1: \begin{abstract}

2: 

3: The problem of reconstructing functions from their asymptotic expansions 

4: in powers of a small variable is addressed by deriving a novel type of 

5: approximants. The derivation is based on the self-similar approximation 

6: theory, which presents the passage from one approximant to another as 

7: the motion realized by a dynamical system with the property of group 

8: self-similarity. The derived approximants, because of their form, are 

9: named the self-similar factor approximants. These complement the 

10: obtained earlier self-similar exponential approximants and self-similar 

11: root approximants. The specific feature of the self-similar factor 

12: approximants is that their control functions, providing convergence of 

13: the computational algorithm, are completely defined from the 

14: accuracy-through-order conditions. These approximants contain the Pad\'e

15: approximants as a particular case, and in some limit they can be 

16: reduced to the self-similar exponential approximants previously

17: introduced by two of us. It is proved that 

18: the self-similar factor approximants are able to reproduce exactly a wide 

19: class of functions which include a variety of transcendental functions. 

20: For other functions, not pertaining to this exactly reproducible class, 

21: the factor approximants provide very accurate approximations, whose 

22: accuracy surpasses significantly that of the most accurate Pad\'e 

23: approximants. This is illustrated by a number of examples showing the

24: generality and accuracy of the factor approximants even when conventional

25: techniques meet serious difficulties.

26: 

27: \end{abstract}

28: