I've come across this problem in the course of my work, and I'm a bit stuck on it.

Suppose I have an $n$-dimensional smooth dynamical system $$ \dot{x_i} = f_i(\mathbf{x}), $$ and suppose there is a fixed point $\hat{\mathbf{x}}$, i.e. $f_i(\hat{\mathbf{x}})=0$ for every $i$.

Now let us define a new dynamical system $$ \dot{x_i} = f'_i(\mathbf{x}), $$ where $f_i'(\mathbf{x}) = f_i(\mathbf{x}) + \epsilon\delta f_i(\mathbf{x})$, for some set of arbitrary smooth functions $\delta f_i$.

It's possible that the fixed point of the original system was a bifurcation point, so that there will be no corresponding fixed point in the new system, no matter how small we make $\epsilon$. But supposing this isn't the case, the fixed point should just move to a slightly different position. That is, $\hat{\mathbf{x}}' = \hat{\mathbf{x}} + \delta \hat{\mathbf{x}}$, where every $f_i'(\hat{\mathbf{x}}')=0$ and $\delta\hat{\mathbf{x}}$ will be of the same order as $\epsilon$.

What I want is an expression for $\delta\hat{\mathbf{x}}$ (in the small $\epsilon$ limit) in terms of $\epsilon$ and the partial derivatives of the functions $f_i$ and $\delta f_i$ at the original fixed point. It seems like it should be possible, but after going cross-eyed scribbling on the whiteboard I haven't come up with the answer. I'm probably just missing some simple trick - can anyone help me out?


I will in the following do nothing but simple calculus trick. Assume $f'(x)=f(x)+\epsilon g(x)$ where $g(x)$ is arbitrary smooth function. Denote fixed points of two dynamical systems by $x_0,x_0'$ so that $f(x_0)=f'(x_0')=0$ and $\delta x=x_0'-x_0$. Then we have $$\begin{align}f'(x'_0)&=f(x_0+\delta x)+\epsilon g(x_0+\delta x)\\ &=f(x_0)+Jf(x_0)\delta x+\epsilon\left(g(x_0)+Jg(x_0)\delta x\right)\\ &=Jf(x_0)\delta x+\epsilon g(x_0)\\ &=0\end{align}$$ Hence $\delta x=-\epsilon(Jf)^{-1}(x_0)g(x_0)$ where $Jf$ is Jacob matrix. Note that in the second line, $f(x_0)=0$ and $\epsilon\delta xJg(x_0)$ is infinitesimal compared to $\epsilon$.

  • $\begingroup$ Thanks - I knew it would be simple, I just wasn't looking at it in the right way. In the multivariate case it becomes $\delta \mathbf{x} = -\epsilon J^{-1}\mathbf{g}(\mathbf{x}_0)$, where $J$ is the Jacobian matrix, i.e. $J_{ij} = \partial f_i/\partial x_j$, or $(J^{-1})_{ij} = \partial x_i/\partial f_j$. $\endgroup$ – Nathaniel Nov 19 '13 at 4:38
  • $\begingroup$ @Nathaniel Oh yes, it should be an inverse Jacob matrix. My naive symbols misled myself. $\endgroup$ – Shuchang Nov 19 '13 at 5:23

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.