I have the following set of ordinary differential equations: $$ \begin{cases} x_1'(s) &= -e^{-s} (1-x_1(s)) - x_2(s) + x_1(s) x_3(s)\\ x_2'(s) &= -x_2(s) + x_1(s)^2\\ x_3'(s) &= -x_3(s) + x_1(s), \end{cases} $$ with boundary condition $x_2(0)=x_3(0)=0$.

I am looking for a method to find an exact solution for this type of ODE. In particular I am interested in finding a solution for $x_1$ which satisfies $x_1(\infty) = 0$, but it already excelent if the solution of $x_1$ is simply given in function of a general boundary condition $x_1(0) = a$.

EDIT I have found this link of methods for solving this type of problems, but my problem does not seem to fit in any of the suggested methods.

  • $\begingroup$ We can always count on the numerical method... $\endgroup$ – Cesareo Sep 18 '18 at 15:27
  • $\begingroup$ Yes I know, I can of course solve it numerically. $\endgroup$ – Darkwizie Sep 19 '18 at 6:52

Some terms where missing in the first edition of the question. The wording becomes clear with the original integro-differential equation from which the system of ODEs is derived. $$x'(s)=e^{-s}(1-x(s))+\int_0^se^{-(s-u)}x^2(u)du-\int_0^se^{-(s-u)}x(u)x(s)du $$ $$\begin{cases} x_1=x(s)\\ x_2=\int_0^se^{-(s-u)}x^2(u)du \\ x_3=\int_0^se^{-(s-u)}x(u)du\ \end{cases}$$ $$ \begin{cases} x_1'(s) &= e^{-s} (1-x_1(s)) + x_2(s) - x_1(s) x_3(s)\\ x_2'(s) &= -x_2(s) + x_1(s)^2\\ x_3'(s) &= -x_3(s) + x_1(s) \end{cases} $$ We express $x_1$ and $x_2$ as functions of $x_3$. For simplification let : $$x_3=X$$ From the third equation : $$x_1=X'+X$$

From the first equation $$x_2 =(X''+X') -e^{-s} (1-X'-X) + (X'+X)X$$

From the second equation : $$ (X'''+X'')+e^{-s}(1-X'-X)-e^{-s}(-X''-X')+(X''°X')X+(X'+X) = -(X''+X')+e^{-s}(1-X'-X)-(X'+X)X+(X'+X)^2 $$ Now we have a third order ODE with only one function $X(s)$, after simplification : $$ X'''+(2+X+e^{-s})X'' +(1+X+e^{-s})X' =0 $$ This is a non-linear third order ODE. It can be simplified with the change : $$t=e^{-s}\quad;\quad X(s)=Y(t)$$ $X'(s)=Y'(t)\frac{dt}{ds}=Y'(t)(-t)=-tY'(t)$


$X'''(s)=(Y'(t)+3tY''(t)+t^2Y'''(t))(-t)= -tY'(t)-3t^2Y''(t)-t^3Y'''(t)$

$ (-tY'-3t^2Y''-t^3Y''')+(2+Y+t)(tY'+t^2Y'') +(1+Y+t)(-tY') =0 $

After simplification : $$tY'''+(-Y-t+1)Y''=0$$

The general solution seems arduous to find. But we don't need it to find some particular solutions. Obviously $Y''=0$ leads to solutions of the form : $$Y(t)=c_1t+c_2$$ where $c_1$ and $c_2$ are arbitrary constants. $$X(s)=c_1\;e^{-s}+c_2$$ Putting it into the above equations for $x_1$ and $x_2$ leads to $$\begin{cases} x_1=c_2 \\ x_2=c_1\;e^{-s}+c_2\\ x_3=c_1\;e^{-s}+c_2 \end{cases}$$ The conditions $x_2(0)=x_3(0)=0$ leads to $c_1+c_2=0$. Then putting it into the integro-differential equation shows that $c_2=1$. Thus the answer is $$\begin{cases} x_1(s)=1 \\ x_2(s)=1-e^{-s}\\ x_3(s)=1-e^{-s} \end{cases}$$ This is a very simple solution.

This rightly answers to the question insofar the third condition is $x_1(0)=1$. No condition on $x_1$ was specified in the wording of the problem.

Note : If there was a third specified condition $x_1(0)\neq 1$ the above very simple solution would be no longer convenient. In this case we have to fully solve the ODE $\:tY'''+(-Y-t+1)Y''=0$ in order to have the general solution which must involve three arbitrary constants $c_1$ , $c_2$ and $c_3$ to be determined according to the boundary conditions. This would be more arduous.

