Fact-checked by Grok 2 weeks ago

Dual space

In linear algebra, the dual space V^* of a vector space V over a field \mathbb{F} (such as \mathbb{R} or \mathbb{C}) is the vector space consisting of all linear functionals on V, that is, all linear maps from V to \mathbb{F}. The dual space inherits the structure of a vector space under pointwise addition and scalar multiplication of functionals. If V has finite dimension n, then V^* also has dimension n, establishing a natural isomorphism between V and V^*. Given a basis \{e_1, \dots, e_n\} for V, there exists a corresponding dual basis \{e^1, \dots, e^n\} for V^* satisfying e^i(e_j) = \delta_{ij}, the Kronecker delta, which uniquely determines coordinates in the dual space. The bidual V^{**}, or dual of the dual, contains V via the natural evaluation map \hat{v}: V^* \to \mathbb{F} defined by \hat{v}(\phi) = \phi(v) for \phi \in V^*, and this embedding is an isomorphism for finite-dimensional spaces. In the context of linear transformations, the dual space facilitates the definition of the dual map or transpose: for a linear map T: V \to W, its dual T^*: W^* \to V^* is given by (T^* \psi)(v) = \psi(T v) for \psi \in W^* and v \in V, preserving the structure of homomorphisms between spaces. Beyond algebra, in functional analysis, the continuous dual refers to the space of continuous linear functionals on a topological vector space, which coincides with the algebraic dual for finite-dimensional normed spaces but differs in infinite dimensions, underpinning concepts like weak topologies and reflexivity in Banach spaces. Dual spaces are fundamental in applications ranging from optimization and quantum mechanics (via bra-ket notation) to representation theory, where they encode covectors and multilinear forms.

Algebraic dual space

Definition

In linear algebra, given a vector space V over a field K, the algebraic dual space, denoted V^*, is defined as the set of all linear functionals on V, that is, all linear maps \phi: V \to K. The elements of V^* are called linear functionals, and V^* itself forms a vector space under the operations of pointwise addition and scalar multiplication: for \phi, \psi \in V^*, v \in V, and c \in K, (\phi + \psi)(v) = \phi(v) + \psi(v), \quad (c \phi)(v) = c \cdot \phi(v). These operations endow V^* with the structure of a vector space over K. To confirm that V^* is indeed a vector space, note that the zero element is the zero functional \mathbf{0}: V \to K defined by \mathbf{0}(v) = 0 for all v \in V, which is linear since it preserves addition and scalar multiplication in the codomain. Addition and scalar multiplication in V^* preserve linearity because if \phi and \psi are linear, then for any v_1, v_2 \in V and c \in K, (\phi + \psi)(v_1 + v_2) = \phi(v_1 + v_2) + \psi(v_1 + v_2) = \phi(v_1) + \phi(v_2) + \psi(v_1) + \psi(v_2) = (\phi + \psi)(v_1) + (\phi + \psi)(v_2), and similarly for scalar multiplication, ensuring the result remains linear. The additive inverses and other vector space axioms follow from those of K. The concept of the dual space originated in the late 19th century within the development of linear algebra, but it was formalized in the early 20th century, particularly by Hans Hahn in 1927, who introduced it in the context of normed linear spaces as part of his work leading to the Hahn-Banach theorem. This algebraic construction laid the groundwork for later extensions in functional analysis by figures such as Stefan Banach. As a concrete example, consider V = K^n, the vector space of n-tuples over K. Each linear functional \phi \in (K^n)^* corresponds to a row vector \mathbf{a} = (a_1, \dots, a_n) \in K^n, where \phi(\mathbf{x}) = \mathbf{a} \cdot \mathbf{x} = \sum_{i=1}^n a_i x_i for \mathbf{x} = (x_1, \dots, x_n) \in K^n, and the vector space structure on (K^n)^* matches that of row vectors under componentwise operations.

Finite-dimensional case

When the vector space V over a field K is finite-dimensional with \dim V = n < \infty, the algebraic dual space V^* is also finite-dimensional, and \dim V^* = n. This equality of dimensions follows from the fact that the dual space consists of all linear functionals from V to K, and the space of such functionals has a basis in one-to-one correspondence with a basis of V. Moreover, V is isomorphic to V^*, though the isomorphism is not canonical and depends on the choice of a basis for V. A key construction in this setting is the dual basis. Let \{e_1, \dots, e_n\} be a basis for V. Then there exists a unique basis \{e^1, \dots, e^n\} for V^*, called the dual basis, such that e^i(e_j) = \delta_{ij} for all i, j = 1, \dots, n, where \delta_{ij} is the Kronecker delta (equal to 1 if i = j and 0 otherwise). This dual basis provides a concrete way to express any linear functional \phi \in V^* as a unique linear combination \phi = \sum_{i=1}^n \phi(e_i) e^i. In coordinates with respect to the basis \{e_1, \dots, e_n\}, linear functionals in V^* can be represented as row vectors. Specifically, for \phi \in V^*, its coordinate representation is the row vector (\phi(e_1), \dots, \phi(e_n)), and if v \in V has coordinate column vector (v_1, \dots, v_n)^T with v = \sum_{j=1}^n v_j e_j, then \phi(v) = (\phi(e_1), \dots, \phi(e_n)) \begin{pmatrix} v_1 \\ \vdots \\ v_n \end{pmatrix}. This matrix multiplication interprets the action of the dual space on V in a familiar computational form. The finite-dimensional structure also implies reflexivity: the natural evaluation map \mathrm{ev}: V \to V^{**} defined by \mathrm{ev}_v(\phi) = \phi(v) for v \in V and \phi \in V^* is a linear isomorphism. Here, V^{**} is the double dual, the dual of V^*. The map is injective because if \mathrm{ev}_v = 0, then v = 0 by the separating property of V^* on V, and surjective because \dim V = \dim V^{**}. This isomorphism identifies V with a subspace of V^{**}, embedding vectors as evaluation functionals.

Infinite-dimensional case

In the infinite-dimensional case, the algebraic dual space V^* of a vector space V over a field K with |K| \geq 2 exhibits significantly different properties compared to the finite-dimensional setting, primarily due to cardinal arithmetic. If V has a Hamel basis of infinite cardinality \kappa, then the dimension of V^* is |K|^\kappa, which strictly exceeds \kappa. This follows from the isomorphism V^* \cong K^B, where B is a basis of V with |B| = \kappa, as linear functionals are uniquely determined by their arbitrary values on B, yielding a vector space of dimension |K|^\kappa. Consequently, there exists no isomorphism between V and V^* in the infinite-dimensional case, as their dimensions differ: \dim V = \kappa < |K|^\kappa = \dim V^*. Unlike the finite-dimensional scenario, where a natural dual basis provides a canonical isomorphism, no such canonical identification is possible here without imposing additional structure, such as a topology. A concrete example illustrates this disparity: consider V = K^{(\mathbb{N})}, the vector space of sequences in K with only finitely many nonzero terms, which has dimension \aleph_0. The dual V^* then consists of all (arbitrarily supported) sequences in K^\mathbb{N}, so \dim V^* = |K|^{\aleph_0} > \aleph_0. In this setup, a dual basis to the standard countable basis of V exists but spans only the subspace of functionals with finite support relative to that basis; the full Hamel basis for V^* must have cardinality |K|^{\aleph_0}, which is uncountable and typically non-constructive to specify explicitly. Further pathologies arise in the double dual V^{**}. The dimension of V^{**} is |K|^{|K|^\kappa}, vastly larger than \dim V^*, and the canonical injection V \to V^{**} given by v \mapsto \hat{v}, where \hat{v}(f) = f(v) for f \in V^*, is injective but not surjective. This embedding identifies V with a proper subspace of V^{**}, highlighting the absence of reflexivity in the purely algebraic setting for infinite dimensions.

Bases and dimension

Given a basis B = \{e_i\}_{i \in I} of a vector space V over a field K, the dual basis \{e^i\}_{i \in I} of the algebraic dual space V^* is defined by the property that e^i(e_j) = \delta_{ij} for all i, j \in I, where \delta_{ij} is the Kronecker delta (1 if i = j, 0 otherwise). This construction ensures that the dual basis is linearly independent and spans V^* in the algebraic sense. Every linear functional \phi \in V^* admits a unique representation in coordinates with respect to the dual basis: \phi = \sum_{i \in I} \phi(e_i) e^i, where the sum is finite (i.e., has finite support) even if I is infinite, reflecting the algebraic nature of the dual space. This uniqueness follows from the linear independence of the dual basis and the fact that any linear functional is determined by its values on the basis of V. In the finite-dimensional case, if \dim V = n < \infty, then \dim V^* = n as well, since the dual basis has the same cardinality as the original basis. For infinite-dimensional V, the algebraic dimension satisfies \dim V^* = |K|^{\dim V} in the sense of cardinal arithmetic, which equals $2^{\dim V} when K = \mathbb{R} or \mathbb{C} and \dim V is infinite. More precisely, if \dim V is infinite and |K| \geq 2, the cardinality of V^* is |K|^{\dim V}, reflecting that linear functionals are determined by arbitrary assignments of values from K to the basis elements of V, with finite support in linear combinations. A concrete example arises with the vector space V = K of polynomials over K, which has Hamel basis B = \{1, x, x^2, \dots \} with \dim V = \aleph_0. The corresponding dual basis consists of linear functionals e^n for n \geq 0 such that e^n(p) extracts the coefficient of x^n in p(x) = \sum_{k=0}^\infty a_k x^k, satisfying e^n(x^m) = \delta_{nm}. Thus, \dim V^* = |K|^{\aleph_0}, which is the cardinality of all sequences in K.

Bilinear forms and pairings

A bilinear form on a vector space V over a field K is a function B: V \times V \to K that is linear in each argument separately, meaning B(av + bw, u) = a B(v, u) + b B(w, u) and B(v, au + bw) = a B(v, u) + b B(v, w) for all scalars a, b \in K and vectors v, w, u \in V. This structure allows bilinear forms to encode inner product-like behaviors in abstract vector spaces. Equivalently, any bilinear form B corresponds to a unique linear map \phi_B: V \to V^*, where V^* denotes the algebraic dual space of V, defined by B(v, w) = \langle v, \phi_B(w) \rangle for all v, w \in V, with \langle \cdot, \cdot \rangle the natural evaluation pairing between V and V^*. This identification highlights the intimate connection between bilinear forms and dual spaces, as B essentially "lifts" elements of V into functionals via the second argument. The bilinear form B is called non-degenerate if the associated map \phi_B: V \to V^* is injective (meaning if B(v, w) = 0 for all w \in V implies v = 0, and similarly for the left action). In the finite-dimensional case, non-degeneracy implies \phi_B is bijective, thereby inducing a natural isomorphism V \cong V^*. In infinite dimensions, non-degeneracy embeds V injectively into a proper subspace of V^*. Specific classes of bilinear forms include symmetric forms, where B(v, w) = B(w, v) for all v, w \in V, and alternating forms, where B(v, v) = 0 for all v \in V. A canonical example is the standard dot product on \mathbb{R}^n, defined by B(v, w) = v_1 w_1 + \cdots + v_n w_n, which is symmetric and non-degenerate, yielding the isomorphism \mathbb{R}^n \cong (\mathbb{R}^n)^* via coordinate functionals. More broadly, a pairing refers to a bilinear map \langle \cdot, \cdot \rangle : V \times W \to K between two vector spaces V and W, linear in each factor. If this pairing separates points in W—that is, if \langle v, w \rangle = 0 for all v \in V implies w = 0—then it induces an injective linear map W \to V^* given by w \mapsto (v \mapsto \langle v, w \rangle ), allowing W to be identified as a subspace of V^*. In the finite-dimensional setting, suppose V has basis \{ e_1, \dots, e_n \}; then the matrix representation of B is the Gram matrix G with entries G_{ij} = B(e_i, e_j), and B is non-degenerate if and only if \det G \neq 0. This determinant condition provides a concrete algebraic test for the isomorphism V \cong V^*.

