I am trying to find the solution to the following equation, $\epsilon x^3 -x^2 +x-\epsilon^{\frac{1}{2}}=0$, for the first two non-zero solutions as $\epsilon \to 0^+$.

I have used the principal of dominant balance (by a Kruskal Newton Graph) to find that my ansatz should be of the form $x(\epsilon)=\dfrac{z(\epsilon)}{\epsilon^\alpha}$ where $\alpha=0,\dfrac{1}{2},1$.

Now if we substitute these into our equation we get, for each $\alpha$:

$\alpha=0 : \ \epsilon x^3 -x^2 +x-\epsilon^{\frac{1}{2}}=0 $ - noting $x=z$;

$\alpha=\dfrac{1}{2}: z\epsilon^{\frac{1}{2}}-z^2+z\epsilon^\frac{1}{2}-\epsilon^\frac{3}{2}$;

$\alpha=1: \ z^3-z^2+z\epsilon-\epsilon^\frac{1}{2}$.

Now it is clear that if we perturb our equations in the form $z(\epsilon)=z_0+\epsilon z_1 +...$ then we have, for $\alpha=0 \ \& \ 1$, an $\epsilon^\frac{1}{2}$ term which can only be equated to zero. Does this mean that we do not have a valid solution for these choices of $\alpha$?

Whereas, if we consider $\alpha=\dfrac{1}{2}$ then we have an equation of the form: $(z_0+\epsilon z_1)^3\epsilon^\frac{1}{2} -(z_0+\epsilon z_1)^2 +(z_0+\epsilon z_1)\epsilon^\frac{1}{2}-\epsilon^\frac{3}{2}=0$ whose coefficients can all be equated, thus valid solutions.

As a result is the only case where we can find solutions at $\alpha=\dfrac{1}{2}$?

  • 1
    $\begingroup$ There are 6 possible balances, three involve $\alpha<0$. $\endgroup$ – David Oct 31 '18 at 21:55
  • $\begingroup$ Yes, you're right - I missed $\alpha =-0.25,-0.5,\dfrac{1}{6}$. I will update this post sometime in the next two days. I assume that I should then get a balance for each order of \epsilon somewhere? $\endgroup$ – KieranSQ Oct 31 '18 at 22:01

Your original equation is $$\epsilon x^3 -x^2 +x-\epsilon^{\frac{1}{2}}=0. \tag1$$ To simplify a bit we define $\, \delta := \sqrt{\epsilon}.\,$ The equation now becomes $$ \delta^2 x^3 - x^2 + x - \delta = 0. \tag2$$

For any $\, \delta \ne 0, \,$ there are three rots of the cubic. If we try a small value for $\, \delta, \,$ such as $\, \delta = 0.01, \,$ the three roots are easily determined to be approximately $$ x_1 \approx \delta + \delta^2, \quad x_2 \approx 1 - \delta, \quad x_3 \approx \delta^{-2} - 1. \tag3$$

It is now easy to find further terms of the Puiseux series expansions. Note that for any $\, \epsilon>0, \,$ there are two values for its square root $\, \delta,\,$ and this gives six roots of equation $\,(1)\,$ since its roots in terms of $\, \delta \,$ are different if we take the negative square root.

You may be interested in ways to increase precision of approximations to the roots. For example, for the root $\,x_1,\,$ the function $\, f_1(t) := \delta + t^2 - \delta^2t^3 \,$ has $\,x_1\,$ as a fixed point. Using iteration we get the sequence of approximations to $\,x_1\,$ as follows: $$ \delta + O(\delta^2) \, \mapsto \, \delta + \delta^2 + O(\delta^3) \, \mapsto \, \delta + \delta^2 + 2\delta^3 + O(\delta^4) \, \dots. $$

Similarly, for the root $\,x_2,\,$ the function $\, f_2(t) := (1+\sqrt{1 - 4\delta + 4\delta^2 t^3})/2 \,$ has $\,x_2\,$ as a fixed point. Using iteration we get the sequence of approximations to $\,x_2\,$ as follows: $$ 1 + O(\delta) \, \mapsto \, 1 - \delta + O(\delta^3) \, \mapsto \, 1 - \delta - 3 \delta^3 - 3\delta^4 + O(\delta^5) \, \dots.$$

The remaining root $\, x_3 = \delta^{-2} - x_1 - x_2 = \delta^{-2} - 1 - \delta^2 + \delta^3 - 2\delta^4 + O(\delta^5). \,$


Following your approach $x=ϵ^{-α}z$ gives the equation for $z$ as $$ ϵ^{1-α}z^3-z^2+ϵ^{α}z-ϵ^{2α+\frac12} $$ The corners of the concave curve $\min\{1-α,0,α,2α+\frac12\}$ are at $α=-\frac12, 0, 1$. plot of the minimum curve

  • For $α=-\frac12$ we get in separation of medium and small terms $$ z=1-ϵ^2z^3+ϵ^{\frac12}z^2 $$ which successively gives $z_0=1$, $z_1=1+ϵ^{\frac12}$, $z_2=1+ϵ^{\frac12}+2ϵ$, ... so that $$x=ϵ^{\frac12}+ϵ+2ϵ^{\frac32}+\dots$$

  • For $α=0$, $$z=1-ϵ^{\frac12}z^{-1}+ϵz^2$$ so that again adding a term per iteration $z_0=1$, $z_1=1-ϵ^{\frac12}$, $$x=z_2+O(ϵ^2)=1-ϵ^{\frac12}-3ϵ^{\frac32}+...$$

  • For $α=1$, $$z=1-ϵz^{-1}+ϵ^{\frac52}z^{-2}$$ so that $z_1=1-ϵ$, $z_2=1-ϵ-ϵ^2$, $$x=ϵ^{-1}-1-ϵ+...$$


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.