Natural log integral inequality

261 Views Asked by At

I'm trying to show that for $j\geq 2$,

$$\ln(j)\geq \int_{j - 1/2}^{j+1/2} \ln(x) dx$$

After about 7 pages of attempts, I'm just sick of looking at this. I tried just powering through the integral and using a ton of laws of logs and what not, which all seemed futile. I've also tried exponentiating and dealing with it that way, which did not seem to work. I even tried flipping the inequality and trying to get a contradiction and still have no intuition for what might actually lead to the correct solution.

Thank you in advance for any help.

2

There are 2 best solutions below

0
On BEST ANSWER

Note: This was done independently of Jack D'Aurizio's answer. I'm submitting it because it goes into more detail.

I show that

$\ln(j)-\frac{1}{24j^2} -\frac{1}{160j^4} \lt \int_{j - 1/2}^{j+1/2} \ln(x) dx \lt \ln(j)-\frac{1}{24j^2} $.

More generally, for $m \ge 1$,

$\begin{array}\\ \int_{j - 1/2}^{j+1/2} \ln(x) dx &= \ln(j)-\sum_{n=1}^{m-1} \frac{1}{j^{2n}(2n+1)2^{2n+1}} -\frac{c}{j^{2m}(2m+1)2^{m+1}}\\ &=\ln(j) - \frac1{24 j^2} - \frac1{160 j^4}- \frac1{896 j^6}-\frac1{4608 j^8}-...\\ \end{array} $

with the numerator of the last term from $1$ to $2$.

Here is a simple proof that doesn't use convexity. Instead, it splits the integral into two halves and recombines them.

$\begin{array}\\ \int_{j - 1/2}^{j+1/2} \ln(x) dx &=\int_{j - 1/2}^{j} \ln(x) dx+\int_{j}^{j+1/2} \ln(x) dx\\ &=\int_{ -1/2}^{0} \ln(x+j) dx+\int_{0}^{1/2} \ln(x+j) dx\\ &=\int_{ 0}^{1/2} \ln(-x+j) dx+\int_{0}^{1/2} \ln(x+j) dx\\ &=\int_{ 0}^{1/2} (\ln(-x+j)+\ln(x+j)) dx\\ &=\int_{ 0}^{1/2} \ln(j^2-x^2) dx\\ &=\int_{ 0}^{1/2} (\ln(j^2)+\ln(1-(x/j)^2)) dx\\ &=\ln(j)+\int_{ 0}^{1/2} \ln(1-(x/j)^2) dx\\ &<\ln(j) \qquad\text{since } \ln(1-(x/j)^2) < 0 \text{ for } 0 < x \le 1/2\\ \end{array} $

Additionally, since $\ln(1-z) < -z$ for $0 < z < 1$,

$\begin{array}\\ \int_{ 0}^{1/2} \ln(1-(x/j)^2) dx &\lt \int_{ 0}^{1/2} -(x/j)^2 dx\\ &= -\frac{x^3}{3j^2}|_0^{1/2}\\ &= -\frac{1}{24j^2}\\ \end{array} $

Therefore $\int_{j - 1/2}^{j+1/2} \ln(x) dx \lt \ln(j)-\frac{1}{24j^2} $.

For a lower bound, for $0 < z < \frac12$,

$\begin{array}\\ -\ln(1-z) &=\sum_{n=1}^{\infty} \frac{z^n}{n}\\ &=z+\sum_{n=2}^{\infty} \frac{z^n}{n}\\ &=z+\frac12\sum_{n=2}^{\infty} \frac{z^n}{n/2}\\ &<z+\frac12\sum_{n=2}^{\infty} z^n\\ &<z+\frac{z^2}{2(1-z)}\\ &<z+z^2\\ \end{array} $

so

$\begin{array}\\ \int_{ 0}^{1/2} \ln(1-(x/j)^2) dx &\gt \int_{ 0}^{1/2} (-(x/j)^2-(x/j)^4) dx\\ &= -(\frac{x^3}{3j^2}+\frac{x^5}{5j^4})|_0^{1/2}\\ &= -\frac{1}{12j^2} -\frac{1}{160j^4}\\ \end{array} $

Therefore $\int_{j - 1/2}^{j+1/2} \ln(x) dx \gt \ln(j)-\frac{1}{12j^2} -\frac{1}{160j^4} $.

