), the difference \(x-P_Mx\) is orthogonal to \(M\). When a functional \(f\) is represented as \(f(u)=\langle u,y\rangle\), choosing \(y\) appropriately allows one to express \(f(P_Mx)\) and \(f(x-P_Mx)\) in terms of inner products with the projected components, clarifying when \(f\) annihilates residual terms.

The Riesz theorem yields multiple canonical identifications. These results are often framed as “maps” between \(H\) and its dual, and they support operator-theoretic formulations.

4.1 Riesz map (isometric isomorphism) viewpoint

Define the Riesz map \(J:H\to H^*\) by \[ (Jy)(x)=\langle x,y\rangle \] (again, with the appropriate convention). The theorem states that \(J\) is surjective and injective, so it is an isomorphism between \(H\) and \(H^*\). It is also an isometry, meaning \[

\|Jy\|=\|y\|,

\] so the dual norm corresponds exactly to the original norm.

4.2 Identification of the dual space with the space itself

Because \(J\) is an isometric isomorphism, one can often treat \(H^*\) as “the same” as \(H\), with the caveat that scalar multiplication and conjugation in the complex case must be interpreted correctly. This identification simplifies many arguments: duality pairing becomes inner product pairing.

4.3 Adjoint operators and the Riesz correspondence

Operator adjoints are defined using inner products; the Riesz theorem clarifies how dual spaces enter those definitions. If \(T:H\to H\) is bounded, the adjoint \(T^*\) is characterized by \[ \langle Tx, y\rangle=\langle x, T^*y\rangle \quad\text{for all }x,y\in H. \] Through the Riesz map, one can translate between functionals \(x\mapsto \langle Tx, y\rangle\) and vectors \(T^*y\).

4.3.1 Self-adjointness under inner product representations

An operator \(T\) is self-adjoint when \(T=T^*\), equivalently when \[ \langle Tx,y\rangle=\langle x,Ty\rangle. \] The Riesz representation ensures that such identities are meaningful because every continuous linear functional is realized by an inner product. As a result, statements about adjoints can be verified by testing them on inner products against arbitrary vectors.

4.4 The Lax–Milgram theorem linkage (overview-level)

In applications to partial differential equations, one often seeks a solution \(u\in H\) such that a bounded sesquilinear form \(a(u,v)\) matches a linear functional \(f(v)\) for all \(v\in H\). The Riesz theorem underpins this framework by ensuring that \(f\) corresponds to an inner-product pairing with some vector, and by enabling the translation of existence/uniqueness statements into operator invertibility of a bounded coercive operator. (The full Lax–Milgram theorem is typically presented separately, but its mechanics rely on Riesz-type correspondences.)

5 Variants and extensions

The theorem adapts to other settings, including changes in linearity type and additional structure on the domain space.

5.1 Bounded linear functionals on pre-Hilbert spaces

A pre-Hilbert space is an inner product space that may not be complete. Bounded linear functionals can still be defined, but the representation theorem may fail without completeness. Completing the space often restores the setting in which Riesz representation applies. In practice, one extends the functional continuously to the completion and then represents it there, restricting back when needed.

5.2 Riesz representation for anti-linear functionals

In complex Hilbert spaces, one can consider anti-linear functionals \(g:H\to \mathbb{C}\) satisfying \(g(ax)=\overline{a}\,g(x)\). A parallel representation holds: such functionals can be represented using the inner product with a conjugation pattern matched to anti-linearity. This variant is useful when dealing with conjugate-linear mappings or sesquilinear forms.

5.3 Applications to weak formulations

In weak formulations of problems (notably in analysis and mathematical physics), one seeks \(u\) such that \[ a(u,v)=\ell(v)\quad\text{for all }v\in H, \] where \(\ell\) is a bounded linear functional. Through Riesz representation, \(\ell\) can be written as an inner product with a vector, often simplifying the algebraic structure of the variational problem and clarifying how right-hand sides act in the space.

