2
$\begingroup$

I must be a terrible googling searcher but I cannot find a reference to the following inequality:

$$ \forall_{\phi\in(0;\frac \pi 4)}\ \ln(\cot(\phi)))\, <\, \cot(2\!\cdot\!\phi) $$

I have just obtained this, it seems to have nice potential, and now I would appreciate a reference. (As a minimum, if this is new to you, I hope that you enjoed it).


REMARK: I apologize for my trivial question; I'd be more than willing to remove it once I get a respective link or a paper.

$\endgroup$
8
$\begingroup$

Plug $x=\cot(\phi)$ and turn your inequality into $$ x \ln x < \frac{x^2-1}{2},$$ where $x>1$. The RHS is the second order Taylor approximation of the LHS around $x=1$. Hence the inequality follows from concavity of the derivative of $x \ln x$, which is $1+\ln x$.

This argument has led me to the following elegant solution: Integrate the following well-known inequality between 1 and $x$:

$$ \ln t \le t-1,$$ and get $$ x \ln x - x + 1 \le \frac{(x-1)^2}{2} \implies $$ $$x\ln x \le \frac{x^2-1}{2}.$$ (Equality iff $x=1$).


[EDIT] Turns out there's a reference for the inequality $\log(1+x) < x-\frac{x^2}{2(1+x)}$, which is equivalent to your inequality.

Hermite–Hadamard inequality states the following: If $f:[a,b]\to \mathbb{R}$ is convex and continuous, then $$f\left( \frac{a+b}{2}\right) \le \frac{1}{b - a}\int_a^b f(x)\,dx \le \frac{f(a) + f(b)}{2}.$$ The proof is geometric - one compares the integral to a trapezoid and to a rectangle. Applying this to $f(x)=-\frac{1}{x}$ (as you have done below), one obtains the desired inequality.

In fact, this was the first application of this inequality. In the letter "Sur deux limites d'une intégrale d´e finie", published by Hermite in Mathesis 3 (1883, p. 82), he proves the above inequality and applies it to $f(x)=\frac{1}{x+1}, a=0$ and gets $$x −\frac{x^2}{2+x^2} < \log(1+x) < x-\frac{x^2}{2(1+x)}.$$

I was not able to locate the letter myself, but the relevant excerpt appears right in the beginning of the introduction to this monograph by Dragomir and Pearce.

$\endgroup$
  • $\begingroup$ Hm, thank you. I already had a proof, of course, but my question was about a reference. I'll up-vote you anyway. I'll post my original proof too. These proofs are like walking from a vertex of a rectangle on the edges toward the opposite vertex, you go either East then North, or first you go North, and then East. $\endgroup$ – Włodzimierz Holsztyński Dec 25 '15 at 7:28
  • 1
    $\begingroup$ Since this is a consequence of a famous inequality ($\ln x \le x-1$), I will be (pleasently) surprised to discover a more specific reference. Looking forward to read your proof! $\endgroup$ – Ofir Gorodetsky Dec 25 '15 at 7:33
  • $\begingroup$ Well, all properties of the real logarithm follow from the homomorphic property and that one inequality (simple axioms). $\endgroup$ – Włodzimierz Holsztyński Dec 25 '15 at 7:53
  • $\begingroup$ Let me be a little pedantic, sorry. The inequality, $\ \log(x) \le x-1\ $ is only logically strong since everything else follows. However, mathematically speaking, in the everyday mathematical sense, the straight trapezoid provides an immediately sharper inequality than the rectangular $\ \log(x) \le x-1\ $. $\endgroup$ – Włodzimierz Holsztyński Dec 25 '15 at 8:05
3
$\begingroup$

Looking at the integral of $\frac 1x$, i.e. at the respective curvilinear trapezoid, and at the straight polygonal trapezoid, one gets:

$$ \ln(b) - \ln(a)\ <\ \frac 12\cdot\left(\frac 1a+\frac 1b\right)\cdot(b-a) $$

whenever $\ 0<a<b;\ $ i.e.

$$ \ln\left(\frac ba\right)\ <\,\ \frac {b^2-a^2}{2\cdot a\cdot b} $$

Now let $\ 0<\phi<\frac \pi 4.\ $ Let $\ a:=\sin(\phi)\,$ and $\, b:=\cos(\phi),\ $ so that $0<a<b$. Then indeed we get the required inequality:

$$ \ln(\cot(\phi))\ <\ \cot(2\cdot \phi) $$

Have fun! This is the time of the year... ok, any time is great.

$\endgroup$
  • 1
    $\begingroup$ Nice proof! Now when I see the proof I think it might be possible to find a reference. For starters, this follows from the error analysis in the Trapezoidal rule: en.wikipedia.org/wiki/Trapezoidal_rule $\endgroup$ – Ofir Gorodetsky Dec 25 '15 at 8:03
  • 2
    $\begingroup$ The Trapezoidal rule applied to convex\concave functions yields an inequality known as "Hermite–Hadamard inequality". In the first paper (by Hermite) which proves this inequality, he applies it to $f(x)=\frac{1}{x+1}$ and obtains your inequality. See my edit to my original answer. $\endgroup$ – Ofir Gorodetsky Dec 25 '15 at 8:47
  • 1
    $\begingroup$ @OfirGorodetsky -- once again, many-many thanks. I consider writing elementary articles for amateurs and youngsters. I will certainly make a reference to He-Ha (and will mention that the info came to me from you). $\endgroup$ – Włodzimierz Holsztyński Dec 25 '15 at 9:09
  • 1
    $\begingroup$ My pleasure. I enjoy learning about 19th century papers. I also want to correct myself - one side of the He-Ha inequality is by the "trapezoid rule", but the other side is by the "midpoint rule". $\endgroup$ – Ofir Gorodetsky Dec 25 '15 at 9:19
  • 1
    $\begingroup$ @OfirGorodetsky -- I expected that the other side estimate would call for a different approach (naturally, due to the nature of the convexity). I was about to write that other side inequalityt but MO slowed me down (LaTeX etc takes a lot of energy, especially on MO). I'd call the other approach "tangential" or "derivative". It's 4:34am now. But possibly later I may add the other side inequality (no firm promises, just a vague expectation :-) $\endgroup$ – Włodzimierz Holsztyński Dec 25 '15 at 9:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.