1: \documentclass[12pt]{article}
2: \usepackage[intlimits]{amsmath}
3: \usepackage{amsfonts,amssymb,amscd,amsthm}
4: \input xy
5: \input epsf
6: \xyoption{all}
7: %\usepackage[cp866]{inputenc}
8: %\usepackage[russian]{babel}
9:
10: \setlength{\textheight}{22cm}
11: \setlength{\textwidth}{16cm}
12: \setlength{\oddsidemargin}{0.1cm}
13: \setlength{\topmargin}{-1cm}
14:
15: \newtheorem{theorem}{Theorem}[section]
16: \newtheorem{lemma}{Lemma}[section]
17: \newtheorem{proposition}{Proposition}[section]
18: \newtheorem{remark}{Remark}[section]
19: \newtheorem{example}{Example}[section]
20: \newtheorem{definition}{Definition}[section]
21: \newtheorem{corollary}{Corollary}[section]
22: \newtheorem{assumption}{Assumption}[section]
23:
24: \newlength{\rig}
25: \newlength{\rigg}
26: \newlength{\hei}
27: \newcommand{\dokaend}{\hfill$\square$ \vskip6truept}
28:
29: \newcommand{\fgr}[3]{
30: \setlength{\hei}{7.8cm} \setlength{\rig}{0.1\textwidth}
31: \setlength{\rigg}{0.8\textwidth}
32: \begin{figure}
33: \rule{\rig}{0in}
34: %\special{em:graph #1}\rule[-\hei]{0pt}{\hei}
35: \epsfxsize=13cm
36: \epsffile[1 1 953 638]{#1} \caption{#3}\label{#2}
37: \end{figure}
38: }
39:
40: \numberwithin{equation}{section}
41: \newcommand{\func}[1]{{\rm #1} \,}
42:
43: \newcommand{\limfunc}[1]{{\rm{#1}}}
44: %\newcommand{\text}[1]{{\mbox {#1}}}
45:
46: %\input tcilatex
47: \def\stackunder#1#2{\mathrel{\mathop{#2}\limits_{#1}}}%
48:
49: \title{\bf Index Defects in the Theory of Non-local
50: Boundary Value Problems and the $\eta$-Invariant}
51:
52: \author{{\bf A.~Yu.~Savin and B.~Yu.~Sternin}}
53:
54: \begin{document}
55: \maketitle
56:
57: \begin{abstract}
58: The paper deals with elliptic theory of boundary value problems on manifolds
59: whose boundary is represented as a covering space. We compute the index for a
60: class of non-local boundary value problems on such manifolds. For a non-trivial
61: covering, the index defect of the Atiyah--Patodi--Singer boundary value problem
62: is computed. Poincar\'{e} duality in $K$-theory of the corresponding manifolds
63: with singularities is obtained.
64: \end{abstract}
65:
66:
67:
68:
69: \section*{Introduction}
70: \addcontentsline{toc}{section}{Introduction}
71:
72: This paper deals with boundary value problems for elliptic operators on a
73: manifold whose boundary is the total space of a finite-sheeted covering. On
74: such manifolds, we consider boundary value problems for operators that do not
75: satisfy the Atiyah--Bott condition (i.e., have no well-posed classical boundary
76: value problems). Recall that this condition does not hold, in particular, for
77: the Hirzebruch and Dirac operators as well as some other related geometric
78: operators.
79:
80: We consider the following two classes of boundary value problems.
81:
82:
83:
84: \textbf{1. Non-local boundary value problems}. Let $M$ be a smooth manifold
85: such that the boundary $\partial M$ is a finite-sheeted covering with
86: projection $ \pi :\partial M\longrightarrow X.$ Then there is an isomorphism
87: $$ C^\infty \left( \partial M\right)
88: \stackrel{\beta }{\simeq }C^\infty \left( X,\pi _{!}1\right)
89: $$
90: between the space $C^\infty \left( \partial M\right) $ of smooth functions on
91: $\partial M$ and the space of sections of the vector bundle $\pi _{!}1\in
92: \limfunc{Vect}\left( X\right) $ on the base of the covering. Here $\pi _{!}1$
93: is the direct image of the trivial line bundle.
94:
95: For a scalar elliptic operator $D$ on $M$, the simplest non-local boundary
96: value problem of the type considered in this paper is
97: \begin{equation}
98: \left\{
99: \begin{array}{l}
100: Du=f, \\
101: B\beta \left. u\right| _{\partial M}=g.
102: \end{array}
103: \right. \label{sisa2}
104: \end{equation}
105: Here $u$ and $f$ are functions on $M$, $g$ is a function on $X$, and the
106: operator $B$ of boundary conditions acts also on $X.$ In terms of the original
107: manifold $M$, the boundary conditions in (\ref{sisa2}) are {\em non-local},
108: since they relate the values of functions at distinct points of $M$.
109:
110: We prove a finiteness theorem and in the case of regular coverings obtain an
111: index formula for this class of non-local boundary value problems. Without
112: going into detail at the moment, let us mention two essential features of the
113: theory.
114:
115: First, in the proof of the index theorem we embed our manifolds in the
116: classifying space of a finite group, while in the classical index theorem it
117: is suffices to use embeddings in $\Bbb{R}^N$.
118:
119: Second, the analogue of the Atiyah--Singer difference element for a non-local
120: boundary value problem is an element of the $K$-group of a {\em non-commutative
121: $C^{*}$-algebra} associated with the cotangent bundle and the covering. Recall
122: that in the classical index theorem it suffices to use topological $K$-theory.
123:
124: The index formula of this paper is given in a form resembling the $K$-theoretic
125: statement of the Atiyah--Singer theorem. Local index formulae will appear
126: elsewhere.\vspace{1mm}
127:
128:
129:
130:
131: \textbf{2. Spectral problems on manifolds with a covering}. The first
132: generalization of classical boundary value problems that is free of the
133: Atiyah--Bott obstruction is due to Atiyah, Patodi, and Singer \cite{APS1}. For
134: a class of first-order elliptic operators, one has so-called \emph{spectral
135: boundary value problems} denoted by $\left(D,\Pi_{+}\right).$ Spectral boundary
136: value problems enjoy the Fredholm property. However, their index is not
137: determined by the principal symbol of $D$.
138:
139: Interesting invariants arise if the boundary has the structure of a covering.
140: Here we consider a class of elliptic operators that are lifted from the base of
141: the covering in a neighbourhood of the boundary. In this case, the principal
142: symbol of an elliptic operator $D$ defines an element
143: \[
144: \left[ \sigma \left( D\right) \right] \in K^0\left( \overline{T^{*}M}^\pi
145: \right)
146: \]
147: in the $K$-group of the singular space $\overline{T^{*}M}^\pi $ obtained from
148: the cotangent bundle $T^{*}M$ if we identify all points in each fiber of the
149: covering (for details, see Section \ref{defe3}). The element $\left[ \sigma
150: \left( D\right) \right] $ has a topological index
151: \[
152: \limfunc{ind}_t\left[ \sigma \left( D\right) \right] \in \Bbb{Q}/n%
153: \Bbb{Z},
154: \]
155: where $n$ is the number of sheets. However, the analytical and the topological
156: index coincide only for trivial coverings. For a general covering, we obtain
157: the \emph{index defect formula}
158: \begin{equation}
159: \func{mod}n\text{-}\limfunc{ind}\left( D,\Pi _{+}\right) -\limfunc{ind}%
160: _t\left[ \sigma \left( D\right) \right] =\eta \left( \left.
161: D\right| _{X}\otimes 1_{n-\pi _{!}1}\right) \in \Bbb{Q}/n\Bbb{Z}.
162: \label{def2}
163: \end{equation}
164: The index defect (the difference between the analytical index modulo $n$ and
165: the topological index) is equal to the relative Atiyah--Patodi--Singer
166: $\eta$-invariant of the restriction of $D$ to the boundary with coefficients in
167: the flat bundle $\pi _{!}1.$ For a trivial covering, the relative $\eta
168: $-invariant is zero, and the index defect formula becomes the index formula
169: \[
170: \func{mod}n\text{-}\limfunc{ind}\left( D,\Pi _{+}\right) =\limfunc{ind}%
171: _t\left[ \sigma \left( D\right) \right]
172: \]
173: due to Melrose and Freed \cite{FrMe1} (see also \cite{Hig1,Zha1,Bot1,Ros2}).
174: However, our proof is new even in this case. It is interesting to note that the
175: main step in the proof is to realize the fractional analytic invariant
176: \begin{equation}
177: \func{mod}n\text{-}\limfunc{ind}\left( D,\Pi _{+}\right) -\eta \left( \left.
178: D\right| _{X}\otimes 1_{n-\pi _{!}1}\right) \in \Bbb{Q}/n\Bbb{Z}
179: \label{in7}
180: \end{equation}
181: as the index of some non-local boundary problem of the form (\ref{sisa2}) (in a
182: suitable elliptic theory with coefficients).
183:
184: There is also a deeper relation between the two elliptic theories described in
185: Subsec.~1 and 2. \vspace{1mm}
186:
187: {\bf 3. Poincar\'{e} isomorphism and duality.} We establish {\em Poincar\'{e} isomorphisms} on
188: the singular spaces $\overline{T^*M}^\pi$ and $\overline{M}^\pi$. For the identity covering
189: $\pi=Id,X=\partial M$, these isomorphisms are just the well-known isomorphisms (e.g., see
190: \cite{Con1,Kas3,MePi2})
191: \begin{equation}
192: \label{alox2}
193: K^0(T^*M)\simeq K_0(M,\partial M), \quad
194: K^0(T^*(M\setminus\partial M))\simeq K_0(M).
195: \end{equation}
196: (For non-compact spaces, we use $K$-theory with compact supports.) In contrast
197: to the smooth case, the Poincar\'{e} isomorphisms for singular spaces relate
198: the $K$-groups of a commutative algebra of functions to those of a dual
199: non-commutative algebra. They are defined on the elements as {\em
200: quantizations}, i.e., take symbols to operators. More precisely, the analogue
201: of the first isomorphism in (\ref{alox2}) is defined in terms of the operators
202: described in Subsec.~2, while in the second case one uses non-local problems
203: introduced in Subsec.~1.
204: %It should be mentioned that in the $C^*$-algebra $K$-theory one knows an abstract form of
205: %duality (see \cite{Hig2,Pash1}).
206:
207: Let us outline the contents of the paper. The first section contains the
208: definition of the class of non-local boundary value problems on manifolds with
209: a covering on the boundary and a proof of the Fredholm property. The index
210: formula is obtained in Sec.~2. By way of example, we define a non-local
211: boundary value problem for the Hirzebruch operator on a manifold with
212: reflecting boundary. In Sec.~\ref{fama1}, we give the homotopy classification
213: of non-local problems. The index defect formula (\ref{def2}) is proved in
214: Sec.~\ref{defe3}. This is one of the central results of the paper. Section
215: \ref{exa4} contains applications to the computation of the fractional part of
216: the $\eta $-invariant. It is also shown that the invariant (\ref{in7}) can be
217: computed by the Lefschetz formula. Poincar\'{e} isomorphisms in $K$-theory of
218: the singular spaces corresponding to manifolds whose boundary bears the
219: structure of a covering are constructed in the last two sections.
220:
221:
222: There are other interesting classes of non-local boundary value problems
223: arising if the projection has singularities (e.g., the projection on the
224: quotient by a non-free action of a finite group). Index theory of such boundary
225: value problems is apparently related to index theory on orbifolds (\cite{Kaw1,
226: Far1}). Our approach is advantageous in that if the base of the covering is
227: smooth, then there are no additional analytic and topological difficulties
228: related to the singularities of the covering. More general classes of non-local
229: boundary value problems (e.g., see \cite{Ant1}) are beyond the scope of this
230: paper.\vspace{1mm}
231:
232:
233: \textbf{Acknowledgements}. The results were announced at the conferences
234: ``Spring School 2001'' in Potsdam, Germany, ``Topology, analysis, and related
235: topics'' in Moscow, 2001, and at the International Congress of Mathematicians
236: in Beijing, 2002. We are grateful to V.E.~Nazaikinskii and V.~Nistor for
237: helpful discussions. The work was supported in part by RFBR grants Nos.
238: 03-02-16336, 02-01-00118, and 02-01-00928. We are grateful to A.S.~Mishchenko,
239: for finding an error in the original version of the paper and to the referee
240: for helpful remarks.
241:
242: \section{Non-local boundary value problems}\label{para1}
243:
244: \textbf{1}. \textbf{Coverings and non-local operators}. Let $Y$ be a finite
245: covering over a manifold $X$ with projection $\pi :Y\longrightarrow X.$ The
246: projection defines the direct image mapping
247: \[
248: \pi _{!}:\limfunc{Vect}\left( Y\right) \longrightarrow \limfunc{Vect}\left(
249: X\right)
250: \]
251: that takes each vector bundle $E\in \limfunc{Vect}\left(Y\right) $ to the
252: bundle
253: \[
254: \text{ }\pi _{!}E\in \limfunc{Vect}\left( X\right) ,\quad \left( \pi
255: _{!}E\right) _x=C^\infty \left( \pi ^{-1}\left( x\right) ,E\right) ,\quad x\in
256: X.
257: \]
258: This clearly gives an isomorphism
259: $\beta _E:C^\infty \left( Y,E\right) \stackrel{\simeq }{\longrightarrow }%
260: C^\infty \left( X,\pi _{!}E\right)$ of section spaces on $Y$ and $X$, while
261: permits one to identify operators defined on the total space and on the base.
262: More precisely, the direct image
263: \[
264: \pi _{!}D=\beta _ED\beta _E^{-1}:C^\infty \left( X,\pi _{!}E\right)
265: \longrightarrow C^\infty \left( X,\pi _{!}E\right)
266: \]
267: of a differential operator
268: \[
269: D:C^\infty \left( Y,E\right) \longrightarrow C^\infty \left( Y,E\right)
270: \]
271: on $Y$ is a differential operator. However, the following example shows that
272: the inverse image
273: \begin{equation}
274: \pi ^{!}D^{\prime }=\beta _E^{-1}D^{\prime }\beta _E:C^\infty \left(
275: Y,E\right) \longrightarrow C^\infty \left( Y,E\right) \label{nona}
276: \end{equation}
277: of a differential operator $D^{\prime } $ on $X$ may well be a \emph{non-local
278: operator}. (It is not even pseudolocal.)
279:
280: \begin{example}
281: \emph{For the trivial covering}
282: \[
283: Y=\stackunder{n\;\;{\emph{copies}}}{\underbrace{X\sqcup X\sqcup ...\sqcup
284: X}}\longrightarrow X
285: \]
286: \emph{\ and the trivial bundle }$E=\Bbb{C},$\emph{\ we have }$\pi
287: _{!}E=\Bbb{C}^n.$\emph{\ The direct image}
288: \[
289: \pi _{!}D=\limfunc{diag}\left( \left. D\right| _{X_1},...,\left. D\right|
290: _{X_n}\right) :C^\infty \left( X,\Bbb{C}^n\right) \longrightarrow C^\infty
291: \left( X,\Bbb{C}^n\right)
292: \]
293: {\em of a differential operator on }$Y$\emph{\ is always a diagonal operator,
294: and hence the inverse image of a non-diagonal operator can not be a
295: differential operator. The off-diagonal entries produce non-local operators on}
296: $Y$\emph{, since they interchange the values of functions on different leaves
297: of the covering.}
298: \end{example}
299:
300: \textbf{2. Non-local boundary value problems}. Let $M$ be a smooth compact
301: manifold with boundary $\partial M.$ Suppose that the boundary is a covering
302: space over a smooth closed manifold $X$ with projection
303: \[
304: \pi :\partial M\longrightarrow X.
305: \]
306: We fix a collar neighbourhood $\partial M\times \left[0,1\right) $ of the
307: boundary. The normal coordinate will be denoted by $t$.
308:
309: For a smooth function $u\in C^\infty \left( M\right) $, let
310: \[
311: j_{\partial M}^{m-1}u=\left( \left. u\right| _{\partial M},\left. -i\frac
312: \partial {\partial t}u\right| _{\partial M},\ldots ,\left. \left( -i\frac
313: \partial {\partial t}\right) ^{m-1}u\right| _{\partial M}\right)
314: \]
315: be the restriction of its $(m-1)$st jet in the normal direction to the
316: boundary. The operator $j_{\partial M}^{m-1}$ is continuous in the Sobolev
317: spaces
318: \[
319: j_{\partial M}^{m-1}:H^s\left( M\right) \longrightarrow
320: \bigoplus_{k=0}^{m-1}H^{s-1/2-k}\left( \partial M\right) ,\qquad s>m-1/2.
321: \]
322: Throughout the paper we assume that for vector bundles $E$ on manifolds with
323: boundary there are given isomorphisms $p^*(\left.E\right|_{\partial M})\simeq
324: \left. E\right|_{\partial M\times[0,1]}$ in the collar neighbourhood of the
325: boundary, where $p:\partial M\times[0,1]\to
326: \partial M$ is the natural projection. In this case, the normal jet of a section of
327: $E$ is also well defined.
328: \begin{definition}
329: A non-local boundary value problem\emph{\ for a differential operator}
330: \[
331: D:C^\infty \left( M,E\right) \longrightarrow C^\infty \left( M,F\right)
332: \]
333: \emph{of order }$m$\emph{\ is a system of equations}
334: \begin{equation}
335: \left\{
336: \begin{array}{ll}
337: Du=f, & u\in H^s\left( M,E\right) ,f\in H^{s-m}\left( M,F\right) ,%
338: \vspace{2mm} \\
339: B\beta _Ej_{\partial M}^{m-1}u=g,\qquad & g\in H^\delta \left( X,G\right) ,
340: \end{array}
341: \right. \label{non1}
342: \end{equation}
343: \emph{\ where the boundary condition is defined by a pseudodifferential operator}
344: \[
345: B:\bigoplus_{k=0}^{m-1}H^{s-1/2-k}\left( X,\pi _{!}\left. E\right|
346: _{\partial M}\right) \longrightarrow H^\delta \left( X,G\right)
347: \]
348: \emph{\ on }$X.$\emph{\ We assume that the component}
349: \[
350: B_k:H^{s-1/2-k}\left( X,\pi _{!}\left. E\right| _{\partial M}\right)
351: \longrightarrow H^\delta \left( X,G\right)
352: \]
353: \emph{of }$B$\emph{\ has the order} $s-1/2-k-\delta. $
354: \end{definition}
355:
356: \begin{remark}
357: {\em One can also consider problems similar to (\ref{non1}) in which the
358: components of the vector function $g$ belong to Sobolev spaces of different
359: orders. The case in which all components have the same order $\delta$ is more
360: convenient and can always be achieved by order reduction. }
361: \end{remark}
362:
363:
364: Note that for the identity covering $\pi =Id,$ $X=\partial M$, problem
365: (\ref{non1}) is just a classical boundary value problem
366: (e.g., see \cite{Hor3}).\vspace{2mm}%
367:
368: \textbf{3. Relation to classical boundary value problems. Finiteness theorem}.
