5

This is a question from a mock exam of national engineering test in my country.

Firstly, we can't take limit inside the integral as the limits of integral are not independent.

I also tried applying the property: $$\int_{a}^{b}f(x)dx=\int_{a}^{b}f(a+b-x)dx$$ but that also doesn't give anything significant which can help in simplifying the problem further.

Leibniz integral rule also doesn't seem to give anything useful. I would be grateful if I could get more ideas from the community on how to approach this.

Aayush
  • 431
  • Is the given answer $1.5$? – DatBoi Jul 29 '22 at 16:29
  • 2
    $\displaystyle\lim_{k\to \infty} \left(1 - \frac xk\right)^k = e^{-x}$ does it help? –  Jul 29 '22 at 16:30
  • @DatBoi Yeah, its 1.5. – Aayush Jul 29 '22 at 16:31
  • Yes this would be my idea. Take $k$ big enough that the difference between $(1 - \frac{x}{k})^k$ and $e^{-x}$ is small. The problem is that I don't now a good estimate for how small that difference is – Charles Hudgins Jul 29 '22 at 16:31
  • @HelpMeToUnderstandContours As I stated too, I don't think sending limit inside the integral is right ? as the upper limit of the integral is also dependent on the limit and not constant. Isn't limits of integration being constant, a necessary condition to send limit inside integral? – Aayush Jul 29 '22 at 16:32
  • 2
    You wouldn't be sending the limit inside, you would be bounding the difference between the two expressions to obtain an error term that vanishes in the limit. – Charles Hudgins Jul 29 '22 at 16:34
  • 1
    For $x\in [0,k]$, you can show that $\left(1-\frac{x}{k}\right)^k\leq e^{-x}$ and also $\left(1-\frac{x}{k}\right)^k\to e^{-x}$, so you can only compute: $$\int_{0}^{k}e^{-\frac23 x}=\frac32$$ – Matteo Jul 29 '22 at 16:49
  • @Aayush check this out: https://math.stackexchange.com/questions/71447/large-n-asymptotic-of-int-0-infty-left-1-x-n-rightn-1-exp-x – insipidintegrator Jul 29 '22 at 20:47
  • 1
    Assuming that you mean JEE when you say "national engineering test", you will always be safe to assume the fact that limit and integration operations are interchangeable in all cases (The syllabus doesn't include any reference of non-converging functions, so you can make do without any knowledge of it). Now, the whole idea of interchanging the operations, makes the question solvable as it a very elementary limit that you have to solve. The community has done very well in providing the conditions to carry out this interchanging, though- so make sure to check that out. – Arsenic Jul 30 '22 at 15:42
  • @Aayush I can say the first part to be true quite confidently, as I have come across this form of questions in many mocks myself. So, for now you can very well take it for granted and interchange the operations. – Arsenic Jul 30 '22 at 15:43

3 Answers3

5

Actually you can exchange $\lim$ and $\int$.

The sequence of non-negative functions

$$ g_k(x) = \left\{\begin{array}{rcl}\left(1-\frac{x}{k}\right)^k&\text{if}&0\leq x\leq k\\0&\text{if}&x\geq k\end{array}\right. $$ converges monotonically to $e^{-x}$. By the dominated/monotone convergence theorem it follows that

$$ \lim_{k\to +\infty}\int_{0}^{+\infty} g_k(x) e^{x/3}\,dx =\int_{0}^{+\infty}e^{-2x/3}\,dx=\color{red}{\frac{3}{2}}.$$


Provided that $g_k(x)\leq e^{-x}$, we can also study how fast the sequence converges to $\frac{3}{2}$. Indeed

$$ \int_{0}^{+\infty}(e^{-x}-g_k(x))e^{x/3}\,dx = \int_{0}^{k}\left((e^{-x/k})^k-\left(1-\frac{x}{k}\right)^k\right)e^{x/3}\,dx + \int_{k}^{+\infty}e^{-2x/3}\,dx $$ where the last integral in the RHS equals $\frac{3}{2}e^{-2k/3}$ and the first one is $$ I_k=k\int_{0}^{1}\left(\left(e^{-x}\right)^k-(1-x)^k\right) e^{kx/3}\,dx. $$ Over $[0,1]$ we have $e^{-x}-(1-x)\leq \frac{1}{2}x^2$, and $a\geq b\geq 0$ implies $$ a^n-b^n = (a-b)(a^{n-1}+\ldots+b^{n-1}) \leq n(a-b)a^{n-1}, $$ so $I_k\geq 0$ is bounded by

$$ k^2\int_{0}^{1}\frac{x^2 e^x}{2}e^{-kx} e^{kx/3}\,dx\leq \frac{ek^2}{2}\int_{0}^{1}x^2 e^{kx/3-kx}\,dx\leq\frac{ek^2}{2}\int_{0}^{+\infty}x^2 e^{-2kx/3}\,dx $$ i.e. $$ I_k \leq \frac{ek^2}{2}\cdot\frac{27}{4k^3}=O\left(\frac{1}{k}\right). $$

