4

Question:

let $$f(0)=0,f(1)=1, f''(x)>0,x\in (0,1)$$ let $k>2$ are real numbers,show that $$4(k+1)\int_{0}^{1}(f(x))^kdx\le 1+3k\int_{0}^{1}(f(x))^{k-1}dx$$

This problem is from china Analysis problem book excise by (Min Hui XIE) ,analysis problem bookI think we can use $$F(x)=\int_{0}^{x}f(t)dt$$ so $$F(x)=x\int_{0}^{1}f(ux)du=x\int_{0}^{1}f[ux+(1-u)\cdot 0]du\ge x\int_{0}^{1}[uf(x)dx+(1-u)]du=\dfrac{x}{2}(f(x)+1)$$ then I can't,Thank you

math110
  • 93,304
  • Is $f$ additionally assumed to be non-negative? – Qmechanic Dec 29 '14 at 13:47
  • 5
    Interesting one ! I got $\displaystyle k^2\int_0^1 f^{k-1}(x),dx \le 1 + (k^2-1)\int_0^1 f^k(x),dx$, but I don't see how to get the constants $3,4$ right away ! :) – r9m Jan 01 '15 at 17:42
  • If $k\gt2$ is assumed to be any real, then $f$ would need to be non-negative. If $k$ is assumed to be an integer, then $f$ could be negative. – robjohn Jan 23 '15 at 20:07

3 Answers3

4

Sign of $\boldsymbol{f}$

If $k\gt2$ is not assumed to be an integer, then we need to assume that $f$ is non-negative for $f(x)^k$ to make sense. Moreover, if $k=4$ and $f(x)=2x^6-x$, then $$ 4(k+1)\int_0^1f(x)^k\,\mathrm{d}x=\frac45\gt\frac{102}{133}=1+3k\int_0^1f(x)^{k-1}\,\mathrm{d}x\tag{1} $$ Therefore, I think it is safe to assume that $f$ needs to be non-negative.


Variational Argument

To maximize $\int_0^1\left(4(k+1)f(x)^k-3kf(x)^{k-1}\right)\,\mathrm{d}x$ for all $f$ so that $f(0)=0$ and $f(1)=1$ we need $$ \begin{align} 0 &=\delta\int_0^1\left[4(k+1)f(x)^k-3kf(x)^{k-1}\right]\,\mathrm{d}x\\ &=\int_0^1\left[4(k+1)kf(x)^{k-1}-3k(k-1)f(x)^{k-2}\right]\delta f(x)\,\mathrm{d}x\tag{2} \end{align} $$ for every $\delta f$ that fixes $\int_0^1f'(x)\,\mathrm{d}x$; that is, $$ \begin{align} 0 &=\delta\int_0^1f'(x)\,\mathrm{d}x\\ &=\int_0^1f''(x)\delta f(x)\,\mathrm{d}x\tag{3} \end{align} $$ To satisfy these conditions for all $\delta f$, we need a constant $\lambda$ so that $$ \begin{align} f''(x) &=\lambda\left[4(k+1)kf(x)^{k-1}-3k(k-1)f(x)^{k-2}\right]\\ &=\lambda 4k(k+1)f(x)^{k-2}\left[f(x)-\frac34\frac{k-1}{k+1}\right]\tag{4} \end{align} $$ Now we can use the condition that $f''(x)\ge0$. Note that if $\lambda\ne0$, then the right side of $(4)$ changes sign as $f(x)-\frac34\frac{k-1}{k+1}$ does, which it must since $f(0)=0$ and $f(1)=1$. Therefore, we must have that $\lambda=0$. This in turn implies that $f''(x)=0$. Thus, we must have that $$ f(x)=x\tag{5} $$ which means $$ \begin{align} \int_0^1\left(4(k+1)f(x)^k-3kf(x)^{k-1}\right)\,\mathrm{d}x &\le\int_0^1\left(4(k+1)x^k-3kx^{k-1}\right)\,\mathrm{d}x\\ &=1\tag{6} \end{align} $$ which is equivalent to the condition sought.

robjohn
  • 345,667
2

It seems that the statement is false even if $f\ge0$ is assumed.

Let $f(x)=x^{10}$. Then $LHS=4(k+1)\cdot\frac1{10k+1}<1<RHS$.

G.Kós
  • 14,297
1

@china math: The analysis problem book doesn't contain this exercise.