1: \documentclass[12pt]{article}
2:
3: \usepackage{amssymb,amsmath,amsthm,graphics}
4: \newtheorem{defi}{Definition}
5: \newtheorem{thr}{Theorem}
6: \newtheorem{exa}[defi]{Example}
7: \newtheorem{lem}[defi]{Lemma}
8: \newtheorem{cor}[defi]{Corollary}
9: \newtheorem{pro}[defi]{Proposition}
10: \newtheorem{con}[defi]{Conjecture}
11:
12: \newcommand{\eps}{\varepsilon}
13: \renewcommand{\phi}{\varphi}
14: \newcommand{\A}{AP}
15: \newcommand{\F}{\mathcal{F}}
16: \renewcommand{\H}{\mathcal{H}}
17: \newcommand{\U}{\mathfrak{U}}
18: \renewcommand{\H}{\mathcal{H}}
19: \newcommand{\T}{\mathbb{T}}
20: \newcommand{\C}{\mathbb{C}}
21: \newcommand{\Z}{\mathbb{Z}}
22: \newcommand{\N}{\mathbb{N}}
23: \newcommand{\R}{\mathbb{R}}
24: \renewcommand{\b}{{}^{\flat}\!}
25: \newcommand{\Eins}{\mbox{1\hspace{-.9mm}\mbox{I}}}
26:
27: \begin{document}
28: \title{Filters and subgroups associated \\
29: with Hartman measurable functions}
30: \author{Gabriel Maresch\footnote{The author would like to thank the FWF for
31: financial support through grant S8312}}
32: \maketitle
33: \begin{abstract}
34: A bounded function $\phi: G\to \C$ on an LCA group $G$ is called Hartman measurable
35: if it can be extended to a Riemann integrable function $\phi^*: X\to \C$ on some group
36: compactification $(\iota_X,X)$, i.e. on
37: a compact group $X$ such that $\iota_X: G\to X$ is
38: a continuous homomorphism with image $\iota_X(G)$ dense in $X$ and
39: $\phi=\phi^*\circ\iota_X$. The concept of Hartman measurability of functions is a generalization
40: of Hartman measurability of sets, which was introduced - with different
41: nomenclature - by S. Hartman
42: to treat number theoretic problems arising
43: in diophantine approximation and equidistribution.
44: We transfer certain results concerning Hartman sets to this more
45: general setting. In particular we assign to each Hartman measurable function
46: $\phi$ a filter $\F(\phi)$ on $G$ and a subgroup $\Gamma(\phi)$ of the dual $\hat{G}$ and
47: show how these objects encode information about the involved
48: group compactification. We present
49: methods how this information can be recovered.\par\vspace{1em}
50:
51: \noindent Key words: almost
52: periodic function, Hartman measurable function, Hartman set,
53: filter, Fourier series.\par\vspace{1em}
54: \noindent 2000 Mathematics Subject Classification
55: 37A45, %Relations with number theory and harmonic analysis
56: 43A60, %Almost periodic functions on groups and semigroups and their generalizations
57: (11K70)%Harmonic analysis and almost periodicity
58: \end{abstract}
59:
60: \newpage \section{Introduction}
61: \subsection{Motivation:} In \cite{FP} the investigation of finitely additive measures in number
62: theoretic context led to the concept of a Hartman measurable subset $H\subseteq G$ of
63: a discrete abelian group $G$. By definition, $H$ is Hartman
64: measurable if it is the preimage $H=\iota_X^{-1}(M)$ of a continuity set $M\subseteq X$
65: in a group compactification $(\iota_X,X)$ of $G$. This, more explicitly, means that $\iota_X: G\to C$
66: is a group homomorphism with $\iota(G)$ dense in the compact group $X$ and that
67: $\mu_X(\partial M)=0$. Here $\partial M$ denotes the topological boundary of $M$ and
68: $\mu_X$ the normalized Haar measure on $X$. By putting $m_G(H)=\mu_X(M)$ one can define a
69: finitely additive measure on the Boolean set algebra of all Hartman measurable
70: sets in $G$. For the special case $G=\Z$ a Hartman set $H\subseteq \Z$, by identification
71: with its characteristic function, can be considered to be a two-sided infinite
72: binary sequence, called a Hartman sequence. Certain number theoretic, ergodic and
73: combinatorial aspects of Hartman sequences have been studied in \cite{SchW} and \cite{StW},
74: while \cite{Wi} presents a method to reconstruct the group compactification
75: $(\iota_X,X)$ for given $H$.
76:
77: In order to benefit from powerful tools from functional and harmonic analysis it
78: is desirable to study appropriate generalizations of Hartman measurable sets by replacing
79: their characteristic functions by complex valued functions not only taking
80: the values 0 and 1. The natural definition of a Hartman measurable function $\phi: G \to \C$
81: is the requirement $\phi^*\circ\iota_X$, where $(\iota_X,X)$ is a group compactification
82: of $G$ and $\phi^*$ is integrable in the Riemann sense, i.e. its points of
83: discontinuity form a null set with respect to the Haar measure on $X$. This
84: definition is equivalent to the one of R-almost periodicity, introduced
85: in \cite{H} by S. Hartman. The investigation
86: of the space $\H(G)$ of all Hartman measurable functions on $G$ is the content of \cite{MW}.
87: Here we are going to
88: transfer ideas from \cite{Wi} into this context. Thus our
89: main topic is to describe $(\iota_X,X)$ only in terms of $\phi$. In particular we establish
90: further connections to Fourier analysis. The natural framework for our investigation
91: is that of LCA (locally compact abelian) groups.
92:
93: \subsection{Content of the paper} After the introduction we collect in section 2 the necessary
94: preliminary definitions and facts about Hartman measurable sets and functions.
95:
96: Section 3 treats the following situation: Given a Hartman measurable function
97: $\phi: G\to X$ on an LCA group $G$, we know by the very definition of Hartman
98: measurability that there is some group compactification $(\iota_X,X)$ of $G$ such that
99: $\phi=\phi^*\circ\iota_X$ for some Riemann integrable function $\phi^*: X\to \C$.
100: We say that $\phi$ can be realized in $(\iota_X,X)$ resp. by $\phi^*$.
101: It is easy to see
102: that in this case $\phi$ can be realized as well on any "bigger" compactification
103: $(\iota_{\tilde{X}},\tilde{X})$. The notion of "bigger" and "smaller"
104: is made more precise in the next section.
105:
106: In particular every Hartman measurable function can be realized
107: in the maximal group compactification of $G$, the Bohr compactification $(\iota_b,bG)$.
108: The question arises if there is
109: a realization of $\phi$ in a group compactification that is as "small" as possible.
110: If a Hartman measurable function $\phi$ possesses a so called aperiodic realization
111: then the group compactification on which this aperiodic realization can be obtained
112: is minimal (Theorem 1).
113: This approach works for arbitrary Hartman measurable $\phi$
114: if one allows "almost realizations", i.e. if one demands $\phi=\phi^*\circ\iota_X$
115: almost everywhere with respect to the finitely additive Hartman measure $m_G$ on $G$ rather
116: than $\phi=\phi^*\circ\iota_X$ everywhere on $G$ (Theorem \ref{weilthr}). Whenever $\phi$ is
117: even almost periodic one can guarantee $\phi=\phi^*\circ\iota_X$ everywhere on $G$.
118: The group compactification on which the minimal realization of $\phi$ occurs is
119: unique up to
120: equivalence of group compactifications.
121: It can be obtained by a method involving filters on $G$ similar to that
122: presented in \cite{Wi}.
123:
124: The content of section 4 is motivated by the following reasoning: Every group compactification
125: of the LCA group $G$ corresponds to a (discrete) subgroup $\Gamma$ of the dual $\hat{G}$ in
126: such a way that it is equivalent to the group compactification $(\iota_{\Gamma}, C_{\Gamma})$ defined by
127: $\iota_{\Gamma}: g\to (\chi(g))_{\chi \in \Gamma}$, $C_{\Gamma}:=
128: \overline{\iota_{\Gamma}(G)} \le \T^{\Gamma}$.
129: If $(\iota_X, X)$ is a group
130: compactification admitting an aperiodic almost realization of the Hartman measurable
131: function $\phi$, the corresponding subgroup $\Gamma\le\hat{G}$ contains all characters $\chi$
132: such that the corresponding Fourier coefficient $m_G(\phi\cdot\overline{\chi})$ does not
133: vanish. If $\phi$ is almost periodic or if $\phi$ can be realized on a finite dimensional
134: compactification this result is sharp in the sense that the subgroup ${\Gamma}$ is minimal
135: with the above property (Theorem \ref{mincorap}). For general Hartman measurable
136: functions the situation is more difficult. This is discussed and illustrated by an example.
