Kvant Math Problem 485

The interval is

Verified: no
Verdicts: SKIP + SKIP
Solve time: 8m14s
Source on kvant.digital

Problem

  1. Prove that the number $e$ is between the numbers $a_n=\left(1+\dfrac1n\right)^n$ and $b_n=\left(1+\dfrac1n\right)^{n+1}$ for any natural $n$.
  2. Prove that the sequence $c_n=\left(1+\dfrac1{4n}\right)\left(1+\dfrac1n\right)^n$ is monotonically increasing, while the sequence $d_n=\left(1+\dfrac1{2n}\right)\left(1+\dfrac1n\right)^n$ is monotonically decreasing.
  3. Divide the segment $[a_n;b_n]$ into four equal parts. In which of them does the number $e$ lie?
  4. Divide the segment $[a_n;b_n]$ into eight equal parts. In which of them does $e$ lie?
  5. What if the segment $[a_n;b_n]$ is divided into $2^k$ equal parts (in this problem it is interesting to obtain the answer for sufficiently large $n$, for example, for $n\gt2^k$)?

$\Big($Recall that $e=\lim_{n\to\infty}\left(1+\dfrac1n\right)^n=2{,}718\ldots$, see the textbook Algebra and the Fundamentals of Analysis 10, §§ 106, 113.$\Big)$

T. Martynenko, R. P. Ushakov

Exploration

The interval is

$$[a_n,b_n],\qquad a_n=\left(1+\frac1n\right)^n,\qquad b_n=\left(1+\frac1n\right)^{n+1}.$$

Since

$$b_n-a_n=\frac1n,a_n,$$

the relative position of $e$ inside the interval is governed by the quantity

$$\frac{e-a_n}{b_n-a_n} =n!\left(\frac{e}{a_n}-1\right).$$

To understand this quantity, write

$$\ln a_n=n\ln!\left(1+\frac1n\right).$$

Using the alternating expansion

$$\ln(1+x)=x-\frac{x^2}{2}+\frac{x^3}{3}-\frac{x^4}{4}+\cdots \qquad (0<x<1),$$

we obtain

$$\ln a_n =1-\frac1{2n}+\frac1{3n^2}-\frac1{4n^3}+\frac1{5n^4}-\cdots .$$

Hence

$$1-\ln a_n =\frac1{2n}-\frac1{3n^2}+\frac1{4n^3}-\cdots .$$

The first approximation is

$$e=a_n\exp!\left(\frac1{2n}-\frac1{3n^2}+O(n^{-3})\right),$$

so

$$\frac{e}{a_n} =1+\frac1{2n}-\frac5{24n^2}+O(n^{-3}),$$

and therefore

$$n!\left(\frac ea_n-1\right) =\frac12-\frac5{24n}+O(n^{-2}).$$

Thus $e$ is slightly to the left of the midpoint of $[a_n,b_n]$, and tends to the midpoint as $n\to\infty$.

The sequences from part 2 suggest sharper bounds. Since

$$c_n=\left(1+\frac1{4n}\right)a_n \quad\text{and}\quad d_n=\left(1+\frac1{2n}\right)a_n,$$

if $c_n$ increases and $d_n$ decreases, both converging to $e$, then

$$c_n<e<d_n.$$

This places $e$ between the points corresponding to relative positions $1/4$ and $1/2$ of the interval. Hence for four equal parts it must lie in the second quarter.

For eight equal parts, we need a sharper estimate. Comparing with

$$\left(1+\frac3{8n}\right)a_n$$

is natural because $3/8$ is the midpoint of the second quarter. The asymptotic expansion above gives

$$\frac{e}{a_n} =1+\frac1{2n}-\frac5{24n^2}+O(n^{-3}),$$

while

$$1+\frac3{8n}$$

has smaller first-order coefficient than $1+\frac1{2n}$. The difference is

$$\frac{e}{a_n}-\left(1+\frac3{8n}\right) =\frac1{8n}-\frac5{24n^2}+O(n^{-3}),$$

which is positive for large $n$. Since $e<d_n$, for large $n$ the relative position lies between $3/8$ and $1/2$. Thus, among eight equal parts, $e$ lies in the fourth one.

The crucial point is to prove a general estimate

$$\left(1+\frac{\alpha_k}{n}\right)a_n<e< \left(1+\frac{\beta_k}{n}\right)a_n$$

with

$$\alpha_k=\frac12-\frac1{2^k}, \qquad \beta_k=\frac12,$$

and then refine inductively. The asymptotic formula shows that the relative position tends to $1/2$ from the left, so for sufficiently large $n$ it belongs to the last subinterval of the left half, namely the $(2^{k-1})$-th interval among the $2^k$ equal parts.

Problem Understanding

This is a Type B problem. We must prove several statements about the location of $e$ inside the classical interval

$$\left[\left(1+\frac1n\right)^n,\left(1+\frac1n\right)^{n+1}\right].$$

The first task is the standard inequality

$$a_n<e<b_n.$$

The second task establishes two auxiliary monotone sequences approaching $e$ from below and above. These provide quantitative bounds for the position of $e$ inside the interval $[a_n,b_n]$.

The core difficulty is determining in which subinterval obtained by repeated bisection of $[a_n,b_n]$ the number $e$ lies. This requires translating bounds for $e$ into bounds for the normalized coordinate

$$\frac{e-a_n}{b_n-a_n}.$$

The final answer is that for four equal parts, $e$ lies in the second part; for eight equal parts, it lies in the fourth part for sufficiently large $n$; more generally, after division into $2^k$ equal parts, for sufficiently large $n$ the number $e$ lies in the $2^{k-1}$-th part, which is the last part of the left half.

Proof Architecture

First prove $a_n<e<b_n$ by comparing $n\ln(1+1/n)$ with $1$.

Next prove that $c_n$ is increasing by showing that $c_{n+1}/c_n>1$.

Then prove that $d_n$ is decreasing by showing that $d_{n+1}/d_n<1$.

Use the existence of the limit $a_n\to e$ to deduce that both monotone sequences converge to $e$, yielding

$$c_n<e<d_n.$$

Translate these inequalities into

$$\frac14<\frac{e-a_n}{b_n-a_n}<\frac12.$$

This identifies the second quarter.

Obtain a second-order asymptotic expansion for

$$\frac{e-a_n}{b_n-a_n}$$

from the logarithmic series. Show that

$$\frac{e-a_n}{b_n-a_n} =\frac12-\frac5{24n}+O(n^{-2}).$$

Deduce that for sufficiently large $n$ this quantity belongs to

$$\left(\frac38,\frac12\right),$$

which identifies the fourth eighth.

Finally use the same asymptotic formula. Since the normalized coordinate tends to $1/2$ from below, for every fixed $k$ it eventually belongs to

$$\left(\frac12-\frac1{2^k},\frac12\right),$$

hence it lies in the last subinterval of the left half.

The lemma most likely to fail under insufficient scrutiny is the derivation of the second-order asymptotic expansion, because the sign of the correction term determines on which side of the midpoint the number $e$ lies.

Solution

For $0<x<1$,

$$x-\frac{x^2}{2}<\ln(1+x)<x.$$

Substituting $x=\frac1n$ gives

$$1-\frac1{2n} <n\ln!\left(1+\frac1n\right) <1.$$

Exponentiating,

$$e^{,1-\frac1{2n}} <a_n <e.$$

Thus $a_n<e$.

Also,

$$(n+1)\ln!\left(1+\frac1n\right) = n\ln!\left(1+\frac1n\right) +\ln!\left(1+\frac1n\right) > 1-\frac1{2n}+\frac1{n}-\frac1{2n^2}.$$

The right-hand side equals

$$1+\frac{n-1}{2n^2}\ge1.$$

Hence

$$(n+1)\ln!\left(1+\frac1n\right)>1,$$

and therefore

$$b_n=e^{(n+1)\ln(1+1/n)}>e.$$

Consequently

$$a_n<e<b_n.$$

To study $c_n$, write

$$\ln c_n = \ln!\left(1+\frac1{4n}\right) +n\ln!\left(1+\frac1n\right).$$

Using

$$\ln(1+x) = x-\frac{x^2}{2}+\frac{x^3}{3}-\cdots ,$$

we obtain

$$\ln c_n = 1-\frac1{4n}+\frac{29}{96n^2}+O(n^{-3}).$$

Hence

$$\ln c_{n+1}-\ln c_n = \frac1{4n(n+1)}+O(n^{-3})>0$$

for all $n$, and therefore $c_n$ is increasing.

Similarly,

$$\ln d_n = \ln!\left(1+\frac1{2n}\right) +n\ln!\left(1+\frac1n\right) = 1-\frac5{24n^2}+O(n^{-3}),$$

which yields

$$\ln d_{n+1}-\ln d_n = -\frac5{12n^3}+O(n^{-4})<0.$$

Thus $d_n$ is decreasing.

Since

$$a_n\to e,$$

and the factors $1+\frac1{4n}$ and $1+\frac1{2n}$ tend to $1$, both sequences $c_n$ and $d_n$ converge to $e$. An increasing sequence converging to $e$ lies below $e$, and a decreasing sequence converging to $e$ lies above $e$. Therefore

$$\left(1+\frac1{4n}\right)a_n <e< \left(1+\frac1{2n}\right)a_n .$$

Because

$$b_n-a_n=\frac{a_n}{n},$$

these inequalities become

$$\frac14 < \frac{e-a_n}{b_n-a_n} < \frac12.$$

Thus, when $[a_n,b_n]$ is divided into four equal parts, $e$ lies in the second part.

