10 12 Probability generating functions Flashcards
Assume X
assume that X is a discrete random variable on the support (state space) of non-negative integers, {0, 1, 2, . . .} (typically representing the number of individuals in a
population)
pgf of X
φ(z) =
E(zˣ) ≡ Σ{k=0,..∞} P {X = k}zᵏ
with |z| ≤ 1
E(zˣ) deontes the mean of random var zˣ
(φₓ(z))
This means that φ(z) is an increasing function when we do not have pk = 0 for all 0 < k < ∞.
pₖ
pₖ
Probability(X = k) ≡ P {X = k} = pₖ
with all 0 ≤ pₖ ≤ 1
Typical graph of pgf φₓ(z) of RV X
φ(z) against z
φₓ(0)= p₀
curve, increases as z increases
φₓ(1)=1
φₓ’(1)=E[X]
values with pgf φₓ(z)
** φₓ(0) = p₀**
** φₓ(1) = 1**
** φₓ’(1) = E[X]**
φₓ(z) =p₀+p₁z+p₂z²+p₃z³+p₄z⁴+….
φ’(z) = p₁+2p₂z+3p₃z²+4p₄z³+….
sum from k=0 to infinity of kp_k
pgf φₓ(z) for variance of RV X
** φₓ’(1) = E[X]**
** φₓ”(1)+φₓ’(1) = E[X²]**
φ’(z) = p₁+2p₂z+3p₃z²+4p₄z³+….
φ”(z) = 2p₂+6p₃z+12p₄z²+….
Var[X]=E[X²] - E[X]²
Var[X] = [ φₓ”(1)+φₓ’(1)] - (φₓ’(1))²
=φₓ”(1) + E[X][1-E[X]]
standard dev sqrt of Var
If X is the number obtained when rolling a dice, pₖ ≡ P {X = k} with k ∈ {1, . . . , 6}
What is φ(z)?
E[X]
Var[X]
pₖ = 1/6 for k=1,2,..6
φ(z) = (1/6)z+(1/6)z² +…+ (1/6)z⁶
check
φ(1)=1
φ’(1)=21/6
φ”(1)=
E[X]=21/6
Var[X]=35/12
s.d= sqrt(35/12)
RVs stochastic process
A stochastic process
{Xₜ: t ∈ T} is a collection of random variables labelled by time t (T can be discrete or continuous). For instance, {X₀, X₁, . . . } may describe the random # individuals in a population over time, and corresponds to the sample path (or stochastic realization) of the process Xₜ at time t = 0, 1, . . . . To characterize the properties
of Xₜ, we thus need a generating function that is a probability generating function of time t
and z.
probability generating function of the stochastic process
function of two variables, z and t
G(z, t) =
Σ_{k=0,..∞} pₖ(t)zᵏ
|z| ≤ 1
absolutely convergent
time dependent
probability generating function of the stochastic process time dependent useful formulae
PARTIAL DERIVATIVES
Partial derivative wrt t and z:
∂G(z, t)/∂t =
Σ_{k=0,..∞} [dpₖ(t)/dt]zᵏ
∂G(z, t)/∂z =
Σ_{k=1,..∞} pₖ(t)zᵏ ⁻¹
useful formulae:
g(z,t) = E(zˣᵗ) ( E( (z^{X_t}))
∂G(z, t)/∂z |z=1
≡ [∂G(1, t)/∂z)
= Σ_{k=1,..∞} kpₖ(t)=E(Xₜ)
why are pgfs useful?
if we can find it,
then all the p_k(t) can be deduced from G(z, t). Typically, G(z, t) is obtained by solving a suitable partial differential equation derived from the properties of G(z, t) and of its derivatives
Important values for stochastic pgf
z=0: only p₀(t) contributes to the pgf G(0,t)= p₀(t) = P(Xₜ=0) there are no individuals at time t (population extinction probability at time t)
z=1 G(1,t)=Σ_{k=0,..∞} pₖ(t = 1 by probability conservation requires finding pop at any of its states at time t =1
(when we set z=1 we set the limit from below z → 1−)
(when we set z=0 we set z → 0+ from above)
MASTER EQUATION
Example. The simple birth process
where each cell gives birth to another cell (cell division) with a rate β. When the population consists of k cells, the probability that a cell division (birth) occurs in a small interval ∆t is βk∆t.
Initially n₀ cells (pₙ₀(0) = 1)
{pₖ(t) = 0 if k < n₀,
{dpₙ₀(t)dt = −βn₀,pₙ₀(t) &
{dpₖ(t)/dt= −kβpₖ(t) + (k−1)βpₖ₋₁(t) if k > n₀,
master equation
master equation” of the stochastic process, and describes via an ODE how the probabilities of the process evolve over time
PGF
Example. The simple birth process
where each cell gives birth to another cell (cell division) with a rate β. When the population consists of k cells, the probability that a cell division (birth) occurs in a small interval ∆t is βk∆t.
G(z,t) =Σ_{k=0,..∞} pₖ(t)zᵏ found from solving below (not covered in this course)
obeys the pde:
∂G(z, t)/∂t = −βz(1 − z) (∂G(z, t)/∂z)
with initial condition
G(z,0)= zⁿ⁰ since pₖ(0)=0 for all k≠n₀