137:
138: Section 5 summarizes the main results and includes an illustrating diagram.
139:
140: \section{Preliminaries and Notation} Throughout this paper
141: $G$ denotes always an LCA (locally compact abelian) group.
142: For group compactifications of $G$ let us
143: write $(\iota_{X_1}, X_1)\le (\iota_{X_2}, X_2)$
144: iff there is a continuous group homomorphism $\pi: X_2\to X_1$ such that the diagram
145: \begin{center}
146: \includegraphics*{diagram1.eps}
147: \end{center}
148: commutes. In this situation we say that
149: $(\iota_{X_1}, X_1)$ is covered by $(\iota_{X_2}, X_2)$.
150: If $(\iota_{X_1}, X_1)$ is covered by $(\iota_{X_2}, X_2)$ (via $\pi_1$)
151: and $(\iota_{X_2}, X_2)$ is covered by $(\iota_{X_1}, X_1)$ (via $\pi_2$)
152: then $(\iota_{X_1}, X_1)$ and $(\iota_{X_2}, X_2)$ are called
153: equivalent. In this case compactness of $X_1$ and $X_2$ implies
154: that $\pi_1$ and $\pi_2$
155: are both topological and algebraic isomorphisms between $X_1$ and $X_2$.
156: $"\le"$ is a partial order on
157: the class of group compactifications
158: modulo equivalence. The maximal element with respect to this order is $(\iota_b,bG)$,
159: the Bohr compactification of the topological
160: group $G$. Recall that $AP(G)$, the set of almost periodic functions on $G$,
161: is isometrically isomorphic to $C(bG)$, the set of continuous functions
162: on $bG$. The mapping $\iota_b^*: C(bG)\to AP(G)$, defined via
163: $f\mapsto f\circ\iota_b$, is an isometry. Note that this is just a different way to characterize those continuous functions
164: on $G$, which can be extended to continuous functions on $bG$.
165: This definition (which is best suited for our purposes) is equivalent
166: to the notion of almost periodicity established by Bohr resp. Bochner.
167:
168: For a locally compact abelian (LCA) group $G$ let us denote by $\hat{G}$ the set
169: of all continuous homomorphisms $\chi: G \to \T\cong \R/\Z$. We will occasionally identify
170: $\T$ with the unimodular group $\{z\in \C: |z|=1\}$. This will cause no confusion.
171:
172: $\hat{G}$ endowed
173: with the compact-open topology is an LCA group in its own right. $\hat{G}$
174: is the Pontryagin dual of $G$. Every subset $\Gamma\subseteq \hat{G}$ induces a
175: group compactification $(\iota_{\Gamma},C_{\Gamma})$ of $G$ via
176: $\iota_{\Gamma}(g):=(\chi(g))_{\chi\in \Gamma}\in
177: \T^{\Gamma}$ and
178: $C_{\Gamma}:=\overline{\iota_{\Gamma}(G)} \le \T^{\Gamma}$
179:
180: One can show that every group compactification $(\iota_X,X)$ of an LCA group $G$
181: is equivalent to the group compactification induced by the subgroup
182: $\{\iota_X\circ\eta: \eta\in\hat{X}\}\le \hat{G}$ (Theorem 26.13 in \cite{HR}).
183: Thus group compactifications of LCA groups
184: can be described by subgroups of the dual and vice versa.
185:
186:
187: The system $\Sigma(G)\subseteq \mathfrak{P}(G)$ of all
188: Hartman measurable sets on $G$, i.e. the system
189: of all preimages $\iota_b^{-1}(M)$ of $\mu_b$-continuity
190: sets in the Bohr compactification $(\iota_b,bG)$ of $G$,
191: is a Boolean set algebra and enjoys the
192: property that there exists a unique translation invariant finitely additive
193: probability measure $m_G$ on $\Sigma(G)$:
194: $m_G(\iota_X^{-1}(M^*))=\mu_X(M^*)$. For details we refer to \cite{FP}.
195:
196: Let us denote by $\Delta$ the symmetric difference of sets and by $\tau_g$
197: the translation operator on an abelian group defined by $\tau_g (h):= g+h$.
198: We introduce two mappings:
199: \begin{itemize}
200: \item for a Hartman measurable set $M$
201: denote by $d_M: G \to [0,1]$ the mapping $g\mapsto m_G( M \Delta \tau_g M )$,
202: \item for a $\mu_X$-continuity set $M^*$ on some group compactification $(\iota_X,X)$
203: denote by $d_{M^*}: X \to [0,1]$ the mapping $g\mapsto \mu_X( M^* \Delta \tau_g M^* )$.
204: \end{itemize}
205:
206: Note that the mapping $d_{M^{*}}$ (and similarly the mapping
207: $d_{M}$)
208: can be used to define a translation invariant pseudometric by letting
209: $\rho_{M^{*}}(g,h):=d_{M^{*}}(g-h)$. The set of zeros $\{g: d_{M^{*}}(g)=0\}$ is always a
210: closed subgroup. We will denote this subgroup by $\ker d_{M^{*}}$.
211:
212: Now consider sets of the form
213: $F(M,\eps):=\{g\in G: d_M(g)<\eps\}$
214: and denote by $\F(M)$ the filter on $G$ generated by $\{F(M,\eps):\eps > 0\}$, i.e.
215: the set of all $F\subseteq G$ such that there exists an $\eps>0$ with $F(M,\eps)\subseteq F$.
216: When we have a realization $M^*$ of $M$ on some group compactification $(\iota_X, X)$
217: we can transfer the topological data encoded in the neighborhood filter of the unit $0_X$ in $X$
218: to $G$ by considering the pullback induced by $\iota_X$.
219:
220: To be precise: Let $(\iota_X,X)$ be a group compactification and $\U(X,0_X)$
221: the filter of all neighborhoods
222: of the unit $0_X$ in $X$. By $\U_{(\iota_X,X)}$ we denote the filter on $G$
223: generated by $\iota_X^{-1}\left(\U(X,0_X)\right)$.
224: Note that if the mapping $\iota_X$ is one-one, $\iota_X^{-1}\left(\U(X,0_X)\right)$ is already a filter.
225:
226: For $\Z$, the group of integers,
227: Theorem 2 in \cite{Wi} states that for any Hartman set $M\subseteq \Z$
228: there is a group compactification
229: $(\iota_X,X)$ such that $\F(M)$ and $\U(X,0_X)$ coincide. Hence the
230: filter $\F(M)$ on $\Z$ contains much information about
231: the group compactification $(\iota_X,X)$:
232: {\em If $M\subseteq \Z$ is a Hartman measurable set and $(\iota_X, X)$ is a group compactification of
233: the integers such that $M$ can be realized on $X$ via the continuity set $M^*$
234: then $H= \ker d_{M^*}$ is a closed subgroup of $X$ and
235: $\F(M)=\U_{(\pi_H\circ \iota_X,X/H)}$, for $\pi_H: X\to X/H$ the canonical quotient mapping.}
236:
237: In what follows we need to generalize this result to arbitrary (LCA) groups. This
238: poses no problem since the proof
239: given in \cite{Wi} for $G=\Z$ applies
240: verbatim to an arbitrary topological group.
241:
242:
243: Recall that for a filter $\F\subseteq \mathfrak{P}(X)$ on some set $X$
244: and a function $\phi: X\to \C$ the filter-limit
245: $\F\!\!-\!\!\lim_{x\in X} \phi(x)$ is defined to be the unique $\lambda\in\C$ such
246: that $\forall \eps> 0$ we have $\{x\in X: |\phi(x)-\lambda|<\eps\}\in \F$.
247: In \cite{Wi} the filter $\F=\F(M)$ is also used to define the subgroup Sub$(M)$ of
248: $\T$ consisting of all those elements $\alpha$ such that
249: $\F\!\!-\!\!\lim_{n\in\Z} \lfloor n\alpha\rfloor=0$
250: (or equivalently: $\F\!\!-\!\!\lim_{n\in\Z} e^{2\pi i n\alpha}=1$).