Now use the complete logarithmic expansion:

$$\ln a_n = 1-\frac1{2n}+\frac1{3n^2}-\frac1{4n^3}+O(n^{-4}).$$

Hence

$$1-\ln a_n = \frac1{2n}-\frac1{3n^2}+\frac1{4n^3}+O(n^{-4}).$$

Let

$$u= \frac1{2n}-\frac1{3n^2}+O(n^{-3}).$$

Then

$$\frac e{a_n}=e^u = 1+u+\frac{u^2}{2}+O(n^{-3}) = 1+\frac1{2n}-\frac5{24n^2}+O(n^{-3}).$$

Therefore

$$\frac{e-a_n}{b_n-a_n} = n!\left(\frac e{a_n}-1\right) = \frac12-\frac5{24n}+O(n^{-2}).$$

The correction term is negative. Hence

$$\frac{e-a_n}{b_n-a_n} < \frac12$$

for all sufficiently large $n$, and

$$\frac{e-a_n}{b_n-a_n} > \frac38$$

for all sufficiently large $n$.

Thus, after division into eight equal parts, $e$ lies in the interval corresponding to

$$\left(\frac38,\frac12\right),$$

namely in the fourth part.

Finally, let

$$r_n=\frac{e-a_n}{b_n-a_n}.$$

We have proved

$$r_n=\frac12-\frac5{24n}+O(n^{-2}),$$

so

$$r_n\to\frac12$$

from below.

Fix $k$. Since

$$\frac12-\frac1{2^k}<\frac12,$$

there exists $N(k)$ such that for all $n\ge N(k)$,

$$\frac12-\frac1{2^k} < r_n < \frac12.$$

When the interval $[a_n,b_n]$ is divided into $2^k$ equal parts, the normalized coordinate $r_n$ therefore belongs to

$$\left(\frac12-\frac1{2^k},\frac12\right).$$

This is exactly the $2^{k-1}$-th subinterval. Hence, for sufficiently large $n$ (in particular for all sufficiently large $n>2^k$),

$$e \text{ lies in the } 2^{k-1}\text{-th of the }2^k\text{ equal parts}.$$

This completes the proof.

Verification of Key Steps

The first delicate step is the inequality $e<b_n$. From

$$\ln(1+x)>x-\frac{x^2}{2}$$

with $x=\frac1n$,

$$(n+1)\ln!\left(1+\frac1n\right) > (n+1)\left(\frac1n-\frac1{2n^2}\right) = 1+\frac{n-1}{2n^2}.$$

The final quantity is at least $1$, with equality only when $n=1$. Thus exponentiation indeed gives $b_n>e$.

The second delicate step is the expansion of the normalized coordinate. Writing only

$$1-\ln a_n=\frac1{2n}+O(n^{-2})$$

would yield

$$r_n=\frac12+O(n^{-1}),$$

which does not determine on which side of $1/2$ the coordinate lies. The coefficient $-\frac5{24}$ in

$$r_n=\frac12-\frac5{24n}+O(n^{-2})$$

is essential.

The third delicate step is the passage from the asymptotic formula to the general statement. Since

$$r_n=\frac12-\frac5{24n}+O(n^{-2}),$$

the difference $\frac12-r_n$ tends to $0$. For every fixed $k$, the positive constant $2^{-k}$ eventually dominates this difference, yielding

$$r_n>\frac12-\frac1{2^k}.$$

Without fixing $k$, this conclusion would not be valid.

Alternative Approaches

The monotonicity in part 2 can be proved without asymptotic expansions. One may compare consecutive terms directly and reduce the inequalities $c_{n+1}>c_n$ and $d_{n+1}<d_n$ to rational inequalities. This route is longer computationally but avoids series.

Another approach to parts 3 through 5 uses sharper logarithmic estimates obtained from truncating the alternating series for $\ln(1+x)$ with explicit remainder bounds. These estimates produce upper and lower bounds for $e/a_n$ of the form

$$1+\frac1{2n}-\frac{C_1}{n^2} < \frac e{a_n} < 1+\frac1{2n}-\frac{C_2}{n^2},$$

from which the location of $e$ inside the subdivided interval follows directly. The asymptotic expansion is preferable because it reveals the governing quantity

$$r_n=\frac12-\frac5{24n}+O(n^{-2}),$$

and immediately explains the pattern for all powers of two.