369: Note that $\partial M\times \left[ 0,1\right) $ is also a covering with
370: projection
371: \[
372: \pi \times 1:\partial M\times \left[ 0,1\right) \longrightarrow X\times
373: \left[ 0,1\right) .
374: \]
375: The induced isomorphism of function spaces will be denoted by
376: \[
377: \beta _E^{\prime }:C^\infty \left( \partial M\times \left[ 0,1\right)
378: ,E\right) \longrightarrow C^\infty \left( X\times \left[ 0,1\right) ,\pi
379: _{!}E\right) .
380: \]
381: The non-local problem $\left( D,B\right) $ can be represented in a
382: neighbourhood of the boundary as the inverse image of the classical boundary
383: value problem
384: \[ \left(
385: \begin{array}{cc}
386: \beta _F^{\prime } & 0 \\
387: 0 & 1
388: \end{array}
389: \right) \circ \left( %
390: \begin{array}{c}
391: D \\
392: B\beta _Ej_{\partial M}^{m-1}
393: \end{array}
394: \right) \circ \left( \beta _E^{\prime }\right) ^{-1}=\left( %
395: \begin{array}{c}
396: \beta _F^{\prime }D\left( \beta _E^{\prime }\right) ^{-1}\vspace{2mm} \\
397: Bj_X^{m-1}
398: \end{array}
399: \right).
400: \]
401: More specifically, this is the boundary value problem
402: \begin{equation}
403: \left(
404: \begin{array}{c}
405: \beta _F^{\prime }D\left( \beta _E^{\prime }\right) ^{-1}\vspace{2mm} \\
406: Bj_X^{m-1}
407: \end{array}
408: \right) :C^\infty \left( X\times \left[ 0,1\right) ,\pi _{!}E\right)
409: \longrightarrow
410: \begin{array}{c}
411: C^\infty \left( X\times \left[ 0,1\right) ,\pi _{!}F\right) \\
412: \oplus \\
413: C^\infty \left( X,G\right) ,
414: \end{array}
415: \label{cla2}
416: \end{equation}
417: for the differential operator $\left( \pi \times 1\right) _{!}D=\beta
418: _F^{\prime }D\left( \beta _E^{\prime }\right) ^{-1}$ on the cylinder $X\times
419: \left[ 0,1\right) .$ In local coordinates, this operator is represented by a
420: diagonal matrix with elements acting on different leaves of the covering.
421: Problem (\ref{cla2}) will be denoted by $\left( \left( \pi \times 1\right)
422: _{!}D,B\right)$ for short.
423:
424: We point out that the \emph{classical boundary value problem}(\ref{cla2})%
425: \emph{\ is defined only in a neighbourhood of the boundary}, since the covering
426: is defined only near the boundary.
427:
428: \begin{definition}
429: \emph{Problem }$\left( D,B\right) $\emph{\ is said to be }elliptic\emph{ if
430: }$D$\emph{\ is elliptic and} $\left( \left( \pi \times 1\right) _{!}D,B\right)
431: $\emph{\ is elliptic, i.e., satisfies the Shapiro--Lopatinskii
432: condition (e.g., see \cite{Hor3}).}%
433: \end{definition}
434:
435: The proof of the following finiteness theorem is standard.
436:
437: \begin{theorem}
438: \label{fini}An elliptic boundary value problem $\mathcal{D}=\left(
439: D,B\right) $ defines a Fredholm operator.
440: \end{theorem}
441:
442: \emph{Proof}. Let $D^{-1}$ be the parametrix of $D$ in the interior of the
443: manifold. Similarly, the parametrix of the classical boundary value problem on
444: $X\times \left[0,1\right) $ will be denoted by $(L,K).$ They can be pasted
445: together globally on $M$ by the formula
446: \[
447: \mathcal{D}^{-1}=\left(\psi _1D^{-1}\varphi _1+\psi _2\left( \pi \times 1\right) ^{!}L\varphi
448: _2,\psi_2 K\right).
449: \]
450: Here
451: \begin{equation}
452: \begin{array}{c}
453: \varphi_1+\varphi_2=1,\quad \psi_j\varphi_j=\varphi_j\\
454: \psi_1=0 \text{ near the boundary }, \psi_2=0 \text{ far from the boundary.}
455: \end{array}
456: \label{usa2}
457: \end{equation}
458: Furthermore, $\psi_2$ is assumed to be constant in the fiber of $\pi\times 1$.
459: Obviously, $\mathcal{D}^{-1}$ is a two-sided parametrix of $\mathcal{D}$. The
460: proof is complete. \dokaend
461:
462:
463: \section{The index of non-local problems\label{hom2}
464: }
465:
466: In the previous section, non-local boundary value problems were represented
467: near the boundary in terms of equivalent classical boundary value problems.
468: Therefore, we can apply well-known topological methods (e.g., see
469: \cite{SaScS4}) to compute the index of non-local boundary value
470: problems.\vspace{1mm}
471:
472:
473: \textbf{1. Reduction to zero-order operators}. We introduce a class of
474: operators that are non-local in a neighbourhood of the boundary. A linear
475: operator
476: \[
477: D:C^\infty \left( M,E\right) \longrightarrow C^\infty \left( M,F\right)
478: \]
479: will be called an \emph{admissible operator of order $m$} if it can be
480: represented modulo operators with smooth kernels as
481: \begin{equation}
482: D=\psi _1D^{\prime }\varphi _1+\psi _2\left( \pi \times 1\right) ^{!}D^{\prime
483: \prime }\varphi _2 \label{dek1}
484: \end{equation}
485: for cutoff functions $\varphi _{1,2},\psi _{1,2}$ as in the proof of Theorem
486: \ref{fini}, a pseudodifferential operator $D^{\prime }:C^\infty \left(
487: M,E\right) \rightarrow C^\infty \left( M,F\right),$ and an operator
488: \[
489: D^{\prime \prime }:C^\infty \left( X\times \left[ 0,1\right) ,\pi _{!}E\right)
490: \longrightarrow C^\infty \left( X\times \left[ 0,1\right) ,\pi _{!}F\right)
491: \]
492: that is a sum of a pseudodifferential operator with compactly supported kernel
493: on $X\times(0,1)$ and a differential operator
494: \begin{equation}
495: \sum_{k=0}^mD_k\left( t\right) \left( -i\frac \partial {\partial t}\right)
496: ^{m-k} \label{alef1}
497: \end{equation}
498: with respect to the normal variable $t$.
499:
500: Here the $D_k\left( t\right) $\ are smooth families of pseudodifferential
501: operators on $X$\ of order $k$ and $D_0\left( t\right) $ is induced by a vector
502: bundle isomorphism.
503:
504: To this class of operators, one can extend the notion of ellipticity, the
505: statement of non-local boundary value problems, and the finiteness theorem (cf.
506: a similar generalization in \cite{Hor3} for the classical case). In particular,
507: the \emph{symbol of an admissible operator} is a pair $(\sigma_M,\sigma_X),$
508: where $\sigma_M:p^*E\to p^*F$ is defined over $M\setminus(\partial
509: M\times[0,\varepsilon))$ ($p:S^*M\to M$ is the natural projection) and
510: $\sigma_X:p^*_0(\pi_!E)\to p^*_0(\pi_!F)$ ($p_0:S^*(X\times[0,1])\to
511: X\times[0,1]$) is defined over $X\times[0,1]$. Moreover, the symbols are smooth
512: and satisfy the compatibility condition
513: $$
514: ({\pi_0})_!\left.\sigma_M\right|_{\partial M\times(\varepsilon,1)}= \sigma_X,
515: $$
516: where the direct image is induced by the natural projection $\pi_0:\left.T^*M\right|_{\partial
517: M\times(\varepsilon,1)}\to T^*(X\times (\varepsilon,1)).$
518:
519:
520:
521: \begin{example}
522: \label{exo1}\emph{Let }$E\in \limfunc{Vect}\left( M\right) $\emph{\ be a vector
523: bundle. Suppose that its direct image over $U_{\partial M}$ is decomposed as a
524: sum of two subbundles}
525: \[
526: \pi _{!}\left. E\right| _{U_{\partial M}}=E_{+}\oplus E_{-},\qquad E_{\pm }\in
527: \limfunc{Vect}\left( X\times[0,\varepsilon)\right) .
