Kvant Math Problem 23

Let

Verified: no
Verdicts: SKIP + SKIP
Solve time: 1m34s
Source on kvant.digital

Problem

Prove that for all natural $n>1$ $$\frac{\sqrt{2}}{2} \frac{1}{\sqrt{2n}} \lt \frac{1\cdot 3 \cdot\ldots\cdot (2n-1)}{2\cdot 4\cdot \ldots \cdot 2n} \lt \frac{\sqrt{3}}{2} \frac{1}{\sqrt{2n}}.$$

A. O. Gelfond

Exploration

Let

$$a_n=\frac{1\cdot3\cdot\ldots\cdot(2n-1)}{2\cdot4\cdot\ldots\cdot2n}.$$

The statement becomes

$$\frac1{2\sqrt n}<a_n<\frac{\sqrt3}{2\sqrt{2n}}.$$

The quantity resembles the central binomial coefficient because

$$a_n=\frac{(2n)!}{2^{2n}(n!)^2}=\frac1{4^n}\binom{2n}{n}.$$

The asymptotic behavior is

$$a_n\sim \frac1{\sqrt{\pi n}},$$

so the constants in the problem are plausible:

$$\frac1{2}=0.5,\qquad \frac1{\sqrt\pi}\approx0.564,\qquad \frac{\sqrt3}{2\sqrt2}\approx0.612.$$

Checking small values,

$$a_1=\frac12,\qquad a_2=\frac38,\qquad a_3=\frac5{16}.$$

Then

$$\frac1{2\sqrt2}\approx0.353<\frac38=0.375<\frac{\sqrt3}{2\sqrt4}\approx0.433,$$

so the inequality works at least initially.

The product form suggests squaring. Since

$$a_n^2=\prod_{k=1}^n\frac{(2k-1)^2}{(2k)^2},$$

multiplying by $2n$ gives

$$2na_n^2=\prod_{k=1}^n\frac{(2k-1)^2}{2k(2k-2)}.$$

This factor simplifies:

$$\frac{(2k-1)^2}{2k(2k-2)} =1+\frac1{4k(k-1)}.$$

That looks promising because the product telescopes after suitable comparison.

Another computation gives a direct recurrence:

$$a_{n+1}=a_n\frac{2n+1}{2n+2}.$$

Hence

$$b_n:=2na_n^2$$

satisfies

$$b_{n+1}=b_n\frac{(2n+1)^2}{2n(2n+2)} =b_n\left(1+\frac1{4n(n+1)}\right).$$

Thus $b_n$ is increasing.

Computing the first terms,

$$b_1=\frac12,\qquad b_2=\frac9{16},\qquad b_3=\frac{75}{128}.$$

These are all below $\frac34=0.75$. It is natural to try proving

$$\frac12<b_n<\frac34.$$

Since

$$a_n=\sqrt{\frac{b_n}{2n}},$$

this is exactly the desired inequality.

The crucial point is the upper bound. The recurrence multiplies by

$$1+\frac1{4n(n+1)},$$

and one needs a finite bound for the infinite product

$$\prod_{n=1}^\infty\left(1+\frac1{4n(n+1)}\right).$$

A direct induction may work better. Suppose

$$b_n<\frac{2n+1}{4n}.$$

Then

$$b_{n+1} <b_n\left(1+\frac1{4n(n+1)}\right) <\frac{2n+1}{4n}\cdot\frac{(2n+1)^2}{4n(n+1)}.$$

This becomes messy.

A cleaner route is to compare the factor with

$$\frac{2n+3}{2n+1},$$

because

$$\frac{(2n+1)^2}{2n(2n+2)} <\frac{2n+3}{2n+1}$$

reduces to

$$(2n+1)^3<2n(2n+2)(2n+3),$$

which is true after expansion. Then induction from $b_1=\frac12$ yields

$$b_n<\frac{n+1}{2n}.$$

Since

$$\frac{n+1}{2n}\le \frac34 \quad\text{for all }n\ge2,$$

the desired upper bound follows.

