Skip to content
Topos Uranos

Summary

The lesson on higher-order derivatives and the differential proved two facts that this lesson brings together: a polynomial is determined by the values of its derivatives at a point, and the tangent line approximates a differentiable function with an error negligible compared with the increment of the variable. The Taylor polynomial of order nn of a function ff at a point aa, Pn,aP_{n,a}, is the polynomial of degree at most nn that has at aa the same derivatives as ff up to order nn. Taylor’s formula with the Peano remainder proves, under the exact hypotheses (n−1n - 1 derivatives on an open interval and the one of order nn at the point), that the error Rn,a(x)=f(x)−Pn,a(x)R_{n,a}(x) = f(x) - P_{n,a}(x) is negligible compared with (x−a)n(x - a)^n, and that Pn,aP_{n,a} is the only polynomial of degree at most nn with that property. From uniqueness follows an algebra of expansions: sums, products, compositions, quotients and antiderivatives of known expansions give new expansions without computing derivatives, and their coefficients give the derivatives at the point. With them one computes limits of functions and of sequences that equivalences cannot reach and that L’Hôpital’s rule solves only with many differentiations. The following lesson will bound the error.

Learning objectives

  1. Define the Taylor polynomial Pn,aP_{n,a} and prove that it is the only polynomial of degree at most nn whose difference from the function is negligible compared with (x−a)n(x - a)^n.
  2. Prove Taylor’s formula with the Peano remainder under its exact hypotheses, and recognize that its converse is false.
  3. Obtain the expansions of the elementary functions and, with the algebra of expansions, those of their sums, products, compositions, quotients and antiderivatives.
  4. Compute with expansions limits of functions and of sequences, orders of infinitesimals and derivatives at a point.

The Taylor polynomial

The corollary on higher-order derivatives of a polynomial, from the lesson on higher-order derivatives and the differential, reconstructs a polynomial from the values of its derivatives at 00, and that lesson announced that the same sum, truncated, applied to a function that is not a polynomial, is the Taylor polynomial. The theorem according to which the differential is the principal part of the increment, from the same lesson, says that, if ff is differentiable at aa, the increment f(x)−f(a)f(x) - f(a) differs from f′(a) (x−a)f'(a) \, (x - a) by an amount negligible compared with x−ax - a: the tangent line is a polynomial of degree at most 11 whose error, compared with x−ax - a, tends to 00. This lesson replaces the line by a polynomial of degree at most nn, and the error is compared with (x−a)n(x - a)^n. The comparisons of the definition of comparison of functions, from the lesson on equivalent infinitesimals and infinitely large functions, are used with the target aa: “near aa” means in a punctured interval centered at aa, and since (x−a)n≠0(x - a)^n \neq 0 for x≠ax \neq a, it makes sense to compare with (x−a)n(x - a)^n. As in the definition of comparison of sequences, from the lesson on finer convergence tests, an equality such as f(x)=g(x)+o⁡((x−a)n)f(x) = g(x) + \operatorname{o}\left((x - a)^n\right) means that f(x)−g(x)=o⁡((x−a)n)f(x) - g(x) = \operatorname{o}\left((x - a)^n\right), and likewise with O⁡\operatorname{O}; the power (x−a)0(x - a)^0 equals 11, also at x=ax = a. As in the lesson on higher-order derivatives, functions are taken to be defined on an open interval. First, two lemmas on translations and polynomials.

LemmaDerivatives of a translate

Let FF be a function defined on an open interval JJ, cc a real number and II the open interval of the xx with x+cx + c in JJ, and let G(x)=F(x+c)G(x) = F(x + c) for xx in II. If FF is kk times differentiable on JJ, then GG is kk times differentiable on II and G(k)(x)=F(k)(x+c)G^{(k)}(x) = F^{(k)}(x + c). If FF is n−1n - 1 times differentiable on JJ and nn times differentiable at a point b+cb + c, with bb in II, then GG is nn times differentiable at bb and G(n)(b)=F(n)(b+c)G^{(n)}(b) = F^{(n)}(b + c). Translating the variable does not change the derivatives: the inner function x+cx + c has derivative 11.

Proof

  1. G′(x)=F′(x+c)⋅1=F′(x+c)\dato{tras}{G'(x) = F'(x + c) \cdot 1 = F'(x + c)}

    Let xx be a point of II such that FF is differentiable at x+cx + c. The inner function x+cx + c is differentiable at xx, with derivative 11, and its values on II lie in JJ; by the chain rule theorem, from the lesson on the chain rule, GG is differentiable at xx.

  2. G(j)(x)=F(j)(x+c)→G(j+1)(x)=F(j+1)(x+c)G^{(j)}(x) = F^{(j)}(x + c) \rightarrow \resaltar{G^{(j + 1)}(x) = F^{(j + 1)}(x + c)}

    If G(j)(x)=F(j)(x+c)G^{(j)}(x) = F^{(j)}(x + c) on II for some jj less than the assumed order of differentiation, the preceding move, applied to F(j)F^{(j)} in place of FF, gives the derivative of the following order at each point xx such that F(j)F^{(j)} is differentiable at x+cx + c: on all of II, or only at bb if j=n−1j = n - 1. Induction on jj, starting from G(0)=GG^{(0)} = G, gives the two statements.

LemmaPolynomials in powers of x minus a

Let aa be a real number and QQ a polynomial of degree at most nn. Then there are numbers c0,c1,…,cnc_0, c_1, \ldots, c_n, and only one choice of them, such that

Q(x)=∑k=0nck(x−a)kQ(x) = \sum_{k=0}^{n} c_k (x - a)^k

for every real number xx, and they are the quotients ck=Q(k)(a)k!c_k = \frac{Q^{(k)}(a)}{k!}, for 0≤k≤n0 \leq k \leq n. Writing x=(x−a)+ax = (x - a) + a, each power of xx is expanded in powers of x−ax - a; the derivatives at aa are then computed as at 00, by translating the variable.

Proof

  1. xj=((x−a)+a)j=∑i=0j(ji)aj−i(x−a)ix^j = \left((x - a) + a\right)^j = \resaltar{\sum_{i=0}^{j} \binom{j}{i} a^{j - i} (x - a)^i}

    Let Q(x)=b0+b1x+⋯+bnxnQ(x) = b_0 + b_1 x + \cdots + b_n x^n. By the binomial theorem, from the lesson on combinatorics and the binomial theorem, with x−ax - a and aa as summands, each power xjx^j is a sum of multiples of the powers (x−a)i(x - a)^i with i≤ji \leq j; collecting, by the linearity in the theorem on properties of summation, from the lesson on summations, progressions and induction, the terms of each power of x−ax - a, one obtains the ckc_k.

  2. S(k)(0)=k! ck,0≤k≤n\dato{coef}{S^{(k)}(0) = k! \, c_k, \qquad 0 \leq k \leq n}

    Let S(t)=∑k=0ncktkS(t) = \sum_{k=0}^{n} c_k t^k, so that Q(x)=S(x−a)Q(x) = S(x - a) for every real xx. By the corollary on higher-order derivatives of a polynomial, from the lesson on higher-order derivatives and the differential, SS is infinitely differentiable and its derivatives at 00 give its coefficients.

  3. Q(k)(a)=S(k)(0)=k! ck→ck=Q(k)(a)k!Q^{(k)}(a) = S^{(k)}(0) = k! \, c_k \rightarrow \resaltar{c_k = \frac{Q^{(k)}(a)}{k!}}

    By the lemma on derivatives of a translate, with SS in place of FF and c=−ac = -a, the derivative of order kk of QQ at aa is that of SS at 00. Therefore, any way of writing QQ in powers of x−ax - a has the coefficients of the statement: they are unique.

DefinitionTaylor polynomial and remainder

Let ff be a function defined on an open interval II, n≥0n \geq 0 an integer and aa a point of II at which ff is nn times differentiable. The Taylor polynomial of order nn of ff at aa is

Pn,a(x)=∑k=0nf(k)(a)k!(x−a)k=f(a)+f′(a) (x−a)+f′′(a)2(x−a)2+⋯+f(n)(a)n!(x−a)nP_{n,a}(x) = \sum_{k=0}^{n} \frac{f^{(k)}(a)}{k!} (x - a)^k = f(a) + f'(a) \, (x - a) + \frac{f''(a)}{2} (x - a)^2 + \cdots + \frac{f^{(n)}(a)}{n!} (x - a)^n

and the Taylor remainder of order nn is the difference Rn,a(x)=f(x)−Pn,a(x)R_{n,a}(x) = f(x) - P_{n,a}(x), for xx in II. If a=0a = 0 and there is no doubt about the point, one writes Pn(x)P_n(x) and Rn(x)R_n(x).

The Taylor polynomial has degree at most nn, and less if f(n)(a)=0f^{(n)}(a) = 0; that is why one says “of order nn”. The one of order 00 is the constant P0,a(x)=f(a)P_{0,a}(x) = f(a), and the one of order 11, P1,a(x)=f(a)+f′(a) (x−a)P_{1,a}(x) = f(a) + f'(a) \, (x - a), is the tangent line. By the lemma on polynomials in powers of xx minus aa, applied to Pn,aP_{n,a}, whose coefficients are the f(k)(a)k!\frac{f^{(k)}(a)}{k!}, the derivatives of Pn,aP_{n,a} at aa coincide with those of ff up to order nn, and if ff is a polynomial of degree at most nn, then Pn,a=fP_{n,a} = f, for every aa. The remainder RnR_n of a function must not be confused with the remainder Rn=S−SnR_n = S - S_n of a series, from the lesson on numerical series and their convergence, which this lesson does not use. For example, by the theorem on derivatives of order nn of powers and of the exponential, all the derivatives of exe^x equal 11 at 00, and Pn(x)=∑k=0nxkk!P_n(x) = \sum_{k=0}^{n} \frac{x^k}{k!}. The higher-order derivatives of the cosine are minus the sine, minus the cosine, the sine and the cosine, and they repeat every four, by the proposition on derivatives of order nn of sine and cosine; at 00 they equal 00, −1-1, 00 and 11, and therefore P2(x)=P3(x)=1−x22P_2(x) = P_3(x) = 1 - \frac{x^2}{2}: this is the parabola from the lesson on the squeeze theorem, which announced that Taylor’s theorem would make precise how closely the cosine resembles it. The question is how close to ff the polynomial Pn,aP_{n,a} is, and settling it requires a definition.

DefinitionContact of order n and expansion

Let ff and gg be functions defined near aa, and n≥0n \geq 0 an integer. The functions ff and gg have contact of order nn at aa if f(x)−g(x)=o⁡((x−a)n)f(x) - g(x) = \operatorname{o}\left((x - a)^n\right) as xx tends to aa. An expansion of order nn of ff at aa is an equality

f(x)=Q(x)+o⁡((x−a)n)f(x) = Q(x) + \operatorname{o}\left((x - a)^n\right)

where QQ is a polynomial of degree at most nn that has contact of order nn with ff at aa; QQ is the polynomial part of the expansion.

An expansion of order nn gives those of lower order: if m<nm < n and the polynomial part is written in powers of x−ax - a, as the lemma on polynomials in powers of xx minus aa allows, each term ck(x−a)kc_k (x - a)^k with m<k≤nm < k \leq n is o⁡((x−a)m)\operatorname{o}\left((x - a)^m\right), because its quotient by (x−a)m(x - a)^m is ck(x−a)k−mc_k (x - a)^{k - m}, which tends to 00; and o⁡((x−a)n)\operatorname{o}\left((x - a)^n\right) is o⁡((x−a)m)\operatorname{o}\left((x - a)^m\right), because its quotient by (x−a)m(x - a)^m is its quotient by (x−a)n(x - a)^n multiplied by (x−a)n−m(x - a)^{n - m}. By the proposition on properties of the comparison, from the lesson on equivalent infinitesimals and infinitely large functions, applied by induction on the number of summands, it suffices to remove those terms from the polynomial part.

LemmaUniqueness of the expansion

If QQ is a polynomial of degree at most nn and Q(x)=o⁡((x−a)n)Q(x) = \operatorname{o}\left((x - a)^n\right) as xx tends to aa, all the coefficients of QQ are zero. Consequently, a function has at most one expansion of order nn at aa. If some coefficient in powers of x−ax - a were not zero, the first of them would give the quotient by a power of x−ax - a a limit different from 00.

Proof

  1. Q(x)(x−a)j=cj+∑k=j+1nck(x−a)k−j→cj\frac{Q(x)}{(x - a)^j} = c_j + \sum_{k=j + 1}^{n} c_k (x - a)^{k - j} \to \dato{cj}{c_j}

    By the lemma on polynomials in powers of xx minus aa, Q(x)=∑k=0nck(x−a)kQ(x) = \sum_{k=0}^{n} c_k (x - a)^k. Suppose that some ckc_k is not zero, and let jj be the smallest of the indices kk between 00 and nn with ck≠0c_k \neq 0, which exists because there are finitely many. For x≠ax \neq a, the quotient by (x−a)j(x - a)^j is a polynomial, and by the corollary on the limit of a polynomial and of a rational function, from the lesson on the algebra of limits, it tends to its value at aa.

  2. Q(x)(x−a)j=Q(x)(x−a)n (x−a)n−j→0→cj=0\frac{Q(x)}{(x - a)^j} = \frac{Q(x)}{(x - a)^n} \, (x - a)^{n - j} \to 0 \rightarrow \resaltar{c_j = 0}

    On the other hand, the same quotient is the quotient by (x−a)n(x - a)^n, which tends to 00 by hypothesis, multiplied by (x−a)n−j(x - a)^{n - j}, which tends to 00 if j<nj < n and equals 11 if j=nj = n; by the theorem on the limit of a product, it tends to 00. By the theorem on uniqueness of the limit, from the lesson on the limit of functions of a real variable, cj=0c_j = 0, contrary to the choice of jj. Therefore, all the ckc_k are zero.

  3. Q1(x)−Q2(x)=(f(x)−Q2(x))−(f(x)−Q1(x))=o⁡((x−a)n)→Q1=Q2Q_1(x) - Q_2(x) = \left(f(x) - Q_2(x)\right) - \left(f(x) - Q_1(x)\right) = \operatorname{o}\left((x - a)^n\right) \rightarrow \resaltar{Q_1 = Q_2}

    If f(x)=Q1(x)+o⁡((x−a)n)f(x) = Q_1(x) + \operatorname{o}\left((x - a)^n\right) and f(x)=Q2(x)+o⁡((x−a)n)f(x) = Q_2(x) + \operatorname{o}\left((x - a)^n\right), the difference Q1−Q2Q_1 - Q_2 is a polynomial of degree at most nn, and by the proposition on properties of the comparison it is negligible compared with (x−a)n(x - a)^n. By what precedes, it is the zero polynomial.