| cite | improve this answer | |
  • $\begingroup$ The original problem statement is as follows: $x'(s) = e^{-s}(1-x(s)) + \int_0^s e^{-(s-u)} x^2(u) du - \int_0^s e^{-(s-u)} x(u) du x(s)$ I then defined $x_1(s)=x(s)$, $x_2(s) = \int_0^s e^{-(s-u)} x_1^2(u) du$ and $x_3(s)=\int_0^s e^{-(s-u)} x(u) du$, which allows to reduce the given integro differential equation to an ordinary differential equation. $\endgroup$ – Darkwizie Sep 28 '18 at 10:24
  • $\begingroup$ OK, the integro-differential equation is solvable. Tomorrow I will re-edit my answer: A very simple solution. $\endgroup$ – JJacquelin Sep 28 '18 at 21:43
  • $\begingroup$ That is very good news! If you can just give me a hint (or several hints) such that I can derive the solution myself, I would be very happy! You can of course simply edit your answer and give the whole solution if you prefer to do that. $\endgroup$ – Darkwizie Sep 29 '18 at 5:57
  • $\begingroup$ In order to avoid mismatch of signs in the first equation, I prefer to re-start all from the integro-differential equation. See my updated answer. $\endgroup$ – JJacquelin Sep 29 '18 at 8:20
  • $\begingroup$ I am sorry but you made a mistake in the IDE: the last term should read $\int _0^s e^{-(s-u)} x(u) du x(s)$. I Think you neglected the $x(s)$ $\endgroup$ – Darkwizie Sep 29 '18 at 8:37

$$ \begin{cases} x_1'(s) &= x_2(s) - x_1(s) x_3(s)\\ x_2'(s) &= -x_2(s) + x_1(s)^2\\ x_3'(s) &= -x_3(s) + x_1(s), \end{cases} $$ We express $x_1$ and $x_2$ as functions of $x_3$. For simplification let : $$x_3=X$$ From the third equation : $$x_1=X'-X$$ From the first equation $$x_2=X''-X'+(X'-X)X$$ From the second equation : $$X'''-X''+(X''-X')X+(X'-X)X'=-(X''-X'+(X'-X)X)+(X'-X)^2$$ Now we have a third order ODE with only one function $X(s)$. $$X'''+X''X+(-X'+X-1)X'-X^2=0$$ This is an autonomous ODE. The change of function is : $$X'(s)=F(X) \quad;\quad X''(s)=F'(X)F(X) \quad;\quad X'''(s)=F''F^2+F'^2F$$ $$F''F^2+F'^2F+XF'F+(-F+X-1)F-X^2=0$$ This is a second order non-linear ODE difficult to solve. Probably there is no closed form general solution with the standard functions.

Some particular solutions might be found. For example $F(X)=X$ is a particular solution. But this solution which is not compatible with the boundary condition $x_3(0)=0$ must be rejected.

If the question comes from an academic exercise, a typo might be suspected in the given equations. Especially the third equation is doubtful. Are you sure that nothing is missing or wrong in the third equation ?

| cite | improve this answer | |
  • $\begingroup$ Thank you for this very nice answer! I had indeed forgotten a term, see my edited question. Also I added some more information : I would mainly like to find the solution for $x_1$ and $x_1$ can have a general boundary condition not necessarily $x_1(0)=1/2$ (Of course, I think that if you can solve it for $x_1(0)=1/2$ I can continue from there..) I am sorry that I have edited the question. If no answer is given in the coming 5 days which are better than yours I will give you the bounty. $\endgroup$ – Darkwizie Sep 27 '18 at 22:08
  • $\begingroup$ The corrections were made in the equation for $x_1$. This equation should have read: $x_1'(s)= - e^{-s} (1-x_1(s)) - x_2(s) +x_1(s) x_3(s)$ $\endgroup$ – Darkwizie Sep 27 '18 at 22:11

Just a beginning, you can eliminate $x_3$ $$ \begin{cases} x_1' &= x_2 - x_1 x_3\\ x_2' &= -x_2 + x_1^2\\ x_3' &= -x_3 + x_1, \end{cases} $$

Sum the first and the second: $$x_1'+x_2'=x_1^2-x_1 x_3=x_1(x_1-x_3)$$

So we have

$$ \begin{cases} x_3' &= {x_1' +x_2' \over x_1} \\ x_3 &= x_1 - {x_1' +x_2' \over x_1} , \end{cases} $$

| cite | improve this answer | |
  • $\begingroup$ This is indeed a nice start, I will see if I can build on this! $\endgroup$ – Darkwizie Sep 19 '18 at 6:52
  • $\begingroup$ After taking a second look, I don't see how this eliminates $x_3$, if we substitute the found expression for $x_3$ into the equation for $x_1'$ I simply obtain the equation $x_2'=-x_2+x_1^2$? $\endgroup$ – Darkwizie Sep 19 '18 at 11:07
  • $\begingroup$ As the system is autonomous this is expected. You should first differentiate one of the other two expressions and then substitute. Differentiating the second I arrived at $\sqrt {x_2'+x_2}={x_2''+x_2' \over x_2-x_2'}$ but I'm not sure of this result. $\endgroup$ – N74 Sep 19 '18 at 11:13
  • $\begingroup$ But then the order of the ODE increases by one, which is classically resolved by introducing a new function $x_3=x_2'$? $\endgroup$ – Darkwizie Sep 19 '18 at 11:18
  • $\begingroup$ It is one way to solve it, but there are also methods to solve higher order equations directly. For your system I think that is the best way. $\endgroup$ – N74 Sep 19 '18 at 11:41

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.