math.wiki
Linear algebra › Linear maps › Invertibility and isomorphic spaces
Sign in

Invertibility and isomorphic spaces

When can a linear map be undone, and when are two vector spaces the same space under different names?

Where you are 5 in Linear maps

Why this exists

When a matrix or transformation can be undone showed that \mathbf{x} \mapsto A\mathbf{x} on \mathbb{R}^n can be undone exactly when it is one-to-one and onto, that only square matrices qualify, and that for them one product AB = I is enough. This lesson proves these facts for abstract spaces, finds where the one-product shortcut breaks, and draws the structural conclusion: an invertible linear map, an isomorphism, matches two spaces so closely that sums and scalar multiples cannot tell them apart, and finite-dimensional spaces over one field match exactly when their dimensions agree.

Naomi and Andre run a smokehouse. The chamber’s controller logs the temperature at hours 0, 2 and 4 of a smoking segment, while Andre’s planning software describes a segment by a quadratic q(t) = a + bt + ct^2. Can every three readings be turned into exactly one quadratic, and back? That is the question whether one linear map is invertible (Example 1).

Definition

Let V and W be vector spaces over \mathbb{F}, and write I_V for the identity operator on V. A linear map T \in \mathcal{L}(V, W) is invertible if there is a linear map S \in \mathcal{L}(W, V) with

ST = I_V \quad \text{and} \quad TS = I_W.

Such an S is unique (Result 1); it is the inverse of T, written T^{-1}. An invertible linear map is an isomorphism, and V is isomorphic to W, written V \cong W, if some isomorphism maps V to W.

A linear S with ST = I_V alone is a left inverse of T; one with TS = I_W alone is a right inverse.

Intuition

An isomorphism T: V \to W is a dictionary. Each vector of V has exactly one translation in W, each vector of W translates exactly one vector of V, and T(a\mathbf{u} + \mathbf{v}) = aT\mathbf{u} + T\mathbf{v} says that translating then combining gives the same as combining then translating. So everything defined from sums and scalar multiples (span, independence, bases, dimension) passes through unchanged (Result 3). Isomorphic spaces differ only in what their vectors are called.

A dictionary carries a basis to a basis, so isomorphic spaces have equal dimension; and pairing two bases of equal length term by term builds a dictionary. So \mathcal{P}_3(\mathbb{R}), \mathbb{R}^{2 \times 2} and \mathbb{R}^4 are one space under three names, and the coordinate map of The matrix of a linear map is the dictionary a basis provides, one per basis.

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 (invertible means injective and surjective). T \in \mathcal{L}(V, W) has at most one inverse, and T is invertible if and only if it is injective and surjective (One-to-one, onto and bijective functions). Then the inverse function of T is linear, and it is T^{-1}.

Result 2 (inverses of products). If T \in \mathcal{L}(U, V) and S \in \mathcal{L}(V, W) are invertible, then ST is invertible with (ST)^{-1} = T^{-1}S^{-1}, and T^{-1} is invertible with (T^{-1})^{-1} = T. Hence V \cong V; V \cong W implies W \cong V; and U \cong V with V \cong W implies U \cong W.

Result 3 (isomorphisms carry lists). Let T \in \mathcal{L}(V, W) be an isomorphism and (\mathbf{v}_1, \dots, \mathbf{v}_n) a list in V. The list is independent if and only if (T\mathbf{v}_1, \dots, T\mathbf{v}_n) is independent; it spans V if and only if (T\mathbf{v}_1, \dots, T\mathbf{v}_n) spans W; so it is a basis of V if and only if its image is a basis of W.

Result 4 (dimension decides). Let V be finite-dimensional. Then V \cong W if and only if W is finite-dimensional and \dim W = \dim V. In particular, if \dim V = n, every coordinate map is an isomorphism V \to \mathbb{F}^n.

Result 5 (equal dimensions: one condition is enough). Let \dim V = \dim W, finite, and T \in \mathcal{L}(V, W). These are equivalent:

  • T is invertible;
  • T is injective;
  • T is surjective;
  • ST = I_V for some S \in \mathcal{L}(W, V);
  • TS = I_W for some S \in \mathcal{L}(W, V).

An S as in either of the last two is T^{-1}.

Result 6 (inverses through matrices). Let V, W be finite-dimensional with bases \mathcal{B}, \mathcal{C}, and T \in \mathcal{L}(V, W). Then T is invertible if and only if [T]_{\mathcal{C} \leftarrow \mathcal{B}} is square and invertible, and then

[T^{-1}]_{\mathcal{B} \leftarrow \mathcal{C}} = \big([T]_{\mathcal{C} \leftarrow \mathcal{B}}\big)^{-1}.

Result 7 (the space of linear maps). Let \dim V = n and \dim W = m, with bases \mathcal{B} and \mathcal{C}. Then T \mapsto [T]_{\mathcal{C} \leftarrow \mathcal{B}} is an isomorphism of \mathcal{L}(V, W) onto \mathbb{F}^{m \times n}, and \dim \mathcal{L}(V, W) = mn.

Why it is true

Result 1. If S and S' are inverses, associativity (Result 4 of Linear maps between vector spaces) gives S = SI_W = S(TS') = (ST)S' = I_VS' = S'. If T is invertible and T\mathbf{v} = \mathbf{0}, then \mathbf{v} = T^{-1}T\mathbf{v} = \mathbf{0}, so T is injective (Result 2 of Null space, range and the fundamental theorem); and each \mathbf{w} = T(T^{-1}\mathbf{w}) is an output. Conversely, let T be injective and surjective. Each \mathbf{w} \in W is T\mathbf{v} for exactly one \mathbf{v}; call it S\mathbf{w}. This inverse function (Inverse functions) has ST = I_V and TS = I_W. It is linear: for \mathbf{w}_1, \mathbf{w}_2 \in W and a \in \mathbb{F},

\begin{aligned} T(aS\mathbf{w}_1 + S\mathbf{w}_2) &= aTS\mathbf{w}_1 + TS\mathbf{w}_2 \\ &= a\mathbf{w}_1 + \mathbf{w}_2 \\ &= T\big(S(a\mathbf{w}_1 + \mathbf{w}_2)\big), \end{aligned}

and T is injective, so aS\mathbf{w}_1 + S\mathbf{w}_2 = S(a\mathbf{w}_1 + \mathbf{w}_2).

Result 2. By associativity, (ST)(T^{-1}S^{-1}) = S(TT^{-1})S^{-1} = SS^{-1} = I_W, and (T^{-1}S^{-1})(ST) = I_U in the same way. The equations T^{-1}T = I_U and TT^{-1} = I_V say that T is an inverse of T^{-1}. So I_V, the inverse of an isomorphism and a composite of two are isomorphisms.

Result 3. Let T be any linear map. If (T\mathbf{v}_k) is independent and \sum_k a_k\mathbf{v}_k = \mathbf{0}, applying T gives \sum_k a_kT\mathbf{v}_k = \mathbf{0}, so every a_k = 0. If T is injective and (\mathbf{v}_k) independent, then \sum_k a_kT\mathbf{v}_k = \mathbf{0} means T\big(\sum_k a_k\mathbf{v}_k\big) = \mathbf{0}, so \sum_k a_k\mathbf{v}_k = \mathbf{0} and every a_k = 0. If (\mathbf{v}_k) spans V, then \operatorname{im} T = \operatorname{span}\{T\mathbf{v}_1, \dots, T\mathbf{v}_n\} (Result 4 of Null space, range and the fundamental theorem), which is W when T is surjective. For the converse about spanning, apply this to the isomorphism T^{-1} (Result 2) and the list (T\mathbf{v}_k).

Result 4. If T: V \to W is an isomorphism and (\mathbf{v}_1, \dots, \mathbf{v}_n) a basis of V, Result 3 makes (T\mathbf{v}_1, \dots, T\mathbf{v}_n) a basis of W, so \dim W = n. Conversely, let (\mathbf{v}_1, \dots, \mathbf{v}_n) and (\mathbf{w}_1, \dots, \mathbf{w}_n) be bases of V and W. The linear map lemma (Result 5 of Linear maps between vector spaces) gives T \in \mathcal{L}(V, W) with T\mathbf{v}_k = \mathbf{w}_k. Its range is \operatorname{span}\{\mathbf{w}_1, \dots, \mathbf{w}_n\} = W, so by the fundamental theorem of linear maps (Result 5 of Null space, range and the fundamental theorem) \dim \ker T = n - n = 0. So T is injective and surjective, an isomorphism by Result 1. The coordinate map is linear, injective and surjective by Result 1 of The matrix of a linear map.

Result 5. The second and third conditions are equivalent by Result 6 of Null space, range and the fundamental theorem, and together they are equivalent to the first by Result 1. The first gives the last two with S = T^{-1}. If ST = I_V and T\mathbf{v} = \mathbf{0}, then \mathbf{v} = ST\mathbf{v} = \mathbf{0}, so T is injective. If TS = I_W, each \mathbf{w} = T(S\mathbf{w}) is an output, so T is surjective. Once T is invertible, ST = I_V gives S = (ST)T^{-1} = T^{-1}, and TS = I_W gives S = T^{-1}(TS) = T^{-1}.

Result 6. Let \dim V = n, \dim W = m. An identity has matrix I_n relative to one basis at both ends, since I\mathbf{v}_k = \mathbf{v}_k has coordinates \mathbf{e}_k. If T is invertible, m = n by Result 4, and by Result 4 of The matrix of a linear map,

\begin{aligned} [T^{-1}]_{\mathcal{B} \leftarrow \mathcal{C}}\,[T]_{\mathcal{C} \leftarrow \mathcal{B}} &= [T^{-1}T]_{\mathcal{B} \leftarrow \mathcal{B}} \\ &= I_n, \end{aligned}

and likewise in the other order. Conversely, let A = [T]_{\mathcal{C} \leftarrow \mathcal{B}} be square and invertible. Result 3 of The matrix of a linear map gives S \in \mathcal{L}(W, V) with [S]_{\mathcal{B} \leftarrow \mathcal{C}} = A^{-1}. Then [ST]_{\mathcal{B} \leftarrow \mathcal{B}} = A^{-1}A = I_n, the matrix of I_V, and since each matrix belongs to exactly one map, ST = I_V; likewise TS = I_W.

Result 7. The map is linear because [S + T] = [S] + [T] and [aT] = a[T], and it is injective and surjective because each A \in \mathbb{F}^{m \times n} is the matrix of exactly one T (Result 3 of The matrix of a linear map). By Result 4 here and \dim \mathbb{F}^{m \times n} = mn (Result 4 of Bases and dimension, proved), \dim \mathcal{L}(V, W) = mn.

Worked examples

Example 1

Readings and coefficients in the smokehouse

Let E: \mathcal{P}_2(\mathbb{R}) \to \mathbb{R}^3, Eq = (q(0), q(2), q(4)). Show that E is an isomorphism, find the matrix of E^{-1} relative to the standard basis \mathcal{S} of \mathbb{R}^3 and \mathcal{M} = (1, t, t^2), and find the profile with readings (120, 200, 240).

Solution

E is linear, since evaluation at a point respects sums and scalar multiples. If Eq = \mathbf{0}, then q has roots 0, 2 and 4; a nonzero polynomial of degree at most 2 has at most two roots (The remainder and factor theorems), so q = 0 and E is injective. Both spaces have dimension 3, so E is invertible by Result 5: every three readings come from exactly one quadratic.

The columns are E(1) = (1, 1, 1), E(t) = (0, 2, 4) and E(t^2) = (0, 4, 16), so

A = [E]_{\mathcal{S} \leftarrow \mathcal{M}} = \begin{bmatrix} 1 & 0 & 0 \\ 1 & 2 & 4 \\ 1 & 4 & 16 \end{bmatrix}.

By Result 6, [E^{-1}]_{\mathcal{M} \leftarrow \mathcal{S}} = A^{-1}, and row reduction of [\,A \mid I_3\,] (The invertible matrix theorem) gives

A^{-1} = \frac18\begin{bmatrix} 8 & 0 & 0 \\ -6 & 8 & -2 \\ 1 & -2 & 1 \end{bmatrix}.

So readings (y_0, y_2, y_4) belong to the quadratic with a = y_0, b = \tfrac14(-3y_0 + 4y_2 - y_4) and c = \tfrac18(y_0 - 2y_2 + y_4). For (120, 200, 240) this is a = 120, b = 50, c = -5:

q(t) = 120 + 50t - 5t^2.

Check: q(2) = 200 and q(4) = 240. The chamber is still warming at the end of the segment, at q'(4) = 10 degrees per hour.

t
q(t)
0 120 2 200 4 240
Figure 1The profile q(t) = 120 + 50t - 5t^2 of Example 1, in degrees Fahrenheit, is the only quadratic through the readings (120, 200, 240) taken at hours 0, 2 and 4. The readings map E and its inverse turn three readings into three coefficients and back.
Example 2

Six spaces, two classes

Sort these real vector spaces into classes of isomorphic spaces, and write down an isomorphism from (c) to (b): (a) \mathbb{R}^{2 \times 3}; (b) \mathcal{P}_5(\mathbb{R}); (c) the upper-triangular matrices in \mathbb{R}^{3 \times 3}; (d) the even polynomials in \mathcal{P}_6(\mathbb{R}), those with p(-x) = p(x); (e) \mathbb{C}^3 with scalars restricted to \mathbb{R}; (f) \mathbb{R}^{S} for a set S of four elements.

Solution

By Results 2 and 4, two of these are isomorphic exactly when their dimensions agree, so it is enough to count. (a) and (b) have dimension 6 (Result 4 of Bases and dimension, proved), and (c) has basis the six E_{jk} with j \le k. Comparing coefficients of p(-x) and p(x), an even p has no odd powers, so (d) has basis (1, x^2, x^4, x^6) and dimension 4. (e) has dimension 2 \cdot 3 = 6 (Result 7 of Bases and dimension, proved). In (f) the functions \delta_s, equal to 1 at s and 0 elsewhere, form a basis, since f = \sum_{s \in S} f(s)\delta_s in exactly one way; the dimension is 4.

The classes are \{(a), (b), (c), (e)\} and \{(d), (f)\}. For (c) to (b), pair the bases as in the proof of Result 4:

\begin{bmatrix} a_0 & a_1 & a_2 \\ 0 & a_3 & a_4 \\ 0 & 0 & a_5 \end{bmatrix} \mapsto a_0 + a_1x + \dots + a_5x^5.

Any other pairing gives another isomorphism.

Example 3

Undoing p plus its derivative

Let T \in \mathcal{L}(\mathcal{P}_2(\mathbb{R})), Tp = p + p'. Show that T is invertible, find T^{-1}, and find the quadratic p with p + p' = x^2 + 3x.

Solution

Write T = I + D with D differentiation. On \mathcal{P}_2(\mathbb{R}), D^3 = 0. Since I commutes with D, the distributive laws of Result 4 of Linear maps between vector spaces expand the product as for numbers:

(I + D)(I - D + D^2) = I + D^3 = I.

That is one product. The space is finite-dimensional, so by Result 5, T is invertible with T^{-1} = I - D + D^2. Hence the equation has exactly one quadratic solution,

p = (x^2 + 3x) - (2x + 3) + 2 = x^2 + x - 1,

and indeed p' = 2x + 1, so p + p' = x^2 + 3x.

Example 4

One product is not enough on all polynomials

On \mathcal{P}(\mathbb{R}) let M be multiplication by x, as in Linear maps between vector spaces, and let L drop the constant term and lower every power by one:

\begin{aligned} &L(a_0 + a_1x + \dots + a_nx^n) \\ &\quad = a_1 + a_2x + \dots + a_nx^{n-1}. \end{aligned}

Show that LM = I but ML \neq I, and that neither map is invertible.

Solution

L is linear, acting on coefficients one at a time. M sends a_0 + \dots + a_nx^n to a_0x + \dots + a_nx^{n+1}, and L returns it, so LM = I. But ML replaces p by p - p(0); for instance ML(1) = 0. Every output of M has constant term 0, so 1 \notin \operatorname{im} M; and L(1) = 0. Neither map is invertible (Result 1), though M has a left inverse: Result 5 needs finite dimension, as differentiation and integration on \mathcal{P}(\mathbb{R}) showed in Null space, range and the fundamental theorem.

Example 5

Products that can be undone

Let V be finite-dimensional and S, T \in \mathcal{L}(V). Prove that ST is invertible if and only if S and T both are.

Solution

If both are invertible, so is ST by Result 2. Conversely, let R = (ST)^{-1}. If T\mathbf{v} = \mathbf{0}, then \mathbf{v} = R(ST)\mathbf{v} = R\mathbf{0} = \mathbf{0}, so T is injective. For \mathbf{w} \in V, \mathbf{w} = (ST)R\mathbf{w} = S(TR\mathbf{w}), so S is surjective. By Result 5 both are invertible. Finite dimension was used: in Example 4, LM = I is invertible but neither L nor M is.

Example 6

A trace correction on 2 by 2 matrices

For real a, let T \in \mathcal{L}(\mathbb{R}^{2 \times 2}), TX = X + a(\operatorname{tr} X)I_2. For which a is T invertible, and what is T^{-1}? For the other a, find \ker T and \operatorname{im} T. Then compute T^{-1}\begin{bmatrix} 4 & 1 \\ 0 & 2 \end{bmatrix} for a = 1.

Solution

T is linear because the trace is, and taking traces, \operatorname{tr}(TX) = (1 + 2a)\operatorname{tr} X. Try an inverse of the same shape, SX = X + b(\operatorname{tr} X)I_2:

\begin{aligned} TSX &= SX + a\operatorname{tr}(SX)I_2 \\ &= X + \big(b + a(1 + 2b)\big)(\operatorname{tr} X)I_2. \end{aligned}

This is X for every X when b(1 + 2a) = -a. So if a \neq -\tfrac12, take b = -\tfrac{a}{1 + 2a}; by Result 5 one product suffices, and

T^{-1}X = X - \frac{a}{1 + 2a}(\operatorname{tr} X)I_2.

If a = -\tfrac12, then TX = \mathbf{0} means X = \tfrac12(\operatorname{tr} X)I_2, a multiple of I_2, and T(cI_2) = cI_2 - \tfrac12(2c)I_2 = \mathbf{0}; so \ker T = \operatorname{span}\{I_2\}. Every TX has trace 0. The trace-zero matrices form the null space of the trace map onto \mathbb{R}, of dimension 3, and \operatorname{rank} T = 4 - 1 = 3 by the fundamental theorem, so \operatorname{im} T is all of them (Result 6 of Bases and dimension, proved).

Result 6 agrees: relative to \mathcal{E} = (E_{11}, E_{12}, E_{21}, E_{22}),

[T]_{\mathcal{E}} = \begin{bmatrix} 1 + a & 0 & 0 & a \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ a & 0 & 0 & 1 + a \end{bmatrix},

with determinant (1 + a)^2 - a^2 = 1 + 2a (Properties of determinants: row operations, products, transposes). For a = 1, T^{-1}X = X - \tfrac13(\operatorname{tr} X)I_2, and the trace is 6:

T^{-1}\begin{bmatrix} 4 & 1 \\ 0 & 2 \end{bmatrix} = \begin{bmatrix} 2 & 1 \\ 0 & 0 \end{bmatrix}.

Boundaries

  • Finite dimension is needed for Result 5. Example 4 gives a left inverse without an inverse, and Exercise 17 shows \mathbb{F}^{\infty} isomorphic to a proper subspace of itself. Nor are all infinite-dimensional spaces isomorphic (\mathcal{P}(\mathbb{R}) and \mathbb{R}^{\infty} are not), but the proof is beyond this lesson.
  • One field at both ends. \mathbb{C}^3 over \mathbb{C} and \mathbb{R}^6 over \mathbb{R} are not compared at all; \mathbb{C}^3 over \mathbb{R} is isomorphic to \mathbb{R}^6.
  • Isomorphic is not equal. Results 4 and 7 build isomorphisms from chosen bases; Dual spaces meets two spaces of equal dimension with no preferred isomorphism between them.
  • Unequal dimensions allow one-sided inverses only (Exercises 12 and 13), as no non-square matrix is invertible (When a matrix or transformation can be undone).

Where this goes next

Dual spaces applies Result 7 with W = \mathbb{F}. Invariant subspaces asks when T - \lambda I fails to be invertible, which by Result 5 is when its null space is nonzero: the definition of an eigenvalue. Exercise 15 foreshadows The minimal polynomial.

Exercises

Deciding invertibility and finding inverses

  1. Let E: \mathcal{P}_2(\mathbb{R}) \to \mathbb{R}^3, Eq = (q(-1), q(0), q(1)). Show that E is an isomorphism and find the q with Eq = (4, 1, 2).
  2. On \mathcal{P}_2(\mathbb{R}), decide whether each map to \mathbb{R}^3 is invertible. If it is, give its inverse; if not, find its null space and range. (a) Tp = (p(0), p'(0), p(1)). (b) Sp = (p(0), p(2), p'(1)).
  3. Let T \in \mathcal{L}(\mathcal{P}_3(\mathbb{R})), Tp = p - 2p''. Find T^{-1} and the cubic p with p - 2p'' = x^3 - x.
  4. Let T \in \mathcal{L}(\mathcal{P}_2(\mathbb{R})), (Tp)(x) = p(x + 1) + p(x). Find [T]_{\mathcal{M}} for \mathcal{M} = (1, x, x^2), show T is invertible, and find T^{-1}(x^2).
  5. For c \in \mathbb{C} let T \in \mathcal{L}(\mathbb{C}^2), T(z, w) = (z + cw,\ cz + w). For which c is T invertible? For c = i, find T^{-1}(1, 0).
  6. For real a let T \in \mathcal{L}(\mathcal{P}_2(\mathbb{R})), (Tp)(x) = p(x) + a\,p(1). For which a is T invertible, and what is T^{-1}? Find T^{-1}(x^2) for a = 2, and \ker T for the excluded a.
  7. Let T: \mathcal{P}_3(\mathbb{R}) \to \mathbb{R}^{2 \times 2}, Tp = \begin{bmatrix} p(0) & p'(0) \\ p''(0) & p'''(0) \end{bmatrix}. Show T is an isomorphism and find T^{-1}\begin{bmatrix} 1 & 2 \\ 3 & 4 \end{bmatrix}.

Isomorphic spaces

  1. Sort into classes of isomorphic real spaces: (a) the symmetric matrices in \mathbb{R}^{3 \times 3}; (b) \mathcal{L}(\mathbb{R}^3, \mathbb{R}^2); (c) \mathbb{R}^{S} with S = \{1, 2, 3\}; (d) \mathcal{P}_2(\mathbb{R}); (e) \{p \in \mathcal{P}_4(\mathbb{R}) : p(1) = 0\}; (f) \mathbb{R}^{2 \times 2}; (g) \mathbb{C}^2 with real scalars.
  2. Give an explicit isomorphism from U = \{p \in \mathcal{P}_4(\mathbb{R}) : p(1) = 0\} to \mathbb{R}^{2 \times 2}.
  3. Find the dimensions of (a) \mathcal{L}(\mathcal{P}_3(\mathbb{R}), \mathbb{R}^{2 \times 3}); (b) \mathcal{L}(\mathbb{R}^{2 \times 2}, \mathcal{P}_2(\mathbb{R})); (c) \{T \in \mathcal{L}(\mathbb{R}^3, \mathbb{R}^2) : T(1, 1, 1) = \mathbf{0}\}.
  4. Andre tries to store a cubic profile by its three smokehouse readings, Rp = (p(0), p(2), p(4)) for p \in \mathcal{P}_3(\mathbb{R}). Explain why R is not an isomorphism, and find \ker R.

