math.wiki
Linear algebra › Abstract vector spaces › Polynomials and functions as vectors
Sign in

Polynomials and functions as vectors

Polynomials, continuous functions and sequences can be added and scaled like columns of numbers. How do span and independence work when the vectors are functions, and when do a polynomial’s values pin down its coefficients?

Where you are 5 in Abstract vector spaces

Why this exists

Layla and Hiro are drawing a font in many weights. Each edge of a letter is a curve traced as a parameter t runs from 0 to 1, and its horizontal coordinate is a polynomial in t of degree at most 3. They draw two masters, a light curve L and a bold curve B; every other weight is a blend (1 - s)L + sB. The font file stores each curve as its four coefficients and blends those lists.

Is blending lists the same as blending curves point by point? Which curves can two masters reach? Could two lists draw the same curve? These are questions about the vector space of cubic polynomials. This lesson studies spaces of polynomials, functions and sequences, and how span and independence work in them.

Definition

Fix \mathbb{F}, meaning \mathbb{R} or \mathbb{C}. By Result 8 of Vector spaces from axioms, the functions \mathbb{F}^{S} on a nonempty set S form a vector space under pointwise operations. Every space below sits inside some \mathbb{F}^{S}, so the subspace test (Result 9 there) is the only check needed.

A polynomial over \mathbb{F} (Polynomials) is a function p: \mathbb{F} \to \mathbb{F} for which there are an integer m \ge 0 and coefficients a_0, \dots, a_m \in \mathbb{F} such that, for every z \in \mathbb{F},

p(z) = a_0 + a_1z + \dots + a_mz^m.

\mathcal{P}(\mathbb{F}) is the set of all polynomials. If p \neq 0, its degree \deg p is the largest k with a_k \neq 0; the zero polynomial has degree -\infty, less than every integer (Degree, terms and standard form). Result 2 shows that the coefficients, so the degree, depend only on the function p. \mathcal{P}_m(\mathbb{F}) is the set of polynomials of degree at most m; it contains 0.

For real a < b, C[a, b] is the set of continuous functions f: [a, b] \to \mathbb{R} (Continuity). Put C^{0}[a, b] = C[a, b] and, for k \ge 1, let C^{k}[a, b] be the set of f differentiable on [a, b] (one-sided at the ends) with f' \in C^{k-1}[a, b]. C(\mathbb{R}) and C^{k}(\mathbb{R}) are defined the same way on the whole line.

In a vector space V, a linear combination of a list (\mathbf{v}_1, \dots, \mathbf{v}_m) is a vector a_1\mathbf{v}_1 + \dots + a_m\mathbf{v}_m with a_j \in \mathbb{F}. The span \operatorname{span}\{\mathbf{v}_1, \dots, \mathbf{v}_m\} is the set of all of them, and the span of the empty list is \{\mathbf{0}\}. The list is linearly independent if such a combination is \mathbf{0} only when every a_j = 0, and linearly dependent otherwise. These are the definitions of Linear combination, Span and Linear independence word for word. The span of an infinite set is the set of combinations of its finite lists, so \mathcal{P}(\mathbb{F}) = \operatorname{span}\{1, z, z^2, \dots\}. For functions, \mathbf{0} is the zero function 0, so a_1f_1 + a_2f_2 = 0 means zero at every point.

Intuition

A polynomial in \mathcal{P}_3(\mathbb{R}) behaves like the list of its four coefficients: adding polynomials adds coefficients (Result 4). A function on [a, b] behaves like a list with one entry for each point, added entry by entry, as in the figure.

0 0.25 0.5 0.75 1 1 2 3 t L M B
Figure 1The light master L(t) = 1 + 6t - 6t^2, the bold master B(t) = 3 - 2t + 2t^3 and their blend M = \tfrac{1}{2}L + \tfrac{1}{2}B. At every t, such as t = \tfrac{1}{4}, the value of M is the average of the values of L and B.

Two things are new beside \mathbb{F}^n. Some of these spaces no finite list spans (Result 5). And with infinitely many entries, one row reduction cannot settle independence; instead compare coefficients (Result 2) or evaluate at a few points, which gives a square linear system (Result 7).

Results

Results 1, 2 and 7 hold over any field, and Results 3 to 5 over any infinite field, such as \mathbb{Q}.

Result 1 (roots). Let m \ge 0 and a_0, \dots, a_m \in \mathbb{F} with a_m \neq 0. Then p(z) = a_0 + a_1z + \dots + a_mz^m has at most m roots in \mathbb{F}.

Result 2 (coefficients are unique). If \sum_{k=0}^{m} a_kz^k and \sum_{k=0}^{m} b_kz^k agree at m + 1 distinct points of \mathbb{F}, then a_k = b_k for every k. In particular, over \mathbb{R} or \mathbb{C}, two coefficient lists that give the same function are equal.

Result 3 (polynomial spaces). \mathcal{P}(\mathbb{F}) is a subspace of \mathbb{F}^{\mathbb{F}}, and \mathcal{P}_m(\mathbb{F}) is a subspace of \mathcal{P}(\mathbb{F}) for every m \ge 0. For p, q \in \mathcal{P}(\mathbb{F}) and c \neq 0, \deg(p + q) \le \max(\deg p, \deg q) and \deg(cp) = \deg p.

Result 4 (the monomial list). The list (1, z, \dots, z^m) spans \mathcal{P}_m(\mathbb{F}) and is linearly independent. Each p \in \mathcal{P}_m(\mathbb{F}) is a combination a_0 + a_1z + \dots + a_mz^m in exactly one way, and matching p with (a_0, \dots, a_m) \in \mathbb{F}^{m+1} sends sums to sums and multiples to multiples.

Result 5 (no finite list spans the polynomials). For every list (p_1, \dots, p_n) in \mathcal{P}(\mathbb{F}), \operatorname{span}\{p_1, \dots, p_n\} \neq \mathcal{P}(\mathbb{F}).

Result 6 (continuous and differentiable functions). For a < b and k \ge 1, C[a, b] is a subspace of \mathbb{R}^{[a, b]} and C^{k}[a, b] is a subspace of C^{k-1}[a, b]; the same holds on \mathbb{R}. Restricting each real polynomial to [a, b] puts \mathcal{P}(\mathbb{R}) inside every C^{k}[a, b], and distinct polynomials have distinct restrictions.

Result 7 (the evaluation test). Let f_1, \dots, f_n \in \mathbb{F}^{S}. If there are points s_1, \dots, s_n \in S for which the n \times n matrix M with entries M_{ij} = f_j(s_i) is invertible, then (f_1, \dots, f_n) is linearly independent.

Why it is true

Result 1. By Proof by induction on m. If m = 0, p is the nonzero constant a_0 and has no roots. Let m \ge 1, assume the result for m - 1, and suppose p has a root r (if none, there is nothing to prove). For k \ge 1,

\begin{aligned}z^k - r^k &= (z - r)\,(z^{k-1} + z^{k-2}r \\ &\qquad + \dots + r^{k-1}),\end{aligned}

so p(z) = p(z) - p(r) = \sum_{k=1}^{m} a_k(z^k - r^k) equals (z - r)q(z) for a polynomial q whose coefficient of z^{m-1} is a_m \neq 0, since only the term k = m reaches z^{m-1}. This is the The remainder and factor theorems. If s \neq r is another root of p, then (s - r)q(s) = 0 with s - r \neq 0, so q(s) = 0, because a product of nonzero elements of a field is nonzero. By the induction hypothesis q has at most m - 1 roots, so p has at most m.

Result 2. Let c_k = a_k - b_k; then \sum_k c_kz^k is zero at the m + 1 points. If some c_k \neq 0, let j \le m be the largest such k. Result 1 allows at most j \le m roots, a contradiction. Over \mathbb{R} or \mathbb{C}, two lists giving the same function agree at infinitely many points.

Result 3. The zero function is a polynomial with every coefficient 0. Let p = \sum_{k=0}^{N} a_kz^k and q = \sum_{k=0}^{N} b_kz^k, padding the shorter list with zeros. By the distributive and commutative laws in \mathbb{F},

(cp + q)(z) = \sum_{k=0}^{N} (ca_k + b_k)z^k,

so cp + q is a polynomial and the subspace test applies. With coefficients well defined (Result 2): for k > \max(\deg p, \deg q), p + q has coefficient a_k + b_k = 0; and for c \neq 0, ca_k = 0 exactly when a_k = 0. So degree at most m survives sums and multiples, and \mathcal{P}_m(\mathbb{F}) passes the test too.

Result 4. Each p \in \mathcal{P}_m(\mathbb{F}) is a_0 + a_1z + \dots + a_mz^m by definition, so the list spans. If such a combination is the zero function, Result 2 (comparing with the list of zeros) gives every a_k = 0. Two expressions for one p have equal coefficients by Result 2, and the display in the proof of Result 3 shows that coefficients add and scale.

Result 5. Choose an integer N \ge 0 larger than every \deg p_j. By the degree rules of Result 3, every combination of p_1, \dots, p_n has degree less than N (the empty combination is 0), but z^N has degree N. So z^N is not in the span.

Result 6. The zero function is continuous, and if f, g are continuous at x, so is af + g, by the limit laws (Continuity). For k \ge 1, induct on k. A differentiable function is continuous, so C^{1} \subseteq C^{0}; for k \ge 2, f' \in C^{k-1} \subseteq C^{k-2} by the induction hypothesis, so C^{k} \subseteq C^{k-1}. If f, g \in C^{k}[a, b], then (af + g)' = af' + g' (Definition of the derivative) lies in the subspace C^{k-1}[a, b]. On \mathbb{R} the proof is the same. A polynomial has a polynomial derivative (the power rule), so its restriction lies in every C^{k}[a, b]; two polynomials with the same restriction agree at infinitely many points, so they are equal by Result 2.

Result 7. Suppose c_1f_1 + \dots + c_nf_n = 0. Evaluating at each s_i gives \sum_j f_j(s_i)c_j = 0, that is, M\mathbf{c} = \mathbf{0} for \mathbf{c} = (c_1, \dots, c_n). Since M is invertible, \mathbf{c} = \mathbf{0} (The invertible matrix theorem).

Worked examples

Example 1

Blending two masters

Layla’s masters are L(t) = 1 + 6t - 6t^2 and B(t) = 3 - 2t + 2t^3 in \mathcal{P}_3(\mathbb{R}). Find (1 - s)L + sB in general and at s = \tfrac{1}{2}, \tfrac{1}{4} and the extra-bold \tfrac{3}{2}. When has the blend degree less than 3? Why may the software blend coefficient lists?

Solution

By Result 3, combinations are computed coefficient by coefficient: the constant term is (1 - s) \cdot 1 + s \cdot 3, and so on. So

\begin{aligned}(1 - s)L + sB &= (1 + 2s) + (6 - 8s)t \\ &\quad + (-6 + 6s)t^2 + 2st^3.\end{aligned}

At s = \tfrac{1}{2} this is M(t) = 2 + 2t - 3t^2 + t^3, the curve in the figure. At s = \tfrac{1}{4} it is \tfrac{3}{2} + 4t - \tfrac{9}{2}t^2 + \tfrac{1}{2}t^3, and at s = \tfrac{3}{2} it is 4 - 6t + 3t^2 + 3t^3, still in \mathcal{P}_3(\mathbb{R}): a subspace does not stop at the masters. The coefficient of t^3 is 2s, zero only at s = 0, so only the light master has degree less than 3.

By Result 4, matching curves with coefficient lists in \mathbb{R}^4 respects blends, and no two lists give one curve.

Example 2

Membership in a span by comparing coefficients

In \mathcal{P}_2(\mathbb{R}) let p_1 = 1 + 2z - z^2 and p_2 = 3 - z + z^2. Are r_1 = 5 + 3z - z^2 and r_2 = 1 + z + z^2 in \operatorname{span}\{p_1, p_2\}? Describe the span by an equation.

Solution

By Result 3 the coefficients of ap_1 + bp_2 are a + 3b, 2a - b and -a + b. By Result 2, the combination equals c_0 + c_1z + c_2z^2 exactly when

\begin{aligned} a + 3b &= c_0, \\ 2a - b &= c_1, \\ -a + b &= c_2. \end{aligned}

For r_1, the first two equations give a = 2, b = 1, and -a + b = -1 matches, so r_1 = 2p_1 + p_2. For r_2, the first two give a = \tfrac{4}{7}, b = \tfrac{1}{7}, but then -a + b = -\tfrac{3}{7} \neq 1, so r_2 is not in the span.

Since (a + 3b) - 4(2a - b) - 7(-a + b) = 0 for all a, b, every member of the span has c_0 - 4c_1 - 7c_2 = 0. Conversely, if c_0 = 4c_1 + 7c_2, then a = c_1 + c_2, b = c_1 + 2c_2 solve all three equations. So the span is this plane: 5 - 12 + 7 = 0 for r_1, while 1 - 4 - 7 = -10 for r_2.

Example 3

Independence by evaluation

In C[0, \pi], is (1, \cos x, \cos 2x) linearly independent? Is (1, \cos^2 x, \cos 2x)?

Solution

Use Result 7 with the points 0, \tfrac{\pi}{2}, \pi; row i lists the three functions at the i-th point:

M = \begin{bmatrix} 1 & 1 & 1 \\ 1 & 0 & -1 \\ 1 & -1 & 1 \end{bmatrix}, \qquad \det M = -4.

So the first list is independent. The second is dependent: \cos 2x = 2\cos^2 x - 1, so 1 - 2\cos^2 x + \cos 2x is the zero function, with coefficients 1, -2, 1. Its matrix at the same points has equal first and third rows, which fits dependence but would not prove it (see Boundaries).

Example 4

A solution space of a differential equation

Let U be the set of y \in C^{2}(\mathbb{R}) with y'' - 3y' + 2y = 0. Show that U is a subspace containing e^{x} and e^{2x}, find the member ae^{x} + be^{2x} with y(0) = 1 and y'(0) = 4, and decide whether the solutions of y'' - 3y' + 2y = 6 form a subspace.

Solution

C^{2}(\mathbb{R}) is a vector space by Result 6, and 0 \in U. For y, w \in U and a scalar a, the sum and constant multiple rules for derivatives give

\begin{aligned}&(ay + w)'' - 3(ay + w)' + 2(ay + w) \\ &= a(y'' - 3y' + 2y) + (w'' - 3w' + 2w) = 0,\end{aligned}

so U is a subspace. Substituting (Differential equations and their solutions), e^{x} gives (1 - 3 + 2)e^{x} = 0 and e^{2x} gives (4 - 6 + 2)e^{2x} = 0, so their whole span lies in U.

For y = ae^{x} + be^{2x}, y(0) = a + b = 1 and y'(0) = a + 2b = 4, so b = 3, a = -2: y = -2e^{x} + 3e^{2x}.

The zero function gives 0 \neq 6, so the solutions of y'' - 3y' + 2y = 6 are not a subspace. The constant 3 is one solution, and the difference of two solutions lies in U, so every solution is 3 plus a member of U.

Example 5

Even polynomials, a short proof

Call p \in \mathcal{P}(\mathbb{F}) even if p(-z) = p(z) for every z \in \mathbb{F}. Prove that the even polynomials form a subspace E of \mathcal{P}(\mathbb{F}) and that E = \operatorname{span}\{1, z^2, z^4, \dots\}.

Solution

The zero polynomial is even. If p, q are even and a \in \mathbb{F}, then (ap + q)(-z) = ap(-z) + q(-z) equals ap(z) + q(z), so E passes the subspace test.

Each z^{2j} is even, since (-z)^{2j} = z^{2j}, and a subspace is closed under linear combinations, so the span lies in E. Conversely, let p = \sum_{k=0}^{m} a_kz^k be even. For every z,

\begin{aligned}0 &= p(z) - p(-z) \\ &= \sum_{k=0}^{m} a_k\bigl(1 - (-1)^k\bigr)z^k \\ &= \sum_{k \text{ odd}} 2a_kz^k.\end{aligned}

By Result 2 every coefficient of this zero function is 0, so 2a_k = 0 for odd k, and a_k = 0 because 2 \neq 0 in \mathbb{F}. So p is a combination of even powers.

Example 6

Sequences given by a rule

Let W be the set of real sequences \mathbf{x} = (x_1, x_2, \dots) with x_{k+2} = 5x_{k+1} - 6x_k for every k \ge 1. Find the geometric sequences (r^k) in W. Prove that W = \operatorname{span}\{\mathbf{g}, \mathbf{h}\} for \mathbf{g} = (2^k) and \mathbf{h} = (3^k), with (\mathbf{g}, \mathbf{h}) linearly independent. Find the member with x_1 = 1, x_2 = 5.

Solution

W is a subspace of \mathbb{R}^{\infty}: the zero sequence obeys the rule, and if \mathbf{x}, \mathbf{y} do, then ax_{k+2} + y_{k+2} equals 5(ax_{k+1} + y_{k+1}) - 6(ax_k + y_k). A geometric sequence obeys it exactly when r^k(r^2 - 5r + 6) = r^k(r - 2)(r - 3) = 0 for every k, so r = 2 or r = 3 (r = 0 gives the zero sequence). So \mathbf{g}, \mathbf{h} \in W, and their span lies in W.

Independence: Result 7 with S = \{1, 2, 3, \dots\} and the points 1, 2 gives

M = \begin{bmatrix} 2 & 3 \\ 4 & 9 \end{bmatrix}, \qquad \det M = 6 \neq 0.

Spanning: for \mathbf{x} \in W, as M is invertible there are a, b with 2a + 3b = x_1 and 4a + 9b = x_2. Then \mathbf{y} = \mathbf{x} - a\mathbf{g} - b\mathbf{h} lies in W with y_1 = y_2 = 0, and by induction on k, y_k = y_{k+1} = 0 forces y_{k+2} = 0. So \mathbf{y} is the zero sequence and \mathbf{x} = a\mathbf{g} + b\mathbf{h}.

For x_1 = 1, x_2 = 5, the system gives b = 1, a = -1, so x_k = 3^k - 2^k, running 1, 5, 19, 65, \dots; indeed 5 \cdot 19 - 6 \cdot 5 = 65.

Boundaries

Polynomial functions need enough points. Over the integers mod 2 (Modular arithmetic), z^2 + z is 0 at both elements: the zero function, with nonzero coefficients. The last claim of Result 2 fails, and Results 4 and 5 with it, since the polynomial functions there are just 0, 1, z and 1 + z. Exercise 11 shows the same effect on three points of \mathbb{R}.

The evaluation test runs one way. A singular matrix proves nothing. At 0 and \pi, both \sin x and \sin 2x vanish, so the matrix is zero, yet the pair is independent: at \tfrac{\pi}{2} and \tfrac{\pi}{4} the matrix is \displaystyle \begin{bmatrix} 1 & 0 \\ \frac{\sqrt{2}}{2} & 1 \end{bmatrix}, with determinant 1.

Infinite sums are not linear combinations. e^{x} is the sum of its Taylor and Maclaurin series, yet it is not in \operatorname{span}\{1, x, x^2, \dots\} (Exercise 13): a span uses finite lists only. Infinite sums need limits, so a distance, which Inner products and norms supply and Fourier series use.

Where this goes next

Span and independence, proved proves the facts about span and independence in any vector space. With Bases and dimension, proved, Result 4 gives \dim \mathcal{P}_m(\mathbb{F}) = m + 1 and Result 5 makes \mathcal{P}(\mathbb{F}) infinite-dimensional. Differentiation and evaluation are Linear maps between vector spaces, and Example 4’s U is a null space (Null space, range and the fundamental theorem).

Exercises

Polynomials

  1. Hiro’s second stroke has light master P(t) = 2 + 3t - t^3 and bold master Q(t) = 4 - 3t + 3t^2. (a) For which s has (1 - s)P + sQ no t term, and what is that blend? (b) For which s is its degree less than 3? (c) Show that every blend has value 4 at t = 1.
  2. Which are subspaces of \mathcal{P}_3(\mathbb{R})? (a) the p with p(0) = p(1); (b) the polynomials of degree exactly 3; (c) the polynomials with integer coefficients.
  3. In \mathcal{P}_2(\mathbb{R}) let p_1 = 1 - z + 2z^2 and p_2 = 2 + z - z^2. Decide whether 4 + 5z - 7z^2 and 3 + z lie in \operatorname{span}\{p_1, p_2\}, and find a linear equation in c_0, c_1, c_2 that describes the span.
  4. Write z^2 and 3 - z + 2z^2 as combinations of (1, z - 1, (z - 1)^2), and show that this list is linearly independent in \mathcal{P}_2(\mathbb{R}).
  5. Show that (1 + z, 1 - z, z + z^2) is linearly independent in \mathcal{P}_2(\mathbb{R}), and write 2 + 3z^2 as a combination of it.

