Skip to content
Topos Uranos

Series, integrals and notable constants

Summary

This lesson closes the course with what series make it possible to compute or study. The lesson on the Taylor series of the elementary functions proved that the exponential, the sine and the cosine are the sum of their Taylor series on the whole line, and the logarithm of 1+x1 + x and the arctangent up to the endpoints of their interval; the lesson on sequences and series of functions, that a uniformly convergent series can be integrated term by term. With these two facts, the antiderivative of e−t2e^{-t^2}, which is not elementary, is written as an alternating series with its error bound, and the integral of x−xx^{-x} on [0,1]\icc{0}{1} turns out to equal the sum of the reciprocals of nnn^n. The sums ln⁡2\ln 2 and π4\frac{\pi}{4}, already proved with Abel’s theorem, are compared with the other route of the unit on numerical series, and with Machin’s formula they give ten decimal places of π\pi with ten terms. The exponential series reveals the nature of two constants: with it we return to the irrationality of ee, and with a polynomial that vanishes at 00 and at π\pi, Niven’s proof shows that π\pi is irrational. Finally, the exponential series converges absolutely at every complex number, and its sum is the complex exponential of Algebra and Geometry II: Euler’s formula remains a definition there, and the computation with admitted series that motivated it is now proved.

Learning objectives

  1. Write as series antiderivatives that are not elementary and integrals such as that of x−xx^{-x}, justifying term-by-term integration and bounding the error.
  2. Obtain ln⁡2\ln 2 and π\pi with series, compare the routes and accelerate the convergence with a guaranteed error bound.
  3. Prove, with Niven’s proof, that π\pi is irrational, and relate the irrationality of ee to the exponential series.
  4. Prove that the exponential series converges absolutely at every complex number and that its sum gives Euler’s formula, distinguishing what is defined from what is proved.

Antiderivatives that are not elementary

By the corollary according to which every function continuous on an interval has an antiderivative, from the lesson on the fundamental theorem of calculus, the function x↦∫0xe−t2 dtx \mapsto \int_0^x e^{-t^2} \, \mathrm{d}t is an antiderivative of e−x2e^{-x^2} on the whole line. No finite combination of powers, exponentials, logarithms, trigonometric functions and their inverses is an antiderivative of e−x2e^{-x^2}; this fact is proved in later courses, and nothing in this course depends on it. The integration techniques of the first unit are therefore of no use, but the exponential series, integrated term by term, gives the values of that antiderivative to any desired precision.

PropositionThe bell curve integral as a series

For every real number xx:

∫0xe−t2 dt=∑n=0∞(−1)nx2n+1n! (2n+1)\int_0^x e^{-t^2} \, \mathrm{d}t = \sum_{n=0}^{\infty} \frac{(-1)^n x^{2n + 1}}{n! \, (2n + 1)}

Moreover, if 0<x≤10 < x \leq 1 and Tn(x)=∑k=0n(−1)kx2k+1k! (2k+1)T_n(x) = \sum_{k=0}^{n} \frac{(-1)^k x^{2k + 1}}{k! \, (2k + 1)}, then for every integer n≥0n \geq 0:

0<(−1)n+1(∫0xe−t2 dt−Tn(x))<x2n+3(n+1)! (2n+3)0 < (-1)^{n + 1} \left(\int_0^x e^{-t^2} \, \mathrm{d}t - T_n(x)\right) < \frac{x^{2n + 3}}{(n + 1)! \, (2n + 3)}

The exponential series at −t2-t^2 converges uniformly on each bounded interval, by the M-test, and can be integrated term by term; the resulting series is alternating, and the Leibniz test bounds its remainder.

Proof

  1. 0≤t≤x→∣(−1)nt2nn!∣≤x2nn!,∑n=0∞x2nn!=ex2\dato{m}{0 \leq t \leq x \rightarrow \abs{\frac{(-1)^n t^{2n}}{n!}} \leq \frac{x^{2n}}{n!}}, \qquad \sum_{n=0}^{\infty} \frac{x^{2n}}{n!} = e^{x^2}

    Let x>0x > 0. For tt in [0,x]\icc{0}{x}, the theorem on the series for the exponential, the sine and the cosine, from the lesson on the Taylor series of the elementary functions, with −t2-t^2 in place of xx, writes e−t2e^{-t^2} as the sum of the series of the functions fn(t)=(−1)nt2nn!f_n(t) = \frac{(-1)^n t^{2n}}{n!}, because (−t2)n=(−1)nt2n(-t^2)^n = (-1)^n t^{2n}. Since 0≤t≤x0 \leq t \leq x, we have t2n≤x2nt^{2n} \leq x^{2n}, and the series of the numbers x2nn!\frac{x^{2n}}{n!} converges, with sum ex2e^{x^2}, by the same theorem. The theorem on the Weierstrass M-test, from the lesson on sequences and series of functions, after shifting the index, gives uniform convergence on [0,x]\icc{0}{x}.

  2. ∫0xe−t2 dt=∑n=0∞(−1)nn!∫0xt2n dt=∑n=0∞(−1)nx2n+1n! (2n+1)\int_0^x e^{-t^2} \, \mathrm{d}t = \sum_{n=0}^{\infty} \frac{(-1)^n}{n!} \int_0^x t^{2n} \, \mathrm{d}t = \resaltar{\sum_{n=0}^{\infty} \frac{(-1)^n x^{2n + 1}}{n! \, (2n + 1)}}

    Each fnf_n is a polynomial, continuous and integrable on [0,x]\icc{0}{x}, by the theorem according to which continuous functions are integrable, from the lesson on integrable functions. The second statement of the corollary on term-by-term continuity, integration and differentiation, from the lesson on sequences and series of functions, allows the series to be integrated term by term, and the corollary on Barrow’s rule for continuous functions, with the antiderivative t2n+12n+1\frac{t^{2n + 1}}{2n + 1}, gives each integral.

  3. x<0→∫0xe−t2 dt=−∫x0e−t2 dt=∑n=0∞(−1)nx2n+1n! (2n+1)x < 0 \rightarrow \int_0^x e^{-t^2} \, \mathrm{d}t = -\int_x^0 e^{-t^2} \, \mathrm{d}t = \resaltar{\sum_{n=0}^{\infty} \frac{(-1)^n x^{2n + 1}}{n! \, (2n + 1)}}

    If x<0x < 0, the two preceding moves, on [x,0]\icc{x}{0}, where t2n≤x2nt^{2n} \leq x^{2n}, give ∫x0e−t2 dt=∑n=0∞(−1)n(0−x2n+1)n! (2n+1)\int_x^0 e^{-t^2} \, \mathrm{d}t = \sum_{n=0}^{\infty} \frac{(-1)^n (0 - x^{2n + 1})}{n! \, (2n + 1)}; by the definition of the integral with equal or reversed limits, from the lesson on the properties of the definite integral, and the proposition on the linearity and monotonicity of series, from the lesson on numerical series and their convergence, changing the sign yields the equality in the statement. If x=0x = 0, both sides equal 00.

  4. bn+1bn=x2 (2n+1)(n+1)(2n+3)<1,bn→0\frac{b_{n + 1}}{b_n} = \frac{x^2 \, (2n + 1)}{(n + 1)(2n + 3)} < 1, \qquad \resaltar{b_n \to 0}

    Let 0<x≤10 < x \leq 1 and bn=x2n+1n! (2n+1)b_n = \frac{x^{2n + 1}}{n! \, (2n + 1)}, which is positive. Since x2≤1x^2 \leq 1 and 2n+1<(n+1)(2n+3)2n + 1 < (n + 1)(2n + 3), the ratio of two consecutive terms is less than 11, and the sequence is strictly decreasing. Moreover, 0<bn≤x (x2)nn!0 < b_n \leq x \, \frac{(x^2)^n}{n!}, which tends to 00 by the lemma on powers divided by the factorial, from the lesson on the Taylor series of the elementary functions, with y=x2y = x^2; by the squeeze theorem, bn→0b_n \to 0.

  5. 0<(−1)n+1(∫0xe−t2 dt−Tn(x))<x2n+3(n+1)! (2n+3)0 < (-1)^{n + 1} \left(\int_0^x e^{-t^2} \, \mathrm{d}t - T_n(x)\right) < \resaltar{\frac{x^{2n + 3}}{(n + 1)! \, (2n + 3)}}

    Written from index 11, with am=bm−1a_m = b_{m - 1}, the series is ∑m=1∞(−1)m+1am\sum_{m=1}^{\infty} (-1)^{m + 1} a_m; its partial sum up to index n+1n + 1 is Tn(x)T_n(x), and its remainder of index n+1n + 1 is the difference between the integral and Tn(x)T_n(x). The theorem on the Leibniz test, from the lesson on alternating series and conditional convergence, with the strict inequalities, gives the sign and the bound of the remainder, an+2=bn+1a_{n + 2} = b_{n + 1}.

With x=1x = 1, the bound with n=7n = 7 equals 18!⋅17=1685440\frac{1}{8! \cdot 17} = \frac{1}{685440}, which exceeds 10−610^{-6}, and with n=8n = 8 it equals 19!⋅19=16894720\frac{1}{9! \cdot 19} = \frac{1}{6894720}, which is less than 10−610^{-6}: nine terms give T8(1)=0.7468242…T_8(1) = 0.7468242\ldots, and since (−1)9=−1(-1)^9 = -1, the integral is less than T8(1)T_8(1) and greater than T8(1)−16894720T_8(1) - \frac{1}{6894720}; its value is 0.7468241…0.7468241\ldots The function e−x2e^{-x^2} is even, and the series reflects this: it has only odd powers of xx, and its sum is an odd function. If x>1x > 1, the first terms may grow before they decrease, and the remainder bound holds only from some index on, as the solved problem “The bell curve far from the origin” shows.