Proofs

  1. Let V, W be finite-dimensional and T \in \mathcal{L}(V, W). Prove that T is injective if and only if it has a left inverse.
  2. Let V, W be finite-dimensional and T \in \mathcal{L}(V, W). Prove that T is surjective if and only if it has a right inverse.
  3. Let V be any vector space and T \in \mathcal{L}(V) with T^k = 0 for some k \ge 1. Prove that I - T is invertible with inverse I + T + \dots + T^{k-1}.
  4. Let V be any vector space and T \in \mathcal{L}(V) with T^2 = 3T - 2I. Prove that T is invertible and express T^{-1} as a combination of I and T.
  5. Let T: V \to W be an isomorphism. Prove that \Phi(S) = TST^{-1} defines an isomorphism \Phi: \mathcal{L}(V) \to \mathcal{L}(W) with \Phi(S_1S_2) = \Phi(S_1)\Phi(S_2).
  6. Prove that \mathbb{F}^{\infty} is isomorphic to its subspace \{(0, x_2, x_3, \dots)\}, and explain why no finite-dimensional space is isomorphic to a proper subspace of itself.
Answers
  1. E is linear. If Eq = \mathbf{0}, the quadratic q has three roots, so q = 0; both dimensions are 3, so E is invertible (Result 5). With q = a + bx + cx^2: a = q(0) = 1, 2b = q(1) - q(-1) = -2 and 2a + 2c = q(1) + q(-1) = 6, so q = 1 - x + 2x^2.
  2. (a) a + bx + cx^2 \mapsto (a, b, a + b + c) has matrix \begin{bmatrix} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 1 & 1 & 1 \end{bmatrix} of determinant 1, so T is invertible (Result 6), with T^{-1}(y_1, y_2, y_3) = y_1 + y_2x + (y_3 - y_1 - y_2)x^2. (b) Sp = (a,\ a + 2b + 4c,\ b + 2c) has matrix \begin{bmatrix} 1 & 0 & 0 \\ 1 & 2 & 4 \\ 0 & 1 & 2 \end{bmatrix} of rank 2, so S is not invertible; \ker S = \operatorname{span}\{x^2 - 2x\}. Since p(2) - p(0) = 2p'(1) for every quadratic, \operatorname{im} S is the plane y_2 = y_1 + 2y_3 (both have dimension 2).
  3. T = I - 2D^2 and D^4 = 0 on \mathcal{P}_3(\mathbb{R}), so (I - 2D^2)(I + 2D^2) = I - 4D^4 = I and, by Result 5, T^{-1} = I + 2D^2. Then p = (x^3 - x) + 2(6x) = x^3 + 11x; check: p'' = 6x and p - 2p'' = x^3 - x.
  4. T(1) = 2, T(x) = 2x + 1, T(x^2) = 2x^2 + 2x + 1, so [T]_{\mathcal{M}} = \begin{bmatrix} 2 & 1 & 1 \\ 0 & 2 & 2 \\ 0 & 0 & 2 \end{bmatrix}, triangular with nonzero diagonal, hence invertible (Result 6). Solving Tp = x^2 from the bottom row up gives c = \tfrac12, b = -\tfrac12, a = 0: T^{-1}(x^2) = \tfrac12(x^2 - x).
  5. The standard matrix \begin{bmatrix} 1 & c \\ c & 1 \end{bmatrix} has determinant 1 - c^2, so T is invertible exactly when c \neq 1 and c \neq -1 (Result 6). For c = i the inverse matrix is \tfrac12\begin{bmatrix} 1 & -i \\ -i & 1 \end{bmatrix}, so T^{-1}(1, 0) = (\tfrac12, -\tfrac{i}{2}).
  6. Try Sq = q - b\,q(1): TSq = q + (a - b - ab)q(1), which is q when b = \tfrac{a}{1 + a}. So for a \neq -1, T is invertible (Result 5) with T^{-1}q = q - \tfrac{a}{1 + a}q(1). For a = 2, T^{-1}(x^2) = x^2 - \tfrac23. For a = -1, Tp = p - p(1) vanishes exactly when p is constant, so \ker T = \operatorname{span}\{1\}.
  7. a_0 + a_1x + a_2x^2 + a_3x^3 goes to \begin{bmatrix} a_0 & a_1 \\ 2a_2 & 6a_3 \end{bmatrix}, so relative to (1, x, x^2, x^3) and \mathcal{E} the matrix is \operatorname{diag}(1, 1, 2, 6), invertible (Result 6). Hence T^{-1}\begin{bmatrix} 1 & 2 \\ 3 & 4 \end{bmatrix} = 1 + 2x + \tfrac32x^2 + \tfrac23x^3.
