On the Garden of Eden theorem for endomorphisms of symbolic algebraic varieties (Q2189528)

One of the fundamental properties of finite sets is: ``a self-map is injective if and only if it is surjective''. The famous Ax's theorem (``any injective regular selfmapping of a complex algebraic variety is surjective'') was deeply investigated by \textit{M. Gromov} [J. Eur. Math. Soc. (JEMS) 1, No. 2, 109--197 (1999; Zbl 0998.14001)], who extended it to the proalgebraic setting (inverse limits of algebraic varieties). In this manner, it connects to the theory of cellular automata, which are none but continuous equivariant self-maps on a space \(A^G\) for \(A\) a finite set and \(G\) a group (typically \(\mathbb Z^2\)). Indeed \(A^G\) is the inverse limit of \(A^X\) where \(X\) runs over finite subsets of \(G\). The classical results in the universe of cellular automata are variations on this theme: due to \textit{E. F. Moore} [in: Proc. Sympos. Appl. Math. 14, 17--33 (1962; Zbl 0126.32408)] and \textit{J. Myhill} [Proc. Am. Math. Soc. 14, 685--686 (1963; Zbl 0126.32501)], they come under the name of ``Garden of Eden theorem'': for \(G=\mathbb Z^n\), a cellular automaton is surjective if and only if it is \emph{pre-injective}, namely it distinguishes configurations which differ in finitely many coordinates. A pioneering work by the first author [\textit{T. G. Ceccherini-Silberstein} et al., Ann. Inst. Fourier 49, No. 2, 673--685 (1999; Zbl 0920.43001)] proved that this holds whenever \(G\) is \emph{amenable}. A common setting encompassing both these results is hinted at in Gromov's article cited above, who asks ``Does the Garden of Eden theorem generalize to the proalgebraic category? First, one asks if preinjectivity implies surjectivity, while the reverse implication needs further modification of definitions.'' This question is answered in the present article when the proalgebraic variety is of the form \(A^G\) for \(A\) being an irreducible, complete algebraic variety over an algebraically closed field. The endomorphism \(\tau\colon A^G\to A^G\) is defined by a morphism \(A^X\to A\) for a finite subset \(X\) of \(G\) (the ``local rule''), and extended by \(G\)-equivariance. The authors need to redefine pre-injectivity of \(\tau\) appropriately: it is the \emph{inexistence} of a finite subset \(Y\subset G\) and a proper, Zariski-closed subset \(H\subset A^Y\) such that \(\tau(A^Y\times\{p\})=\tau(H\times\{p\})\) for all \(p\in A^{G\setminus Y}\). In the case of \(A\) finite, this definition coincides with the old one (i.e., if \(p,q\in A^G\) differ on \(Y\subset G\) finite and have same image under \(\tau\), consider \(H=A^Y\setminus\{p\restriction Y\}\); if \(\tau(A^Y\times\{p\})=\tau(H\times\{p\})\), extend \(p\) to \(G\) in two manners, within and without \(H\)). Once the proper definition is set up, the proof proceeds along the usual strategy of relating both pre-injectivity and surjectivity to preservation of the ``mean dimension'', a limiting, scaled dimension taken along restrictions to a Følner sequence. The article ends with an interesting collection of counterexamples showing that the hypotheses of their main result cannot be easily weakened.