11
$\begingroup$

My question is to find the solutions to the following

$\frac{df(x)}{dx} = f^{-1} (x)$

where $f^{-1} (x)$ refers to the inverse of the function f. The domain really isn't important, though I am interested in either (-inf, inf) or (0, inf), so if any solutions are known for more restricted domains then they are welcome.

I cannot find any material relating to this type of question in any of my calculus and differential equations textbooks and references; it seems quite unorthodox. Any material which covers this type of diff equation would be wlecome

$\endgroup$
  • $\begingroup$ Let $g(x)=f^{-1}(x)$, then $$ g'(x)=f'(g(x))^{-1}=g(g(x))^{-1}, $$ As such, it connects values and derivatives of the function $g$ at two different locations and is thus loosely related to delay-differential equations. $\endgroup$ – Lutz Lehmann Sep 16 '15 at 21:34
  • $\begingroup$ @LutzL To be clear for readers, the term $g(g(x))^{-1}$ is the reciprocal of $g(g(x))$, and not the inverse of $g(g(x))$. $\endgroup$ – Mark Viola Sep 16 '15 at 21:37
  • $\begingroup$ @LutzL Interesting comparison (will have to read up on delay differential equations). $\endgroup$ – Cicero Sep 16 '15 at 23:23
11
$\begingroup$

Look for solutions of the form $f(x)=A\,x^a$. The equation reads then $$ A\,a\,x^{a-1}=\Bigl(\frac{x}{A}\Bigr)^{1/a}. $$ You can find $A$ and $a$ fron here.

$\endgroup$
  • 5
    $\begingroup$ And the Golden ratio appears magically again! $\endgroup$ – Mark Viola Sep 16 '15 at 22:06
  • $\begingroup$ Thanks for the interesting example (the golden ratio does appear!) $\endgroup$ – Cicero Sep 16 '15 at 23:22
  • 1
    $\begingroup$ Is this solution unique (by a variation of the uniqueness theorem for certain differential eqs)? $\endgroup$ – Cicero Sep 16 '15 at 23:26
  • 1
    $\begingroup$ This answer assumes a form without proof that it's the only possible form. Are there other forms that might satisfy this diff eq? $\endgroup$ – Jerry Guern Sep 17 '15 at 6:38
  • $\begingroup$ I do not know if there are other solutions or if there is uniqueness. $\endgroup$ – Julián Aguirre Sep 17 '15 at 8:57
5
+50
$\begingroup$

Here is a partial solution using power series composition.


To solve $$ f(f'(x))=x\tag{1} $$ take the derivative of $(1)$ $$ f'(f'(x))\,f''(x)=1\tag{2} $$ If we can find a point $a\gt0$ so that $f'(a)=a$, we can expand $f'(a+x)$ around $x=0$. Let $$ g(x)=f'(a+x)-a\tag{3} $$ Applying $f'(a+x)=a+g(x)$ twice gives $$ \begin{align} f'(f'(a+x)) &=f'(a+g(x))\\ &=a+g(g(x))\tag{4} \end{align} $$ Differentiating $(3)$ yields $$ g'(x)=f''(a+x)\tag{5} $$ Plugging $(4)$ and $(5)$ into $(2)$, we get $$ g'(x)=\frac1{a+g(g(x))}\tag{6} $$ Iterating $(6)$ gives $$ \begin{align} g(x) &=\frac xa-\frac{x^2}{2a^4}+\frac{(1+3a)x^3}{6a^8}-\frac{(1+3a+10a^2+15a^3)x^4}{24a^{13}}\\ &+\frac{\left(1+3a+10a^2+30a^3+55a^4+105a^5+105a^6\right)x^5}{120a^{19}}+\dots\tag{7} \end{align} $$ Therefore, $$ \begin{align} f(x) &=a+a(x-a)+\frac{(x-a)^2}{2a}-\frac{(x-a)^3}{6a^4}+\frac{(1+3a)(x-a)^4}{24a^8}\\[6pt] &-\frac{(1+3a+10a^2+15a^3)(x-a)^5}{120a^{13}}\\ &+\frac{\left(1+3a+10a^2+30a^3+55a^4+105a^5+105a^6\right)(x-a)^6}{720a^{19}}+\dots\tag{8} \end{align} $$


Comparison with Julián Aguirre's Answer $$ \begin{align} f(x) &=\phi^{-1/\phi}x^\phi\\ &=\phi+\phi(x-\phi)+\frac{\phi-1}2(x-\phi)^2-\frac{5-3\phi}6(x-\phi)^3+\frac{18\phi-29}{24}(x-\phi)^4\\ &-\frac{217-134\phi}{120}(x-\phi)^5+\frac{1219\phi-1972}{720}(x-\phi)^6+\dots\tag{9} \end{align} $$ which matches $(8)$ with $a=\phi$.

$\endgroup$
1
$\begingroup$

It's a bit contrived, but you could take a power series approach. First, write the original ODE as \begin{equation} f(f'(x)) = x. \tag{1} \end{equation} Then, assume a solution to $(1)$ exists, and has a power series expansion around $x=0$. That is, we write \begin{equation} f(x) = \sum_{n=0}^\infty a_n x^n\quad\text{and}\quad g(x):= f'(x) = \sum_{m=0}^\infty (m+1)\, a_{m} x^{m}. \end{equation} Then, using Faà di Bruno's formula, we can write the power series of the composition \begin{equation} f(g(x)) = \sum_{n=0}^\infty c_n x^n, \end{equation} where $c_n$ is expressed in terms of $a_n$ and $n$ through combinatorial expressions. Now, using $(1)$, we see that \begin{equation} c_0 = 0,\; c_1 = 1,\;\text{and}\; c_n = 0\;\text{for}\;n\geq 2. \end{equation} This yields a system of equations for $a_n$. Again, it's a bit ugly, but might give some insight.

$\endgroup$

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.