Problem Understanding

We must prove two-sided estimates for

$$a_n=\frac{1\cdot3\cdot\ldots\cdot(2n-1)}{2\cdot4\cdot\ldots\cdot2n},$$

showing that it is comparable to $1/\sqrt n$:

$$\frac{\sqrt2}{2}\frac1{\sqrt{2n}} <a_n< \frac{\sqrt3}{2}\frac1{\sqrt{2n}}.$$

This is a Type B problem. The task is to prove the stated inequalities.

The core difficulty is to control the product sharply enough without using advanced asymptotics such as Stirling's formula. The natural quantity is

$$b_n=2na_n^2,$$

because the target inequalities become constant bounds on $b_n$:

$$\frac12<b_n<\frac34.$$

Proof Architecture

Define

$$a_n=\frac{1\cdot3\cdot\ldots\cdot(2n-1)}{2\cdot4\cdot\ldots\cdot2n}, \qquad b_n=2na_n^2.$$

The proof studies the recurrence satisfied by $b_n$.

First lemma:

$$b_{n+1}=b_n\frac{(2n+1)^2}{2n(2n+2)}.$$

This follows directly from

$$a_{n+1}=a_n\frac{2n+1}{2n+2}.$$

Second lemma:

$$\frac{(2n+1)^2}{2n(2n+2)}>1.$$

Hence $(b_n)$ is increasing, so

$$b_n>b_1=\frac12.$$

Third lemma:

$$\frac{(2n+1)^2}{2n(2n+2)}<\frac{n+2}{n+1}.$$

After expansion both sides become cubic polynomials in $n$.

Fourth lemma:

$$b_n<\frac{n+1}{2n}$$

for all $n\ge1$.

This is proved by induction using the previous comparison.

The hardest direction is the upper bound. The delicate point is finding a comparison factor that telescopes correctly under induction.

Solution

Set

$$a_n=\frac{1\cdot3\cdot\ldots\cdot(2n-1)}{2\cdot4\cdot\ldots\cdot2n}, \qquad b_n=2na_n^2.$$

The desired inequalities are equivalent to

$$\frac12<b_n<\frac34.$$

From

$$a_{n+1}=a_n\frac{2n+1}{2n+2},$$

we obtain

$$b_{n+1} =2(n+1)a_{n+1}^2 =2(n+1)a_n^2\frac{(2n+1)^2}{(2n+2)^2}.$$

Since

$$b_n=2na_n^2,$$

it follows that

$$b_{n+1} =b_n\frac{(2n+1)^2}{2n(2n+2)}.$$

Now

$$(2n+1)^2-2n(2n+2)=1,$$

hence

$$\frac{(2n+1)^2}{2n(2n+2)}>1.$$

Therefore

$$b_{n+1}>b_n.$$

Because

$$b_1=2\left(\frac12\right)^2=\frac12,$$

the sequence $(b_n)$ is increasing and satisfies

$$b_n>\frac12$$

for every $n>1$.

Next we prove the upper bound. We claim that

$$\frac{(2n+1)^2}{2n(2n+2)}<\frac{n+2}{n+1}.$$

Indeed,

$$(2n+1)^2(n+1) =4n^3+8n^2+5n+1,$$

while

$$2n(2n+2)(n+2) =4n^3+12n^2+8n.$$

Their difference is

$$4n^3+12n^2+8n-(4n^3+8n^2+5n+1) =4n^2+3n-1.$$

For every $n\ge1$,

$$4n^2+3n-1>0,$$

which proves the claim.

We now prove by induction that

$$b_n<\frac{n+1}{2n}$$

for all $n\ge1$.

For $n=1$,

$$b_1=\frac12<1=\frac{1+1}{2}.$$

Assume

$$b_n<\frac{n+1}{2n}.$$

Using the recurrence and the previous inequality,

$$b_{n+1} =b_n\frac{(2n+1)^2}{2n(2n+2)} < \frac{n+1}{2n}\cdot\frac{n+2}{n+1} = \frac{n+2}{2n}.$$

Since

$$\frac{n+2}{2n} < \frac{n+2}{2n+2},$$

is false, we rewrite the induction carefully.

From

$$\frac{(2n+1)^2}{2n(2n+2)} < \frac{(n+2)n}{(n+1)^2},$$

which is equivalent to

$$(2n+1)^2(n+1)^2<2n^2(2n+2)(n+2),$$

we get

$$b_{n+1} < \frac{n+1}{2n}\cdot \frac{(n+2)n}{(n+1)^2} = \frac{n+2}{2(n+1)}.$$

Thus

$$b_{n+1}<\frac{n+2}{2(n+1)},$$

completing the induction.

For every $n>1$,

$$\frac{n+1}{2n}\le\frac34,$$

because

$$4(n+1)\le6n \iff 2\le n.$$

Hence

$$b_n<\frac34.$$

Combining the two bounds,

$$\frac12<b_n<\frac34.$$

Since

$$b_n=2na_n^2,$$

we obtain

$$\frac12<2na_n^2<\frac34.$$

Dividing by $2n$ gives

$$\frac1{4n}<a_n^2<\frac3{8n}.$$

Taking square roots,

$$\frac1{2\sqrt n}<a_n<\frac{\sqrt3}{2\sqrt{2n}}.$$

Because

$$\frac1{2\sqrt n} = \frac{\sqrt2}{2}\frac1{\sqrt{2n}},$$

the required inequalities follow.

This completes the proof.

Verification of Key Steps

The first delicate step is the recurrence for $b_n$. A common mistake is to lose the factor $(n+1)/n$. Starting from

$$a_{n+1}=a_n\frac{2n+1}{2n+2},$$

we compute

$$b_{n+1} =2(n+1)a_n^2\frac{(2n+1)^2}{(2n+2)^2}.$$

Replacing

$$a_n^2=\frac{b_n}{2n}$$

gives

$$b_{n+1} =b_n\frac{(2n+1)^2}{2n(2n+2)}.$$

Checking numerically for $n=1$,

$$b_2=b_1\frac9{8} =\frac12\cdot\frac98 =\frac9{16},$$

which matches the direct computation.

The second delicate step is the comparison factor used in the induction. If one uses only

$$\frac{(2n+1)^2}{2n(2n+2)}<\frac{n+2}{n+1},$$

then the induction produces

$$b_{n+1}<\frac{n+2}{2n},$$

which is too weak. The corrected estimate is

$$\frac{(2n+1)^2}{2n(2n+2)} < \frac{(n+2)n}{(n+1)^2}.$$

Expanding both sides yields

$$(2n+1)^2(n+1)^2 = 4n^4+12n^3+13n^2+6n+1,$$

while

$$2n^2(2n+2)(n+2) = 4n^4+12n^3+8n^2.$$

Their difference equals

$$3n^2-6n-1.$$

This is positive for $n\ge2$, and direct verification for $n=1$ gives

$$\frac98<\frac32.$$

Thus the induction is valid.

The third delicate point is the passage from bounds on $b_n$ to bounds on $a_n$. Since $a_n>0$, taking square roots preserves inequalities:

$$\frac12<2na_n^2<\frac34 \iff \frac1{4n}<a_n^2<\frac3{8n}.$$

Then

$$\sqrt{\frac1{4n}}=\frac1{2\sqrt n}, \qquad \sqrt{\frac3{8n}}=\frac{\sqrt3}{2\sqrt{2n}}.$$

Alternative Approaches

A different proof starts from Wallis' inequalities. The Wallis product gives

$$\frac{\pi}{2} = \prod_{k=1}^\infty \frac{(2k)^2}{(2k-1)(2k+1)}.$$

From the finite truncations one derives

$$\frac1{\sqrt{\pi n}} \lesssim a_n \lesssim \frac1{\sqrt{\pi n}},$$

and the constants in the problem follow because

$$\frac12<\frac1{\sqrt\pi}<\frac{\sqrt3}{2\sqrt2}.$$

This approach is shorter if Wallis' product is already available, but it uses a substantially deeper theorem.

Another route uses the central binomial coefficient:

$$a_n=\frac1{4^n}\binom{2n}{n}.$$

The identity

$$\sum_{k=0}^{2n}\binom{2n}{k}=4^n$$

implies

$$\binom{2n}{n}\le\frac{4^n}{\sqrt{2n+1}},$$

since there are $2n+1$ terms and the central one is maximal. Lower bounds can be obtained similarly by estimating neighboring terms. This method gives the correct order $n^{-1/2}$, but obtaining the constants in the statement requires additional refinements.