251:
252:
253: All three objects - filter, compactification and subgroup - carry the same
254: information regarding a fixed Hartman set $M$. It is interesting to note
255: that any subgroup of a compact abelian group $G$ can be written as
256: $\{g\in G: \F\!\!-\!\!\lim_{\chi\in\hat{G}} \chi(g)=1\}$ for some filter
257: $\F$ on $\hat{G}$ (cf. \cite{BS}).
258:
259: We transfer these concepts into our more general context.
260: To that cause we need the following definitions. Recall that
261: a bounded function $f$ on a group compactification $(\iota_X,X)$
262: is called Riemann integrable iff the set $\mbox{disc}(f)$ of points of discontinuity
263: is a $\mu_X$-null set, for $\mu_X$ the normalized Haar measure on $X$.
264: Let us denote the set of all such functions by $R_{\mu_X}(X)$ or,
265: simply, $R(X)$. We use the
266: following characterization, a proof of which can be found in \cite{Ta}.
267:
268: \begin{pro}\label{riemannchar}
269: Let $X$ be a compact space and $\mu_X$ a finite positive Borel measure
270: on $X$. For a bounded real-valued $\mu_X$-measurable function $f$ the following
271: assertions are equivalent:
272: \begin{enumerate}
273: \item $f$ is Riemann integrable.
274: \item For every $\eps>0$ there exist continuous functions $g_{\eps}$ and
275: $h_{\eps}$ such that $g_{\eps}\le f \le h_{\eps}$ and
276: $\int_X (h_{\eps}-g_{\eps}) d\mu_X< \eps$.
277: \end{enumerate}
278: \end{pro}
279:
280: Let $\phi$ be a function defined on a topological group $G$ and $(\iota_b,bG)$ the
281: Bohr compactification of $G$. We call
282: a function ${\phi}_b$ defined on $bG$
283: an extension (or realization) of $\phi$ iff $\phi={\phi}_b\circ\iota_b$.
284: For example: The set of almost periodic functions on $G$ coincides with
285: the set of those functions that can be extended to continuous functions on the Bohr
286: compactification.
287:
288: \begin{defi} Let $(\iota_b,bG)$ be the Bohr compactification of the topological group $G$.
289: We call a bounded function $\phi$ on $G$ Hartman measurable
290: iff $\phi$ can be extended to a Riemann integrable function $\phi_b$ on $bG$.
291: The set of Hartman measurable functions $\{\phi^*\circ\iota_b: \phi^*\ \in R_{\mu_b}(bG)\}$
292: is denoted by $\H(G)$.
293: \end{defi}
294:
295: Given a Hartman measurable
296: function $\phi$, we say that \emph{$\phi^*$ realizes $\phi$} if $\phi^*$ is a Riemann integrable
297: function defined on some
298: group compactification $(\iota_X, X)$ such that $\phi=\phi^*\circ\iota_X$, cf. the diagram below:
299: \begin{center}
300: \includegraphics*{diagram2.eps}
301: \end{center}
302:
303: In this situation we also say that $\phi$ can be realized on $(\iota_X, X)$.
304: Most of this paper is devoted to the task of finding a minimal group compactification
305: on which a given $\phi\in\H(G)$ can be realized. Note that $\phi^* \in R(X)$ implies
306: $\phi_b=\phi^*\circ\pi \in R(bG)$ (cf. \cite{MW}).
307:
308:
309:
310: \section{Filters associated with Hartman measurable functions}
311: By definition every $\phi \in \H(G)$ has a realization on $bG$
312: by a Riemann integrable function $\phi^*\in R_{\mu_b}(bG)$.
313: The mapping
314: \[d_{\phi^*}: x\mapsto \|\phi^* -\tau_x \phi^*\|_1:=\int_{bG} |\phi^* -\tau_x \phi^*|d\mu\]
315: is continuous (cf. \cite{El}, Corollary 2.32).
316: This implies that $d_\phi:=d_{\phi^*} \circ \iota_b$ is
317: an almost periodic function on $G$.
318:
319: The finitely additive invariant measure $m_G$ can be extended to an invariant mean
320: on $\H(G)$, i.e. to an invariant and non-negative normalized linear functional on $\H(G)$.
321: It will cause no confusion if we denote this invariant mean again by $m_G$ (cf. \cite{MW}).
322: Thus we can also write
323: $d_{\phi}(g)= m_G(|\phi -\tau_g\phi|)$. It is then obvious
324: to define $F(\phi,\eps):=\{g\in G: d_{\phi}(\tau_g \phi)<\eps\}$
325: and denote by $\F(\phi)$ the filter on $G$ generated by $\{F(\phi,\eps):\eps > 0\}$.
326:
327:
328: In the LCA setting, we can apply the tools developed in \cite{Wi} to conclude a functional
329: analogue of Theorem 2 in \cite{Wi}.
330:
331: \begin{defi}
332: Let $\phi\in \H(G)$ be realized by $\phi^*$ on the
333: group compactification $(\iota_X,X)$. $\phi^*$ is called an aperiodic
334: realization iff $\ker d_{\phi^*}:=\{x\in X: \|\phi^*-\tau_x \phi^*\|_1=0\}= \{0_X\}$.
335: \end{defi}
336:
337: \begin{thr} \label{ganzleicht}
338: Let $\phi\in \H(G)$ be realized by $\phi^*$ on the
339: group compactification $(\iota_X,X)$. Then $\F(\phi)\subseteq \U_{(\iota_X,X)}$.
340: Furthermore
341: $\F(\phi)=\U_{(\iota_X,X)}$ if $\phi^*$ is an aperiodic realization.
342: \end{thr}
343: \begin{proof}
344: Suppose $\phi=\phi^*\circ\iota_X$ with $\phi^*\in R_{\mu_X}(X)$ for a group
345: compactification $(\iota_X, X)$. For any set $A\in \F(\phi)$
346: there exists $\eps>0$ such that $d_{\phi}(x)<\eps$ implies $x\in A$.
347: Using almost periodicity
348: of $d_{\phi}$, i.e. continuity of $d_{\phi^*}$, we find a
349: neighborhood $U\in \U(X,0_X)$ such that
350: $d_{\phi^*}(U)\subseteq [0,\eps)$. For every
351: $x\in\iota_X^{-1}(U)$ we have $d_{\phi}(x)<\eps$. Consequently
352: $\iota_X^{-1}(U)\subseteq A \in \U_{(\iota_X,X)}$
353: and hence $\F(\phi)\subseteq \U_{(\iota_X,X)}$.
354:
355:
356: Suppose that $\phi^*\in R_{\mu_X}(X)$ is aperiodic, i.e.
357: $d_{\phi^*}(x)=0$ iff $x=0_X$, the unit in $X$.
358: Let
359: $A \in \U_{(\iota_X,X)}$ be arbitrary; w.l.o.g. we
360: can assume $A\supseteq \iota_X^{-1}(U)$
361: for an open neighborhood $U\in \U(X,0_X)$.
362: Due to the continuity of $d_{\phi^*}$ and compactness
363: of $X$ we have $d_{\phi^*}(x)\ge\eps>0$ for $x\in X\setminus U^{\circ}$.
364: This implies
365: $\iota_X (\{g\in G: d_{\phi}(g)<\eps\}) \subseteq U$ and hence
366: $\{g\in G: d_{\phi}(g)<\eps\}\subseteq\iota_X^{-1}(U)
367: \subseteq A \in \F(\phi)$.
368: Thus $\U_{(\iota_X,X)}\subseteq \F(\phi)$ and
369: consequently $\U_{(\iota_X,X)}= \F(\phi)$.
370: \end{proof}
371:
372: \begin{defi} Let $\phi\in \H(G)$ and let $(\iota_X, X)$
373: be a group compactification of $G$. A
374: function $\psi^*\in R_{\mu_X}(X)$ is called an almost
375: realization of $\phi$ iff
376: $m_G(|\phi-\psi|)=0$ for $\psi:=\psi^*\circ\iota_X$ and
377: $m_G$ the unique invariant mean on $\H(G)$.
378: \end{defi}
379:
380:
381: \begin{thr}\label{weilthr} Every $\phi\in\H(G)$ has
382: an aperiodic almost realization on some group compactification
383: $(\iota_X,X)$.
384: If $\phi^*: X\to \C$ is any aperiodic almost realization of $\phi$
385: then $\F(\phi)=\U_{(\iota_X,X)}$.
386: \end{thr}
387:
388: \begin{proof}
389: We only have to prove that an aperiodic almost realization exists, the rest follows from
390: Theorem \ref{ganzleicht}.
