Handout on a Critical Branching Process


Problem. Consider a branching process with the following probabilities: pj=P(N1=j)=(½)j+1j=0,1,2,.... Find the probability ej that the process dies out by generation j,  and also find the distribution of Nj.
Compute the conditional istribution of
Nj conditioned on the event {Nj > 0}, and find the conditional expected number of members of generation j, that is E(Nj | Nj > 0).

Solution
First, note that this is a critical branching process - compute the average number of particles
   
E N1=1/2×0 + 1/4 ×1 +1/8 ×2+1/16 × 3+ .... = 1  (exercice!)
Hence the process must eventually die out.

Next, observe that
    G(s)=1/2 + 1/4
s+1/8 s2+1/16 s3+ .... = 1/2 ( 1+ [s/2] +1/8 [s/2]2+1/16 [s/2]3+....=1/2 (1-s/2)-1=1/(2-s)

Therefore,
    G2(s)=G(G(s))=(2-s)/(3-2s)
    G3(s)=G(G(G(s)))=(3-2s)/(4-3s)

and so on - by induction it is easy to show that
    Gj(s)=[j-(j-1)s] / [(j+1)-js]
Hence
    jGj(s)-(j-1) = (j+1-js)-1 = (j+1)-1 (1 - rjs)-1[1+(rjs)+(rjs)2+(rjs)3+...] / (j+1)
where
    rj=j/(j+1)
Therefore,
    Gj(s)=[j2+ (rjs)+(rjs)2+(rjs)3+... ] / (j+j2)
Now we can "decode" the p.m.f. of
Nj using the p.g.f. Namely:
ej = P(Nj=0)=j2 / (j+j2)  = j / (j+1)
which approaches 1.
Moreover,
   
P(Nj=1)=(rj)  / (j+j2
    P(Nj=2)=(rj)2 / (j+j2
and so on, thus giving
    P(Nj=k)=(rj)k  / (j+j2
for k=1,2,3,....

Finally, since
    1 - P(Nj=0) = sum ( P(Nj=k), k=1,2,3,... )  = sum ( (rj)k  / (j+j2) , k=1,2,3,...)
                        =
1 / (j+j2) × 1 / (1-rj) = 1 / (j+1)
we have
   
P(Nj=k | Nj>0 )=(rj)k  / (j+j2)  / P(Nj>0) = (rj)k  / j
for
k=1,2,3,.... and hence
  
E (Nj=k | Nj>0 ) = sum ( P(Nj=k | Nj>0 ) × k , k=1,2,3,...) = sum ( k (rj)k  / j , k=1,2,3,...) = j+1

Thus the conditional (given survival) expected number of particles goes to infinity!

Note: useful formula:
    sum
( ak × k , k=1,2,3,...= a sum ( ak-1 × k , k=1,2,3,...) = a sum ( ak , k=1,2,3,...)' =
                                               =
a sum ( ak , k=0,1,2,3,...)' a ( (1-a)-1 )' = a / (1-a)2