Dual mappings and constructions

Transpose of a linear map

Given a linear map T: V \to W between vector spaces over a field F, the transpose (or dual map) T^*: W^* \to V^* is the linear map defined by (T^* \psi)(v) = \psi(T v) for all \psi \in W^* and v \in V, where V^* and W^* denote the algebraic dual spaces. This construction preserves linearity: for scalars \alpha, \beta \in F and \psi_1, \psi_2 \in W^*, (T^*(\alpha \psi_1 + \beta \psi_2))(v) = \alpha (T^* \psi_1)(v) + \beta (T^* \psi_2)(v). The transpose exhibits several key properties relating the structure of T to that of T^*. Specifically, if T is injective, then T^* is surjective; conversely, if T is surjective, then T^* is injective. In the finite-dimensional case, the dimensions satisfy \dim(\ker T^*) = \dim(\coker T), where \coker T = W / \im T, following from the rank-nullity theorem applied to the dual spaces, since \dim V^* = \dim V and \dim W^* = \dim W. In terms of bases, suppose \{v_i\} and \{w_j\} are bases for V and W, respectively, with dual bases \{v_i^*\} and \{w_j^*\}. If the matrix of T with respect to these bases is A (whose columns are the coordinates of T(v_i) in the \{w_j\}-basis), then the matrix of T^* with respect to the dual bases is the transpose A^T. For the concrete case where V = F^m and W = F^n with standard bases, T is represented by an n \times m matrix A, and T^* acts by pre-multiplication with A^T, so T^* \psi = \psi \circ A for \psi: F^n \to F. Regarding the double dual, the transpose of T^* is a map T^{**}: V^{**} \to W^{**}. In the algebraic setting, T^{**} coincides with the action of T under the canonical injection V \to V^{**}, but this identification yields an isomorphism only when V and W are finite-dimensional; in infinite dimensions, the canonical map is injective but generally not surjective.

Annihilators and quotient spaces

In the context of a vector space V over a field K and a subspace U \subseteq V, the annihilator of U, denoted U^\perp, is defined as the set of all linear functionals in the dual space V^* that vanish on every element of U: U^\perp = \{ \phi \in V^* \mid \phi(u) = 0 \ \forall \, u \in U \}. This set forms a subspace of V^*. When V is finite-dimensional, the dimension of the annihilator satisfies the relation \dim U^\perp = \dim V - \dim U. This follows from the fact that the annihilator corresponds to the kernel of the restriction map from V^* to U^*, which has dimension \dim U, and the first isomorphism theorem for vector spaces yields the codimension. The annihilator also provides a natural isomorphism with the dual space of the quotient V/U. Specifically, there is a canonical linear map \pi^*: (V/U)^* \to V^* defined by (\pi^*(\varphi))(v) = \varphi(\pi(v)), where \pi: V \to V/U is the projection. This map is injective, and its image is precisely U^\perp, establishing the isomorphism (V/U)^* \cong U^\perp. Equivalently, the map sending \psi \in U^\perp to the functional on V/U given by \tilde{\psi}(v + U) = \psi(v) is an isomorphism. Applying the annihilator construction twice yields further insights. In general, (U^\perp)^\perp \cong U^{**} via the dual of the quotient isomorphism V^*/U^\perp \cong U^*, and via the canonical injection j: V \to V^{**}, we have j(U) \subseteq (U^\perp)^\perp \subseteq V^{**}. When U (or equivalently V) is finite-dimensional, j is an isomorphism, so (U^\perp)^\perp = j(U) \cong U. In the infinite-dimensional case where \dim U = \infty, the inclusion j(U) \subsetneq (U^\perp)^\perp is generally strict. As an illustrative example, consider V = \mathbb{R}^n with the standard dual basis, and let U = \operatorname{span}\{e_1\}, where e_1 = (1, 0, \dots, 0). The annihilator U^\perp consists of all linear functionals \phi \in (\mathbb{R}^n)^* such that \phi(e_1) = 0, which are precisely those that vanish on the first coordinate. Identifying (\mathbb{R}^n)^* \cong \mathbb{R}^n via the standard inner product, U^\perp corresponds to the hyperplane orthogonal to e_1, having dimension n-1.

Double dual and canonical injection

The double dual space of a vector space V over a field k, denoted V^{**}, is defined as the algebraic dual of the dual space V^*, consisting of all k-linear functionals from V^* to k. A canonical linear map j: V \to V^{**}, known as the canonical injection (or natural embedding), is given by j(v)(\phi) = \phi(v) for all v \in V and \phi \in V^*. This construction identifies each vector v with the evaluation functional on V^* that picks out the value of any functional at v. The map j is always linear by the linearity of evaluation. The injectivity of j follows from the fact that V^* separates points on V: if j(v) = 0, then \phi(v) = 0 for every \phi \in V^*. Using the axiom of choice, one can construct a Hamel basis for V containing v (assuming v \neq 0) and define a functional \phi that is 1 on v and 0 on the rest of the basis, yielding \phi(v) = 1, a contradiction. Thus, the algebraic dual V^* always admits a separating family of functionals under the axiom of choice, ensuring j is injective. When V is finite-dimensional, j is an isomorphism, establishing that V \cong V^{**} as vector spaces. This reflexivity holds because the dimensions satisfy \dim V = \dim V^* = \dim V^{**}, and j is both injective and surjective in this case. In the infinite-dimensional case, however, j(V) forms a proper subspace of V^{**}, so V is not isomorphic to its double dual. For a concrete example, let V = \bigoplus_{n=1}^\infty k be the vector space of sequences in k with only finitely many nonzero terms (direct sum of countably many copies of k). Then V^* \cong \prod_{n=1}^\infty k, the space of all sequences in k (direct product), via the action on basis vectors. The double dual V^{**} consists of all linear functionals on this product space, which has cardinality larger than that of V (assuming k is infinite), containing elements beyond the image of j, such as functionals that sum over infinite supports in a linear fashion not representable by finite-support evaluations.

Continuous dual space

Definition and properties

In the context of a topological vector space V over a field K (typically \mathbb{R} or \mathbb{C}), the continuous dual space, denoted V', consists of all continuous linear functionals \phi: V \to K. These are linear maps that are continuous with respect to the topology on V, forming a vector subspace of the algebraic dual V^*, which comprises all linear functionals regardless of continuity. When V is a normed space, a linear functional \phi is continuous if and only if it is bounded, meaning there exists M > 0 such that |\phi(v)| \leq M \|v\| for all v \in V. The space V' inherits the structure of a topological vector space under pointwise operations: addition (\phi + \psi)(v) = \phi(v) + \psi(v) and scalar multiplication (c\phi)(v) = c \phi(v) for \phi, \psi \in V', c \in K, and v \in V. If V is normed, V' is closed under uniform convergence on the unit ball. In a Hausdorff topological vector space V, the continuous dual V' separates points: for any distinct v, w \in V, there exists \phi \in V' such that \phi(v) \neq \phi(w). For example, when V = \mathbb{R}^n is equipped with the Euclidean topology, every linear functional is continuous, so V' coincides with the algebraic dual V^*. The Hahn-Banach theorem ensures the existence of non-zero continuous linear functionals on subspaces and their extensions to the whole space while preserving boundedness, providing a foundational tool for constructing elements of V'.

Topologies on the dual space

In the context of a locally convex topological vector space V over \mathbb{R} or \mathbb{C}, the continuous dual space V' can be equipped with several standard topologies derived from the duality pairing \langle \phi, v \rangle = \phi(v) for \phi \in V' and v \in V. These topologies facilitate the study of continuity and convergence in infinite-dimensional settings, where the norm topology on V' may be too coarse or restrictive. The weak* topology on V', denoted \sigma(V', V), is the coarsest topology making all evaluation maps ev_v: V' \to \mathbb{K}, \phi \mapsto \phi(v), continuous for each v \in V. It is generated by the seminorms p_v(\phi) = |\phi(v)|, v \in V. This topology renders V' Hausdorff provided that V' separates points in V, which holds for Hausdorff locally convex spaces. A net (\phi_\alpha) in V' converges to \phi \in V' in the weak* topology if and only if \phi_\alpha(v) \to \phi(v) for every v \in V. If V is a separable Banach space, then the weak* topology restricted to the closed unit ball of V' is metrizable. The compact-open topology, or topology of uniform convergence on compact subsets of V, denoted \tau_c, has as a subbasis the sets \{\phi \in V' : \sup_{v \in K} |\phi(v) - \psi(v)| < \varepsilon\} for compact K \subset V, \varepsilon > 0, and \psi \in V'. For locally convex V, this topology coincides with the Mackey topology in certain cases and ensures continuity of linear maps when V has a rich supply of compact sets. It is finer than the weak* topology and locally convex. The strong dual topology, denoted \beta(V', V), is the topology of uniform convergence on bounded subsets of V. It is generated by the seminorms p_B(\phi) = \sup_{v \in B} |\phi(v)|, where B \subset V ranges over the bounded sets. This topology is finer than both the weak* and compact-open topologies, making V' a complete locally convex space when V is a Fréchet space, and it aligns with the norm topology on V' when V is a Banach space. Among these, the weak* topology is the weakest, ensuring V' is always locally convex, while the strong dual topology is the strongest polar topology compatible with the duality. In Banach spaces, Alaoglu's theorem asserts that the closed unit ball \{\phi \in V' : \|\phi\| \leq 1\} is compact in the weak* topology, providing a fundamental tool for existence results in optimization and approximation. This compactness fails in the norm or strong topologies.

Reflexivity and double dual

In the context of continuous dual spaces, the double dual V'' of a topological vector space V is endowed with the weak^* topology, defined as the weakest topology making all evaluation maps \hat{\phi}: V'' \to \mathbb{R} (or \mathbb{C}), \hat{\phi}(F) = F(\phi) for \phi \in V', continuous. The canonical embedding j: V \to V'', given by j(v)( \phi ) = \phi(v) for v \in V and \phi \in V', is linear and continuous from the original topology on V to the weak^* topology on V''; moreover, j is injective whenever V is a Hausdorff topological vector space. A topological vector space V is said to be reflexive if the canonical embedding j: V \to j(V) is a topological isomorphism, where V carries its original topology and j(V) is equipped with the subspace topology induced by the weak^* topology on V''. For Banach spaces, reflexivity is equivalent to j being surjective (hence an isometric isomorphism), in which case the original norm topology on V coincides with the weak topology on bounded sets and also with the pullback of the weak^* topology via j. By Kakutani's theorem, a Banach space V is reflexive if and only if its closed unit ball is weakly compact. Hilbert spaces provide a canonical example of reflexive spaces: the Riesz representation theorem identifies the continuous dual H' of a Hilbert space H isometrically with H itself via the inner product, \phi(v) = \langle u, v \rangle for some u \in H, implying that the canonical embedding j: H \to H'' is surjective and thus an isomorphism. More generally, by James' theorem, a Banach space V is reflexive if and only if every continuous linear functional on V attains its norm on the closed unit ball of V. Reflexive Banach spaces exhibit several key properties related to their topologies and geometry. The closed unit ball B_V of a reflexive Banach space V is weakly compact, and by the Krein-Milman theorem, B_V equals the weak closure of the convex hull of its extreme points; moreover, infinite-dimensional reflexive spaces possess uncountably many extreme points in B_V. A prominent example of a non-reflexive Banach space is \ell^1, the space of absolutely summable sequences: its continuous dual is \ell^\infty, the space of bounded sequences, but the double dual (\ell^\infty)^* properly contains \ell^1 as the image j(\ell^1), consisting of functionals representable by absolutely summable sequences, while (\ell^\infty)^* includes additional singular functionals on \ell^\infty that vanish on c_0 (the subspace of sequences converging to zero). Thus, j: \ell^1 \to (\ell^1)'' is a proper embedding, confirming non-reflexivity.

Examples and applications

Finite-dimensional examples

In the finite-dimensional Euclidean space \mathbb{R}^n, the algebraic dual space (\mathbb{R}^n)^* consists of all linear functionals on \mathbb{R}^n, which can be identified with \mathbb{R}^n itself through the standard dot product \langle x, y \rangle = x \cdot y, where each functional \phi_y(z) = y \cdot z for y \in \mathbb{R}^n. This identification is canonical and preserves the vector space structure, with the dimension of the dual space equaling n. In the presence of the standard Euclidean topology, the continuous dual space coincides with the algebraic dual, as every linear functional is continuous. For the space of m \times n matrices M_{m \times n}(\mathbb{K}) over a field \mathbb{K} (such as \mathbb{R} or \mathbb{C}), the dual space has dimension mn and can be identified with M_{n \times m}(\mathbb{K}) via the trace inner product \langle A, B \rangle = \operatorname{Tr}(A^T B), where each functional is given by \phi_B(A) = \operatorname{Tr}(B^T A). This pairing induces an isomorphism, and under the Frobenius norm (derived from the trace inner product), all algebraic dual elements are continuous. The space of polynomials P_n(\mathbb{K}) of degree at most n over \mathbb{K} provides another example, where the dual space P_n(\mathbb{K})^* admits bases formed by evaluation functionals at n+1 distinct points s_0, \dots, s_n, defined by \phi_{s_j}(p) = p(s_j) for p \in P_n(\mathbb{K}). Alternatively, coefficient functionals extracting the coefficients in the monomial basis also span the dual. Equipped with the supremum norm on a compact interval, all these functionals are continuous, making the continuous dual identical to the algebraic dual. Over the complex numbers, the dual of \mathbb{C}^m is naturally represented by row vectors, where a functional \phi acts as \phi(z) = w z for a row vector w \in \mathbb{C}^{1 \times m} and column vector z \in \mathbb{C}^m, under the standard topology. This representation highlights the isomorphism between \mathbb{C}^m and its dual, facilitated by the Hermitian inner product. In finite-dimensional optimization, elements of the dual space appear as Lagrange multipliers, where for a constrained problem \min f(x) subject to g(x) = 0, the multipliers \lambda lie in the dual of the constraint space, enabling the formulation of the Lagrangian \mathcal{L}(x, \lambda) = f(x) + \lambda \cdot g(x).

Common infinite-dimensional examples

In infinite-dimensional settings, the continuous dual spaces of classical Banach spaces often admit explicit identifications that differ markedly from their algebraic duals, which consist of all linear functionals without continuity requirements. A prominent example is the sequence space \ell^p for $1 \leq p < \infty, whose continuous dual (\ell^p)^* is isometrically isomorphic to \ell^q, where \frac{1}{p} + \frac{1}{q} = 1. The duality pairing is given by \langle x, y \rangle = \sum_{n=1}^\infty x_n y_n for x = (x_n) \in \ell^p and y = (y_n) \in \ell^q, with the operator norm on \ell^q ensuring boundedness: for any \phi \in (\ell^p)^*, there exists a = (a_n) \in \ell^q such that \phi(x) = \sum_{n=1}^\infty a_n x_n and \|\phi\| = \|a\|_q < \infty. When p=1, the conjugate exponent q = \infty, so (\ell^1)^* \cong \ell^\infty, where functionals act via summation against bounded sequences. A parallel duality holds for the function spaces L^p(\mu) over a \sigma-finite measure space (X, \mathcal{A}, \mu), where $1 \leq p < \infty: the continuous dual (L^p(\mu))^* is isometrically isomorphic to L^q(\mu) under the pairing \langle f, g \rangle = \int_X f \, g \, d\mu for f \in L^p(\mu) and g \in L^q(\mu). This identification follows from the Riesz representation theorem for L^p spaces, which characterizes bounded linear functionals as integration against L^q functions, again with \|\phi\| = \|g\|_q < \infty. For p=1, the dual is L^\infty(\mu). Another key example is the space C(K) of continuous real- or complex-valued functions on a compact Hausdorff space K, equipped with the sup norm. Its continuous dual C(K)^* is isometrically isomorphic to the space M(K) of finite signed regular Borel measures on K, via the pairing \langle f, \mu \rangle = \int_K f \, d\mu for f \in C(K) and \mu \in M(K). This follows from the Riesz–Markov–Kakutani representation theorem, which ensures every bounded linear functional on C(K) arises uniquely from such a measure, with the total variation norm \|\mu\|_{TV} matching the dual norm. The weak* topology on M(K) is particularly useful for studying convergence of sequences of measures in this dual space. In the context of Sobolev spaces, the continuous dual of H^1(\Omega) for a bounded domain \Omega \subset \mathbb{R}^n with smooth boundary involves distributions, but it can be identified with H^{-1}(\Omega), the completion of L^2(\Omega) under the graph norm involving the Laplacian. Elements of H^{-1}(\Omega) act on H^1(\Omega) via integration: for u \in H^1(\Omega) and v \in H^{-1}(\Omega), the duality is \langle v, u \rangle = \int_\Omega v u \, dx + \int_\Omega \nabla v \cdot \nabla u \, dx when v \in H^1(\Omega), extended by density. This structure highlights how boundary conditions influence the dual, with H_0^1(\Omega)^*\cong H^{-1}(\Omega) for functions vanishing on \partial \Omega. A notable distinction in infinite dimensions is the size of the algebraic versus continuous duals. For instance, the algebraic dual of \ell^2 comprises all linear functionals on sequences, which is vastly larger (in cardinality and without norm structure) than the continuous dual (\ell^2)^* \cong \ell^2 under the Hilbert space inner product \langle x, y \rangle = \sum_{n=1}^\infty x_n \overline{y_n}. Continuity imposes the boundedness condition \sup_{\|x\|_2 \leq 1} |\phi(x)| < \infty, restricting functionals to those representable by \ell^2 sequences.

Hahn-Banach theorem and extensions

The Hahn-Banach theorem provides a fundamental tool for extending linear functionals on normed vector spaces while preserving bounds, playing a central role in the study of dual spaces. In its analytic form for normed spaces over the real or complex field, the theorem states that if V is a normed vector space, M a subspace of V, and \phi: M \to \mathbb{K} (where \mathbb{K} = \mathbb{R} or \mathbb{C}) a bounded linear functional satisfying |\phi(x)| \leq \|x\| for all x \in M, then there exists a bounded linear extension \psi: V \to \mathbb{K} such that |\psi(x)| \leq \|x\| for all x \in V and \|\psi\| = \|\phi\|. This extension is achieved without increasing the norm of the functional, ensuring the continuous dual space V^* inherits key structural properties from subspaces. A key corollary is that every non-trivial normed space has a non-trivial continuous dual space. For any x \in V with x \neq 0, consider the one-dimensional subspace M = \operatorname{span}\{x\} and define \phi(tx) = t \|x\| for t \in \mathbb{K}; this satisfies |\phi(y)| \leq \|y\| for y \in M, so Hahn-Banach extends it to \psi \in V^* with \psi(x) = \|x\| and \|\psi\| = 1. Consequently, the dual separates points: for distinct x, y \in V, there exists f \in V^* such that f(x) \neq f(y), as the functional separating x - y from 0 witnesses this. In the complex case, the theorem adapts by treating the real and imaginary parts separately or directly using a sublinear majorant p(x) = \|x\|. Specifically, if \phi: M \to \mathbb{C} is \mathbb{C}-linear and bounded, it extends to \psi: V \to \mathbb{C} preserving the bound, since the real part \operatorname{Re} \phi is a real-linear functional dominated by \|\cdot\|, extendable via the real version, and the imaginary part follows analogously or via phase adjustment. This ensures the result holds uniformly for both real and complex scalars without altering the core extension mechanism. Applications include the existence of bounded functionals attaining their norms: for any x \in V, the extension constructed above yields f \in V^* with \|f\| = 1 and f(x) = \|x\|, confirming that the unit ball of V^* attains its supremum on the unit ball of V. The geometric form further asserts that if A and B are disjoint convex sets in a normed space with A open, there exists f \in V^* with \|f\| = 1 and \operatorname{Re} f(a) < \alpha \leq \operatorname{Re} f(b) for all a \in A, b \in B, and some \alpha \in \mathbb{R}; this implies supporting hyperplanes for convex sets at boundary points, where a hyperplane \{z \in V : \operatorname{Re} f(z) = \operatorname{Re} f(x_0)\} touches the set at x_0 without crossing it. The standard proof relies on Zorn's lemma for the extension aspect. First, a base step extends any bounded linear functional from a subspace to a one-codimensional enlargement by solving for the value on a new vector v satisfying the bound via the inequality p(v - m) \geq \phi(m) for m \in M, yielding a unique choice in [\inf (\phi(m) + p(v - m)), \sup (-\phi(m) + p(v + m))]. The collection of all such partial extensions, ordered by inclusion of domains, forms a partially ordered set where chains have upper bounds (unions), so Zorn's lemma guarantees a maximal extension, which must cover V by density or direct construction. Analytic extensions in several complex variables build on this, adapting to plurisublinear majorants for holomorphic functionals, though the linear case remains the foundation. A notable application in infinite-dimensional dual spaces is the construction of functionals beyond standard representations. In L^\infty(\mu), Hahn-Banach yields continuous linear functionals not representable as integration against L^1(\mu) elements; for instance, Banach limits on \ell^\infty \subset L^\infty([0,1]) extend the limsup functional from the subspace of convergent sequences to a shift-invariant positive functional on all bounded sequences, with \liminf x_n \leq L(x) \leq \limsup x_n and L(x) = \lim x_n for convergent x, but not arising from any g \in \ell^1.