  8. Dimensions: (a) 6; (b) 6 by Result 7; (c) 3; (d) 3; (e) 4, the null space of evaluation at 1, which maps onto \mathbb{R}; (f) 4; (g) 4. By Result 4 the classes are \{(a), (b)\}, \{(c), (d)\} and \{(e), (f), (g)\}.
  9. The list (1, x - 1, \dots, (x - 1)^4) has degrees 0 to 4, so it is independent (the top power in a combination survives) and, of length 5, a basis of \mathcal{P}_4(\mathbb{R}). The coefficient of 1 is p(1), so U has basis ((x - 1), \dots, (x - 1)^4). Sending a_1(x - 1) + a_2(x - 1)^2 + a_3(x - 1)^3 + a_4(x - 1)^4 to \begin{bmatrix} a_1 & a_2 \\ a_3 & a_4 \end{bmatrix} carries this basis to \mathcal{E}, an isomorphism by the proof of Result 4.
  10. (a) 4 \cdot 6 = 24. (b) 4 \cdot 3 = 12. (c) By Results 7 and 3 this subspace has the dimension of \{A \in \mathbb{R}^{2 \times 3} : A(1, 1, 1) = \mathbf{0}\}, the null space of A \mapsto A(1, 1, 1), which maps onto \mathbb{R}^2 (row sums are arbitrary). So the dimension is 6 - 2 = 4.
  11. \dim \mathcal{P}_3(\mathbb{R}) = 4 \neq 3, so by Result 4 no isomorphism exists. R is surjective, since on \mathcal{P}_2(\mathbb{R}) it is the isomorphism E of Example 1, so \dim \ker R = 4 - 3 = 1. The cubic x(x - 2)(x - 4) lies in it, so \ker R = \operatorname{span}\{x^3 - 6x^2 + 8x\}.
  12. If ST = I_V and T\mathbf{v} = \mathbf{0}, then \mathbf{v} = ST\mathbf{v} = \mathbf{0}. Conversely, let T be injective and (\mathbf{v}_1, \dots, \mathbf{v}_n) a basis of V. Then (T\mathbf{v}_1, \dots, T\mathbf{v}_n) is independent (proof of Result 3), so it extends to a basis (T\mathbf{v}_1, \dots, T\mathbf{v}_n, \mathbf{w}_1, \dots, \mathbf{w}_k) of W (Result 3 of Bases and dimension, proved). The linear map lemma gives S \in \mathcal{L}(W, V) with S(T\mathbf{v}_j) = \mathbf{v}_j and S\mathbf{w}_i = \mathbf{0}. Then ST and I_V agree on a basis, so ST = I_V by the uniqueness in the lemma.
  13. If TS = I_W, each \mathbf{w} = T(S\mathbf{w}) is an output. Conversely, let T be surjective and (\mathbf{w}_1, \dots, \mathbf{w}_m) a basis of W. Choose \mathbf{v}_k with T\mathbf{v}_k = \mathbf{w}_k, and by the linear map lemma let S \in \mathcal{L}(W, V) have S\mathbf{w}_k = \mathbf{v}_k. Then TS and I_W agree on a basis, so TS = I_W.
  14. Let R = I + T + \dots + T^{k-1}. By the distributive laws, (I - T)R is (I + \dots + T^{k-1}) - (T + \dots + T^k), which is I - T^k = I, and R(I - T) expands to the same sum. Both products are I, so no dimension count is needed.
  15. T(3I - T) = 3T - T^2 = 2I, and (3I - T)T = 2I as well. With both products checked, T^{-1} = \tfrac32I - \tfrac12T in any dimension.
  16. \Phi(S) is a composite of linear maps, so it lies in \mathcal{L}(W). By the distributive laws, T(aS_1 + S_2)T^{-1} = aTS_1T^{-1} + TS_2T^{-1}, so \Phi is linear, and by associativity \Phi(S_1)\Phi(S_2) = TS_1(T^{-1}T)S_2T^{-1} = \Phi(S_1S_2). Let \Psi(R) = T^{-1}RT. Then \Psi(\Phi(S)) = (T^{-1}T)S(T^{-1}T) = S and likewise \Phi(\Psi(R)) = R, so \Phi is an isomorphism with inverse \Psi.
  17. The right shift R(x_1, x_2, \dots) = (0, x_1, x_2, \dots) is linear, injective, and its range is exactly U = \{(0, x_2, x_3, \dots)\}, so it is an isomorphism onto U, and U is proper since (1, 0, 0, \dots) \notin U. If V is finite-dimensional and V \cong U for a subspace U, then \dim U = \dim V by Result 4, and Result 6 of Bases and dimension, proved gives U = V.
math.wiki · CC-BY-SA-4.0