391: Let $\phi^*$ be a realization of $\phi$ on $X$. The reader will easily
392: check that
393: $H:=\ker d_{\phi^*}=\{x\in X: d_{\phi^*}(x) = 0\}$
394: is a closed subgroup of the compact abelian group $X$.
395:
396: Weil's formula for continuous functions on quotients (Theorem 3.22 in \cite{El}) states
397: that there exists $\alpha>0$ such that for every $f\in C(X)$
398: \begin{eqnarray}\label{Weil}
399: \int_{X/H}\Big(\underbrace{\int_H f(s+t) d\mu_H(t)}_{=\b f(s)}\Big) d\mu_{X/H}(s)&=&\alpha \int_X f(u)d\mu_X(u)
400: \end{eqnarray}
401: holds. This implies that the canonical mapping
402: $\b: C(X)\to C(X/H)$, $f\to \b f$ defined by $\b f(s+H)=\int_H f(s+t)d\mu_H(t)$ satisfies
403: $\|\b f\|_1\le \alpha \|f\|_1$. We rescale the Haar measure on $H$ such that
404: $\alpha=1$.
405: Thus we can extend $\b$ to a continuous linear operator
406: $L^1(X)\to L^1(X/H)$. Furthermore
407: positivity of $\b$ enables us to extend $\b$ to a mapping defined
408: on $R_{\mu_X}(X)$ in the following way:
409:
410: According to Proposition \ref{riemannchar} $f \in R_{\mu_X}(X)$ implies that there are $g_n,h_n \in C(X)$ such
411: that $g_n\le f \le h_n$ and $\|h_n-g_n\|_1\to 0 $ as $n\to \infty$. Thus
412: every function $\tilde{f}$ on $X/H$
413: satisfying
414: \begin{equation*}
415: f_{\bullet}:=\sup_{n\ge 0} \b g_n \le \tilde{f} \le \inf_{n\ge 0} \b h_n=:f^{\bullet}
416: \end{equation*}
417: is in $R_{\mu_{X/H}}(X/H)$. Note that $f_{\bullet}$ and $f^{\bullet}$ are $\mu_H$-measurable
418: and coincide $\mu_H$-a.e.; to define $\b f$ we pick
419: any function $\tilde{ f}$
420: satisfying $f_{\bullet}\le \tilde{f}\le f^{\bullet}$. Then Weil's formula (\ref{Weil}) will still be valid,
421: regardless of the particular choice of the $g_n,h_n$ and $\b f$.
422:
423:
424: Since $\phi^*$ is Riemann integrable on $X$,
425: there exist functions $\phi_n\in C(X)$ such that $\|\phi^*-\phi_n\|_1\to 0$.
426: Note that $d_{\phi_n}\to d_{\phi^*}$ even uniformly on $X$:
427: \begin{equation*}
428: |d_{\phi_n}(s)-d_{\phi^*}(s)| = \Big| \|\tau_s \phi_n - \phi_n\|_1 - \|\tau_s \phi^* - \phi^*\|_1\Big|
429: \le 2 \| \phi_n -\phi^*\|_1 \to 0.
430: \end{equation*}
431: Using the continuity of $\b$ as a mapping on $L^1(X)$ the same argument also shows that
432: $|d_{\b \phi^*}(s+H)-d_{\b \phi_n}(s+H)|\le 2 \|\phi_n-\phi^*\|_1\to 0$ uniformly on $X/H$.
433: Suppose $d_{\b \phi}^*(s+H)=0$. Then
434: \begin{equation*}
435: d_{\phi^*}(s)=\lim_{n\to \infty}d_{\phi_n}(s)=\lim_{n\to \infty}d_{\b\phi_n}(s+H)=0
436: \end{equation*}
437: implies $s\in H$, i.e. $s+H=0_X+H \in X/H$. So $\b \phi^*$ is aperiodic.
438:
439: We show that $\phi^*$ being a realization of $\phi$ implies that
440: $\b \phi^*$ is an almost realization of $\phi$.
441: By definition $t\in H$ iff $A_t:=\{ s \in X: \phi^*(s+t)=\phi^*(s)\}$ has $\mu_{X}$-measure $1$.
442: Applying Weil's formula (\ref{Weil}) to the function $f=\Eins_{A_t}\in L^1(X)$ gives
443: \begin{equation}\label{vorher}
444: \int_{X/H}\b f d\mu_{X/H}=\int_{X/H}\b \Eins_{A_t}(s+H) d\mu_{X/H}(s+H)=\int_X fd\mu_X =1.
445: \end{equation}
446: Plugging the definition of $\;\b\;$ into (\ref{vorher}) we get $\mu_{X/H}$-a.e. the identity
447: \begin{equation*}
448: \b \Eins_{A_t}(s+H)= \int_H \Eins_{A_t}(s+u)d\mu_H(u)=1.
449: \end{equation*}
450: So for every $t\in H$ and $\mu_{X/H}$-a.e. $s+H$ we know that
451: the set $\{u\in H: \phi^*(s+t+u)\neq\phi^*(s+u)\}$ is a $\mu_H$-null set. This
452: means
453: \[\tau_{t}(\tau_s{\phi^*}_{|H})=\tau_s{\phi^*}_{|H}\quad \mu_H\mbox{-a.e.}\]
454:
455: Thus $\tau_s\phi^*$ is constant $\mu_H$-a.e. on $H$ and
456: for $\mu_{X/H}$ almost all $s+H$ we have
457: \begin{eqnarray*}
458: \b \phi^* (s+H)&=&\int_H \tau_s\phi^*(t) d\mu_H(t)=\int_H \phi(s)^* d\mu_H(t)=\phi^*(s).
459: \end{eqnarray*}
460:
461: Let $\pi_H: X\to X/H$ be the quotient mapping
462: onto the group compactification $(\iota_{X/H},X/H)$.
463: Let $\psi^*:= \b\phi^*\circ \pi_H $.
464: Since $\b\phi^*$ is Riemann integrable on $X/H$ it is an elementary fact
465: that $\psi^*$ is Riemann integrable
466: on $X$ (cf. \cite{MW}).
467: Once again, Weil's formula (\ref{Weil}) together with the fact
468: that the Haar measure on the quotient $X/H$ is given by $\mu_{X/H}=\pi_H^{-1}(\mu_{X})$
469: implies
470: $\psi^*= \phi^*$ $\mu_{X}$-a.e. Thus the function $\psi$ defined by
471: \[\psi:=\psi^*\circ \iota_X = \b \phi^*\circ\iota_{X/H}\] satisfies
472: $m_G(|\phi-\psi|)=\|\phi^*-\psi^*\|_1=0$ for the unique invariant mean $m_G$.
473: Thus $\psi^*$ is the required almost realization of $\phi$.
474: \end{proof}
475:
476: \begin{cor}
477: \label{apreal}
478: Every $\phi\in\A(G)$ has an aperiodic realization
479: on some group compactification $(\iota_X,X)$.
480: \end{cor}
481: \begin{proof}
482: We use the notation from Theorem \ref{weilthr}.
483: If $\phi$ is almost periodic then $\phi^*$ is continuous.
484: Consequently $\b \phi^*$ and
485: $\psi^*:=\b\phi^*\circ\pi$ are also continuous. Since these functions
486: coincide $\mu_X$-a.e.
487: they coincide everywhere on $X$. This implies that
488: $\phi^*$ is constant on $H$-cosets and
489: $\b \phi^* (s+H)=\phi^*(s)$ for all $s+H \in X/H$. So $\phi^*$ is truly
490: a realization of $\phi$, not only an almost realization.
491: \end{proof}
492:
493: This Corollary is a special case of F\o lner's
494: "Main Theorem for Almost Periodic Functions", for a detailed treatment
495: cf. \cite{Fo}.
496:
497: \emph{Remark:} Note that for any given realization of a Hartman measurable
498: function $\phi\in\H(G)$ on a group compactification $(\iota_X,X)$
499: we can always assume that there exists an aperiodic almost realization of $\phi$ on a
500: group compactification $(\tilde{\iota_X},\tilde{X})$ with
501: $(\tilde{\iota_X},\tilde{X})\le (\iota_X,X)$.
502: Since in \cite{MW} it is shown
503: that every Hartman measurable function on an LCA group
504: with separable dual has a realization on a
505: metrizable group compactification,
506: every Hartman measurable function
507: on such a group has
508: an aperiodic almost realization on a metrizable group compactification.
509:
510:
511: \begin{lem}\label{compsub} Let $G$ be an LCA group and let $(\iota_X,X)$
512: be a group
513: compactification. Then there exists a unique subgroup $\Gamma\le \hat{G}$
514: such that $(\iota_{\Gamma},C_{\Gamma})$ and $(\iota_X,X)$ are equivalent.
515: Furthermore $(\iota_X,X)$ is the supremum of
516: all group compactifications $(\iota_{\gamma},C_{\gamma})$
517: such that $(\iota_{\gamma},C_{\gamma})\le (\iota_X,X)$ (writing
518: in short $(\iota_{\gamma},C_{\gamma})$ for
519: $(\iota_{\langle\gamma\rangle},C_{\langle\gamma\rangle})$).
520:
521: The mapping $(\iota_X,X)\mapsto C_{\Gamma}$ is a bijection between
522: equivalence classes of group compactifications of $G$
523: and subgroups of $\hat{G}$.
524: \end{lem} \begin{proof} See Theorem 26.13 in \cite{HR}. \end{proof}
525:
526: \begin{cor}\label{unique} Let $\phi\in\H(G)$. Any two group compactifications
527: $(\iota_{X_1},X_1)$ and $(\iota_{X_2},X_2)$ on which
528: $\phi$ has an aperiodic almost realization are equivalent. \end{cor}
529: \begin{proof} By Theorem \ref{ganzleicht} we have $\U_{(\iota_{X_1},X_1)}=\F(\phi)
530: =\U_{(\iota_{X_2},X_2)}$.
531: A straight forward generalization of Theorem 1 in \cite{Wi} implies
532: that the mapping $$\Phi: \hat{G}\ge \Gamma \mapsto (\iota_{{\Gamma}},C_{\Gamma})$$ coincides
533: with the composition of the mappings
534: \begin{eqnarray*}
535: \Sigma: (\iota_b,bG)\ge(\iota_X,X) &\mapsto& \U_{(\iota_X,X)},\\
536: \Psi: \mathfrak{P}(G)\supset \F &\mapsto&
537: \{\chi \in \hat{G}: \F\!\!-\!\!\lim_{g\in G}\chi(g) = 0\}.
538: \end{eqnarray*}
539: Since Lemma \ref{compsub} states that $\Phi=\Psi\circ\Sigma$ is invertible,
540: $\Sigma$ must be one-one. In particular
541: $\U_{(\iota_{X_1},X_1)}=\U_{(\iota_{X_2},X_2)}$ implies that
542: $(\iota_{X_1},X_1)$ and $(\iota_{X_2},X_2)$ are
543: equivalent group compactifications.
544: \end{proof}
545:
546:
547: {\em For the rest of this section assume that $G$ is an
548: LCA group with separable dual.}
549:
550:
551: \begin{cor}
552: Every filter $\F(\phi)$ with $\phi\in\H(G)$ coincides with a filter
553: $\U_{(\iota_X,X)}$ for a metrizable
554: group compactification $(\iota_X,X)$. If $\phi^*$ is
555: an arbitrary realization of $\phi$, say on the Bohr compactification
556: $bG$, we can take $X\cong bG/\ker d_{\phi^*}$.
557: \end{cor}
558:
559: \begin{cor}
560: Hartman measurable functions
561: induce exactly the filters coming from metrizable group compactifications.
562: \end{cor}
563: \begin{proof} In Theorem 3 in \cite{Wi}
564: for every metrizable group compactification $(\iota_X,X)$ of the
565: integers $G=\Z$, an aperiodic Hartman periodic function of the form
566: $f=\Eins_A$ is constructed. The same construction can be done in an arbitrary
567: LCA group $G$ as long as the dual $\hat{G}$ contains a countable and dense
568: subset.
569: This shows that any $\U_{(\iota_X,X)}$ with metrizable $X$ can be obtained
570: already by Hartman measurable \emph{sets},
571: i.e. by a filter $\F(\phi)$ with $\phi=\Eins_A$.
572: Since
573: any Hartman measurable function on $G$ can be realized on a metrizable group compactification
574: (cf. \cite{MW}).
575: Thus Theorem \ref{ganzleicht} implies that
576: no filter $\F(\phi)$ can coincide with $\U_{(\iota_X,X)}$ for a
577: non metrizable group compactification $(\iota,C)$.
578: \end{proof}
579:
580:
581: \section{Subgroups associated with Hartman measurable functions}
582:
583:
584: For Hartman measurable $\phi$
585: let us denote by $\Gamma(\phi)$ the (countable) subgroup of $\hat{G}$ generated by
586: the set
587: \[\mbox{spec}\;\phi:=\{\chi\ \in \hat{G}: m_G(\phi\cdot\overline{\chi})\neq 0\}\]
588: of all characters with
589: non vanishing Fourier coefficients. We will prove that $\Gamma=\Gamma(\phi)$
590: determines
591: a group compactification $(\iota_{\Gamma}, C_{\Gamma})$ such that $\phi$ can
592: be realized aperiodically on $C_{\Gamma}$. First we deal with almost periodic functions:
593:
594: \begin{pro}\label{fact1} Let $\phi\in\A(G)$ and
595: $(\iota_X, X)$ a group compactification
596: such that
597: every character $\chi\in \Gamma(\phi)$ has a representation $\chi = \eta \circ \iota_X$ with a
598: continuous character $\eta \in \hat{X}$.
599: Then every function $f\in \overline{\mbox{span}}\,\Gamma(\phi)\subseteq AP(G)$
600: has a realization on $(\iota_X, X)$. \end{pro}
601: \begin{proof}
602: This is essentially a reformulation of Theorem 5.7 in \cite{B}.
603: In fact the Stone-Weierstrass Theorem implies
604: that $\overline{\mbox{span}}\,\Gamma(\phi)= \iota_{\Gamma}^*\;C(X)$. Furthermore
605: $\phi\in\overline{\mbox{span}}\,{\Gamma}(\phi)$, i.e. $\phi$ can be realized by some
606: continuous $\phi^*: X \to \C$. \end{proof}
607:
608: \begin{pro}\label{characterreal1} Let $\phi\in\A(G)$ and
609: $(\iota_{\Gamma}, C_{\Gamma})$
610: the group compactification of $G$ induced
611: by the subgroup $\Gamma=\Gamma(\phi)\le \hat{G}$.
612: Then for every continuous character $\psi\in \Gamma(\phi)$
613: there exists a continuous $\psi^*: C_{\Gamma} \to \C$ such that
614: $\psi=\psi^*\circ\iota_{\Gamma}$. \end{pro}
615: \begin{proof}
616:
617: Given the group compactification $(\iota_{\Gamma},C_{\Gamma})$, then
618: the compact group $C_{\Gamma}$ is
619: by definition topologically isomorphic to
620: $\overline{\{(\chi(g))_{\chi\in \Gamma}: g\in G\}} \le
621: \T^{\Gamma}$.
622:
623: The restriction of each projection
624: \[\pi_{\chi_0}: C_{\Gamma}\le\T^{\Gamma} \to \T,\quad
625: (x_{\chi})_{\chi \in \Gamma}\mapsto x_{\chi_0} \]
626: is a bounded character of $C_{\Gamma}$ for each $\chi_0 \in \Gamma(\phi)$.
627: I.e. $\pi_{\chi_0}$ is an element of $\widehat{C_{\Gamma}}$.
628: Thus $\chi_0= \pi_{\chi_0} \circ \iota_{\Gamma}$ for each
629: $\chi_0\in \Gamma(\phi)$ and
630: we may apply Proposition \ref{fact1} to obtain the assertion.
631: \end{proof}
632:
633: \begin{pro}\label{characterreal2}
634: Let $\phi\in \A(G)$ and let
635: $(\iota_X,X)$ be a
636: group compactification of $G$ such
637: that $\phi$ can be realized by a continuous function $\phi^*: X \to \C$.
638: Then each continuous character $\chi\in \Gamma(\phi)$ has a representation
639: $\chi = \eta \circ \iota_{\Gamma}$ with $\eta \in \hat{X}$. \end{pro}
640:
641: \begin{proof} Obviously it is enough to prove the assertion for
642: a generating subset of $\Gamma(\phi)$. Let
643: $\chi\in \hat{G}$ be such
644: that $m_G(\phi\cdot\overline{\chi})\neq 0$. Define a linear functional
645: $m_{\chi}: C(X) \to \C$
646: via $\psi\mapsto m_{\chi}(\psi)=m_G(( \psi\circ\iota_{\Gamma}) \cdot \overline{\chi})$.
647: It is routine to check
648: that $m_{\chi}$ is bounded and $\|m_{\chi}\|=1$. Since $X$ is compact the
649: complex-valued mapping $\tilde{\eta}:
650: X \mapsto m_{\chi}(\tau_x \phi^*)$
651: is continuous on $X$ (the mapping $x\mapsto \tau_x \phi^*$ is continuous).
652: For $g\in G$ we compute
653: \begin{eqnarray*}
654: \tilde{\eta} \circ \iota_X (g) &=& m_G((\tau_{\iota_X(g)} \phi^*\circ \iota_X) \cdot\overline{\chi})
655: = m_G(\tau_g (\phi^*\circ \iota_X)\cdot\overline{\chi})\\
656: &=& m_G( (\phi^*\circ \iota_X)\cdot \tau_{-g}\overline{\chi})
657: = m_G( (\phi^*\circ \iota_X) \cdot\chi(g)\overline{\chi})\\
658: &=&\chi(g) m_{\chi}(\phi^*)= \chi(g) \tilde{\eta} (0).
659: \end{eqnarray*}
660: Since $\tilde{\eta}(0)=m_{\chi}(\phi^*)=m_G(\phi \cdot \overline{\chi})\neq 0$ we can define
661: $\eta := \tilde{\eta}(0)^{-1}\tilde{\eta}$. The mapping
662: $\eta: X \to \T$ is continuous
663: and satisfies the functional equation
664: \[
665: \eta(\iota_X(g)+\iota_X(h))=
666: \tilde{\eta}(0)^{-1}\tilde{\eta}(\iota_X(g)+\iota_X(h))=
667: \chi(g)\chi(h)=\eta(\iota_X(g))\eta(\iota_X(h))\]
668: on the dense set $\iota_X(G)$.
669: Hence $\eta$ is a bounded character on $X$
670: and $\eta\circ\iota_X = \chi$.
671: \end{proof}
672:
673: \begin{cor}\label{characterreal2a}
674: Let $\phi\in \H(G)$ be realized by $\phi^*$ on the
675: group compactification $(\iota_X,X)$.
676: Then each $\chi\in \Gamma(\phi)$ has a representation
677: $\chi = \eta \circ \iota_X$ with $\eta \in \hat{X}$. \end{cor}
678: \begin{proof} For every $\chi\in\hat{G}$ with
679: $m_G(\phi\cdot\overline{\chi})=\alpha\neq 0$
680: we can pick a continuous function $\psi^*:X\to \C$
681: such that $\|\psi^*-\phi\|_1 < |\alpha|/2$. Then
682: $\psi:=\psi^*\circ\iota_X$ satisfies
683: \[|m_G(\phi\cdot\overline{\chi})-m_G(\psi\cdot\overline{\chi})|\le m_G(|\phi-\psi|)\le
684: \|\psi^*-\phi^*\|_1 < |\alpha|/2.\] In particular $m_G(\psi\cdot\overline{\chi})\neq 0$.
685: Applying Proposition \ref{characterreal2} to the function $\psi\in\A(G)$ yields that
686: the character $\chi$ can be realized on $X$.\end{proof}
687:
688:
689:
690: Thus for almost periodic functions $\phi$ the subgroup $\Gamma(\phi)$ contains all the
691: relevant information to reconstruct $\phi$ from its Fourier-data in a minimal way.
692: It is not obvious how to obtain similar results for
693: Hartman measurable functions that are not
694: almost periodic. The following example illustrates how a straight forward
695: approach may fail.
696:
697:
698: \begin{exa}{\em
699: Let $\phi_n(k):=\prod_{j=1}^n \cos^2\left(2\pi \frac{k}{3^j}\right)$
700: on $G=\Z$. Each $\phi_n$ is a finite product of periodic (and hence almost periodic)
701: functions. Since $\A(\Z)$ is an algebra, $\phi$ is almost periodic.
702: In \cite{MW} it is shown that
703: $\phi(k):=\lim_{n\to \infty} \phi_n(k)$ exists and defines
704: a non negative Hartman measurable function with $m_{\Z}(\phi)=0$.
705: Since $\Gamma(\phi_n)\cong\Z/3^n \Z$ we have (using obvious notation):
706: \[\lim_{n\to\infty} \Gamma(\phi_n)=
707: \bigcup_{n=1}^{\infty} \Gamma(\phi_n)\cong
708: \Z_3^{\infty},\] the Pr\"ufer 3-group (i.e. the subgroup of all
709: complex $3^n$-th roots of unity for $n\in \N$),
710: but
711: \[\Gamma(\lim_{n\to\infty} \phi_n)=\{0\}.\]}\end{exa}
712:
713:
714: \begin{pro}\label{fejer}
715: Let $\{K_n\}_{n=1}^{\infty}$ denote the family of
716: Fej\'er kernels on $\T^k$
717: \[ K_n(\exp(it_1),\ldots, \exp(it_k))=\frac{1}{k}
718: \prod_{j=1}^{k} \left(\frac{\sin( \frac{1}{2}nt_j)}{\sin( \frac{1}{2}t_j)}\right)^2.\]
719: The linear convolution operators on $L^1(\T^k)$ defined
720: by \[\sigma_n: \phi\mapsto K_n*\phi\] are
721: non negative, their norm is uniformly bounded by
722: $\|\sigma_n\|=1$ and $\sigma_n \phi(x) \to \phi(x)$ a.e. for every $\phi\in L^1(\T^k)$.
723: Furthermore $\sigma_n \phi \in \mbox{span } \Gamma(\phi)$ for every $n\in \N$.
724: \end{pro} \begin{proof} This is a reformulation
725: of the results in section 44.51 in \cite{HR}.\end{proof}
726:
727: Let $f$ be Riemann integrable on $X=\T^k$, w.l.o.g.
728: real-valued, and $\phi_{i}, \psi_i\in C(X)$ such that
729: $\phi_{i}\ge f\ge \psi_i$ and $\|\phi_{i}-\psi_i\|_1<\eps_i$ for a sequence
730: $\{\eps_i\}_{i=1}^{\infty}$ of positive real numbers, tending monotonically to $0$.
731: We know that $\sigma_n f(x)\to f(x)$ for a.e. $x\in X$. Thus we have
732: \begin{equation*}\label{rapf}
733: \phi_n^*:=\sigma_n \phi_n \ge \sigma_n f \ge \sigma_n \psi_n=\psi_n^*
734: \end{equation*}
735: and
736: \begin{equation*}
737: \|\phi_n^*-\psi_n^*\|_1\le \|\sigma_n (\phi_n^*-\psi_n^*)\|_1
738: \le \|\sigma_n\|\, \|\phi_n-\psi_n\|_1 \le \eps_n.
739: \end{equation*}
740: Let $\phi^*:=\inf_{n\in\N}\phi_n$ and
741: $\psi^*:=\sup_{n\in\N}\psi_n$. If we assume w.l.o.g. $\psi_n$ to increase and $\phi_n$ to decrease as $n\to\infty$,
742: the same will hold for $\psi_n^*$ and $\phi_n^*$. This implies that in the inequality
743: \[\phi^*(x)=\lim_{n\to\infty} \phi_n^*(x) \ge \limsup_{n\to\infty} \sigma_n f
744: \ge \liminf_{n\to\infty} \sigma_n f \ge
745: \lim_{n\to\infty}\psi_n^*(x)=\psi^*(x)\]
746: actually equality holds $\mu_X$-a.e. on $X$. Thus we can apply
747: Proposition \ref{riemannchar} and conclude that any function $f^*$
748: with $\phi^*\ge f^*\ge\psi^*$ is Riemann integrable (and coincides
749: $\mu_X$-a.e. with $f$).
750: In particular $f^{\bullet}:=\limsup_{n\to \infty} \sigma_n f$ and
751: $f_{\bullet}:=\liminf_{n\to \infty} \sigma_n f$
752: are (lower resp. upper semicontinuous) Riemann integrable
753: functions that coincide $\mu_X$-a.e. with $f$.
754:
755: Let us call
756: a group compactification $(\iota_X,X)$ finite dimensional iff
757: $X$ is topologically isomorphic
758: to a closed subgroup of $\T^n$ for some $n\in \N$.
759: Note that if $(\iota_X,X)$ is finite dimensional, then every
760: group compactification covered by $(\iota_X,X)$ is finite dimensional
761: as well.
