08.10.2016 Views

Foundations of Data Science

2dLYwbK

2dLYwbK

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

for all δ greater than δ critical . If δ is less than δ critical , then all vertices are in finite components.<br />

In this case g(1) = 1 and both the numerator and the denominator approach zero.<br />

Appling L’Hopital’s rule<br />

lim<br />

x→1 g′ (x) = 1 2δ<br />

xg ′ (x)−g(x)<br />

x 2<br />

g ′ (x)<br />

or<br />

(<br />

(g ′ (1)) 2 = 1 2δ g ′ (1) − g(1) ) .<br />

The quadratic (g ′ (1)) 2 − 1 2δ g′ (1) + 1 g(1) = 0 has solutions<br />

2δ<br />

√<br />

1<br />

± 1<br />

− 4<br />

g ′ 2δ 4δ<br />

(1) =<br />

2 2δ<br />

= 1 ± √ 1 − 8δ<br />

. (4.7)<br />

2<br />

4δ<br />

The two solutions given by (4.7) become complex for δ > 1/8 and thus can be valid only<br />

for 0 ≤ δ ≤ 1/8. For δ > 1/8, the only solution is g ′ (1) = 1 and a nonfinite component<br />

2δ<br />

exists. As δ is decreased, at δ = 1/8 there is a singular point where for δ < 1/8 there are<br />

three possible solutions, one from (4.6) which implies a giant component and two from<br />

(4.7) which imply no giant component. To determine which one <strong>of</strong> the three solutions is<br />

valid, consider the limit as δ → 0. In the limit all components are <strong>of</strong> size one since there<br />

are no edges. Only (4.7) with the minus sign gives the correct solution<br />

g ′ (1) = 1 − √ 1 − 8δ<br />

= 1 − ( 1 − 18δ − 1 2 4 64δ2 + · · ·)<br />

= 1 + 4δ + · · · = 1.<br />

4δ<br />

4δ<br />

In the absence <strong>of</strong> any nonanalytic behavior in the equation for g ′ (x) in the region<br />

0 ≤ δ < 1/8, we conclude that (4.7) with the minus sign is the correct solution for<br />

0 ≤ δ < 1/8 and hence the critical value <strong>of</strong> δ for the phase transition is 1/8. As we shall<br />

see, this is different from the static case.<br />

As the value <strong>of</strong> δ is increased, the average size <strong>of</strong> the finite components increase from<br />

one to<br />

1 − √ 1 − 8δ<br />

4δ ∣ = 2<br />

δ=1/8<br />

when δ reaches the critical value <strong>of</strong> 1/8. At δ = 1/8, the average size <strong>of</strong> the finite components<br />

jumps to 1 ∣<br />

2δ δ=1/8<br />

= 4 and then decreases as 1 as the giant component swallows<br />

2δ<br />

up the finite components starting with the larger components.<br />

Comparison to static random graph<br />

Consider a static random graph with the same degree distribution as the graph in the<br />

growth model. Again let p k be the probability <strong>of</strong> a vertex being <strong>of</strong> degree k. From (4.5)<br />

p k =<br />

(2δ) k<br />

(1 + 2δ) k+1 .<br />

121<br />

∣<br />

∣<br />

x=1

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!