1: \begin{abstract}

2: Recently, the scaling limit of cluster sizes for critical inhomogeneous random graphs

3: of rank-1 type having finite variance but infinite third moment degrees was obtained \cite{BhaHofLee09b}.

4: It was proved that when the degrees

5: obey a power law with exponent $\tau\in (3,4)$, the sequence of clusters

6: ordered in decreasing size and multiplied through by $n^{-(\tau-2)/(\tau-1)}$

7: converges as $n\rightarrow \infty$ to a sequence of decreasing non-degenerate random variables.

8: 

9: Here, we study the tails of the limit of the rescaled largest cluster, i.e., the

10: probability that the scaling limit of the largest cluster takes

11: a large value $u$, as a function of $u$. This extends a related result of

12: Pittel \cite{Pitt01} for the Erd\H{o}s-R\'enyi random graph

13: to the setting of rank-1 inhomogeneous random graphs

14: with infinite third moment degrees. We make use of delicate large deviations

15: and weak convergence arguments.

16: 

17: %It is well known that

18: %removing the largest critical cluster in the Erd\H{o}s-R\'enyi random graph shifts

19: %the location in the critical window by an amount proportional to the size of the largest

20: %connected component. This explains that, to leading order, the

21: %probability of having two large clusters of size $u$ and $a u$ with $u$ large

22: %in the Erd\H{o}s-R\'enyi random graph

23: %is similar to the product of the probabilities of having a largest cluster

24: %of sizes $u$ and $a u$ respectively. In our inhomogeneous setting,

25: %the effect of removing a large cluster is far more dramatic, and

26: %thus the probability of having two large clusters is much smaller

27: %than the product of the probabilities of having a largest cluster of size $u$ and $a u$

28: %respectively.

29: \end{abstract}

30: