Measurable tilings by abelian group actions

After the submission of the paper, Tim Austin suggested a simpler proof of a stronger version of Theorem 1.5 saying that (i) and (ii) in the theorem are equivalent to the third statement:

(iii) Let A be the stationary point process on G such that F\oplus A=G. If A is not trivial then A has positive topological entropy.

The direction “(ii) implies (iii)” is similar to our proof of “(ii) implies (i)” in ?. The direction “(iii) implies (ii)” is an immediate corollary of Theorem 1.2, but can also be deduced by a more elementary argument similar to the proof of [cyr-kra, Lemma 2.15].

The dilation lemma fails when the group \Gamma is non-abelian. For instance, consider the group \Gamma = \mathbb{Z} \times G for some (non-abelian) finite group G = (G,\cdot) (and using the additive group law on \mathbb{Z}), acting on X = \Gamma (equipped with counting measure) by left translation; one can also take X to be a quotient \mathbb{Z}/N\mathbb{Z} \times G of \Gamma = \mathbb{Z} \times G if desired to ensure that X has finite measure. Let Ha be some right coset of a proper subgroup H of G, and consider the finite set F \subset \Gamma defined by

Observe that if A \subset X is a set of the form

for some sequence g_n of elements of G, then the translates f \cdot A, f \in F partition X if and only if one has the constraint

for all n \in \mathbb{Z}. If we take X to be \mathbb{Z}/N\mathbb{Z} \times G instead of \mathbb{Z} \times G, the above discussion still applies, but with n now ranging in \mathbb{Z}/N\mathbb{Z} rather than \mathbb{Z}. In the non-abelian setting, one can easily construct examples(For instance, one can take G=S_3, H to be a subgroup of S_3 of order two, and a to be an element not in H.) in which HaHaHa=G, in which case the constraint (2.1) gives no relationship whatsoever between g_{n+r} and g_n for r \ge 3. In particular, for such r there is no dilated tile of the form

for some non-empty E, E' \subset G with the property that F \odot A = X implies the F_r \odot A = X. A similar analysis shows that the assertions F \odot A = X and F^{-1} \odot A = X are inequivalent. This example indicates that no reasonable analogue of the dilation lemma holds in this setting. This example also shows that non-abelian tiling problems with one tile can be “local” in various senses; see, for instance, ? below for a more precise statement.

As showed in [GreenfeldTao, Lemma 3.1], one can generalize the dilation lemma by requiring the tiling to be a periodic level tiling rather than a partition up to null sets, by which we mean that for every 1\le k\le |F|, the level set \{x\colon {\bf 1}_F*{\bf 1}_A(x)=k\} is periodic in X up to \mu-null sets, (where here a periodic set is a set which is \mu-almost everywhere invariant with respect to an action of some lattice). The conclusion is then that there is a number q (depending on F and {\bf 1}_F*{\bf 1}_A) such that if r=1\mod q then {\bf 1}_{F^r}*{\bf 1}_A=_{\mathrm{a.e.}} {\bf 1}_{F}*{\bf 1}_A, but now we permit collisions f_1^r = f_2^r to occur. We leave the details of this generalization to the interested reader.

Theorem 1.2(i) easily extends to the setting in which the action of \Gamma is quasi-invariant rather than invariant, which means that it maps \mu-null sets to \mu-null sets. This can be accomplished simply by replacing \mu with the (non-\sigma-finite) measure \tilde \mu defined by setting \tilde \mu(E) to equal +\infty when \mu(E)>0 and to equal zero otherwise.

As it turns out, one can replace the requirement that the measure \mu be finite to merely \sigma-finite, and also assume that the action is only quasi-invariant rather than measure-preserving, as long as we also drop the conclusion that the \varphi_f have mean \mu(A); see ?.

Using Remark 2.2, one can also extend the above structure theorem to periodic level tilings, and in particular, to level k tilings(Higher level tilings are studied in several places in the literature; see for instance [kol], [gkrs], [swee-hong].) (by which we mean that almost every element of X lies in precisely k of the translates f\cdot A for some k\le |F|), but now replacing {\bf 1}_X in (1.2) with k {\bf 1}_X. However this identity (1.2) is significantly less useful in the k>1 case due to the gap in values between k {\bf 1}_X and {\bf 1}_A, which leaves more room for the functions \varphi_f, f\in F to vary (cf., [GreenfeldTao, Theorem 1.3(ii)]).

In the case when tiling a finitely generated abelian group with a tile which does not contain any non-trivial element of finite order (e.g., tiling \mathbb{Z}^d with a non-trivial tile), Theorem 1.2 implies a stronger conclusion saying that the spectral measure of the tiling is supported on a finite union of subtorii; in particular, the tiling in this case is not weak-mixing in some directions. We thank the referee for this observation.

An alternate way to prove Theorem 1.7(iv) is as follows. In [Lagarias-Wang, Theorem 2] it was proved that if A\subset \mathbb{R} is bounded, Lebesgue measurable and has a zero measure boundary and if A\oplus R=_{\mathrm{a.e.}}\mathbb{R} for some R\subset \mathbb{R}, then the set R'=\mu(A)^{-1}R must be rational, i.e., R'-R'=\{r'-r\colon r',r\in R'\}\subset \mathbb{Q}. However, looking into the proof there, the condition that the set A has boundary of measure zero is used in order to show that any such tiling set R must be periodic, and the rest of the argument, [Lagarias-Wang, Theorem 6 and Section 4], does not use this assumption. Thus, under the assumption that a bounded measurable set A tiles the line by a periodic set R, the argument of Lagarias–Wang gives the rationality of R'=\mu(A)^{-1}R. Since any measurable tiling F \oplus A =_{\mathrm{a.e.}} \mathbb{T} of the torus induces a periodic measurable tiling (F + \mathbb{Z}) \oplus \tilde A =_{\mathrm{a.e.}} \mathbb{R} of the real line by a bounded measurable set \tilde A of rational measure (defined as the image of A under the identification of the circle \mathbb{T} with the [0,1)), we conclude Theorem 1.7(iv). In particular, the conjecture from [conley_grebik_pikhurko2020divisibility_of_spheres] may also be deduced from the results in [Lagarias-Wang].

In fact, it was shown in [Kolountzakis-Lagarias, Theorem 6.1] that any tiling of \mathbb{R} with a bounded measurable set A is periodic. Thus, combining this result with [Lagarias-Wang, Section 4] we have that every tiling of \mathbb{R} by a bounded measurable set is periodic and rational.(We remark that classifying bounded measurable tiles A\subset \mathbb{R} is a notoriously difficult problem even in the case when A is a finite union of intervals. See, for instance, [N77, tijdeman215decomposition, CM99, LS, LL21] and the references therein.)

Using Theorem 1.7(iv), it is possible to fully describe all measurable tilings of the circle in terms of tilings of finite cyclic groups. Namely, given F=\{f_1,\dots, f_n\}\subseteq \mathbb{Q} and assuming that f_1=0, we find a g\in \mathbb{T}^1\cap \mathbb{Q} such that \langle g\rangle, the cyclic group generated by g, contains F'=\{f_1\mod 1,\dots, f_n\mod 1\} and is minimal with respect to set inclusion; for instance one can take g = \frac{1}{q} \mod 1 where q is the least common multiple of the denominators of elements from F. Let \mathcal{T}_{g,F'} be the set of all A'\subseteq \langle g\rangle such that F'\oplus A'=\langle g\rangle, i.e., \mathcal{T}_{g,F'} consists of all tiles of the finite cyclic group \langle g\rangle using translates F'. Consider the action of \langle g\rangle on \mathbb{T}^1 induced by x\mapsto g+x. As g\in \mathbb{Q}, we infer that orbit of each x\in \mathbb{T}^1 is finite, in fact, of cardinality |\langle g\rangle|. It follows that there is a measurable set X\subseteq \mathbb{T}^1 that intersects each orbit of the \langle g\rangle in exactly one point.

There is a one-to-one correspondence between measurable sets A\subseteq \mathbb{T}^1 that satisfies F\oplus A=_{\mathrm{a.e.}} \mathbb{T}^1 and measurable functions

(where \mathcal{T}_{g,F} is endowed with the discrete \sigma-algebra) that is given as follows: the measurable set A_\psi, that corresponds to \psi, is defined as

Similarly, given a measurable tile A, the function

defined for x\in X, is measurable and \psi_A(x)\in \mathcal{T}_{g,F'} almost surely.

Suppose that F \oplus A =_{\mathrm{a.e.}} \mathbb{T}^d is a measurable tiling of the d-torus, d\geq 2, by some finite subset F = \{f_1,\dots,f_n\} of \mathbb{R}^d and some measurable subset A of \mathbb{T}^d. Consider the velocities v_i as in Theorem 1.7(i), and let F_{(v)} \coloneqq \{ f_i: v_i = v \} be those elements of F that are moving with velocity v. Theorem 1.7(ii) gives that the sets F_{(v)} \oplus A are \mathbb{R} v-invariant for every v \in \mathbb{R}^2. In dimensions d\ge 3 this is no longer true. For instance, if A=[0,1/2]^3 and

for some irrationally numbers \alpha,\beta, and F^0=\{0,1/2\}^3. Then, if v=(\alpha, \beta) or v=(\alpha,0), we have that F_{(v)}\oplus A is not \mathbb{R} v-invariant.

However, we do not know if there is any measurable tiling F\oplus A=_{\mathrm{a.e.}} \mathbb{T}^d, d\ge 3 that does not satisfy the following weaker analogue of Theorem 1.7(ii). The velocities are replaced with piecewise linear functions v_f:[0,1]\to \mathbb{R}^d, for f\in F, such that F^t\oplus A=_{\mathrm{a.e.}} \mathbb{T}^d, where F^t is defined analogously as in Theorem 1.7. Moreover, if we set F^t_{(v)}=\{f\in F:v'_f(t)=v\}, for v\in \mathbb{R}^d, then F^t_{(v)}\oplus A is \mathbb{R} v-invariant for every v\in \mathbb{R}^d and t\in [0,1] whenever all the derivatives exist. In fact, we do not even know whether there exists any measurable tiling F\oplus A=_{\mathrm{a.e.}} \mathbb{T}^d, d\ge 3 such that there is no velocity 0\not= v\in \mathbb{R}^d for which there is a proper non-empty subset F' of F such that F'\oplus A is \mathbb{R} v-invariant. Note that the argument in [Szabo] implies that if the tile A is a cube then every tiling F satisfies this weaker analogue of Theorem 1.7(ii). On the other hand, if we are allowed to use more than one tile, then we can construct such a tiling in \mathbb{T}^3, as follows. Let f \colon \mathbb{T} \to \mathbb{R} be the function

and consider the following subsets A_1, A_2, A_3 of \mathbb{T}^3:

It is a routine matter to verify that

for any t \in \mathbb{R}, but that none of the individual sets A_1, A_2, A_3 enjoy any translational symmetries. Thus we see that there is a non-rigidity to the tiling problem

that cannot be explained purely by sliding each of the A_1,A_2,A_3 separately, or by translating the entire triple A_1,A_2,A_3 by a common shift.

The hypothesis that A is open and connected can be relaxed to the hypothesis that A is “measurably connected” in the following sense: for every \xi \in \mathbb{Z}^2 \backslash \{0\}, the function \psi_\xi \colon \mathbb{T} \to [0,+\infty) defined by

where the integral is with respect to the Haar probability measure on \{ x \in \mathbb{T}^2: \xi \cdot x=t\}, has connected support (modulo null sets). We leave the details of this generalization to the interested reader.

The periodic tiling conjecture [GS-book, Lagarias-Wang] asserts that if one has a tiling F \oplus A = \mathbb{Z}^d for some finite tile F in \mathbb{Z}^d and A\subset \mathbb{Z}^d, then there is also a periodic tiling F \oplus A' = \mathbb{Z}^d, thus A' is a finite union of cosets of some lattice in \mathbb{Z}^d. Currently this conjecture has only been established up to d \leq 2; see [bhattacharya2020periodicity, GreenfeldTao]. A variant of the argument used to prove Theorem 1.7(i) can establish the following partial result towards this conjecture: suppose that there is a homomorphism T \colon \mathbb{Z}^d \to \mathbb{T}^m and a measurable subset E of the torus \mathbb{T}^m and a finite F\subset \mathbb{Z}^d such that one has the measurable tiling T(F) \oplus E =_{\mathrm{a.e.}} \mathbb{T}^m, where an element f of \mathbb{Z}^d acts on \mathbb{T}^m via translation by T(f). Then F admits a periodic tiling F \oplus A' = \mathbb{Z}^d. Indeed, if one defines {\mathcal T}(F) to be the space of homomorphisms \tilde T \colon \mathbb{Z}^d \to \mathbb{T}^m (which one can identify with (\mathbb{T}^m)^d) such that \tilde T(F) \oplus E =_{\mathrm{a.e.}} \mathbb{T}^m, then a variant of the above arguments shows that {\mathcal T}(F) contains the orbit closure

By the above analysis, this orbit closure contains a rational point T^0, and by restricting the measurable tiling T^0(F) \oplus E =_{\mathrm{a.e.}} \mathbb{T}^m to a generic coset of T^0(\mathbb{Z}^d) and pulling back by \mathbb{Z}^d one obtains a periodic tiling F \oplus A' = \mathbb{Z}^d; we leave the details to the interested reader.