5

I ask for an inequality which is a follow up of this question: Prove that $\int_0^\infty\frac1{x^x}\, dx<2$ :

$$\int_{0}^{\infty}x^{-x}dx<\pi-\ln\pi$$

You can find a nice proof @RiverLi among others and also an attempt of mine.

My try

We have:

$$\int_{1}^{\phi}\phi^{\left(a-ax^{2}\right)}dx+\int_{0}^{1}x^{-x}dx+\int_{b}^{\infty}x^{-x}dx>\int_{0}^{\infty}x^{-x}dx$$

Where $\phi=\frac{1+\sqrt{5}}{2},a=1$ because we have the inequality for $x\in[1,b]$:

$$b^{a-ax^{2}}\geq x^{-x}$$

Unfortunetaly it's already bigger...

We also have:

$$\int_{1}^{\frac{1}{2}+\sqrt{\frac{5}{4}}}\left(\frac{1+\sqrt{5}}{2}\right)^{1-x^{2}}dx+\int_{0}^{1}x^{-x}dx+\int_{\frac{1}{2}+\sqrt{\frac{5}{4}}}^{1.8}f\left(x+\frac{1}{9}\right)dx+\int_{1.8}^{\infty}x^{-x}dx<2$$

Where:

$$f\left(x\right)=2^{\frac{1+\sqrt{5}}{2}}\left(x^{\frac{1}{3}}\right)e^{-x^{\frac{4}{3}}}$$

$$f(x+1/9)>x^{-x},x\in[\phi,\infty)$$

Until now all the derivatives are easy and there is one trick .

How to show it analytically?

Kamal Saleh
  • 3,852
  • 1
  • 3
  • 27
Erik Satie
  • 3,376
  • 3
  • 9
  • 36
  • can you show your efforts? – TShiong Dec 28 '22 at 13:10
  • 4
    [Please do not include "thanks" and other chit chat in posts](https://math.stackexchange.com/help/behavior). – Thomas Markov Dec 28 '22 at 13:11
  • @ShivaVenkata See the link there is a try of mine . Do you read carefully ? – Erik Satie Dec 28 '22 at 13:11
  • I think you should show your own thoughts to get an answer or find if the existing asnwers help you – TShiong Dec 28 '22 at 13:12
  • 3
    Is there some significant difference between the linked question and this one that makes it not a duplicate? – Thomas Markov Dec 28 '22 at 13:14
  • @ThomasMarkov There is reasonably more difference with my question and TheSimplifire's question than TheSimplifire's question and this https://math.stackexchange.com/questions/238199/showing-that-int-0-infty-x-x-mathrmdx-leq-2 – Erik Satie Dec 30 '22 at 11:39
  • Are we supposed to know the value of $\sum _{n=1}^{\infty } n^{-n}$ which can be found everywhere ? Cheers & ***Happy New Year !*** – Claude Leibovici Dec 31 '22 at 08:20
  • @Claudecivilebio YESSS! sorry for the anagram – Erik Satie Dec 31 '22 at 11:20
  • I get the statement from the linked one with some accurate approximations for $\sum_{k>0}k^{-k}$ and $\int_{0}^{+\infty}\frac{e^{-x}}{W(x)+1}\,dx.$ So are you asking for formal proof for the sum or for the integral with the Lambert function? – Marius S.L. Jan 05 '23 at 21:38

2 Answers2

2

Firstly, $$I_1=\int\limits_0^e x^{-x}\,\text dx = e\int\limits_0^1 (ex)^{-ex}\,\text dx = e\int\limits_0^1 e^{-ex(1+\ln x)}\,\text dx = e\int\limits_0^1\,\sum_{k=0}^\infty \dfrac{(-ex(1+\ln x))^k}{k!}\,\text dx$$ $$= \sum_{k=0}^\infty \dfrac{(-1)^k e^{k +1}}{k!} \int\limits_0^1\, x^k(1+\ln x)^k\,\text dx = e+\sum_{k=1}^\infty \dfrac{(-1)^k e^{k+1}}{k!} \sum_{j=0}^k\dbinom kj \int\limits_0^1\, x^k(\ln x)^j\,\text dx$$ $$= e+\sum_{k=1}^\infty \dfrac{(-1)^k e^{k+1}}{k!} \sum_{j=0}^k\dbinom kj \left(\dfrac2{k+1}\right)^{j+1} \int\limits_0^1\, \dfrac{k+1}2\,x^k \ln^j\left(x^{\large\frac{k+1}2}\right)\,\text dx$$ $$= e + \sum_{k=1}^\infty \dfrac{(-1)^k e^{k+1}}{k!} \sum_{j=0}^k\dbinom kj \left(\dfrac2{k+1}\right)^{j+1} \int\limits_0^1\, y\ln^j\left(y\right)\,\text dy$$ $$=e+\sum_{k=1}^\infty \dfrac{(-1)^k e^{k+1}}{k!} \sum_{j=0}^k\dbinom kj \left(\dfrac2{k+1}\right)^{j+1} (-1)^j \dfrac{j!}{2^{j+1}}$$ $$=e+\sum_{k=1}^\infty (-1)^k e^{k+1} \sum_{j=0}^k\dfrac{(-1)^j}{ (k-j)!(k+1)^{j+1}}$$ $$=e + \sum_{k=1}^\infty (-1)^k e^{k+1} e^{-k-1}\dfrac{(-1)^k}{(k+1)^{k+1} k!}\Gamma(k+1,-k-1),$$ $$I_1=e + \sum_{k=2}^\infty \dfrac{\Gamma(k,-k)}{k^k \Gamma(k)}$$ received an alternating series, wherein $$e + \sum_{k=2}^{22}\dfrac{\Gamma(k,-k)}{k^k \Gamma(k)} \le 1.96466\,14544\,475,\quad \dfrac{\Gamma(23,-23)}{\Gamma(23)}{23^{-23}}\le 2\cdot 10^{-13}.$$ Therefore, $$I_1\ge 1.96466\,14544\,477.$$

$\color{green}{\textbf{Edit of 21.02.23}}$

Obtained estimation can be significantly improved.

Let $x\in\big(a,(1+\delta a\big),\;\delta\ll1,\;$ then $$x=a(1+\delta z)=a +\delta a z,\qquad \big(z\in(0,1),\;a\in(1,\infty)\big),$$ $$x^{-x}=a^{\large -a-\delta a z}\;\left(1+\delta z \right) ^{\large -a(1+\delta z)},$$

Taking in account the inequality $$(1+t)^{\large-a(1+t)} -e^{-at} \le \dfrac a2 t^2\left(1-\dfrac{1+3a}3 t\right)\tag1$$ and identity $$\int\limits_0^d t^k a^{-a t}\,\text dt =\dfrac{Γ(k+1)-Γ(k+1,ad\ln a)}{(a\ln a)^{k+1}},\tag2$$ one can get

$$I(a, \delta)=\int\limits_a^{a+\delta a} x^{-x}\,\text dx = \delta a\cdot a^{-a} \int\limits_0^1 a^{-\delta az}\left(1+\delta z\right)^{\large- a(1+\delta z)}\,\text dz$$ $$= a^{2-a}\int\limits_0^\delta a^{-at}\left(1+t\right)^{\large-a(1+t)}\,\text dt \le a^{2-a} \int\limits_0^\delta a^{\large-at} \left(e^{-at}+\frac a2 t^2 - \frac {a(3a+1)}6 t^3\right)\,\text dt$$ $$=a^{1-a} \int\limits_0^\delta (ea)^{\large-a t}\,\text dt + \dfrac12a^{2-a} \int\limits_0^\delta t^2 a^{\large-a t}\,\text dt -\dfrac {3a+1}6 a^{2-a} \int\limits_0^\delta t^3 a^{\large-a t}\,\text dt,$$ $$\color{brown}{\mathbf{I(a, \delta)=a^{-a}\left(\dfrac{1-(ea)^{-\large \delta a}}{\ln a} +\dfrac{2-\Gamma(3,a\delta\ln a)}{2a\ln^3a} -(3a+1)\dfrac{6-\Gamma(4,a\delta\ln a)}{6a\ln^4a}\right)}},\tag3$$ $$\int\limits_e^\infty x^{-x} dx = \lim\limits_{\delta \to 0} \sum\limits_{k=0}^\infty I(e\delta^k, \delta).\tag4$$

Comparing of this expression (for $\delta=\sqrt[\large 6000]2-1,\; n=3000$) with the numeric value of the integral is shown in the table below $$\left[\begin{matrix} [a,b] & I(a,\delta) & \text{numeric}\\ \left[e, e\sqrt2\right] & 0.02848\,08301\,58887 & 0.02848\,08301\,58764\\ \left[e\sqrt2, 2e\right] & 0.00227\,71750\,39440 & 0.00227\,71750\,39412\\ \left[2e, 2e\sqrt2\right] & 0.00003\,64476\,51537 & 0.00003\,64476\,51535\\ \left[2e\sqrt2,4e\right] & 0.00000\,00502\,01224 & 0.00000\,00502\,01224\\ [e,4e] & 0.03079\,45030\,51082 & 0.03079\,45030\,50937 \end{matrix}\right]$$

Besides, $$(1+t)^{-a(1+t)}\le e^{-at},$$ $$I(a,\infty)=\int\limits_a^\infty x^{-x}\,\text dx =a^{-a}\int_0^\infty a^{-at}\left(1+t\right)^{\large- a(1+t)}\,\text dt$$ $$\le a^{-a} \int\limits_0^\infty a^{-at} e^{-at}\text dz = \dfrac{a^{-a}}{\ln(ae)} (ae)^{-at}\bigg|_0^\infty = \dfrac{a^{-a}}{\ln(ae)},$$ $$I(4e,\infty) \le \dfrac{(4e)^{-4e}}{\ln(4e^2)} \le 1.592\cdot 10^{-12}.$$

Finally, $$I\le 1.96466\,14544\,477 + 0.03079\,45030\,51082 + 1.5917\cdot 10^{-12} < \color{brown}{\mathbf{1.99545\,59576}}.$$ Numeric calculations give $$I\approx 1.99545\,59575.$$

$\textbf{Previous version.}$

On the other hand, $$I_2=\int\limits_e^\infty x^{-x}\,\text dy =e\int\limits_0^\infty (e(1+y))^{-e(1+y)}\,\text dy =\int\limits_0^\infty \dfrac{e^{1-e(1+y)}}{(1+y)^3} (1+y)^{3-e(1+y)}\,\text dy=I_{21}+I_{22},$$ where $$I_{21}=\int\limits_1^2 \dfrac{e^{1-ez}}{z^3} z^{3-ez}\,\text dz,\quad I_{22}=\int\limits_2^\infty \dfrac{e^{1-ez}}{z^3}\, z^{3-ez}\,\text dz.$$ Applying inequality $z^{3-ez}\le f(z),$ where $$f(z)=\begin{cases} e^{-0.7177ez}(-26.937+49.92z-15.94z^2),\quad\text{if}\quad z\in[1,2)\\[8pt] (z-0.48) e^{6-1.4868ez},\quad\text{if}\quad z\in[2,+\infty) \end{cases}$$ (see also WA plot), one can get $$I_{21}\leq \int\limits_1^2 \dfrac{e^{1-ez}}{z^3}\,(e^{0.7177ez}(-26.937+49.92z-15.94z^2))\,\text dz$$ $$=\int\limits_1^2 \dfrac1{z^3}\,(e^{1-1.7177ez}(-26.937+49.92z-15.94z^2))\,\text dz$$ $$=\left(\dfrac{36.6112 - 306.641z}{z^2}\,e^{-4.66919 z} - 1475.1\operatorname{ Ei}(-4.66919 z)\bigg|_1^2\right)\le 0.03105\,33417\,253,$$(see also WA checking) $$I_{22}\leq \int\limits_2^\infty \dfrac{e^{1-ez}}{z^3}\,((z-0.48) e^{6-1.4868ez})\,\text dz =\int\limits_2^\infty \dfrac{z-0.48}{z^3}\,e^{7-2.4868ez})\,\text dz$$ $$=\dfrac{263.192-2875.76 z}{z^2}e^{-6.75982 z} - 19439.7 Ei(-6.75982 z) \bigg|_2^\infty \le 0.00003\,70357\,268$$ (see also WA checking).

