Generalization of convexity

96 Views Asked by At

This is not a homework, this is just something that came to my mind recently. Assume $f$ is a sufficiently nice function. We know that $$\frac{df}{dx} \geq 0 \iff f(x_2) \geq f(x_1) \text{ for } x_2 \geq x_1$$ $$\frac{d^2f}{dx^2} \geq 0 \iff \frac{f(x_1) + f(x_2)}{2} \geq f(\frac{x_1 + x_2}{2})$$

Is there any nice way to generalize this for for higher derivatives? A generalization of nonnegativity of higher order derivatives being equivalent to some short nice condition not involving any derivatives at all? If not generalization, is there at least a nice extension to $\frac{d^3f}{dx^3}$?

1

There are 1 best solutions below

0
On

I think we can play with the Schwarzian derivative (we work on $[0;\infty[$):

We have : $$(Sf)(x)=\frac{f'''(x)}{f'(x)}-1.5(\frac{f''(x)}{f'(x)})^2$$

Here we assume that the Schwarzian is positive so : $$\frac{f'''(x)}{f'(x)}\geq1.5(\frac{f''(x)}{f'(x)})^2$$

We can rewrite the inequaliy like this : $$\frac{f'''(x)}{f''(x)}\geq1.5(\frac{f''(x)}{f'(x)})$$

We integrate between $0$ and $x$ we have (always) :

$$ln(|f''(x)|)\leq 1.5ln(|f'(x)|)$$

We take the exponential we get :

$$|f''(x)|\geq (|f'(x)|)^{1.5}$$

We can rewrite the inequality like this :

$$\frac{|f''(x)|}{\sqrt{|f'(x)|}}\geq(|f'(x)|)$$

We integrate it give :

$$2\sqrt{|f'(x)|}\geq |f(x)|$$

Or $$4|f'(x)|\geq f(x)^2$$

Or :

$$\frac{4|f'(x)|}{f(x)^2}\geq 1$$

Now we assume that the derivative is always positive we get :

$$\frac{4 f'(x)}{f(x)^2}\geq 1$$

Or :

$$ 4f'(x)\geq f(x)^2$$

And it implies convexity of the function $f(x)$ if you solve the inequality (see the lemma of Gronwall)