By taking more terms in the series for $\ln(1-z)$, we can get a more accurate approximation.


And here's how this is done.

$\begin{array}\\ -\ln(1-z) &=\sum_{n=1}^{\infty} \frac{z^n}{n}\\ &=\sum_{n=1}^{m-1} \frac{z^n}{n}+\sum_{n=m}^{\infty} \frac{z^n}{n}\\ &<\sum_{n=1}^{m-1} \frac{z^n}{n}+\sum_{n=m}^{\infty} \frac{z^n}{m}\\ &<\sum_{n=1}^{m-1} \frac{z^n}{n}+\frac{z^m}{m}\sum_{n=m}^{\infty} z^{n-m}\\ &<\sum_{n=1}^{m-1} \frac{z^n}{n}+\frac{z^m}{m(1-z)}\\ \text{and}\\ -\ln(1-z) &>\sum_{n=1}^{m-1} \frac{z^n}{n}+\frac{z^m}{m}\\ \text{so that}\\ -\ln(1-z) &=\sum_{n=1}^{m-1} \frac{z^n}{n}+\frac{cz^m}{m}\qquad \text{where }1 < c < 2\\ \end{array} $

Therefore, for $m \ge 1$,

$\begin{array}\\ \int_{ 0}^{1/2} \ln(1-(x/j)^2) dx &= \int_{ 0}^{1/2} (\sum_{n=1}^{m-1} \frac{(x/j)^{2n}}{n}+\frac{c(x/j)^{2m}}{m}) dx\\ &= \sum_{n=1}^{m-1} \int_{ 0}^{1/2} \frac{(x/j)^{2n}}{n}dx+\int_{ 0}^{1/2} \frac{c(x/j)^{2m}}{m} dx\\ &= \sum_{n=1}^{m-1} \frac{x^{2n+1}}{j^{2n}(2n+1)} |_{ 0}^{1/2} +\frac{cx^{2m+1}}{j^{2m}(2m+1)} |_{ 0}^{1/2}\\ &= \sum_{n=1}^{m-1} \frac{1}{j^{2n}(2n+1)2^{2n+1}} +\frac{c}{j^{2m}(2m+1)2^{m+1}}\\ \end{array} $

Therefore

$\begin{array}\\ \int_{j - 1/2}^{j+1/2} \ln(x) dx &= \ln(j)-\sum_{n=1}^{m-1} \frac{1}{j^{2n}(2n+1)2^{2n+1}} -\frac{c}{j^{2m}(2m+1)2^{m+1}}\\ &=\ln(j) - \frac1{24 j^2} - \frac1{160 j^4}- \frac1{896 j^6}-\frac1{4608 j^8}-...\\ \end{array} $

with the numerator of the last term from $1$ to $2$.

2
On

$\log(x)$ is a concave function on $\mathbb{R}^+$, hence your inequality is a straightforward consequence of the Hermite-Hadamard inequality. It is not difficult to strengthen, too. For any $j\geq 1$ we have

$$ \int_{j-1/2}^{j+1/2}\log(x)\,dx = \int_{0}^{1/2}\log\left(j^2-x^2\right)\,dx = \log(j) +\underbrace{\int_{0}^{1/2}\log\left(1-\frac{x^2}{j^2}\right)\,dx}_{J} $$ where $J\leq 0$ is trivial and Jensen's inequality gives $J\leq \log\left(\frac{1}{2}-\frac{1}{24j^2}\right).$