Jack D'Aurizio
  • 353,855
4

For $x\le k$, Bernoulli's Inequality yields $$ \begin{align} \frac{\left(1-\frac x{k+1}\right)^{k+1}}{\left(1-\frac xk\right)^k} &=\left(1-\frac xk\right)\left(\frac{k+1-x}{k-x}\frac{k}{k+1}\right)^{k+1}\\ &=\left(1-\frac xk\right)\left(1+\frac{x}{(k-x)(k+1)}\right)^{k+1}\\[3pt] &\ge\left(1-\frac xk\right)\left(1+\frac{x}{k-x}\right)\\[6pt] &=1 \end{align} $$ That is, $\left(1-\frac xk\right)^k$ is increasing. Thus, using either monotone convergence or dominated convergence (with $e^{-2x/3}$ as the dominating function) gives $$ \begin{align} \lim_{k\to\infty}\int_0^k\left(1-\frac xk\right)^ke^{x/3}\,\mathrm{d}x &=\lim_{k\to\infty}\int_0^\infty\overbrace{[0\le x\le k]\left(1-\frac xk\right)^k}^\text{monotonically increases to $e^{-x}$}e^{x/3}\,\mathrm{d}x\\ &=\int_0^\infty e^{-2x/3}\,\mathrm{d}x\\ &=\frac32 \end{align} $$ where $[\cdots]$ are Iverson brackets.

robjohn
  • 345,667
0

Tl;dr: I tried it by “elementary methods”, it didn’t work out (at least till now) I kinda did it, though not convincing.


Applying the property you mentioned, we get $$I=\int_0^k\left(1-\frac xk\right)^k\cdot e^{\tfrac x3}dx= \int_0^k\left(\frac xk\right)^k\cdot e^{\tfrac {k-x}{3}}dx $$$$= \dfrac{e^{\tfrac k3}}{k^k}\int_0^k\left(x\right)^k\cdot e^{-\tfrac x3}dx$$ Let $-\dfrac x3=u$. Then $dx=-3du$ so that we have $$-\frac13I= \dfrac{(-3)^ke^{\tfrac k3}}{k^k}\int_{-\frac k3}^0\left(u\right)^k\cdot e^udu$$

Now I believe it would do you good to remember that $$\displaystyle \int e^u\cdot p(u) du=e^u\sum _{n=0}^k(-1)^np^{(n)}(u)+C$$ where $p(u)$ is a polynomial function and $p^{(n)}(u)$ is the $n$th derivative of $p(u)$, with $p^{(0)}(u)=p(u)$. You can actually prove this by repeated use of Integration by Parts.

Now use another formula: the $n$th derivative of $p(u)=u^k, n<k $ is $$p^{(n)}(u)=\frac{k!}{(k-n)!}u^{k-n}$$ so we get a huge expression: $$-\frac13 I= \dfrac{(-3)^ke^{\tfrac k3}}{k^k} e^u\sum _{n=0}^k(-1)^n \frac{k!}{(k-n)!}u^{k-n}\Bigg|_{-\tfrac k3}^0$$$$= \dfrac{(-3)^ke^{\tfrac k3}}{k^k} \left((-1)^kk!-e^{-\tfrac k3}\sum _{n=0}^k(-1)^n \frac{k!}{(k-n)!}\left(\frac{-k}{3}\right)^{k-n}\right) $$$$= \dfrac{3^kk!e^{\tfrac k3}}{k^k} - \dfrac{(-3)^k}{k^k} \sum _{n=0}^k(-1)^n \frac{k!}{(k-n)!}\left(\frac{-k}{3}\right)^{k-n} $$$$ =\dfrac{(3)^kk!e^{\tfrac k3}}{k^k} - \sum _{n=0}^k(-1)^n \frac{k!}{(k-n)!}\left(\frac{-k}{3}\right)^{-n} $$$$=\dfrac{3^kk!e^{\tfrac k3}}{k^k} - \sum _{n=0}^kn!\ ^kC_n \left(\frac{3}{k}\right)^{n} $$

Therefore $$I=3\left(\sum _{n=0}^kn!\ ^kC_n \left(\frac{3}{k}\right)^{n}-\dfrac{3^kk!e^{\tfrac k3}}{k^k} \right).$$ By these two posts (this and this), there exists no closed form. Also, from the comments to this question, you indeed have to evaluate the error term as k increases arbitrarily: the sum approximates the second term.

UPDATE: So $$I=3\dfrac{e^{\tfrac k3}k!}{(k/3)^k}\left(e^{-\tfrac k3}\sum _{n=0}^k\frac{(k/3)^n}{n!}-1\right).$$ Then, (special thanks to user @Ant), you can use this post and this post, you can find $$|l|=\dfrac 32.$$

I feel that this is the best one can go by using elementary methods.

  • I’ve asked a follow up question here. – insipidintegrator Jul 29 '22 at 18:12
  • You might have a look at the second part of my answer (now updated) to have explicit bounds on $\frac{3}{2}-\int_{0}^{k}\left(1-\frac{x}{k}\right)^k e^{x/3},dx.$ – Jack D'Aurizio Jul 30 '22 at 13:44
  • 1
    The second part of your answer was the only part that was actually intelligible to me :) . Also, I was hell-bent on finding an elementary solution because “ from a mock exam of national engineering test in my country” usually implies that the OP is an Indian High School student. – insipidintegrator Jul 30 '22 at 14:01