Dual spaces

Prices, readings and integrals each turn a vector into a single number, linearly. What do all such measurements of a space look like together, and what does a linear map do to them?

Where you are 5 in Linear maps

Why this exists

Evaluating a polynomial at a point, integrating a function over an interval, taking the trace of a matrix, pricing a list of quantities: each takes a vector, returns one number, and respects sums and scalar multiples. Such a map is a linear functional, and the functionals on V form a new vector space, the dual space V'. Two questions organize this lesson. Which measurements read off the coordinates in a basis? And when a linear map T: V \to W acts first, how does a measurement on W become one on V? The answer to the second gives the transpose of a matrix a meaning beyond flipping rows and columns: it is the matrix of the dual map.

At Amara and Hiro’s meadery a honey blend is a vector of kilograms, its cost is a functional, and the batch count of each house mix is a dual basis functional (Example 1).

Definition

Let V be a vector space over \mathbb{F}. A linear functional on V is an element of \mathcal{L}(V, \mathbb{F}), a linear map to the scalars. The dual space of V is

V' = \mathcal{L}(V, \mathbb{F}),

a vector space over \mathbb{F} with the pointwise operations of Result 3 of Linear maps between vector spaces. Functionals are written \varphi, \psi.

Let (\mathbf{v}_1, \dots, \mathbf{v}_n) be a basis of V. Its dual basis is the list (\varphi_1, \dots, \varphi_n) in V' with

\varphi_j(\mathbf{v}_k) = \delta_{jk} = \begin{cases} 1, & j = k, \\ 0, & j \neq k. \end{cases}

Each \varphi_j exists and is unique by the linear map lemma (Result 5 of Linear maps between vector spaces). For a basis \mathcal{B}, write \mathcal{B}' for its dual basis.

For T \in \mathcal{L}(V, W), the dual map is T': W' \to V',

T'(\psi) = \psi T \quad \text{for } \psi \in W',

the functional on V that applies T and then measures with \psi. For a subspace U of V, the annihilator of U is U^{0} = \{\varphi \in V' : \varphi(\mathbf{u}) = 0 \text{ for all } \mathbf{u} \in U\}.

Intuition

On \mathbb{F}^n a functional is a row: by Result 6 of Linear maps between vector spaces with m = 1, every \varphi \in (\mathbb{F}^n)' is \varphi(\mathbf{x}) = \mathbf{a}\mathbf{x} for exactly one 1 \times n matrix \mathbf{a}, with entries a_k = \varphi(\mathbf{e}_k). Vectors are columns, functionals are rows.

Over \mathbb{R}^2 a nonzero functional shows itself through its level lines \varphi = 0, 1, 2, \dots, parallel and evenly spaced. A dual basis draws the coordinate grid of its basis: in the figure the lines \varphi_1 = k run parallel to \mathbf{b}_2, the lines \varphi_2 = k parallel to \mathbf{b}_1, and \mathbf{v} sits where \varphi_1 = 2 meets \varphi_2 = 1, so \mathbf{v} = 2\mathbf{b}_1 + \mathbf{b}_2. The dual basis functionals are the coordinate functions: \varphi_j(\mathbf{v}) is entry j of [\mathbf{v}]_{\mathcal{B}} (Coordinates in a basis and change of basis).

\varphi_1 = 0
\varphi_1 = 1
\varphi_1 = 2
\varphi_2 = 0
\varphi_2 = 1
\varphi_2 = 2
\mathbf{b}_1
\mathbf{b}_2
\mathbf{v}
\mathbf{0}
Figure 1The dual basis of \mathbf{b}_1 = (2, 1), \mathbf{b}_2 = (-1, 1) from Example 2. Solid lines are the level lines \varphi_1 = 0, 1, 2, dashed lines the level lines \varphi_2 = 0, 1, 2; together they are the coordinate grid of the basis, and \mathbf{v} = (3, 3) has \varphi_1(\mathbf{v}) = 2, \varphi_2(\mathbf{v}) = 1.

The dual map runs backward: “apply T, then measure” turns a measurement on W into one on V. Hence (ST)' = T'S', mirroring (AB)^{T} = B^{T}A^{T} in The transpose and symmetric matrices, which Result 4 explains.

Results

Throughout, U, V, W are vector spaces over \mathbb{F}, and a result that needs finite dimension says so. Every result holds, with the same proof, over any field.

Result 1 (dimension of the dual). If V is finite-dimensional, then so is V', and \dim V' = \dim V. In particular V \cong V'.

Result 2 (the dual basis). Let (\mathbf{v}_1, \dots, \mathbf{v}_n) be a basis of V with dual basis (\varphi_1, \dots, \varphi_n). Then (\varphi_1, \dots, \varphi_n) is a basis of V', and for every \mathbf{v} \in V and \varphi \in V',

\begin{aligned} \mathbf{v} &= \varphi_1(\mathbf{v})\mathbf{v}_1 + \dots + \varphi_n(\mathbf{v})\mathbf{v}_n, \\ \varphi &= \varphi(\mathbf{v}_1)\varphi_1 + \dots + \varphi(\mathbf{v}_n)\varphi_n. \end{aligned}

If V = \mathbb{F}^n and B = [\,\mathbf{v}_1 \ \cdots \ \mathbf{v}_n\,], then \varphi_j(\mathbf{x}) is row j of B^{-1} times \mathbf{x}.

Result 3 (the dual map). For T \in \mathcal{L}(V, W), T' is linear from W' to V'. For S, T \in \mathcal{L}(V, W) and a \in \mathbb{F}, (S + T)' = S' + T' and (aT)' = aT'. For T \in \mathcal{L}(U, V) and S \in \mathcal{L}(V, W), (ST)' = T'S'; and I' = I.

Result 4 (the matrix of the dual map is the transpose). Let V, W be finite-dimensional with bases \mathcal{B}, \mathcal{C}, and T \in \mathcal{L}(V, W). Then

[T']_{\mathcal{B}' \leftarrow \mathcal{C}'} = \big([T]_{\mathcal{C} \leftarrow \mathcal{B}}\big)^{T}.

Result 5 (annihilators). For every subspace U of V, U^{0} is a subspace of V'. If V is finite-dimensional, then \dim U^{0} = \dim V - \dim U.

Result 6 (null space and range of the dual map). For T \in \mathcal{L}(V, W), \ker T' = (\operatorname{im} T)^{0}. If V and W are finite-dimensional, then also \operatorname{im} T' = (\ker T)^{0} and \operatorname{rank} T' = \operatorname{rank} T; so T' is injective if and only if T is surjective, and surjective if and only if T is injective.

Result 7 (row rank equals column rank). For every A \in \mathbb{F}^{m \times n}, \operatorname{rank} A^{T} = \operatorname{rank} A: the rows of A span a space of the same dimension as its columns.

Why it is true

Result 1. V' = \mathcal{L}(V, \mathbb{F}) and \dim \mathbb{F} = 1. If \dim V = n, Result 7 of Invertibility and isomorphic spaces gives \dim V' = n \cdot 1 = n, and Result 4 there gives V \cong V'.

Result 2. If a_1\varphi_1 + \dots + a_n\varphi_n = 0, applying it to \mathbf{v}_k gives a_k = 0. So the list is independent, of length n = \dim V' (Result 1), hence a basis (Result 5 of Bases and dimension, proved). Write \mathbf{v} = c_1\mathbf{v}_1 + \dots + c_n\mathbf{v}_n; applying \varphi_j gives c_j = \varphi_j(\mathbf{v}), the first formula. The two sides of the second formula are functionals with the same value \varphi(\mathbf{v}_k) at each \mathbf{v}_k, so they are equal by the uniqueness in the linear map lemma. For V = \mathbb{F}^n, B is invertible since its columns form a basis (The invertible matrix theorem). If \mathbf{r}_j is row j of B^{-1}, entry (j, k) of B^{-1}B = I_n says \mathbf{r}_j\mathbf{v}_k = \delta_{jk}, and \mathbf{x} \mapsto \mathbf{r}_j\mathbf{x} is linear, so it is \varphi_j.

Result 3. For \psi \in W', \psi T is a composite of linear maps, so it lies in V' (Result 4 of Linear maps between vector spaces). By the distributive laws of that result, for \psi_1, \psi_2 \in W' and a \in \mathbb{F},

\begin{aligned} T'(a\psi_1 + \psi_2) &= (a\psi_1 + \psi_2)T \\ &= a\,\psi_1T + \psi_2T, \end{aligned}

so T' is linear. Likewise (S + T)'(\psi) = \psi S + \psi T and (aT)'(\psi) = a\,\psi T. By associativity, (ST)'(\psi) = (\psi S)T = T'(S'(\psi)), and I'(\psi) = \psi I = \psi.

Result 4. Let \mathcal{B} = (\mathbf{v}_1, \dots, \mathbf{v}_n) and \mathcal{C} = (\mathbf{w}_1, \dots, \mathbf{w}_m) have dual bases (\varphi_k) and (\psi_j), and let A = [T]_{\mathcal{C} \leftarrow \mathcal{B}} and G = [T']_{\mathcal{B}' \leftarrow \mathcal{C}'}: so T\mathbf{v}_k = \sum_{i} A_{ik}\mathbf{w}_i and T'(\psi_j) = \sum_{k} G_{kj}\varphi_k. By the second formula of Result 2, the coefficient of \varphi_k in a functional is its value at \mathbf{v}_k. Hence

\begin{aligned} G_{kj} &= (T'\psi_j)(\mathbf{v}_k) = \psi_j(T\mathbf{v}_k) \\ &= \psi_j\Big(\sum_{i} A_{ik}\mathbf{w}_i\Big) = A_{jk}, \end{aligned}

so G = A^{T}.

Result 5. The zero functional lies in U^{0}, and if \varphi, \psi \in U^{0}, then (a\varphi + \psi)(\mathbf{u}) = a\varphi(\mathbf{u}) + \psi(\mathbf{u}) = 0 for \mathbf{u} \in U. Now let \dim V = n, take a basis (\mathbf{u}_1, \dots, \mathbf{u}_m) of U (Result 6 of Bases and dimension, proved), extend it to a basis (\mathbf{u}_1, \dots, \mathbf{u}_m, \mathbf{v}_{m+1}, \dots, \mathbf{v}_n) of V (Result 3 there), and let (\varphi_1, \dots, \varphi_n) be its dual basis. For j > m, \varphi_j vanishes at each \mathbf{u}_k, hence on their span U. Conversely, Result 2 expands any \varphi \in U^{0} with coefficients its values at the basis vectors, and the first m are \varphi(\mathbf{u}_k) = 0. So U^{0} = \operatorname{span}\{\varphi_{m+1}, \dots, \varphi_n\}, an independent list, and \dim U^{0} = n - m.

Result 6. \psi \in \ker T' means \psi(T\mathbf{v}) = 0 for every \mathbf{v} \in V, that is, \psi vanishes on \operatorname{im} T. Now let \dim V = n, \dim W = m, r = \operatorname{rank} T. By Result 5, \dim \ker T' = m - r, so the fundamental theorem of linear maps (Result 5 of Null space, range and the fundamental theorem) for T', with \dim W' = m, gives \operatorname{rank} T' = r. If \varphi = \psi T and \mathbf{v} \in \ker T, then \varphi(\mathbf{v}) = \psi(\mathbf{0}) = 0, so \operatorname{im} T' \subseteq (\ker T)^{0}. The fundamental theorem for T gives \dim \ker T = n - r, so \dim (\ker T)^{0} = r by Result 5, and a subspace of equal finite dimension is the whole space (Result 6 of Bases and dimension, proved). Finally, T' is injective exactly when m - r = 0, that is, when T is surjective; and T' is surjective exactly when r = n, that is, when \ker T = \{\mathbf{0}\}, which means T is injective (Result 2 of Null space, range and the fundamental theorem).

Result 7. Let T\mathbf{x} = A\mathbf{x} from \mathbb{F}^n to \mathbb{F}^m. Relative to the standard bases [T] = A, so [T'] = A^{T} by Result 4. A map and its matrix have the same rank, since the coordinate isomorphism carries \operatorname{im} T onto the column space (Result 6 of The matrix of a linear map). So \operatorname{rank} A^{T} = \operatorname{rank} T' = \operatorname{rank} T = \operatorname{rank} A by Result 6, and the columns of A^{T} are the rows of A. This proof of the fact in Rank never reduces a matrix.

Worked examples

Example 1

Reading off batches at the meadery

Amara blends three honeys, orange blossom, clover and buckwheat; a blend is \mathbf{x} = (x_1, x_2, x_3) \in \mathbb{R}^3 in kilograms. Hiro’s house mixes are \mathbf{b}_1 = (1, 1, 0), \mathbf{b}_2 = (0, 1, 1) and \mathbf{b}_3 = (1, 0, 2). (a) Find the dual basis of \mathcal{B} = (\mathbf{b}_1, \mathbf{b}_2, \mathbf{b}_3) and the number of batches of each mix in the order (3, 5, 5). (b) The honeys cost \$6, \$9 and \$12 per kilogram. Write the cost functional c in the dual basis and explain its coefficients.

Solution

(a) B = [\,\mathbf{b}_1 \ \mathbf{b}_2 \ \mathbf{b}_3\,] has determinant 3, so \mathcal{B} is a basis, and by Result 2 the dual basis functionals are the rows of

B^{-1} = \frac13\begin{bmatrix} 2 & 1 & -1 \\ -2 & 2 & 1 \\ 1 & -1 & 1 \end{bmatrix}.

That is,

\begin{aligned} \varphi_1(\mathbf{x}) &= \tfrac13(2x_1 + x_2 - x_3), \\ \varphi_2(\mathbf{x}) &= \tfrac13(-2x_1 + 2x_2 + x_3), \\ \varphi_3(\mathbf{x}) &= \tfrac13(x_1 - x_2 + x_3). \end{aligned}

At (3, 5, 5) they give 2, 3 and 1, so by Result 2 the order is 2\mathbf{b}_1 + 3\mathbf{b}_2 + \mathbf{b}_3; indeed (2, 2, 0) + (0, 3, 3) + (1, 0, 2) = (3, 5, 5).

(b) c(\mathbf{x}) = 6x_1 + 9x_2 + 12x_3 dollars, and by the second formula of Result 2,

\begin{aligned} c &= c(\mathbf{b}_1)\varphi_1 + c(\mathbf{b}_2)\varphi_2 + c(\mathbf{b}_3)\varphi_3 \\ &= 15\varphi_1 + 21\varphi_2 + 30\varphi_3. \end{aligned}

The coefficients are the batch prices of the mixes. For the order, 15 \cdot 2 + 21 \cdot 3 + 30 \cdot 1 = 123, matching c(3, 5, 5) = 18 + 45 + 60 = 123 dollars.

Example 2

The dual basis depends on the whole basis

In \mathbb{R}^2 let \mathbf{b}_1 = (2, 1) and \mathbf{b}_2 = (-1, 1). Find the dual basis of (\mathbf{b}_1, \mathbf{b}_2), then of (\mathbf{b}_1, \mathbf{c}_2) with \mathbf{c}_2 = (0, 1), and compare.

Solution

B = \begin{bmatrix} 2 & -1 \\ 1 & 1 \end{bmatrix} has \displaystyle B^{-1} = \frac13\begin{bmatrix} 1 & 1 \\ -1 & 2 \end{bmatrix}, so by Result 2

\begin{aligned} \varphi_1(x, y) &= \tfrac13(x + y), \\ \varphi_2(x, y) &= \tfrac13(-x + 2y). \end{aligned}

These draw the figure: \varphi_1 = 0 is the line through \mathbf{b}_2, \varphi_1 = 1 the parallel line through \mathbf{b}_1, and \mathbf{v} = (3, 3) has \varphi_1(\mathbf{v}) = 2, \varphi_2(\mathbf{v}) = 1. For the second basis, \displaystyle \begin{bmatrix} 2 & 0 \\ 1 & 1 \end{bmatrix}^{-1} = \frac12\begin{bmatrix} 1 & 0 \\ -1 & 2 \end{bmatrix}, so

\begin{aligned} \tilde{\varphi}_1(x, y) &= \tfrac12 x, \\ \tilde{\varphi}_2(x, y) &= \tfrac12(-x + 2y). \end{aligned}

The first vector did not change, yet \tilde{\varphi}_1 \neq \varphi_1: the first functional must vanish on the second basis vector, so its null space moved. And \tilde{\varphi}_2 = \tfrac32\varphi_2, with the same null space but the value 1 taken at a different vector. So the isomorphism \mathbf{b}_k \mapsto \varphi_k behind Result 1 depends on the whole basis; nothing in V alone selects one.

Example 3

Three readings of a quadratic

On \mathcal{P}_2(\mathbb{R}) let \varepsilon_a(p) = p(a). Show that (\varepsilon_{-1}, \varepsilon_1, \varepsilon_2) is a basis of \mathcal{P}_2(\mathbb{R})', find the basis whose dual basis it is, and write \iota(p) = \int_{-1}^{2} p(x)\,dx in it.

Solution

Each \varepsilon_a is linear. The quadratics that vanish at two of the points and equal 1 at the third (The remainder and factor theorems) are

\begin{aligned} L_{-1}(x) &= \tfrac16(x - 1)(x - 2), \\ L_1(x) &= -\tfrac12(x + 1)(x - 2), \\ L_2(x) &= \tfrac13(x + 1)(x - 1), \end{aligned}

so \varepsilon_a(L_b) is 1 if a = b and 0 otherwise. If aL_{-1} + bL_1 + cL_2 = 0, applying the three evaluations gives a = b = c = 0; so the list is independent of length 3, a basis (Result 5 of Bases and dimension, proved). Its dual basis is the unique list with these values, (\varepsilon_{-1}, \varepsilon_1, \varepsilon_2), a basis of the dual by Result 2.

The integral is linear (The definite integral), so \iota = \iota(L_{-1})\varepsilon_{-1} + \iota(L_1)\varepsilon_1 + \iota(L_2)\varepsilon_2. With \int_{-1}^{2} x^2\,dx = 3, \int_{-1}^{2} x\,dx = \tfrac32 and \int_{-1}^{2} 1\,dx = 3,

\begin{aligned} \iota(L_{-1}) &= \tfrac16\big(3 - \tfrac92 + 6\big) = \tfrac34, \\ \iota(L_1) &= -\tfrac12\big(3 - \tfrac32 - 6\big) = \tfrac94, \\ \iota(L_2) &= \tfrac13(3 - 3) = 0. \end{aligned}

So for every p \in \mathcal{P}_2(\mathbb{R}),

\int_{-1}^{2} p(x)\,dx = \tfrac34\,p(-1) + \tfrac94\,p(1),

an exact rule that ignores the reading at 2. Check with p = x^2: \tfrac34 + \tfrac94 = 3.