528: \]
529: \emph{Consider the operator} $D_\pm:C^\infty(M,E)\to C^\infty(M,E)$ {\em given
530: by}
531: \begin{equation}
532: D_{\pm }=\psi _2\left( \pi \times 1\right) ^{!}\left[ \left( -i\frac
533: \partial {\partial t}+i{\Lambda _{X,E_{+}}}\right) \oplus \left( i\frac
534: \partial {\partial t}+i{\Lambda _{X,E_{-}}}\right) \right] \varphi
535: _2+\psi _1i{\Lambda _M}\varphi _1. \label{kru2}
536: \end{equation}
537: \emph{(Here }$\Lambda $%
538: \emph{\ stands for first-order pseudodifferential operators with principal
539: symbol $|\xi|$ on the corresponding manifolds, and the cutoff functions are
540: chosen as before.) This formula defines an admissible elliptic operator. We
541: equip it with the Dirichlet boundary condition}
542: \[
543: P_{E_{-}}\beta_E \left. u\right| _{\partial M}=g\in C^\infty \left( X,E_{-}\right) ,
544: \]
545: \emph{where }$P_{E_{-}}:\pi _{!}\left. E\right| _{\partial
546: M}\rightarrow \pi _{!}\left. E\right|
547: _{\partial M}$\emph{\ is a projection onto the subbundle }$%
548: E_{-}.$\emph{\ Denote this boundary value problem } \emph{\ by }
549: $\mathcal{D}%
550: _{\pm }.$\emph{\ By analogy with the classical case (e.g., see \cite{Hor3}),
551: one proves that the index of this boundary value problem is zero}.
552:
553: \emph{For example, let} $E_+= \pi _{!}\left. E\right| _{\partial M}$ and
554: $E_{-}=0.$ \emph{Then the operator (\ref{kru2}), which will be denoted by
555: }$D_{+}$\emph{, is Fredholm without any boundary condition.}
556: \end{example}
557:
558: \begin{remark}\emph{
559: Just as in the classical elliptic theory on a closed manifold
560: (see~\cite{AtSi1}), there are two equivalent definitions of homotopy of
561: non-local elliptic problems. First, one can say that two problems are homotopic
562: if they can be connected by a family of non-local elliptic problems continuous
563: in the operator norm (in some given pair of Sobolev spaces). Second, two
564: problems are said to be homotopic if there exists a continuous homotopy of
565: their principal symbols (preserving ellipticity). The equivalence of the two
566: definitions is based on the smoothing of continuous homotopies and the
567: standard norm estimates modulo compact operators, e.g., see \cite{KoNi1}.}
568: \end{remark}
569:
570:
571:
572:
573:
574: Let\footnote{Later on, by $\limfunc{Ell}^m\left( M,\pi \right)$ we denote also
575: the corresponding Grothendieck groups for closed manifolds and for manifolds
576: with boundary with projection $\pi$ defined, possibly, on an open subset. Which
577: group is meant is always clear from the context.} $\limfunc{Ell}^m\left( M,\pi
578: \right), $ $m\geq 1$, be the Grothendieck group of the semigroup of homotopy
579: classes of elliptic boundary value problems for admissible operators of order
580: $m$ modulo boundary value problems of the form $\mathcal{D}_{\pm }\circ
581: D_{+}^{m-1}.$
582:
583: The group of stable homotopy classes of zero-order admissible elliptic
584: operators is denoted by $\limfunc{Ell}^0\left( M,\pi \right) $. Recall that
585: stabilization is taken modulo trivial operators. In this case, by trivial
586: operators we mean operators induced by vector bundle isomorphisms. It should be
587: noted that elliptic operators of order zero do not require boundary conditions,
588: since near the boundary they are induced by vector bundle isomorphisms.
589:
590: Just as in the classical theory (see \cite{SaScS4} or \cite{Hor3}), the order
591: of a non-local boundary value problem can be reduced to zero by stable
592: homotopies. More precisely, the following theorem holds.
593:
594: \begin{theorem}[\rm order reduction]\label{th0q}
595: The composition with the operator $D_{+}$ \emph{(}with
596: coefficients in vector bundles\emph{)} induces an isomorphism
597: \[
598: \begin{array}{ccc}
599: \times D_{+}^m:\limfunc{Ell}^0\left( M,\pi \right) & \longrightarrow
600: & \limfunc{Ell}^m\left( M,\pi \right) ,\vspace{2mm} \\
601: \qquad\left[ D\right] & \mapsto & \left[ D\circ D_{+}^m\right] .%
602: \end{array}
603: \]
604: \end{theorem}
605:
606: The \emph{proof} of this result is a straightforward generalization of the
607: corresponding proof in the classical case (see \cite{SaScS4}) and hence is
608: omitted. \dokaend
609:
610: \begin{remark}
611: {\em \label{remk} Let us explicitly describe order reduction, i.e., the mapping
612: $\left(\times D^m_+\right)^{-1}$, in the important special case of boundary
613: value problems
614: \[
615: \left\{
616: \begin{array}{lc}
617: Du=f, & \\
618: P\beta _E\left( \left. u\right| _{\partial M}\right) =g, & \quad g\in C^\infty(X,{\rm Im} P),
619: \end{array}
620: \right.