Functions and sequences

  1. Which lists are linearly independent in C[-1, 1]? (a) (e^{x}, e^{-x}, \cosh x); (b) (e^{x}, xe^{x}, x^2e^{x}); (c) (x, \lvert x \rvert).
  2. Which are subspaces of C[0, 1]? (a) the f with \int_0^1 f(x)\,dx = 0; (b) the f differentiable at \tfrac{1}{2}; (c) the nondecreasing f.
  3. Show that \cos 3x and \sin 3x solve y'' + 9y = 0, and find the solution a\cos 3x + b\sin 3x with y(0) = 2 and y'(0) = -3. Show that 2 + \cos 3x solves y'' + 9y = 18, and that the solutions of y'' + 9y = 18 do not form a subspace.
  4. Find the geometric sequences (r^k) with x_{k+2} = x_{k+1} + 2x_k, and the member of their span with x_1 = 0, x_2 = 6. Give x_3 and x_4.
  5. Hiro joins p(t) = 2 + t - t^2 on [0, 1] to q(t) = c_0 + c_1(t - 1) + c_2(t - 1)^2 on [1, 2]. Find c_0, c_1, c_2 so that the joined function lies in C^{1}[0, 2] and q(2) = 3.
  6. Let S = \{0, 1, 2\}. Find a, b, c with x^3 = a + bx + cx^2 for every x \in S, and explain why this does not contradict Result 2.

Proofs

  1. Prove that \deg(pq) = \deg p + \deg q for p, q \in \mathcal{P}(\mathbb{F}), where -\infty plus anything is -\infty.
  2. Prove that e^{x} is not a polynomial function on \mathbb{R}.
  3. Prove that the sequences in \mathbb{F}^{\infty} with only finitely many nonzero terms form a subspace that no finite list spans.
  4. Let c_0, \dots, c_m \in \mathbb{F} be distinct and p_j(z) = \prod_{k \neq j}(z - c_k). Prove that (p_0, \dots, p_m) is linearly independent and spans \mathcal{P}_m(\mathbb{F}).
  5. Prove that (\sin x, \sin 2x, \dots, \sin nx) is linearly independent in C[0, \pi] for every n \ge 1, by integrating against \sin kx.
Answers
  1. The blend is (2 + 2s) + (3 - 6s)t + 3st^2 + (-1 + s)t^3. (a) s = \tfrac{1}{2}, giving 3 + \tfrac{3}{2}t^2 - \tfrac{1}{2}t^3. (b) Only s = 1, where the blend is Q. (c) P(1) = Q(1) = 4, so the blend gives (1 - s) \cdot 4 + s \cdot 4 = 4.
  2. (a) Yes: 0 qualifies, and (ap + q)(0) = ap(0) + q(0) equals ap(1) + q(1) = (ap + q)(1). (b) No: z^3 + z and -z^3 have degree 3, but their sum z does not. (c) No: 1 qualifies but \tfrac{1}{2} \cdot 1 does not.
  3. The coefficients of ap_1 + bp_2 are a + 2b, -a + b and 2a - b. For 4 + 5z - 7z^2: a = -2, b = 3, and 2a - b = -7 matches, so it is -2p_1 + 3p_2. For 3 + z: a = \tfrac{1}{3}, b = \tfrac{4}{3}, but 2a - b = -\tfrac{2}{3} \neq 0, so no. The span is c_0 - 5c_1 - 3c_2 = 0: both (1, -1, 2) and (2, 1, -1) satisfy it, and conversely, if c_0 = 5c_1 + 3c_2, then a = c_1 + c_2, b = 2c_1 + c_2 solve all three equations.
  4. z^2 = ((z - 1) + 1)^2 = 1 + 2(z - 1) + (z - 1)^2, and 3 - z + 2z^2 = 4 + 3(z - 1) + 2(z - 1)^2. If a + b(z - 1) + c(z - 1)^2 = 0, expanding gives cz^2 + (b - 2c)z + (a - b + c) = 0, so by Result 2, c = 0, then b = 0, then a = 0.
  5. The combination a(1 + z) + b(1 - z) + c(z + z^2) has coefficients a + b, a - b + c and c. Setting it to 0 gives c = 0, a + b = 0, a - b = 0, so all are 0. For 2 + 3z^2: c = 3, a + b = 2, a - b = -3, so a = -\tfrac{1}{2}, b = \tfrac{5}{2}.
  6. (a) Dependent: \tfrac{1}{2}e^{x} + \tfrac{1}{2}e^{-x} - \cosh x = 0. (b) Independent: at 0, 1, -1 the rows are (1, 0, 0), (e, e, e) and (e^{-1}, -e^{-1}, e^{-1}), with determinant 2. (c) Independent: at 1 and -1 the rows are (1, 1) and (-1, 1), with determinant 2.
  7. (a) Yes: \int_0^1 (af + g) = a\int_0^1 f + \int_0^1 g (The definite integral), and 0 qualifies. (b) Yes: 0 is differentiable, and by the sum and constant multiple rules so is af + g. (c) No: x is nondecreasing, but (-1)x is not.
  8. (\cos 3x)'' = -9\cos 3x and (\sin 3x)'' = -9\sin 3x. Then y(0) = a = 2 and y'(0) = 3b = -3, so y = 2\cos 3x - \sin 3x. For y = 2 + \cos 3x, y'' + 9y is -9\cos 3x + 18 + 9\cos 3x, which is 18. The zero function gives 0 \neq 18.
  9. r^2 = r + 2 gives r = 2 or r = -1. From 2a - b = 0 and 4a + b = 6, a = 1 and b = 2, so x_k = 2^k + 2(-1)^k, with x_3 = 6 and x_4 = 18.
  10. p(1) = 2 and p'(1) = -1, so continuity needs c_0 = 2 and equal slopes c_1 = -1; then q(2) = 2 - 1 + c_2 = 3 gives c_2 = 2. Then both one-sided derivatives at 1 are -1, and the derivative is continuous.
  11. At 0: a = 0. At 1 and 2: b + c = 1 and 2b + 4c = 8, so c = 3, b = -2, and x^3 = -2x + 3x^2 on S. Result 2 needs 4 points for two polynomials of degree at most 3; these agree at only 3, the roots of x^3 - 3x^2 + 2x = x(x - 1)(x - 2).
  12. If p = 0 or q = 0, both sides are -\infty. Otherwise let p = \sum_{j=0}^{m} a_jz^j and q = \sum_{k=0}^{n} b_kz^k with a_m, b_n \neq 0. By the distributive law, pq = \sum_i \bigl(\sum_{j+k=i} a_jb_k\bigr)z^i. A nonzero product a_jb_k has j \le m and k \le n, so j + k \le m + n, with equality only for j = m, k = n. So the coefficient of z^{m+n} is a_mb_n \neq 0 and all later ones are 0. By Result 2, \deg(pq) = m + n.
  13. Suppose e^{x} = p(x) for every real x, with p \in \mathcal{P}(\mathbb{R}). Then p \neq 0, since e^{0} = 1; let m = \deg p. Each derivative lowers a polynomial’s degree by one (the power rule), so the (m + 1)-th derivative of p is 0, while that of e^{x} is e^{x}. So e^{x} = 0 for every x, contradicting e^{0} = 1 (Proof by contradiction).
  14. Call the set U; it contains \mathbf{0}, and if \mathbf{x}, \mathbf{y} are zero after indices N_1, N_2, then a\mathbf{x} + \mathbf{y} is zero after \max(N_1, N_2). Given a list in U, choose N with every member zero after index N; so is every combination, but the sequence with 1 in position N + 1 and 0 elsewhere lies in U but not in the span.
  15. Each p_j has degree m. At the points c_i, p_j(c_i) = 0 for i \neq j, while p_j(c_j) = \prod_{k \neq j}(c_j - c_k) \neq 0. So the matrix of Result 7 is diagonal with nonzero diagonal, and the list is independent. Given p \in \mathcal{P}_m(\mathbb{F}), let \displaystyle q = \sum_j \frac{p(c_j)}{p_j(c_j)}p_j. At each c_i only the term j = i survives, so q(c_i) = p(c_i), and Result 2 gives p = q, a member of the span.
  16. Suppose \sum_{j=1}^{n} a_j\sin jx = 0 on [0, \pi]. Multiplying by \sin kx and integrating gives \sum_j a_jI_{jk} = 0, where I_{jk} = \int_0^{\pi}\sin jx\sin kx\,dx, which equals \tfrac{1}{2}\int_0^{\pi}\cos(j - k)x\,dx - \tfrac{1}{2}\int_0^{\pi}\cos(j + k)x\,dx. For an integer l \neq 0, \displaystyle \int_0^{\pi}\cos lx\,dx = \frac{\sin l\pi}{l} = 0. So I_{jk} = 0 for j \neq k and I_{kk} = \tfrac{\pi}{2}, hence a_k = 0 for every k.
math.wiki · CC-BY-SA-4.0