11
$\begingroup$

Let $f: \mathbb{R} \to \mathbb{R}$ be a non-constant function, integrable on every interval $[x,y] \subset \mathbb{R}$ and $a \in \mathbb{R}$ such that $$f(ax + (1-a)y) = \frac{1}{y-x}\int_x^y{f(t)dt}, \quad \forall x<y$$ Prove that $\displaystyle a = \frac{1}{2}$.

This is a small part from another problem. If I can prove that $\displaystyle{a = \frac{1}{2}}$ then the whole problem is solved, but I don't know how.

Edit:

If $f$ is integrable, then the function $F : \mathbb{R} \to \mathbb{R}, F(x) = \displaystyle{\int_m^x{f(t)dt}}, m \in \mathbb{R}$ is continuous. Since we have that $\displaystyle{f(ax + (1-a)y) = \frac{1}{y-x}\int_x^y{f(t)dt}, \forall x<y}$, then $f$ is continuous and $F$ is differentiable. This implies that $f \in C^{\infty}(\mathbb{R})$.

If $\displaystyle{a = \frac{1}{2}}$, then $\displaystyle{f\left(\frac{x+y}{2}\right) = \frac{1}{y-x}\int_x^y{f(t)dt}, \forall x < y}$. So we have equality in the Hermite-Hadamard inequality, which is achieved when $f$ is linear on that interval.

$\endgroup$
  • $\begingroup$ No \displaystyle in titles, please. (For that matter, no \implies either.) $\endgroup$ – Did Feb 4 '18 at 13:55
  • $\begingroup$ The given condition implies that $f \in C^\infty(\mathbb{R})$. If it's also given that $f'(x_0) ≠ 0$ for one $x_0 \in \mathbb{R}$, it follows that $a = \frac{1}{2}$. $\endgroup$ – Saad Feb 5 '18 at 11:29
  • $\begingroup$ @AlexFrancisco Can you give a more detailed answer? I don't see why the fact that $f \in C^{\infty}(\mathbb{R})$ implies that $a = \frac{1}{2}$. $\endgroup$ – C_M Feb 5 '18 at 14:28
  • $\begingroup$ As answered by Fred $a$ can take any value $\endgroup$ – ibnAbu Feb 6 '18 at 21:32
2
+50
$\begingroup$

$\def\d{\mathrm{d}}$The proof for $f \in C^\infty(\mathbb{R})$ is skipped for simplicity since @C_M already knows how to prove it :) Now define$$ F(x) = \int_0^x f(t) \,\d t, \quad x \in \mathbb{R} $$ then $F \in C^\infty(\mathbb{R})$.

For any $x \in \mathbb{R}$ and $h > 0$, since$$ \int_{x_1}^{x_2} f(t) \,\d t = (x_2 - x_1) f(ax_1 + (1 - a)x_2)), \quad \forall x_1 < x_2 $$ i.e.$$ F(x_2) - F(x_1) = (x_2 - x_1) F'(ax_1 + (1 - a)x_2)), \quad \forall x_1 < x_2 $$ set $(x_1, x_2) = (x - (1 - a)h, x + ah)$ to get$$ F(x + ah) - F(x - (1 - a)h) = h F'(x). \quad \forall x \in \mathbb{R},\ h > 0 $$ Therefore, \begin{align*} 0 &= \frac{\partial^2}{\partial h^2}(h F'(x)) = \frac{\partial^2}{\partial h^2}(F(x + ah) - F(x - (1 - a)h))\\ &= a^2 F''(x + ah) - (1 - a)^2 F''(x - (1 - a)h). \quad \forall x \in \mathbb{R},\ h > 0 \end{align*} For any $x_1 < x_2$, set $(x, h) = (a x_1 + (1 - a)x_2, x_2 - x_1)$ to get$$ (1 - a)^2 F''(x_1) = a^2 F''(x_2). \quad \forall x_1 < x_2 \tag{1} $$

Now, because $f$ is not constant, then $F'' = f'$ is not always zero. Suppose $F''(x_0) \neq 0$. If $a = 0$, set $(x_1, x_2) = (x_0, x_0 + 1)$ in (1), then $F''(x_0) = 0$, a contradiction. Analogously, if $a = 1$, there is also a contradiction. Thus $a \neq 0, 1$, which implies $\dfrac{1 - a}{a} \neq 0$ and $F''(x) \neq 0 \ (\forall x \in \mathbb{R})$.

Next, on the one hand, from (1) there is$$ F''(2) = \left( \frac{1 - a}{a} \right)^2 F''(0). $$ On the other hand, from (1) there is$$ F''(2) = \left( \frac{1 - a}{a} \right)^2 F''(1) = \left( \frac{1 - a}{a} \right)^4 F''(0). $$ Therefore,$$ \left( \frac{1 - a}{a} \right)^4 = \left( \frac{1 - a}{a} \right)^2 \Longrightarrow \frac{1 - a}{a} = \pm 1. $$ Since $\dfrac{1 - a}{a} = 1 \Leftrightarrow a = \dfrac{1}{2}$ and $\dfrac{1 - a}{a} = -1 \Leftrightarrow 1 - a = -a$, then $a = \dfrac{1}{2}$.


From this it can be proved that $f = F'$ is a linear function.

$\endgroup$
3
$\begingroup$

We know $f$ is differentiable on $\mathbb R.$ If $f$ is constant, then any $a$ will work.

If $f$ is not constant, then $f'(x_0)\ne 0$ for some $x_0.$ For simplicity, assume at first that $x_0 = 0$ and $f(0)=0.$ Then $f(y) = f'(0)y + o(y)$ as $y\to 0.$ It follows that

$$\tag 1 f((1-a)y) = f'(0)(1-a)y + o(y)\,\,\text {as } y\to 0.$$

We also have

$$ f((1-a)y) = f(a\cdot 0 +(1-a)y) = \frac{1}{y}\int_0^y f(t)\,dt$$ $$\tag 2 = \frac{1}{y}\int_0^y (f'(0)t + o(t))\,dt = \frac{1}{y}(f'(0)y^2/2 + o(y^2))= f'(0)y/2 +o(y).$$

Equating $(1)$ with the last expression in $(2),$ and finally using $f'(0)\ne 0,$ we obtain $1-a=1/2,$ and thus $a=1/2,$ as desired.

What about the assumption $x_0 =0=f(x_0)?$ This special case implies the general case, simply because if $f$ has the given property, then so does $x\to f(x+d)+c$ for any constants $c,d.$

$\endgroup$
  • $\begingroup$ Very simple argument+1 $\endgroup$ – Paramanand Singh Feb 15 '18 at 4:43
  • $\begingroup$ @ParamanandSingh Thank you. $\endgroup$ – zhw. Feb 15 '18 at 16:57

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.