762: A Hartman measurable function $\phi\in\H(G)$ can be realized
763: finite dimensionally iff there exists a realization of $\phi$ on some
764: finite dimensional group compactification.
765:
766:
767: \begin{pro}
768: For a compact LCA group $C$ the following assertions are equivalent:
769: \begin{enumerate}
770: \item $C$ is finite dimensional,
771: \item $\hat{C}$ is finitely generated,
772: \item $C$ is topological isomorphic to $\T^k\times F$ for $k\in \N$
773: and a finite group $F$ of the
774: form \[F=\prod_{i=1}^N (\Z/n_i\Z)^{p_i},\quad p_i \mbox{ prime}.\]
775: \end{enumerate}
776:
777: \end{pro}
778: \begin{proof} Folklore.
779: \end{proof}
780:
781: \begin{pro}
782: Let $\phi\in\H(G)$. If $\phi$ can be realized finite dimensionally, then there
783: is an almost realization of $\phi$ on the (finite dimensional) compactification
784: induced by $\Gamma:=\Gamma(\phi)$.
785: \end{pro}
786:
787:
788: \begin{proof}
789:
790: Let $\phi$ be realized finite dimensionally on some group compactification
791: $(\iota_X,X)$. Since there exists a group compactification
792: covered by $(\iota_X,X)$, on which $\phi$ can be almost realized aperiodically (cf.
793: Theorem \ref{weilthr}), we can assume w.l.o.g. that $\phi$ can be almost realized aperiodically
794: already on $(\iota_X,X)$. We have to show that $(\iota_X,X)$ and
795: $(\iota_{\Gamma},C_{\Gamma})$ are equivalent.
796:
797: Let $\psi^*$ be an aperiodic realization of $\phi$ on
798: $C_{\Gamma}\cong \T^k\times F$ with $k\in\N$ and $F$ finite.
799: Let us denote the elements of $\T^k\times F$ by tuples $(\vec{\alpha},x)$.
800: For every fixed $\vec{\alpha}\in \T^k$ define
801: a mapping $\psi_{\vec{\alpha}}: F\to \R$ via
802: \[\psi^*_{\vec{\alpha}}(x):=\psi^*(\vec{\alpha},x).\]
803: For each $\overline{\chi} \in \hat{F}$, the dual of the
804: finite group $F$, define the $F$-Fourier coefficient of $\psi^*_{\vec{\alpha}}$ as
805: \[c_{\overline{\chi}}(\vec{\alpha}):= \int_F \psi^*_{\vec{\alpha}}(x)\overline{\chi}(x) dx= \frac{1}{\# F}
806: \sum_{x\in F} \psi^* (\vec{\alpha};x) \overline{\chi}(x)\in\C.\]
807: We want to show that $c_{\overline{\chi}}: \T^k\to \C$ is a Riemann integrable function:
808: The mapping $\gamma_x: \T^k\to X$ defined via
809: $\vec{\alpha}\mapsto(\vec{\alpha};x)$ is continuous and measure-preserving
810: for every $x\in F$. $\psi^*$ is by definition Riemann integrable. Thus the mapping
811: $\psi^*\circ\gamma_x: \T^k\to \C $ is Riemann integrable for each $x\in F$.
812: Note that
813: \[c_{\overline{\chi}}(\vec{\alpha})=
814: \sum_{x\in F} (\psi^*\circ\gamma_x) (\vec{\alpha}) \overline{\chi}(x).\]
815: Hence, for each fixed character $\chi \in \hat{F}$,
816: the mapping $c_{\overline{\chi}}: \T^k \to \C$ defined
817: via $\vec{\alpha}\mapsto \sum_{x\in F} (\psi^*\circ\gamma_x) (\vec{\alpha}) \overline{\chi}(x)$
818: is Riemann integrable on $\T^k$.
819:
820: Thus Proposition \ref{fejer} implies
821: $\sigma_n c_{\chi}(\vec{\alpha})\to c_{\chi}(\vec{\alpha}) $ a.e. on $\T^k$.
822: Taking into account
823: that the Haar measure on $F$ is the normalized counting measure, we get
824: \begin{equation}\label{syn} \psi^*_n(\vec{\alpha};x):=
825: \sum_{\overline{\chi}\in \hat{F}} \left(\sigma_n c_{\overline{\chi}}(\vec{\alpha})\right)
826: \overline{\chi}(x) \to \sum_{\overline{\chi}\in \hat{F}} c_{\overline{\chi}}(\vec{\alpha})\overline{\chi}(x) =
827: \psi^*_{\vec{\alpha}}(x)= \psi^*(\vec{\alpha};x)\end{equation}
828: for almost every $\vec{\alpha}\in \T^k$ and every $x\in F$, as $n\to \infty$.
829: Since Haar measure $\mu_C$ on $C$ is the product measure
830: of the Haar measures on the groups $\T^k$ and $F$, the relation (\ref{syn})
831: holds $\mu_C$-a.e. on $C$.
832: We conclude that
833: any function majorizing $\liminf_{n\to \infty} \psi^*_n$ and minorizing $\limsup_{n\to \infty} \psi^*_n$
834: is an almost realization of $\phi$.
835: Note that according to the properties of the Fej\'er kernels on $\T^k$
836: (see 44.51 in \cite{HR}) for each character
837: $(\eta\times\chi) (\vec{\alpha};x):=\eta(\vec{\alpha})\chi(x)$,
838: $\eta\in \hat{\T^k}$ and $\chi\in\hat{F}$,
839: there exists an $n_0\in N$ such that for $n\ge n_0$
840: in the Fourier expansion of $\psi^*_n$
841: the Fourier coefficient (computed in $C$) associated with the character
842: does not vanish iff the $\T^k$-Fourier coefficient of $c_{\chi}$
843: \[c_{\eta}(c_{\chi})=\int_{\T^k} c_{\chi }(\vec{\alpha})\overline{\eta}(\vec{\alpha})d\vec{\alpha}\]
844: does not vanish.
845: A simple computation shows that
846: the Fourier coefficients of $\psi^*$ are given by
847: \begin{eqnarray*}
848: c_{\eta \times \chi}(\psi^*)&=&
849: \int_{\T^k}\int_F \psi^*(\vec{\alpha},x)\overline{\eta}(\vec{\alpha})\overline{\chi}(x) d\vec{\alpha} dx \\
850: &=& \int_{\T^k} c_{\chi}(\vec{\alpha}) \overline{\eta}(\vec{\alpha}) d\vec{\alpha}
851: = c_{\eta}(c_{\chi}) \end{eqnarray*}
852:
853: So the character $\eta\times\chi$ contributes to the Fourier
854: expansion of $\psi^*$ if and only if
855: $c_{\eta \times \chi}(\psi)\neq 0$. Thus
856: $\psi^*_n\in \mbox{span } \Gamma(\phi)$ for every $n\in\N$,
857: implying that there exist almost realizations of $\phi$ on the
858: group compactification
859: induced by $\Gamma(\phi)$, e.g. $\liminf_{n\to \infty} \psi^*_n$
860: or $\limsup_{n\to \infty} \psi^*_n$.
861: \end{proof}
862:
863: Combining this result with the results of the previous section
864: we obtain
865: \begin{thr}\label{mincorap}
866: Let $\phi\in\H(G)$ and $\Gamma=\Gamma(\phi)\le \hat{G}$. The following assertions hold:
867: \begin{enumerate}
868: \item $(\iota_{\Gamma},C_{\Gamma})\le (\iota_X,X)$ for every
869: compactification $(\iota_X,X)$ on which $\phi$ can be realized.
870: In particular $\F(\phi)\subseteq \U_{(\iota_{\Gamma},C_{\Gamma})}$.
871: \item Assume that $\phi\in \A(G)$ or that $\phi$ can be realized finite
872: dimensionally. Then $\phi$ can be realized aperiodically
873: on $C_{\Gamma}$. In particular $\F(\phi)=\U_{(\iota_{\Gamma},C_{\Gamma})}$.
874: \end{enumerate}
875: \end{thr}
876:
877: We strongly conjecture that the second assertion in Theorem \ref{mincorap}
878: holds for \emph{any}
879: Hartman measurable function, at least on LCA groups $G$ with
880: separable dual $\hat{G}$. A proof of this might
881: utilize more general summation methods
882: (in the flavour of Theorems 44.43 and 44.47 in \cite{HR}) than the Fej\'er summation
883: presented here.
