8
$\begingroup$

Find all differentiable $f:\mathbb R \to \mathbb R$ such that $$\forall (x,y)\in \mathbb R^2, f(x+y)=f(x)f'(y)+f'(x)f(y)$$

It's easy to check that the only constant solution is $0$ and the only polynomial solution is $x\mapsto x$. Besides, it's also easy to check that $\sin$ is a solution, as well as $\sinh$. More generally, $x\mapsto \frac {\sin(ax)}{a}$ and $x\mapsto \frac {\sinh(ax)}{a}$ are solutions.

Setting $x=y=0$ yields $f(0)(1-2f'(0))=0$.
By letting $y=0$, one gets the ODE $ f(0)f'(x)+(f'(0)-1)f(x)=0$.

If $f(0)\neq 0$, then $f'(0)=\frac 12$ and this is easily solved as $x\mapsto \frac {\exp(ax)}{2a}$
If $f(0)=0$, either $f'(0)\neq 1$ and then $f=0$ , or $f'(0)=1$ and the ODE is now useless.

How should I continue ? Are there other solutions ?

$\endgroup$
  • 4
    $\begingroup$ Differentiating w.r.t. $x$ gives $f'(x+y)=f'(x)f'(y)+f''(x)f(y)$. W.r.t. $y$ gives $f'(x+y)=f(x)f''(y)+f'(x)f'(y)$. Therefore $f(x)f''(y)=f(y)f''(x)$ for all $x,y$. Implying that $f(x)/f''(x)$ must be a constant. Probably we can make progress from here on. Also, probably there's a cleaner way. Gotta go, sorry. $\endgroup$ – Jyrki Lahtonen Sep 3 '18 at 15:10
  • $\begingroup$ Looks like a special case of Solution to functional equation. $\endgroup$ – Martin R Sep 3 '18 at 15:11
  • 3
    $\begingroup$ $f$ must be twice differentiable since $f'(x+y)=f'(x)f'(y)+f''(x)f(y)$ proves it. $\endgroup$ – Marco Sep 3 '18 at 15:21
  • 1
    $\begingroup$ @LorenzoQuarisa: Fix any $y$ with $f(y) \ne 0$ and solve the functional equation for $f'(x)$. It follows that $f'$ is differentiable. $\endgroup$ – Martin R Sep 3 '18 at 15:22
  • 1
    $\begingroup$ Possible duplicate of Solution to functional equation $\endgroup$ – Jyrki Lahtonen Sep 3 '18 at 17:59
6
$\begingroup$

With $y=0$ you get $f(x)=f(x)f'(0)+f'(x)f(0)$.

If $f(0)\ne0$, we obtain $f'(x)=\frac{1}{f(0)}f(x)(1-f'(0))$. The case $f'(0)=1$ yields $f'(x)=0$, so $f$ is constant $c$, which implies $c=0$: a contradiction.

If $f'(0)\ne1$, we can write $f'(x)=kf(x)$, with $k\ne0$, so $f(x)=ae^{kx}$ (with $a\ne0$). The main relation now is $$ ae^{k(x+y)}=ae^{kx}ake^{ky}+ake^{kx}ae^{ky} $$ that implies $1=2ak$. This is a solution.

If $f(0)=0$, the equation becomes $f(x)=f(x)f'(0)$ or $(1-f'(0))f(x)=0$. If $f'(0)\ne 1$, we get the constant $0$ function.

If $f(0)=0$ and $f'(0)=1$, the business becomes interesting.

Since $f'(0)\ne1$, the function is not constant, so there is $y_0$ with $f(y_0)\ne0$. In particular, for all $x$, $$ f'(x)=\frac{1}{f(y_0)}(f(x+y_0)-f(x)f'(y_0)) $$ which shows $f'$ is differentiable (and also that $f$ is infinitely differentiable). Thus we can differentiate the main relation with respect to $x$ and $y$: \begin{align} f'(x+y)&=f'(x)f'(y)+f''(x)f(y) \\ f'(x+y)&=f(x)f''(y)+f'(x)f'(y) \end{align} that implies $f(x)f''(y)=f''(x)f(y)$. Therefore, for all $x$, $$ f''(x)=rf(x) $$ where $r=f''(y_0)/f(y_0)$.

This is an easy differential equation.

$\endgroup$
  • $\begingroup$ @GabrielRomon That's not a contradiction: $0=f'(0)\ne1$. $\endgroup$ – egreg Sep 3 '18 at 17:27
1
$\begingroup$

Just suggesting an approach. Its clear that its infinitely differentiable.Let $f$ be analytic. $f(x) = \sum_{i=0}^{\infty} a_i (x-a)^i$.

Then $f(x+y) = f(x) f'(y) + f'(x) f(y)$ implies

$\sum_{i=0}^{\infty} a_i (x+y-a)^i = (\sum_{i=0}^{\infty} a_i (x-a)^i)(\sum_{i=1}^{\infty} i a_i (y-a)^{i-1}) + (\sum_{i=0}^{\infty} a_i (y-a)^i)(\sum_{i=1}^{\infty} i a_i (x-a)^{i-1})$.

Now solve for $a_i$.

For example: Let $a=0$. setting $x=0,y=0$

$a_0 = a_0 a_1 + a_0 a_1$.

Hence $a_1 = \frac{1}{2}$ or $a_0=0$.

Now set $a=0,x=0$: We get $\sum_{i=0}^{\infty} a_i y^i = a_0 \sum_{i=1}^{\infty} i a_i y^{i-1} + a_1 \sum_{i=0}^{\infty} a_i y^{i}$

Hence $a_i = (i+1) a_0 a_{i+1} + a_1 a_i$ Hence $a_{i+1} = \frac{a_i(1-a_1)}{(i+1)a_0}$. (Assuming $a_0 \neq 0$)

$\endgroup$
  • 1
    $\begingroup$ Infinitely differentiable does not imply analytic though. $\endgroup$ – Gabriel Romon Sep 3 '18 at 16:21
  • $\begingroup$ Atleast $a_1$ is solved for... If you take $a=0$, this might help. $\endgroup$ – Balaji sb Sep 3 '18 at 16:28
  • $\begingroup$ I just edited it ..its either $a_1=1/2$ or $a_0=0$ $\endgroup$ – Balaji sb Sep 3 '18 at 16:29
  • $\begingroup$ I updated with calcuation for $a_i$ assuming $f$ is analytic near 0. $\endgroup$ – Balaji sb Sep 3 '18 at 16:47

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.