I am trying to figure out how exactly to determine the rate of convergence for a stochastic dynamical system.

I am given a stochastic equation, i.e., a quadratic SDE, and I am interested to see how to determine the velocity of convergence near a stationary point for which I already know it does not converge exponentially. I can't seem to find a proper definition or even a proper way to determine this.

What my professor has given me riddles me a little. He suggested that I take the SDE, which is given in terms of $\dot{p}$ and take a look at $\frac{d}{dt}\|p - p^*\|^2$ where $p^*$ is a statement to $\dot{p} = H(p)=0$, which makes it an equilibrium or stationary point (but with negative eigenvalues, so it does not attract). On the other hand the internet suggests rather imprecisely that one should regard $$ \| \dot{p}-p^*\|$$ rather than taking the derivative outside the norm. This produces different results.

Although I am quite aware that the differences may well be insignificant, but here is basically where i am stuck. How do I determine a proper convergence rate? I mean I only know that despite having negative eigenvalues my dynamical system converges (with a certain probability, but right now the stochastical perspective is of less interest)

I am thankful for any help

We consider the one dimensional Riccati equation $$\dot{p}= -2ap + r_1 - p^2 \frac{b^2}{r_2},$$ where $b, r_2,r_1$ are positive. I have already considered the case in which a is positive and that leaves a good solution which is exponentially stable and converges at at least exponential rate. But now the case for which a is negative seems tricky, or rather I am not sure what I am to compute here to ascertain a convergence rate.

The stochastic contribution lies in $a$. for now i am considering a very simple case, namely that $a$ is constant. later i want to consider the case where $a$ is the derivative of a differientible function $A$ where it is $A(X_t)$ with a stochastic process $X_t$. $a$ will then be bounded. The last step will be to see the multi dimensional case. My aim is to see whether or not I can get a strong solution as in the paper(see comments) by allowing negative $a$. To research here i need to know what is a decent way to compute the convergence rate.

Also referring to the comment, i have tried to solve the equation by substitution but i am having problems resubstituting. This and the fact that the multidimensional case do no posess known general solutions leave me opting to get a convergence rate here.

  • 1
    For reference, please give the full SDE. Is there a link to the internet source? – LutzL Jun 12 at 12:58
  • Basically it's a Riccati equation: – Danny Jun 12 at 15:14
  • In what sense is that a stochastic DE? I can see no noise term. Is this equation scalar? Then you can solve it via separation of variables. – LutzL Jun 12 at 15:49
  • i have already solved this. the more general setting with the stochastic term is found in this paper arxiv.org/abs/1606.08251 – Danny Jun 12 at 15:53

Your Answer


By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.