884:
885: In \cite{Wi} it is shown that for any Hartman measurable set $M\subseteq G=\Z$ and the
886: induced filter $\F=\F(M)$ there is an aperiodic
887: realization of $\phi_M=\Eins_M$ on the compactification determined by the subgroup
888: Sub$(M)=\{\alpha: \F\!\!-\!\!\lim_{n\in\Z} \lfloor n\alpha\rfloor=0\}$ or, equivalently,
889: Sub$(M)=\{\alpha: \F\!\!-\!\!\lim_{n\in\Z} e^{2\pi i n\alpha}=1\}$.
890:
891: Together with
892: Theorem \ref{mincorap} this implies that for Hartman sets $M$
893: with finite dimensional realization
894: both the group compactifications of $\Z$ induced by the
895: subgroups $\Gamma(\phi_M)$ and Sub$(M)$ admit aperiodic realizations of $\phi_M$.
896: Hence uniqueness of the minimal compactification with
897: aperiodic realization
898: (Corollary \ref{unique})
899: implies that in this special case $\Gamma(\phi_M)=\mbox{Sub}(M)$.
900: In the general situation we can prove up to now far only the following
901:
902: \begin{pro}
903: For a Hartman measurable function $\phi\in\H(G)$ let
904: $\F=\F(\phi)$, $\Gamma=\Gamma(\phi)$ and $\mbox{Sub}(\phi)=\{\chi\in \hat{G}:
905: \F\!\!-\!\!\lim_{g\in G} \chi(g) = 1_{\C}\}$.
906: Then $\Gamma(\phi)\le \mbox{Sub}(\phi)$.
907: \end{pro}
908: \begin{proof}
909: Suppose $\chi \in \Gamma(\phi)$. To prove $\F\!\!-\!\!\lim_{g\in G} \chi(g) = 1_{\C}$
910: (unit element of the multiplicative group of complex numbers)
911: we have to show that for every $\eps>0$ the set
912: $\{g\in G: |1-\chi(g)|<\eps\}$ belongs to the filter $\F(\phi)$,
913: i.e. that there exists $\delta=\delta(\eps)>0$ such that
914: \begin{equation}\label{filterlimes}
915: \{g\in G: m_G(|\tau_g\phi-\phi|)<\delta\}\subseteq \{g\in G: |1-\chi(g)|<\eps\}\in\F(\phi).
916: \end{equation}
917:
918: Using the fact that $m_G$ is an invariant mean and that $\chi$ is a homomorphism,
919: we have
920: \[
921: \chi(g)\,m_G(\phi\cdot \overline{\chi})= m_G(\tau_g\phi\cdot \overline{\chi})=
922: m_G( (\tau_g\phi-\phi)\cdot\overline{\chi})+ m_G(\phi\cdot \overline{\chi}).
923: \]
924: Using $\|\chi\|_{\infty}=1$ this implies
925: \begin{equation*}\label{fcesti}
926: |1-\chi(g)|\cdot |m_G(\phi\cdot \overline{\chi})| = |m_G( (\tau_g\phi-\phi)\cdot\overline{\chi})|
927: \le m_G(|\tau_g\phi-\phi|).
928: \end{equation*}
929: Since $m_G(\phi\cdot \overline{\chi})\neq 0$ we can define
930: $\delta:= \eps\cdot \frac{m_G(|\tau_g\phi-\phi|)}{|m_G(\phi\cdot \overline{\chi})|}>0$.
931: With this choice of $\delta$ indeed
932: $m_G(|\tau_g\phi-\phi|)<\delta$ implies
933: $|1-\chi(g)|<\eps$, i.e. the inclusion (\ref{filterlimes}) holds.
934: \end{proof}
935:
936: \section{Summary}
937: The content of the present paper essentially
938: deals with the definition and properties of the objects occurring
939: in the diagram below. Abusing the terminus technicus of \emph{commutative
940: diagrams} in a kind of sloppy way, the theorems of this paper
941: circle around the question under which assumptions this diagram is commutative:
942: \begin{center}
943: \includegraphics*{diagram3.eps}
944: \end{center}
945: Section 3 deals with the left half of this diagram: to every Hartman measurable function
946: $\phi$ a filter $\F(\phi)$ is associated and to every group compactification $(\iota_X,X)$ on
947: which $\phi$ can be realized a filter $\U(X,0_X)$ is associated. In general $\U(X,0_X)\supseteq
948: \F(\phi)$, and there always exists compactifications such that equality holds (indicated by
949: $\uparrow\downarrow$).
950:
951: Section 4 deals with the right half of the diagram: to every Hartman measurable function
952: $\phi$ a subgroup $\Gamma(\phi)$ of the dual is associated, which in turn induces a group
953: compactification $(\iota_{\Gamma},C_{\Gamma})$. In general
954: $(\iota_{\Gamma},C_{\Gamma})\le (\iota_X,X)$ for every group compactification $(\iota_X,X)$ on
955: which $\phi$ can be realized. If $\phi$ is either almost periodic or can be realized finite
956: dimensionally then $(\iota_A,X_A)$ is itself a group compactification on
957: which $\phi$ can be realized (indicated by
958: {\tiny${\rightarrow\atop\leftarrow}$}) and the filter $\U(X_A,0_{X_A})$ associated with this particular
959: compactification coincides with $\F(\phi)$.
960: The filter $\F(\phi)$ in turn defines a subgroup Sub($\phi$) of the dual $\hat{G}$. While it
961: can be shown that in general $\Gamma(\phi)\le\mbox{Sub}(\phi) $ it is and open problem
962: whether this inclusion can be reversed.
963:
964:
965: \begin{thebibliography}{AA}
966: \bibitem{B} R. Burckel, \emph{Weakly Almost Periodic Functions on Semigroups}, Gordon
967: and Breach, New York, 1970.
968: \bibitem{BS} M. Beiglb\"ock, C. Steineder, R. Winkler
969: \emph{Sequences and filters of characters characterizing subgroups
970: of compact abelian groups}, to appear in Top. Appl.
971: \bibitem{El} J. Elstrodt, \emph{Ma\ss- und Integrationstheorie}, Springer-Verlag
972: Berlin Heidelberg New York, 1999.
973: \bibitem{Fo} E. F\o lner, \emph{A Proof of the Main Theorem for Almost Periodic
974: Functions in an Abelian Group}, Ann. of Math., 50/5 (1949), 559-569.
975: \bibitem{FP} S. Frisch, M. Pa\v{s}teka, R. Tichy, R. Winkler, \emph{Finitely additive measures
976: on groups and rings}, Rend. Circ. Mat. Palermo, Series II, 48 (1999), 323-340.
977: \bibitem{H} S. Hartman, \emph{Remarks on equidistribution on non-compact groups},
978: Compositio Math., 16 (1964), 66-71.
979: \bibitem{HR} E. Hewitt and K. Ross, \emph{Abstract Harmonic Analysis I,II},
980: Springer-Verlag Berlin-Heidelberg-New York, 1963.
981: \bibitem{MW} G. Maresch and R. Winkler, \emph{Hartman measurable functions and
982: related function spaces}, E-print, 2005, available at \emph{www.dmg.tuwien.ac.at/maresch}
983: \bibitem{SchW} J. Schmeling, E. Szab\`o, R. Winkler, \emph{Hartman and Beatty bisequences},
984: Algebraic Number Theory and Diophantine analysis, 405-421, Walter de Gruyter, Berlin New York, 2000.
985: \bibitem{StW} C. Steineder and R. Winkler, \emph{Complexity of Hartman sequences},
986: to appear in Journal de Th\'{e}orie des Nombres de Bordeaux.
987: \bibitem{Ta} M. Talagrand, \emph{Closed convex hull of measurable functions,
988: Riemann measurable functions and measurability of trans\-lations},
989: Ann. Inst. Fourier (Grenoble), 32/1 (1982), 39-69.
990: \bibitem{Wi} R. Winkler, \emph{Ergodic Group Rotations, Hartman Sets and Kronecker
991: Sequences}, Monatsh. Math., 135 (2002), 333-343.
992: \end{thebibliography}
993:
994: \vspace{1em}\begin{center}
995: Author's address:\\
996: Gabriel Maresch\\
997: Technical University Vienna\\
998: Institute of Discrete Mathematics and Geometry\\
999: Wiedener Hauptstra\ss e 8-10\\
1000: 1040 Vienna, Austria\\
1001: email: {\ttfamily gabriel.maresch@tuwien.ac.at}\\
1002: web: http://www.dmg.tuwien.ac.at/maresch\end{center}
1003: \end{document}