5
$\begingroup$

I'm having troubles in simplifying a differential equation to find its solutions. Consider this ODE: $$ \frac{1}{r} \, \frac{d}{dr} \Bigl( \frac{r}{B} \, \frac{d B}{dr} \Bigr) = k \, B^2, \tag{1} $$ where $k$ is just a real constant and $B(r)$ is the function to be found. I already know one solution of this equation (it was found by guessing, trial and errors): $$ B(r) = \frac{\beta}{1 + \lambda r^2}, \tag{2} $$ where $\beta$ and $\lambda = k \beta^2 / 4$ are arbitrary constants. My goal is to transform (1) into an integral so I could find back the solution (2). I'm yet unable to find a change of variable that makes that equation solvable analyticaly. I tried $u = 1/B$, and $B = e^{\phi(r)}$ (and few other trials). Any idea how to solve (1)?

$\endgroup$
4
  • $\begingroup$ You already found a one-parameter family of solutions (which is no small feat). What is left to do is to show that there are no other solutions. Two more remarks: (1) Your solution is for the right hand side $-kB^2$, not $+kB^2$. (b) Set $u(r) = \log B(r)$. Then $u$ is a solution of the Gelfand problem $\Delta u + ke^ {2u}= 0$ in 2 dimensions, using polar coordinates. But you probably knew this already :) $\endgroup$ Jun 2, 2020 at 19:15
  • $\begingroup$ @HansEngler, your comment suggest that there is no procedure or change of variable to transform (1) into an elementary integral, is that right? $\endgroup$
    – Cham
    Jun 2, 2020 at 19:44
  • $\begingroup$ I am not aware of one. The linear version, essentially $\frac{1}{r}\frac{d}{dr}(r u(r)) + k u = 0$, is a Bessel equation with non-elementary solution. $\endgroup$ Jun 2, 2020 at 21:01
  • $\begingroup$ I would be interested in any other analytical solutions to (1), which are not special cases of (2). $\endgroup$
    – Cham
    Jun 3, 2020 at 0:41

1 Answer 1

4
$\begingroup$

Since the equation features only multiplication and division, it is worth checking whether it has some kind of homogeneous continuous group of symmetries. Indeed, it is straight-forward to check that the one-parameter Lie group of transformations \begin{align} & r \, \mapsto \, e^{s}r\\ & B \, \mapsto \, e^{-s}B \end{align} leaves the equation $$\frac{1}{r} \frac{d}{dr}\left(\frac{r}{B} \frac{dB}{dr}\right) \,=\, k\, B^2$$ invariant (unchanged) Furthermore, the quantity $$C = rB$$ is also invariant under the action of the one-parameter group of symmetries. The vector field that generates this group of transformation is $$r\,\frac{\partial }{\partial r} \, -\, B\,\frac{\partial }{\partial B}$$ Now, using the invariant quantity $C$, we can construct a change of variables (a local diffeomorphism) that rectifies the latter vector field. In other words, if we define the change of variables \begin{align} &s = \ln(r)\\ &C = rB \end{align} with inverse \begin{align} &r = e^s\\ &B = e^{-s}\,C \end{align} the vector field above transforms into $$r\,\frac{\partial }{\partial r} \, -\, B\,\frac{\partial }{\partial B} \, =\, \frac{\partial }{\partial s}$$ and so, in the new coordinates $s, \,C$ the original equation should be invariant under the simple translation-transformation \begin{align} &s \mapsto s + \tilde{s}\\ &C \mapsto C \end{align} which means that the transformed equation should not depend explicitly on the new $s$ variable. Indeed, since $dr = e^s ds$ we can write the equation in the new coordinates as follows: $$\frac{1}{r} \frac{d}{dr}\left(\frac{r}{B} \frac{dB}{dr}\right) \,=\, k\, B^2$$ $$\frac{1}{e^s} \, \frac{d}{e^sds}\left(\frac{e^s}{e^{-s}\,C} \frac{d}{e^s ds}\Big(e^{-s}\,C\Big)\right) \,=\, k\, \big(e^{-s}\,C\big)^2$$ $$e^{-2s} \, \frac{d}{ds}\left(\frac{e^s}{C} \frac{d}{ds}\Big(e^{-s}\,C\Big)\right) \,=\, k\, e^{-2s}\,C^2$$ $$\frac{d}{ds}\left(\frac{e^s}{C} \frac{d}{ds}\Big(e^{-s}\,C\Big)\right) \,=\, k\, C^2$$ $$\frac{d}{ds}\left(\frac{e^s}{C} \Big(e^{-s}\,\frac{dC}{ds} - e^{-s}\,C\Big)\right) \,=\, k\, C^2$$ $$\frac{d}{ds}\left(\frac{e^s}{C} \, e^{-s}\, \Big(\frac{dC}{ds} - C\Big)\right) \,=\, k\, C^2$$ $$\frac{d}{ds}\left(\frac{1}{C}\, \Big(\frac{dC}{ds} - C\Big)\right) \,=\, k\, C^2$$ $$\frac{d}{ds}\left(\frac{1}{C}\, \frac{dC}{ds} - 1\right) \,=\, k\, C^2$$ so now it is clear that the equation in the new variables is $$\frac{d}{ds}\left(\frac{1}{C}\, \frac{dC}{ds}\right) \,=\, k\, C^2$$ and does not depend on the variable $s$ explicitly. The next step is to introduce the extra variable $$P \, =\, \frac{1}{C}\, \frac{dC}{ds}$$ which turns the equation into $$\frac{dP}{ds} \, =\, k\,C^2$$ and as a result of this substitution, we obtain the following system of differential equations \begin{align} &\frac{dC}{ds} \, =\, C \, P\\ &\\ &\frac{dP}{ds} \, =\, k\,C^2 \end{align} We can try to find a first integral of the latter system as follows \begin{align} &\frac{dP}{dC} \, =\, \frac{k\,C^2}{C\,P} \end{align} \begin{align} &\frac{dP}{dC} \, =\, \frac{k\,C}{P}\\ \end{align} $$P\,dP \, =\, k\,C \,dC $$ so after integrating both sides $$P^2 \,=\, k\,C^2 \, +\, b$$ where $b$ is an arbitrary constant. Substituting back $$P \,=\, \frac{1}{C} \, \frac{dC}{ds}$$ we obtain the separable, integrable, differential equation $$\left(\frac{1}{C}\,\frac{dC}{ds}\right)^2 \, =\, k\,C^2 \, +\, b$$ $$\frac{1}{C}\,\frac{dC}{ds}\, =\, \pm \sqrt{\,k\,C^2 \, +\, b\,}$$ $$\frac{dC}{ds}\, =\, \pm\, C\,\sqrt{\,k\,C^2 \, +\, b\,}$$ The solution to the latter equation can be obtain by integrating $$\pm\int \, \frac{dC}{C\,\sqrt{\,k\,C^2 \, +\, b\,}} \, =\, \int ds$$ and, consequently, it can be written in implicit form as follows $$\mp\,\frac{1}{\sqrt{b}}\, \tanh^{-1}\left(\sqrt{\,1 \, + \,\frac{k}{b} \,C^2\,}\right) \, =\, s - a$$ where $a$ is another arbitrary constant. Rewrite it $$\sqrt{\,1 \, + \,\frac{k}{b} \,C^2\,} \, =\,\tanh\Big( \mp \sqrt{b}\,(s - a)\,\Big)$$ $$1 \, + \,\frac{k}{b} \,C^2\, =\,\tanh^2\Big( \mp \sqrt{b}\,(s - a)\,\Big)$$ $$\frac{k}{b} \,C^2\, =\,\tanh^2\Big( \mp \sqrt{b}\,(s - a)\,\Big) \, -\, 1$$ and solve for $C$ $$C\, =\,\pm\,\sqrt{\,\frac{b}{k}\,\tanh^2\Big( \mp \sqrt{b}\,(s - a)\,\Big) \, -\, \frac{b}{k}\,}$$ Finally, return to the original variables \begin{align} &s = \ln(r)\\ &C = rB \end{align} which leave us with the solution $$r\,B\, =\,\pm\,\sqrt{\,\frac{b}{k}\,\tanh^2\Big( \mp \sqrt{b}\,\big(\ln(r) - a\big)\,\Big) \, -\, \frac{b}{k}\,}$$ or explicitly $$B\, =\,\pm\,\frac{1}{r}\sqrt{\,\frac{b}{k}\,\tanh^2\Big( \mp \sqrt{b}\,\big(\ln(r) - a\big)\,\Big) \, -\, \frac{b}{k}\,}$$ After relabling the constants $\alpha = \mp\,a \,\sqrt{b}$ and $\beta = \mp\,\sqrt{b}$, one can write the final solution as follows $$B\, =\,\pm\,\frac{1}{r}\sqrt{\,\frac{\beta^2}{k}\,\tanh^2\Big( \beta\,\ln(r) - \alpha\,\Big) \, -\, \frac{\beta^2}{k}\,}$$ $$B\, =\,\pm\,\frac{\beta}{r}\sqrt{\,-\,\frac{1}{k \, \cosh^2\Big( \beta\,\ln(r) - \alpha\,\Big)} \,}$$ $$B\, =\,\pm\,\frac{\beta}{\sqrt{-\,k\,}\,}\,\frac{1}{\,r \, \cosh\Big( \beta\,\ln(r) - \alpha\,\Big)}$$ $$B\, =\,\pm\,\frac{\beta}{\sqrt{-\,k\,}\,}\,\frac{1}{\,r \, \cosh\Big( \beta\,\ln(r) - \alpha\,\Big)}$$ so maybe it simplifies to something like this $$B\, =\,\,\frac{\beta}{\sqrt{-\,k\,}\,}\, \frac{2 \, \lambda \,\, r^{\beta-1}}{\,1 \, + \, \lambda^2 \,\, r^{2\beta}}$$ where the constant $\lambda = \pm e^{-\alpha}$

$\endgroup$
3
  • $\begingroup$ The last expression could be simplified alot! $\endgroup$
    – Cham
    Feb 1 at 18:52
  • 1
    $\begingroup$ @Cham I know. After all, $\tanh$ is a fraction of exponents, but I did not feel like going further. I have focused mostly on the systematic method for solving the equation, as requested. $\endgroup$ Feb 1 at 18:57
  • $\begingroup$ @Cham I included some simplification. $\endgroup$ Feb 1 at 22:22

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.