# Linear birth death process, probability of extinction by time t

I have a linear birth death process with birth rates $\lambda n$ and death rates $\mu n$ .

Let r(t) be the probability of extinction by time t. If there is 1 individual alive at time 0 explain why

r(t) = $\int^t_0 e^{-(\mu+\lambda)s}(\mu+\lambda r(t-s)^2)ds$

With r(0)=0 I have done this but the next bit is to derive a differential equation for r and solve it (at least in the case where $\mu\neq\lambda)$ and show that if $\mu>\lambda$ then r(t)--> 1 whereas if $\mu<\lambda$ then r(t) --> $\mu/\lambda$

I wanted to proceed as we do similar things in lectures.

So step 1 : multiply by $e^{(\lambda +\mu)t}$

This gives $e^{(\lambda +\mu)t}r(t)$ = $\int^t_0 e^{(\mu+\lambda)(t-s)}(\mu+\lambda r(t-s)^2)ds$

Then I let t-s = k to get $e^{(\lambda +\mu)t}r(t)$ = $\int^t_0 e^{(\mu+\lambda)(k)}(\mu+\lambda r(k)^2)dk$

I then differentiate both sides $e^{(\lambda +\mu)t}r'(t)+ (\lambda +\mu)r(t)e^{(\lambda +\mu)t}$ = $e^{(\mu+\lambda)(t)}(\mu+\lambda r(t)^2)$

Divide all terms by $e^{(\mu+\lambda)(t)}$ Rearrange to get $dr/(r(t)(r(t)\lambda -\lambda -\mu)) = dt\mu$

Then separate into partial fractions

This gives $[-1/(\lambda+\mu r(t))] + [\lambda/((\lambda+\mu)(r(t)\lambda-\lambda-\mu))] = dt\mu$

Integrate both sides between t and 0

$[-1/(\lambda+\mu).lnr(t) + 1/(\lambda+\mu).ln(r(t)\lambda-\lambda-\mu)]^{r(t)}_{r(0)}= \mu t$

When I evaluate this at the limits and rearrange I get $[\lambda r(t)-\lambda-\mu]/r(t) = -(\mu + \lambda)e^{\mu t(\lambda + \mu)} = D$

Finally I rearrange to get $r(t) = (\lambda+\mu)/(\lambda - D)$ which does not have the qualities required, what have a done wrong?

-

The trouble is that you rearrange wrong... In fact, $$r'(t)+(\lambda+\mu)r(t)=\mu+\lambda r(t)^2,$$ hence $$\frac{r'(t)}{\lambda r(t)^2-(\lambda+\mu)r(t)+\mu}=1.$$ The decomposition into simple fractions reads $$\frac1{\lambda r^2-(\lambda+\mu)r+\mu}=\frac1{\lambda -\mu}\left(\frac1{r-1}-\frac{\lambda}{\lambda r-\mu}\right).$$ Can you finish?
thanks for your help, in the case of $\mu < \lambda$ is it right to use L'Hopitals rule? Thanks – Rosie Mar 6 '13 at 22:17