10
$\begingroup$

The integrand seems extremely easy:

$$I_n=\int_0^1\exp(x^n)dx$$

I want to determine the asymptotic behavior of $I_n$ as $n\to\infty$. It's not hard to show that $\lim_{n\to\infty}I_n=1$ follows from Lebesgue's monotone convergence theorem. However, when I try to obtain more precise results, I confront difficulties. Since it's not of the canonical form of Laplace's method, I have no powerful tools to estimate $I_n$.

Is there any good approach for that? I will be pleased if we can obtain the asymptotic expansion of $I_n$. Thanks!

$\endgroup$
2
  • $\begingroup$ The existing answers are better, but note that $e^{x^n}$ is concave and so lies above its tangent line (and obviously exceeds $1$ as well). Note that the slope of the tangent line to $y=e^{x^n}$ at $x=1$ equals $ne$. Therefore $I_n$ is larger than the integral of the piecewise linear function going through the three points $(0,1)$, $(1-1/ne,1)$, and $(1,e)$, which is $1+1/2en$. A similar argument involving secant lines would probably yield an upper bound of the same order of magnitude. $\endgroup$ Apr 29, 2013 at 6:59
  • $\begingroup$ The integral seems easy, and the behavior of the expansion is expected, but the coefficients are totally not trivial. This turned out to be a very cool problem - thanks for posting! $\endgroup$
    – Ron Gordon
    Apr 29, 2013 at 13:42

3 Answers 3

13
$\begingroup$

You can expand the exponential in a Taylor series quite accurately:

$$\exp{\left ( x^n \right )} = 1 + x^n + \frac12 x^{2 n} + \ldots$$

Because $x \in [0,1]$, this series converges rapidly as $n \to \infty$.

Then the integral is

$$1 + \frac{1}{n+1} + \frac12 \frac{1}{2 n+1} + \ldots = \sum_{k=0}^{\infty}\frac{1}{k!} \frac{1}{k n+1}$$

We can rewrite this as

$$\begin{align}I_n&=1+\frac{1}{n} \sum_{k=1}^{\infty} \frac{1}{k \cdot k!} \left ( 1+\frac{1}{k n} \right )^{-1}\\ &= 1+\frac{1}{n} \sum_{m=0}^{\infty} \frac{(-1)^m}{n^m} \: \sum_{k=1}^{\infty} \frac{1}{k^{m+1} k!}\\ &= 1+\sum_{m=1}^{\infty} (-1)^{m+1}\frac{K_m}{n^m} \end{align}$$

where

$$K_m = \sum_{k=1}^{\infty} \frac{1}{k^{m} k!}$$

To first order in $n$:

$$I_n \sim 1+\frac{K_1}{n} \quad (n \to \infty)$$

where

$$K_1 = \sum_{k=1}^{\infty} \frac{1}{k\, k!} = \text{Ei}(1) - \gamma \approx 1.3179$$

This checks out numerically in Mathematica.

BONUS

As a further check, I computed the following asymptotic approximation:

$$g(n) = 1+\frac{K_1}{n} -\frac{K_2}{n^2} $$

where

$$K_2 = \sum_{k=1}^{\infty} \frac{1}{k^2 k!} \approx 1.1465$$

I computed

$$\log_2{\left[\frac{\left|g\left(2^m\right)-I_{2^m}\right|}{I_{2^m}}\right]}$$

for $m \in \{1,2,\ldots,9\}$ The results are as follows

$$\left( \begin{array}{cc} 1 & -4.01731 \\ 2 & -6.56064 \\ 3 & -9.26741 \\ 4 & -12.0963 \\ 5 & -15.0028 \\ 6 & -17.9538 \\ 7 & -20.9287 \\ 8 & -23.916 \\ 9 & -26.9096 \\ \end{array} \right)$$

Note that the difference between successive elements is about $-3$; because this is a log-log table, that means that this error is $O(1/n^3)$ and that the approximation is correct.

$\endgroup$
4
  • $\begingroup$ Very good, thanks! $\endgroup$
    – Yai0Phah
    Apr 29, 2013 at 6:59
  • $\begingroup$ Since $\exp(x^n)=\sum_0^\infty x^{nk}/k!$ converges absolutely on $[0,1]$, the interchange of summation and integration is justified. The interchange of summation is valid for the absolute convergence of the double sum. Am I right? $\endgroup$
    – Yai0Phah
    Apr 29, 2013 at 7:18
  • $\begingroup$ @FrankScience: absolutely correct. $\endgroup$
    – Ron Gordon
    Apr 29, 2013 at 7:20
  • $\begingroup$ @Ron Gordon The first few terms are numerically $\left\{\frac{1.3179}{n},-\frac{1.1465}{n^2},\frac{1.0694}{n^3},-\frac{1.03348}{n^4},\frac{1.01635}{n^5}\right\}$ Neglecting the difference from unity in the numerator and summing up the asmptotic series the asymptotic behaviour of the integral can be written in this approximation as $\frac{n+2}{n+1}$ $\endgroup$ Nov 7, 2019 at 10:05
8
$\begingroup$

Let $n \geq 1$. A first crude estimate can be obtained as follows. Substitute $x \leftarrow x^{1/n}$ to get

$$ I_n = \frac{1}{n}\int_0^1 x^{\frac{1}{n} - 1} e^x dx. $$

Now we can estimate $e^x$ on the interval $[0,1]$ to get

$$ \frac{1}{n}\int_0^1 x^{\frac{1}{n}-1}(1+x)\, dx < I_n < \frac{1}{n}\int_0^1 x^{\frac{1}{n} - 1} (1 + (e-1)x)\, dx $$ or $$ 1 + \frac{1}{n+1} < I_n < 1 + \frac{e-1}{n+1}. $$

$\endgroup$
1
  • $\begingroup$ -- very nice! -- $\endgroup$ Apr 29, 2013 at 7:00
2
$\begingroup$

A related technique. Here is a start. Integrating by parts gives, $$I= {{\rm e}}-{\frac {{{\rm e}}\,n}{1+n}}+{\frac {{{\rm e}}\,{n}^{ 2}}{2\,{n}^{2}+3\,n+1}}-{\frac {{{\rm e}}\,{n}^{3}}{6\,{n}^{3}+11\,{ n}^{2}+6\,n+1}}$$

$$+{\frac{{{\rm e}}\,{n}^{4}}{24\,{n}^{4}+50\,{n}^{3}+ 35\,{n}^{2}+10\,n+1}}-\int _{0}^{1}\!{\frac {{n}^{5}{{\rm e}^{{ x}^{n}}}{x}^{5\,n+1}}{x \left( 4\,n+1 \right) \left( 3\,n+1 \right)\left( 1+2\,n \right) \left( 1+n \right) }}{dx}.$$

From the above, we can see that as $n\to \infty$, the integral approaches $1$

$$ I = \rm e( 1 - 1 + \frac{1}{2} - \frac{1}{3!}+\frac{1}{4!} - \dots)=\rm e \rm e^{-1}=1. $$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.