1: \begin{abstract}

2: We propose and test a new method for generating canonical 

3: sequences for analysis by the Coherent Anomaly Method (CAM) 

4: {}from non-mean-field approximations. 

5: By intentionally underestimating the rate of convergence of

6: exact-diagonalization values for the mass or energy gaps 

7: of finite systems, we form families of sequences of gap estimates.

8: The gap estimates cross zero with generically nonzero linear 

9: terms in their Taylor expansions, so that $\nu \! = \! 1$ for 

10: each member of these sequences of estimates.  Thus, the 

11: CAM can be used to determine $\nu$. 

12: Our freedom in deciding exactly how to underestimate the 

13: convergence allows us to choose the sequence that displays 

14: the clearest coherent anomaly. 

15: We demonstrate this approach on the two-dimensional ferromagnetic 

16: Ising model, for which $\nu \! = \! 1$. 

17: We also use it  on the three-dimensional 

18: ferromagnetic Ising model, finding 

19: $\nu \! \approx \! 0.629$, in good agreement with other 

20: estimates. 

21: Finally, we apply it to an antiferromagnetic spin-1 Heisenberg 

22: chain, finding $\nu \! \approx \! 0.987$ at the phase transition 

23: between the Haldane phase and the dimerized phase, in agreement 

24: with the field-theoretic prediction $\nu \! = \! 1$. 

25: Although the specific systems used to test the extrapolation-CAM 

26: procedure involve finite system sizes, the method could be 

27: applied to other finite approximations, such as systematic 

28: variational approximations. 

29: \end{abstract}

30: