An alternative ending to "Pleasant extensions retaining algebraic structure''

Tim Austin

Published 2009-10-06, updated 2010-05-24Version 3

The culmination of the papers (arXiv:0905.0518, arXiv:0910.0909) was a proof of the norm convergence in $L^2(\mu)$ of the quadratic nonconventional ergodic averages \frac{1}{N}\sum_{n=1}^N(f_1\circ T_1^{n^2})(f_2\circ T_1^{n^2}T_2^n)\quad\quad f_1,f_2\in L^\infty(\mu) associated to an arbitrary probability-preserving \bbZ^2-system (X,\mu,T_1,T_2). This is a special case of the Bergelson-Leibman conjecture on the norm convergence of polynomial nonconventional ergodic averages. That proof relied on some new machinery for extending probability-preserving $\bbZ^d$-systems to obtain simplified asymptotic behaviour for various nonconventional averages such as the above. The engine of this machinery is formed by some detailed structure theorems for the `characteristic factors' that are available for some such averages after ascending to a suitably-extended system. However, these new structure theorems underwent two distinct phases of development, separated by the discovery of some new technical results in Moore's cohomology theory for locally compact groups (arXiv:1004.4937). That discovery enabled a significant improvement to the main structure theorem (Theorem 1.1 in (arXiv:0905.0518)), which in turn afforded a much shortened proof of convergence. However, since the proof of convergence using the original structure theorem required some quite different ideas that are now absent from these other papers, I have recorded it here in case it has some independent interest.