Example 4

A dual map between polynomials and pairs

Let T \in \mathcal{L}(\mathcal{P}_2(\mathbb{R}), \mathbb{R}^2), Tp = (p(1), p(-1)). Let (\varphi_0, \varphi_1, \varphi_2) be the dual basis of (1, x, x^2), so \varphi_k(p) is the coefficient a_k of x^k, and let \psi_j(\mathbf{y}) = y_j on \mathbb{R}^2. Find [T'], compute T' at \tfrac12(\psi_1 - \psi_2) and \tfrac12(\psi_1 + \psi_2), and confirm Result 6.

Solution

T(1) = (1, 1), T(x) = (1, -1), T(x^2) = (1, 1), so by Result 4

[T] = \begin{bmatrix} 1 & 1 & 1 \\ 1 & -1 & 1 \end{bmatrix}, \quad [T'] = \begin{bmatrix} 1 & 1 \\ 1 & -1 \\ 1 & 1 \end{bmatrix}.

Directly, T'\psi_1 is p \mapsto p(1) = a_0 + a_1 + a_2, that is \varphi_0 + \varphi_1 + \varphi_2, the first column. Applying [T'] to (\tfrac12, -\tfrac12) and (\tfrac12, \tfrac12) gives (0, 1, 0) and (1, 0, 1):

\begin{aligned} T'\big(\tfrac12(\psi_1 - \psi_2)\big) &= \varphi_1, \\ T'\big(\tfrac12(\psi_1 + \psi_2)\big) &= \varphi_0 + \varphi_2. \end{aligned}

If p(1) = p(-1) = 0, the factor theorem makes p a multiple of x^2 - 1, so \ker T = \operatorname{span}\{x^2 - 1\} and \operatorname{rank} T = 2. T is surjective, so T' is injective, as the independent columns of [T'] confirm. A functional a_0\varphi_0 + a_1\varphi_1 + a_2\varphi_2 takes the value a_2 - a_0 at x^2 - 1, so (\ker T)^{0} is the plane a_0 = a_2, spanned by \varphi_1 and \varphi_0 + \varphi_2: exactly \operatorname{im} T'.

Example 5

Functionals with nested null spaces

Let V be any vector space over \mathbb{F} and \varphi, \psi \in V' with \ker \varphi \subseteq \ker \psi. Prove that \psi = c\varphi for some c \in \mathbb{F}.

Solution

If \varphi = 0, then \ker \varphi = V, so \psi = 0 = 0\varphi. Otherwise choose \mathbf{w} with \varphi(\mathbf{w}) \neq 0 and let \mathbf{u} = \mathbf{w}/\varphi(\mathbf{w}), so \varphi(\mathbf{u}) = 1. For any \mathbf{v} \in V,

\varphi\big(\mathbf{v} - \varphi(\mathbf{v})\mathbf{u}\big) = \varphi(\mathbf{v}) - \varphi(\mathbf{v}) = 0,

so \mathbf{v} - \varphi(\mathbf{v})\mathbf{u} \in \ker \psi, and applying \psi gives \psi(\mathbf{v}) = \psi(\mathbf{u})\varphi(\mathbf{v}). So \psi = c\varphi with c = \psi(\mathbf{u}). Functionals with equal null spaces are multiples, like \tilde{\varphi}_2 and \varphi_2 in Example 2.

Example 6

Where the dual basis stops spanning

On \mathcal{P}(\mathbb{F}) let \varphi_k(p) be the coefficient of x^k, for k \ge 0. Show that (\varphi_0, \dots, \varphi_N) is independent for every N, but that \varepsilon_1(p) = p(1) lies in the span of none of these lists.

Solution

If a_0\varphi_0 + \dots + a_N\varphi_N = 0, applying it to x^j gives a_j = 0. Every combination of \varphi_0, \dots, \varphi_N vanishes at x^{N+1}, while \varepsilon_1(x^{N+1}) = 1. The expansion of Result 2 would be the infinite sum \varphi_0 + \varphi_1 + \varphi_2 + \cdots, and a linear combination is finite.

Boundaries

  • Finite dimension. Results 1 and 2, the formula of Result 5 and the second half of Result 6 need it; Example 6 shows the dual basis failing to span.
  • No preferred isomorphism. V \cong V' in finite dimension, but Example 2 shows the isomorphism comes from a chosen basis. V \to V'' has one built from no choice (Exercise 17), and an inner product supplies one for V', the Riesz representation of Orthonormal bases and Gram-Schmidt, proved.
  • The annihilator lives in V', not in V; it is not an orthogonal complement until an inner product enters.
  • Linear over the given field. On a complex space a functional must be complex-linear, so (z, w) \mapsto z + \bar{w} is not one (Exercise 1).
  • The dual map is not the adjoint. T' maps W' to V' with no inner product; the adjoint of Adjoints and self-adjoint operators maps W to V and needs inner products.

Where this goes next

Orthonormal bases and Gram-Schmidt, proved proves that on a finite-dimensional inner product space every functional is \mathbf{v} \mapsto \langle \mathbf{v}, \mathbf{w} \rangle for exactly one \mathbf{w}, and Adjoints and self-adjoint operators uses that to turn the dual map into a map from W to V. Since every result here holds over any field, field extensions in abstract algebra use them unchanged.

Exercises

Functionals and dual bases

  1. Which of these are linear functionals? (a) \varphi(x, y, z) = 3x - z on \mathbb{R}^3. (b) \varphi(p) = p(0)\,p(2) on \mathcal{P}_2(\mathbb{R}). (c) \varphi(X) = \operatorname{tr} X - 2x_{21} on \mathbb{R}^{2 \times 2}, with x_{21} the lower left entry. (d) \varphi(f) = f(0) + 1 on C[0, 1]. (e) \varphi(f) = \int_{-1}^{1} x f(x)\,dx on C[-1, 1]. (f) \varphi(z, w) = z + \bar{w} on \mathbb{C}^2 over \mathbb{C}.
  2. Find the dual basis of ((1, 3), (2, 5)) in \mathbb{R}^2, and write \psi(x, y) = x + y in it.
  3. Find the dual basis of (1 + x, 1 - x) in \mathcal{P}_1(\mathbb{R}), in terms of p(0) and p(1).
  4. Find a, b, c with \int_0^2 p(x)\,dx = a\,p(0) + b\,p(1) + c\,p(2) for every p \in \mathcal{P}_2(\mathbb{R}). Show that the rule holds on \mathcal{P}_3(\mathbb{R}) but fails for x^4.
  5. Find the dual basis of ((1, i), (1, -i)) in \mathbb{C}^2.
  6. Hiro tries other house mixes, \mathbf{d}_1 = (2, 1, 0), \mathbf{d}_2 = (0, 1, 1), \mathbf{d}_3 = (0, 0, 1). Find the dual basis and the number of batches of each in the order (4, 3, 5).

Dual maps and annihilators

  1. Let T(x, y, z) = (x + 2y,\ 3y - z) from \mathbb{R}^3 to \mathbb{R}^2 and \psi(u, v) = 2u - v. Compute T'\psi and the matrix of T' relative to the dual bases of the standard bases.
  2. Let T \in \mathcal{L}(\mathcal{P}_2(\mathbb{R})), (Tp)(x) = p(2x - 1), and \varepsilon_a(p) = p(a). Find T'\varepsilon_0, T'\varepsilon_1, and the matrix of T' relative to the dual basis of (1, x, x^2).
  3. Let R(c_1, c_2) = c_1\mathbf{d}_1 + c_2\mathbf{d}_2 from \mathbb{R}^2 to \mathbb{R}^3, with \mathbf{d}_1, \mathbf{d}_2 from Exercise 6, turn batches into kilograms of honey. For the prices \psi(\mathbf{x}) = 4x_1 + 7x_2 + 10x_3, find R'\psi and say what it measures. Then find \ker R'.
  4. Let T \in \mathcal{L}(\mathbb{R}^{2 \times 2}), TX = X - X^{T}. Find \dim \ker T' and a basis of \operatorname{im} T'.
  5. Let U = \operatorname{span}\{(1, 2, 0, -1), (0, 1, 1, 1)\} in \mathbb{R}^4. Find a basis of U^{0}.
  6. Find the subspace U of \mathbb{R}^3 with U^{0} = \operatorname{span}\{\varphi, \psi\}, where \varphi(\mathbf{x}) = x_1 + x_2 + x_3 and \psi(\mathbf{x}) = x_1 - x_3.

Proofs

  1. Let V be finite-dimensional, \varphi_1, \dots, \varphi_m \in V', and \Phi \in \mathcal{L}(V, \mathbb{F}^m) the map \Phi\mathbf{v} = (\varphi_1(\mathbf{v}), \dots, \varphi_m(\mathbf{v})). Prove that (\varphi_1, \dots, \varphi_m) is linearly independent if and only if \Phi is surjective.
  2. Let V be finite-dimensional and \mathbf{v} \in V nonzero. Prove that some \varphi \in V' has \varphi(\mathbf{v}) = 1. Conclude that \mathbf{v} = \mathbf{0} if every \varphi \in V' vanishes at \mathbf{v}.
  3. Let U, W be subspaces of V. Prove that U \subseteq W implies W^{0} \subseteq U^{0}, and that (U + W)^{0} = U^{0} \cap W^{0}.
  4. Let V be finite-dimensional and U a subspace. Prove that a vector at which every functional in U^{0} vanishes lies in U.
  5. For \mathbf{v} \in V define \Lambda\mathbf{v} \in V'' by (\Lambda\mathbf{v})(\varphi) = \varphi(\mathbf{v}). Prove that \Lambda is linear, and an isomorphism when V is finite-dimensional.
  6. Let V, W be finite-dimensional. Prove that taking dual maps, T \mapsto T', is a linear bijection from \mathcal{L}(V, W) to \mathcal{L}(W', V').
Answers
  1. (a) Linear, the row [\,3 \ 0 \ {-1}\,]. (b) Not linear: \varphi(2) = 4 but 2\varphi(1) = 2. (c) Linear, a combination of the trace and an entry. (d) Not linear: \varphi(0) = 1 \neq 0. (e) Linear, since the integral is. (f) Not linear over \mathbb{C}: \varphi(0, i) = -i but i\,\varphi(0, 1) = i.
  2. \begin{bmatrix} 1 & 2 \\ 3 & 5 \end{bmatrix}^{-1} = \begin{bmatrix} -5 & 2 \\ 3 & -1 \end{bmatrix}, so \varphi_1(x, y) = -5x + 2y, \varphi_2(x, y) = 3x - y. By Result 2, \psi = \psi(1, 3)\varphi_1 + \psi(2, 5)\varphi_2 = 4\varphi_1 + 7\varphi_2.
  3. a + bx = c_1(1 + x) + c_2(1 - x) gives c_1 = \tfrac12(a + b), c_2 = \tfrac12(a - b), and p(0) = a, p(1) = a + b. So \varphi_1(p) = \tfrac12 p(1) and \varphi_2(p) = p(0) - \tfrac12 p(1).
  4. The rule at 1, x, x^2 gives a + b + c = 2, b + 2c = 2, b + 4c = \tfrac83, so c = \tfrac13, b = \tfrac43, a = \tfrac13. At x^3 both sides are 4, so by linearity the rule holds on \mathcal{P}_3(\mathbb{R}). At x^4: \tfrac{32}{5} \neq \tfrac13(4 + 16) = \tfrac{20}{3}.
  5. \begin{bmatrix} 1 & 1 \\ i & -i \end{bmatrix}^{-1} = \tfrac12\begin{bmatrix} 1 & -i \\ 1 & i \end{bmatrix}, so \varphi_1(z, w) = \tfrac12(z - iw) and \varphi_2(z, w) = \tfrac12(z + iw).
  6. The lower-triangular B = \begin{bmatrix} 2 & 0 & 0 \\ 1 & 1 & 0 \\ 0 & 1 & 1 \end{bmatrix} has inverse rows (\tfrac12, 0, 0), (-\tfrac12, 1, 0), (\tfrac12, -1, 1): \varphi_1(\mathbf{x}) = \tfrac12 x_1, \varphi_2(\mathbf{x}) = x_2 - \tfrac12 x_1, \varphi_3(\mathbf{x}) = \tfrac12 x_1 - x_2 + x_3. At (4, 3, 5): 2, 1 and 4 batches.
  7. (T'\psi)(x, y, z) is 2(x + 2y) - (3y - z), so T'\psi is 2x + y + z. By Result 4, [T'] = \begin{bmatrix} 1 & 0 \\ 2 & 3 \\ 0 & -1 \end{bmatrix}, which sends (2, -1) to (2, 1, 1).
  8. (T'\varepsilon_0)(p) = (Tp)(0) = p(-1), so T'\varepsilon_0 = \varepsilon_{-1}; and (T'\varepsilon_1)(p) = p(1), so T'\varepsilon_1 = \varepsilon_1. Since T(x) = 2x - 1 and T(x^2) = 4x^2 - 4x + 1, [T] has columns (1, 0, 0), (-1, 2, 0), (1, -4, 4), and [T'] = \begin{bmatrix} 1 & 0 & 0 \\ -1 & 2 & 0 \\ 1 & -4 & 4 \end{bmatrix}.
  9. (R'\psi)(c_1, c_2) = \psi(c_1\mathbf{d}_1 + c_2\mathbf{d}_2) = 15c_1 + 17c_2: the cost of c_1 and c_2 batches, at \$15 and \$17 a batch. By Result 6, \ker R' = (\operatorname{im} R)^{0}, the price lists making both mixes free: 2a_1 + a_2 = 0 and a_2 + a_3 = 0, so \ker R' = \operatorname{span}\{x_1 - 2x_2 + 2x_3\}.
  10. \ker T is the symmetric matrices, of dimension 3, so \operatorname{rank} T = 1. By Results 5 and 6, \dim \ker T' = 4 - 1 = 3, and \operatorname{im} T' = (\ker T)^{0} has dimension 1. A functional \sum a_{jk}x_{jk} vanishes on E_{11}, E_{22} and E_{12} + E_{21} exactly when a_{11} = a_{22} = 0 and a_{21} = -a_{12}, so a basis is X \mapsto x_{12} - x_{21}.
  11. \mathbf{a}\mathbf{x} lies in U^{0} when a_1 + 2a_2 - a_4 = 0 and a_2 + a_3 + a_4 = 0, so a_2 = -a_3 - a_4, a_1 = 2a_3 + 3a_4: a basis is \mathbf{x} \mapsto 2x_1 - x_2 + x_3 and \mathbf{x} \mapsto 3x_1 - x_2 + x_4, and \dim U^{0} = 4 - 2 as Result 5 requires.
  12. U = \operatorname{span}\{(1, -2, 1)\}. As \varphi, \psi are independent, Result 5 gives \dim U = 3 - 2 = 1, and U lies in the common null space of \varphi and \psi, the line through (1, -2, 1). Conversely, that line has a 2-dimensional annihilator containing the independent \varphi, \psi, so it equals their span.
  13. Let \psi_k(\mathbf{y}) = y_k on \mathbb{F}^m, so (\psi_1, \dots, \psi_m) is the dual basis of the standard basis. Then (\Phi'\psi_k)(\mathbf{v}) = \psi_k(\Phi\mathbf{v}) = \varphi_k(\mathbf{v}), so \Phi'(a_1\psi_1 + \dots + a_m\psi_m) = a_1\varphi_1 + \dots + a_m\varphi_m. Hence \ker \Phi' = \{0\} exactly when only the trivial combination of the \varphi_k is 0: \Phi' is injective if and only if the list is independent. By Result 6, \Phi' is injective if and only if \Phi is surjective.
  14. The list (\mathbf{v}) is independent, so it extends to a basis (\mathbf{v}, \mathbf{v}_2, \dots, \mathbf{v}_n) of V (Result 3 of Bases and dimension, proved), and the first dual basis functional has \varphi_1(\mathbf{v}) = 1. The conclusion is the contrapositive (Proof by contraposition).
  15. If \varphi \in W^{0}, it vanishes on W \supseteq U, so \varphi \in U^{0}. As U, W \subseteq U + W, this gives (U + W)^{0} \subseteq U^{0} \cap W^{0}. Conversely, if \varphi vanishes on U and on W, then \varphi(\mathbf{u} + \mathbf{w}) = \varphi(\mathbf{u}) + \varphi(\mathbf{w}) = 0 on all of U + W.
  16. Let \mathbf{v} \notin U and (\mathbf{u}_1, \dots, \mathbf{u}_m) a basis of U. Then (\mathbf{u}_1, \dots, \mathbf{u}_m, \mathbf{v}) is independent (Result 3 of Span and independence, proved); extend it to a basis of V. The dual basis functional belonging to \mathbf{v} vanishes at each \mathbf{u}_k, so it lies in U^{0}, and it is 1 at \mathbf{v}, as required.
  17. (\Lambda\mathbf{v})(a\varphi + \psi) = a\varphi(\mathbf{v}) + \psi(\mathbf{v}), so \Lambda\mathbf{v} \in V'', and \big(\Lambda(a\mathbf{v} + \mathbf{w})\big)(\varphi) = a\varphi(\mathbf{v}) + \varphi(\mathbf{w}) for every \varphi, so \Lambda is linear. Now let V be finite-dimensional. If \Lambda\mathbf{v} = 0, every \varphi vanishes at \mathbf{v}, so \mathbf{v} = \mathbf{0} by Exercise 14. By Result 1 twice, \dim V'' = \dim V, so the injective \Lambda is an isomorphism (Result 5 of Invertibility and isomorphic spaces). No basis was chosen.
  18. The map is linear by Result 3. If T' = 0, then \psi(T\mathbf{v}) = 0 for all \psi \in W', so each T\mathbf{v} = \mathbf{0} by Exercise 14 in W, and T = 0: the map is injective. With \dim V = n, \dim W = m, Result 7 of Invertibility and isomorphic spaces and Result 1 give both spaces dimension mn, so it is an isomorphism by Result 5 there.
math.wiki · CC-BY-SA-4.0