Getting closer and closer

Calculus Level 5

f ( a , n ) = r = a 1 2 a 1 k = 1 n k r n r + 1 \displaystyle \large f(a,n) = \sum_{r=a-1}^{2a-1}\displaystyle \sum_{k=1}^{n}\frac{k^{r}}{n^{r+1}}

Find f ( 1 0 7 , 1 0 7 ) f(10^{7},10^{7}) to 1 decimal place.


If you are looking for more such twisted questions, Twisted problems for JEE aspirants is for you!


The answer is 1.3.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

1 solution

Mark Hennings
Feb 3, 2017

This is a complicated piece of convergence and error analysis. We shall do this in stages. First note that ln ( 1 x ) > x -\ln(1-x) > x for all 0 < x < 1 0 < x < 1 and also that ln ( 1 x ) < x + x 2 -\ln(1-x) < x + x^2 for all 0 < x < 1 2 0 < x < \tfrac12 .


We deduce from the above that ( 1 x n ) 2 n e 2 x \big(1 - \tfrac{x}{n}\big)^{2n} \, \le \, e^{-2x} for all 0 < x < n 0 < x < n , and that ( 1 x n ) 2 n > e 2 x 2 x 2 n \big(1 - \tfrac{x}{n}\big)^{2n} \, > \, e^{-2x - \frac{2x^2}{n}} for all 0 < x < 1 2 n 0 < x < \tfrac12n . Let us define α n , k = { 1 k ( 1 k n ) 2 n 1 k n 1 0 k n A = k = 1 e 2 k k = ln ( e 2 e 2 1 ) \alpha_{n,k} \; = \; \left\{ \begin{array}{ll} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{2n} & 1 \le k \le n-1 \\ 0 & k \ge n \end{array} \right. \hspace{2cm} A \; = \; \sum_{k=1}^\infty \frac{e^{-2k}}{k} \; = \; \ln\left(\frac{e^2}{e^2-1}\right) Then we note that 0 α n , k e 2 k k k 1 , n 2 lim n α n , k = e 2 k k k 1 0 \le \alpha_{n,k} \le \tfrac{e^{-2k}}{k} \hspace{1cm} k \ge 1\,,\,n \ge 2 \hspace{2cm} \lim_{n \to \infty} \alpha_{n,k} \; =\; \tfrac{e^{-2k}}{k} \hspace{0.5cm} k \ge 1 Thus the Dominated Convergence Theorem tells us that lim n k = 1 n 1 1 k ( 1 k n ) 2 n = lim n k = 1 α n , k = k = 1 e 2 k k = A \lim_{n \to \infty} \sum_{k=1}^{n-1} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{2n} \;= \; \lim_{n \to \infty} \sum_{k=1}^\infty \alpha_{n,k} \; = \; \sum_{k=1}^\infty \frac{e^{-2k}}{k} \; = \; A Since 3 e 4 ( e 2 1 ) > 1000 3e^4(e^2-1) > 1000 we note that 0 A k = 1 n 1 1 k ( 1 k n ) 2 n = k = 1 ( e 2 k k α n , k ) k = 1 2 ( e 2 k k α n , k ) + k = 3 e 2 k k < k = 1 2 1 k [ e 2 k ( 1 k n ) 2 n ] + 1 0 3 \begin{aligned} 0 \le A - \sum_{k=1}^{n-1} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{2n} & = \sum_{k=1}^\infty \left(\frac{e^{-2k}}{k} - \alpha_{n,k}\right) \\ & \le \sum_{k=1}^2 \left(\frac{e^{-2k}}{k} - \alpha_{n,k}\right) + \sum_{k=3}^\infty \frac{e^{-2k}}{k} \\ & < \sum_{k=1}^2 \tfrac{1}{k}\big[e^{-2k} - \big(1 - \tfrac{k}{n}\big)^{2n}\big] + 10^{-3} \end{aligned} for all n 3 n \ge 3 . But then 0 A k = 1 n 1 1 k ( 1 k n ) 2 n k = 1 2 1 k e 2 k [ 1 e 2 k 2 n ] + 1 0 3 k = 1 2 e 2 k k × 2 k 2 n + 1 0 3 = 2 n k = 1 2 k e 2 k + 1 0 3 < 2 × 1 0 3 \begin{aligned} 0 \le A - \sum_{k=1}^{n-1} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{2n} & \le \sum_{k=1}^2 \tfrac{1}{k} e^{-2k}\big[1 - e^{-\frac{2k^2}{n}}\big] + 10^{-3} \\ & \le \sum_{k=1}^2 \frac{e^{-2k}}{k} \times \frac{2k^2}{n} + 10^{-3} \; = \; \frac{2}{n} \sum_{k=1}^2 ke^{-2k} + 10^{-3} \\ & < 2 \times 10^{-3} \end{aligned} using the Mean Value Theorem, provided that n > 350 n > 350 .


We now define β n , k = { 1 k ( 1 k n ) n 1 1 k n 1 0 k n B = k = 1 e k k = ln ( e e 1 ) \beta_{n,k} \; = \; \left\{ \begin{array}{ll} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{n-1} & 1 \le k \le n-1 \\ 0 & k \ge n \end{array} \right. \hspace{2cm} B \; = \; \sum_{k=1}^\infty \frac{e^{-k}}{k} \; = \; \ln\left(\frac{e}{e-1}\right) Then we note that (we use the fact that n k ( n k ) n n 1 < 2 \tfrac{n}{k(n-k)} \le \tfrac{n}{n-1} < 2 for 1 k n 1 1 \le k \le n-1 ) 0 β n , k 2 e k k k 1 , n 2 lim n β n , k = e k k k 1 0 \le \beta_{n,k} \le 2\tfrac{e^{-k}}{k} \hspace{1cm} k \ge 1\,,\,n \ge 2 \hspace{2cm} \lim_{n \to \infty} \beta_{n,k} \; =\; \tfrac{e^{-k}}{k} \hspace{0.5cm} k \ge 1 Thus the Dominated Convergence Theorem tells us that lim n k = 1 n 1 1 k ( 1 k n ) n 1 = lim n k = 1 β n , k = k = 1 e k k = B \lim_{n \to \infty} \sum_{k=1}^{n-1} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{n-1} \;= \; \lim_{n \to \infty} \sum_{k=1}^\infty \beta_{n,k} \; = \; \sum_{k=1}^\infty \frac{e^{-k}}{k} \; = \; B Since 6 e 5 ( e 1 ) > 1000 6e^5(e-1) > 1000 , we have B k = 1 n 1 1 k ( 1 k n ) n 1 k = 1 5 1 k e k ( 1 k n ) n 1 + k = 6 e k k k = 1 5 n k ( n k ) ( 1 k n ) e k ( 1 k n ) n + 1 0 3 2 n k = 1 5 k e k + 2 k = 1 5 e k ( 1 k n ) n + 1 0 3 2 n k = 1 5 k e k + 2 k = 1 5 e k 1 e k 2 n + 1 0 3 2 n k = 1 5 k e k + 2 k = 1 5 e k × k 2 n + 1 0 3 = 2 n k = 1 5 k ( k + 1 ) e k + 1 0 3 \begin{aligned} \left|B - \sum_{k=1}^{n-1} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{n-1}\right| & \le \sum_{k=1}^5 \tfrac{1}{k}\left| e^{-k} - \big(1 - \tfrac{k}{n}\big)^{n-1}\right| + \sum_{k=6}^\infty \frac{e^{-k}}{k} \\ & \le \sum_{k=1}^5 \frac{n}{k(n-k)}\left|\big(1 - \tfrac{k}{n}\big)e^{-k} - \big(1 - \tfrac{k}{n}\big)^n\right| + 10^{-3} \\ & \le \frac{2}{n}\sum_{k=1}^5 k e^{-k} + 2\sum_{k=1}^5 \left|e^{-k} - \big(1 - \tfrac{k}{n}\big)^n\right| + 10^{-3} \\ & \le \frac{2}{n}\sum_{k=1}^5 k e^{-k} + 2\sum_{k=1}^5 e^{-k}\left| 1 - e^{-\frac{k^2}{n}}\right| + 10^{-3} \\ & \le \frac{2}{n}\sum_{k=1}^5 k e^{-k} + 2\sum_{k=1}^5 e^{-k} \times \frac{k^2}{n} + 10^{-3} \\ & = \tfrac{2}{n} \sum_{k=1}^5 k(k+1)e^{-k} + 10^{-3} \end{aligned} for all n > 10 n > 10 , using the Mean Value Theorem, from which we deduce that B k = 1 n 1 1 k ( 1 k n ) n 1 2 × 1 0 3 n 5500 \left|B - \sum_{k=1}^{n-1} \tfrac{1}{k}\big(1 - \tfrac{k}{n}\big)^{n-1}\right| \; \le \; 2 \times 10^{-3} \hspace{2cm} n \ge 5500


Now for the problem itself. We have (note that the k = n k=n terms have to be handled separately) f ( a , n ) = r = a 1 2 a 1 k = 1 n k r n r + 1 = a + 1 n + r = a 1 2 a 1 k = 1 n 1 k r n r + 1 = a + 1 n + k = 1 n 1 k a 1 n a ( 1 ( k n ) a + 1 ) 1 k n = a + 1 n + k = 1 n 1 1 n k ( ( k n ) a 1 ( k n ) 2 a ) \begin{aligned} f(a,n) & = \sum_{r=a-1}^{2a-1} \sum_{k=1}^n \frac{k^r}{n^{r+1}} \; = \; \frac{a+1}{n} + \sum_{r=a-1}^{2a-1} \sum_{k=1}^{n-1} \frac{k^r}{n^{r+1}} \\ & = \frac{a+1}{n} + \sum_{k=1}^{n-1} \frac{\frac{k^{a-1}}{n^a}\left(1 - \big(\frac{k}{n}\big)^{a+1}\right)}{1 - \frac{k}{n}} \\ & = \frac{a+1}{n} + \sum_{k=1}^{n-1}\frac{1}{n-k}\left(\big(\tfrac{k}{n}\big)^{a-1} - \big(\tfrac{k}{n}\big)^{2a}\right) \end{aligned} and hence f ( n , n ) = n + 1 n + k = 1 n 1 1 n k ( ( k n ) n 1 ( k n ) 2 n ) = n + 1 n + k = 1 n 1 1 k [ ( 1 k n ) n 1 ( 1 k n ) 2 n ] = n + 1 n + k = 1 n 1 1 k ( 1 k n ) n 1 k = 1 n 1 1 k ( 1 k n ) 2 n \begin{aligned} f(n,n) & = \frac{n+1}{n} + \sum_{k=1}^{n-1}\frac{1}{n-k}\left(\big(\tfrac{k}{n}\big)^{n-1} - \big(\tfrac{k}{n}\big)^{2n}\right) \\ & = \frac{n+1}{n} + \sum_{k=1}^{n-1} \frac{1}{k}\left[\big(1 - \tfrac{k}{n}\big)^{n-1} - \big(1 - \tfrac{k}{n}\big)^{2n}\right] \\ & = \frac{n+1}{n} + \sum_{k=1}^{n-1} \frac{1}{k}\big(1 - \tfrac{k}{n}\big)^{n-1} - \sum_{k=1}^{n-1} \frac{1}{k} \big(1 - \tfrac{k}{n}\big)^{2n} \end{aligned} and hence we have shown above that lim n f ( n , n ) = 1 + B A = ln ( 1 + e ) = 1.313261688 \lim_{n \to \infty} f(n,n) \; = \; 1 + B - A \; = \; \ln(1+e) \; = \; 1.313261688 Moreover f ( n , n ) f(n,n) is within 5 × 1 0 3 5 \times 10^{-3} of this limit if n > 5500 n > 5500 . Certainly, then f ( 1 0 7 , 1 0 7 ) f(10^7,10^7) is equal to 1.3 \boxed{1.3} to 1 1 decimal place. Mathematica evaluates f ( 1 0 7 , 1 0 7 ) f(10^7,10^7) as 1.31326 1.31326 to 5 5 decimal places.

Um I don't know how to respond to this really. I'm stunned by your thorough analysis despite it has gaps that my knowledge fails to help me fill. Utterly speechless..

Anirudh Chandramouli - 4 years, 4 months ago

@Rohith M.Athreya How this can be a JEE question ???

Kushal Bose - 4 years, 4 months ago

Log in to reply

By sheer rotten luck :P

Rohith M.Athreya - 4 years, 4 months ago

If u read what I have named the set,it says problems FOR jee aspirants and not jee type problems

Rohith M.Athreya - 4 years, 3 months ago

yes i did like too considered n as infinity :) (but not with this much perfection :P)

A Former Brilliant Member - 4 years, 3 months ago

i was too overconfident. ur=k^r/n^(r+1) i thought u1+u2+......+un can be approximated by integral of x^r from 0 to 1. the integral is 1/(r+1). now we get 1/a +1/a+1+...........1/2a this i approximated by integral of ln(x+1) from 0 to 1. i got a horribly wrong answer. your approach is nice. just i am not understanding why i ended with such a horrible answer. at least my wrong answer should have been closer to the actual answer.

Srikanth Tupurani - 1 year, 10 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...