1: \begin{abstract}

2: The Luria-Delbr\"uck distribution is a classical model of mutations

3: in cell kinetics. It is obtained as a limit when 

4: the probability of mutation tends to zero and the number 

5: of divisions to infinity. It can be interpreted as a compound Poisson

6: distribution (for the number of mutations) of exponential mixtures

7: (for the developing time of mutant clones) of geometric distributions

8: (for the number of cells produced by a mutant clone in a given time).

9: The probabilistic interpretation, and a rigourous proof of

10: convergence in the general case, are deduced from classical results on

11: Bellman-Harris branching processes. The two parameters of the

12: Luria-Delbr\"uck distribution are the expected number of mutations,

13: which is the parameter of interest, and the relative fitness of normal

14: cells compared to mutants, which is the heavy tail exponent. 

15: Both can be simultaneously estimated by the maximum

16: likehood method. However, the computation becomes numerically unstable 

17: when the maximal value of the sample is large, which

18: occurs frequently due to the heavy tail property. Based on the

19: empirical probability generating function, 

20: robust estimators are proposed 

21: and their asymptotic variance is given. 

22: They are comparable in precision to maximum likelihood

23: estimators, with a much broader range of calculability, a better

24: numerical stability, and a negligible computing time. 

25: \end{abstract}

26: