Let $a,b\in\mathbb R_{\geq 0}$ and let $p>1$. Is it true that $$ (a+b)^p\geq a^p+b^p $$ holds? It's obvious for $p\in\mathbb N_{>1}$. I was thinking of making an argument involving derivatives. So for each $p\in\mathbb N_{>1}$ we have that $(a+b)^p-a^p-b^p>0$. De derivative with respect to $p$ then yields $\ln(p)[(a+b)^p-a^p-b^p]>0$. Maybe I can somehow use this fact? In any case, I'm stuck and would appreciate some help.
-
2Well, if $a$ or $b$ (or both) are $0$ then it is false. – lulu Feb 20 '20 at 22:11
-
1If you require $a,b>0$ then you just have to subtract the right hand from the left and remark that the left hand still has some positive terms left over. – lulu Feb 20 '20 at 22:12
-
@lulu Right, I've modified the inequality. – Sha Vuklia Feb 20 '20 at 22:22
-
@lulu Also, how do you mean "positive terms left over"? – Sha Vuklia Feb 20 '20 at 22:24
-
Take $p=2$. then you are trying to prove that $a^2+2ab+b^2≥a^2+b^2$. After subtracting you are left with $2ab≥0$ which is obviously true, That's how it works for all $p$. – lulu Feb 20 '20 at 22:25
-
@lulu I stated that for $p$ a natural number I see the inequality. – Sha Vuklia Feb 20 '20 at 22:26
-
Ah, missed that. Sorry. You could still use the Generalized binomial theorem and apply the same argument. Best to divide by $a^p$ first, assuming $a>b$. That way the series converges. – lulu Feb 20 '20 at 22:29
-
@lulu If $p$ is not an integer then we get an infinite series, which will eventually contain infinitely many terms. The problem is exactly to show that after subtracting RHS the remainder is still positive. – Allawonder Feb 21 '20 at 15:09
-
@Allawonder Yeah, probably not the best way to proceed. I agree. – lulu Feb 21 '20 at 15:29
4 Answers
Note that if $a =0$ or $b = 0$ the inequality fails to hold. So we consider $a,b > 0$. Put $\sin^2(\theta) = \dfrac{a}{a+b}, \cos^2(\theta) = \dfrac{b}{a+b}\implies \sin^{2p}(\theta)+\cos^{2p}(\theta) < 1$. But this is true because $\sin^{2p}(\theta)+\cos^{2p}(\theta) < \sin^2(\theta)+\cos^2(\theta) = 1, \theta \in (0,\frac{\pi}{2})$ .
- 77,651
We may restrict to the case that $a\in(0,\infty)$ and $b\in(0,\infty)$ because $(a+b)^{p}=a^{p}+b^{p}$ if $a=0$ or $b=0$.
By further dividing the whole expression by $a^{p}$, the inequality is equivalent to $\left[1+(\frac{b}{a})\right]^{p}>1+(\frac{b}{a})^{p}$. This suggests us to investigate the function $f:[0,\infty)\rightarrow\mathbb{R}$, $f(x)=(1+x)^{p}-1-x^{p}$. Note that $f(0)=0$. For $x\in(0,\infty)$, we have $f'(x)=p[(1+x)^{p-1}-x^{p-1}].$ Consider three cases.
Case 1: $p\in(0,1)$. Observe that the function $(0,\infty)\rightarrow(0,\infty)$, $y\mapsto y^{p-1}$ is strictly decreasing, so $f'(x)<0$.
Case 2: $p=1$. In this case, $f'(x)=0$.
Case 3: $p\in(1,\infty)$. Observe that the function $(0,\infty)\rightarrow(0,\infty),y\mapsto y^{p-1}$ is strictly increasing, so $f'(x)>0$.
We conclude that:
If $p\in(0,1)$, $f$ is strictly decreasing on $[0,\infty)$ and hence for any $x\in(0,\infty)$, $f(x)<f(0)=0$.
If $p=1$, $f$ is a constant function and hence $f(x)=f(0)=0$.
If $p\in(1,\infty)$, $f$ is strictly increasing and hence for any $x\in(0,\infty)$, $f(x)>f(0)=0$.
- 12,250
-
I also consider the cases that $p\in(0,1)$ and $p=1$. However, you may ignore it. – Danny Pak-Keung Chan Feb 20 '20 at 22:25
The method using derivatives does not work as $$\ln(p)[(a+b)^p-a^p-b^p]>0\iff (a+b)^p-a^p-b^p>0,$$ as $\ln(p)>0$ when $p>1$. So we are back to the original problem. However, we know that when $a,b>0$, $\left(\frac{a}{a+b}\right)^p<\frac{a}{a+b}$, for any $p>1$ because $\frac{a}{a+b}<1$. Thus, $$ \frac{a^p+b^p}{(a+b)^p} = \left(\frac{a}{a+b}\right)^p+\left(\frac{b}{a+b}\right)^p < \frac{a}{a+b}+\frac{b}{a+b}=1\implies a^p+b^p < (a+b)^p.$$
- 3,147
Write your binomial in the form $a(1+x),$ where $0<x<1.$ Then we have that $$a^p(1+x)^p>a^p(1+px)>a^p(1+x^p),$$ from which the result immediately proceeds.
- 13,327