Finally, $$I=I_1+I_{21}+I_{22}\le 1.96466\,14544\,477+0.03105\,33417\,253+0.00003\,70357\,268,$$ $$\color{brown}{\mathbf{I\le1.99575\,18318\,998 \le \pi-\ln\pi.}}$$

Yuri Negometyanov
  • 27,332
  • 1
  • 29
  • 59
  • Impressive. I assume there is an analytic way to guess-timate the coefficients in, e.g., $z^{3-ez} – FShrike Feb 07 '23 at 16:55
  • @FShrike The accuracy of the last integrals can be improved via increasing of the polynomial order, but this way is not trivial for the second integral. Looks ugly the further fragmentation of the intervals, but it can be reserved as brute force approach. – Yuri Negometyanov Feb 07 '23 at 18:02
0

I put here my progress so far we have :

$$\int_{2}^{\pi}k\left(x\right)dx+\int_{1}^{2}f\left(x\right)dx+\int_{0}^{1}h\left(x\right)dx+\int_{\pi}^{4}\frac{3+\frac{3.5}{1000}}{\ln3}x^{\frac{12}{1000}+\frac{2}{5}}\left(\frac{3+\frac{3.5}{1000}}{\ln3}\right)^{-x^{1+\frac{2}{5}+\frac{12}{1000}}}dx+\int_{4}^{5}ex^{\frac{12}{1000}+\frac{2}{5}}e^{-x^{\left(1+\frac{2}{5}+\frac{12}{1000}\right)}}dx+\int_{5}^{\infty}5.25^{-x}e^{\left(-x+5+0.25\right)}dx\simeq 1.996912$$

Where :

$$h\left(x\right)=x^{-x}$$

$$f\left(x\right)=a\left(\left((1-c)^{2}+ac(2-c)-ac(1-c)\ln a\right)\right)$$

$$k\left(x\right)=a^{2}\left(\left((1-d)^{2}+ad(2-d)-ad(1-d)\ln a\right)\right)$$

$$a=1/x,c=x-1,d=x-2$$

For some references see my first answer on TheSimplifire's question .

I got it !!!

$$\int_{0}^{1}x^{-x}dx+\int_{1}^{2}f\left(x\right)dx+\int_{2}^{3-0.75}k\left(x\right)dx+\int_{3-0.75}^{4}\frac{3+\frac{3.5}{1000}}{\ln3}x^{\frac{12}{1000}+\frac{2}{5}}\left(\frac{3+\frac{3.5}{1000}}{\ln3}\right)^{-x^{1+\frac{2}{5}+\frac{12}{1000}}}dx+\int_{4}^{4.8}ex^{\frac{13}{1000}+\frac{2}{5}}e^{-x^{\left(1+\frac{2}{5}+\frac{13}{1000}\right)}}dx+\int_{4.8}^{\infty}5^{-x}e^{-x+5}dx<\pi-\ln \pi$$

Erik Satie
  • 3,376
  • 3
  • 9
  • 36