5.4 Examples on standard function spaces

Many concrete examples rely on identifying the relevant Hilbert space and then computing the representing vector using known inner products.

5.4.1 Typical inner product representations for distributions-free functionals

In settings like \(L^2\)-type spaces, continuous linear functionals can often be realized by inner products with an \(L^2\) function. For example, if \(H=L^2(\Omega)\) with inner product \(\langle u,v\rangle=\int_\Omega u\overline{v}\,dx\), then functionals of the form \[ f(u)=\int_\Omega u\,\overline{g}\,dx \] are represented by the vector \(g\). The key point is that not all linear functionals are continuous; continuity is what forces the representative to belong to the Hilbert space itself.

6 Proof strategies

The theorem can be proved in several standard ways. Different approaches emphasize different geometric or analytic viewpoints.

6.1 Proof using orthonormal bases

One method selects an orthonormal basis \(\{e_n\}\) of \(H\). Using boundedness, one shows that the coefficients \( \overline{f(e_n)} \) form a square-summable sequence and thus define a vector \(y\in H\) via \[ y=\sum_n \overline{f(e_n)}\,e_n \] (up to convention). Then one verifies that for each \(x\in H\), the series expansions and continuity of \(f\) give \(f(x)=\langle x,y\rangle\). Uniqueness follows from orthogonality as in the theorem statement.

6.2 Proof using orthogonal complements

Another approach begins with the subspace \(\ker f\), which is closed. If \(f\neq 0\), then \((\ker f)^\perp\) is nontrivial and one can choose a vector \(y\) in \((\ker f)^\perp\) with the property that \(f(y)\neq 0\). Every vector \(x\) decomposes orthogonally as \(x=x_0+x_1\) with \(x_0\in\ker f\) and \(x_1\in(\ker f)^\perp\). Since \(f(x_0)=0\), the functional depends only on \(x_1\), and the inner-product relation is established by scaling \(y\) appropriately.

6.3 Proof via completion and reduction arguments

For pre-Hilbert spaces or for situations where the Hilbert structure is obtained by completion, one can proceed as follows: extend \(f\) to the completion \(\overline{H}\), represent it there by some \(y\in \overline{H}\), and then analyze whether \(y\) actually lies in the original subspace. This strategy is particularly useful when constructing \(f\) naturally on a dense subspace.

7 Illustrative examples

Concrete examples show how functionals correspond to vectors and how continuity constraints determine which representatives are allowed.

7.1 Representing evaluation functionals in \(L^2\)-type settings

In pure \(L^2(\Omega)\) spaces, point evaluation \(u\mapsto u(x_0)\) is generally not continuous, so it is not representable as an inner product with an \(L^2\) function. This illustrates the role of continuity: only those functionals arising from inner products with elements of the Hilbert space qualify.

7.2 Representing inner-product-induced functionals

For a fixed \(y\in H\), the map \[ f_y(x)=\langle x,y\rangle \]

is always linear (or conjugate-linear, depending on convention) and bounded, with \(\|f_y\|=\|y\|\). The Riesz theorem says that every bounded linear functional arises uniquely in this way. Thus \(f_y\) provides the “model family” for the dual.

7.3 Computing the representing element from a given formula

When a bounded linear functional is presented explicitly, one can often read off the representing vector by matching the functional’s formula to the inner product definition.

7.3.1 Checking continuity and determining the norm

Given a proposed representation \(f(x)=\langle x,y\rangle\), continuity follows immediately from Cauchy–Schwarz: \[

f(x)\le \|x\|\,\|y\|.

\]

To confirm that the functional is genuinely bounded and to find \(\|f\|\), one uses the norm equality \(\|f\|=\|y\|\). If the functional is given without \(y\), continuity can sometimes be tested by attempting to bound it uniformly by \(\|x\|\); when successful, the Riesz theorem guarantees existence of a representing vector, and its norm can be determined from the least bounding constant.