621: \]
622: for a first-order admissible operator $D:C^\infty \left( M,E\right)
623: \rightarrow C^\infty \left( M,F\right) $ that admits a decomposition
624: \[
625: \left( \pi \times 1\right) _{!}\left( \left. D\right| _{U_{\partial M}}\right)
626: =\Gamma \left( \frac \partial {\partial t}+A\left( t\right) \right)
627: \]
628: in a neighbourhood of the boundary, where $A\left( t\right) $ is a smooth
629: operator family on $X$ and $\Gamma:\pi_!E|_{\partial M}\to \pi_!F|_{\partial
630: M}$ is a vector bundle isomorphism. The boundary condition is defined by the
631: projection
632: $P$\ in the bundle $\pi _{!}\left( \left.E\right|
633: _{\partial M}\right).$ We assume for simplicity that the symbol $a(x,\xi)$ of
634: $A\left( 0\right) $ is symmetric and additionally satisfies $a^{*}a=\left| \xi
635: \right| ^2.$ We assume that $P$ is also symmetric.
636:
637: Let $L_{+}\left( A\left( 0\right) \right)\in {\rm Vect}(S^*X) $ be the
638: Calder\'on bundle. For our first-order operator, this is the bundle over
639: $S^{*}X$ generated by eigenvectors of $a\left(x, \xi \right) $ with positive
640: eigenvalues.
641:
642: The ellipticity condition for $\left( D,P\right) $ requires that $P$ define an
643: isomorphism
644: \[
645: L_{+}\left( A\left( 0\right) \right) \stackrel{P}{\longrightarrow }p _0^{*}%
646: \func{Im}P,\qquad p _0:S^{*}X\rightarrow X,
647: \]
648: of subbundles. Consider the principal symbol of our operator on the boundary:
649: \[
650: \sigma \left( \frac \partial {\partial t}+A\left( 0\right) \right) =i\tau +a\left(x, \xi \right)
651: \]
652: (here $\tau$ is dual to $t$). The linear homotopy
653: \[
654: \left( 1-\varepsilon \right) \left( i\tau +a\left(x, \xi \right) \right)
655: +\varepsilon \left( 2P(x)-1\right) ,\quad \varepsilon \in \left[ 0,1\right] ,
656: \]
657: is a homotopy of elliptic symbols for $\tau ^2+\xi ^2=1$ provided that the
658: ellipticity condition for $(D,P)$ is satisfied. Furthermore, at the end of the
659: homotopy (for $\varepsilon=1$) the symbol does not depend on the cotangent
660: variables. Let us treat the homotopy of elliptic symbols on $X$ as an elliptic
661: symbol on $X\times \left[ 0,1\right] $. Then the symbol of $D$ and the homotopy
662: taken together define the symbol of an admissible elliptic operator of order
663: zero on the manifold $M$\ with $\left[ 0,1\right] \times \partial M$ attached.
664:
665:
666: This zero-order symbol can be transferred to $M$ by an obvious diffeomorphism
667: $M\simeq M\cup_{\partial M}\left(\left[0,1\right]\times\partial M\right)$ that
668: is equal to identity far from the boundary. One can show (cf. \cite{SaScS4})
669: that the element defined by this symbol (operator) is precisely the image of
670: the problem
671: $\left( D,P\right) $ under the order reduction mapping $%
672: \left( \times D_{+}\right) ^{-1}$ of Theorem \ref{th0q}. }
673: \end{remark}
674:
675:
676: \textbf{2. Admissible operators on closed manifolds}. Let $\overline{U}$ be a
677: codimension zero submanifold of some closed manifold $M$. We assume that
678: $\overline{U}$ is a covering space
679: \[
680: \overline{U}\stackrel{\pi }{\longrightarrow }\overline{Y}
681: \]
682: with smooth base $\overline{Y}.$ Let $U$ and $Y$ be the corresponding sets of
683: interior points (we allow $\overline{U}$ to have a boundary). Then {\em scalar
684: admissible operators} on $M$ are by definition operators of the form
685: \[
686: D=D^{\prime }+\psi \left( \pi ^{!}D^{\prime \prime }\right) \varphi ,
687: \]
688: where $D^{\prime }$ is a pseudodifferential operator on $M,$ $D^{\prime \prime }$ is a
689: pseudodifferential operator on $Y$ acting on sections of $\pi _{!}1\in \limfunc{%
690: Vect}\left( Y\right) ,$ and the cutoff functions $\varphi$ and $\psi $ are
691: supported in $U.$
692:
693: In the non-scalar case, we consider operators acting in the spaces slightly
694: more general than section spaces of vector bundles.
695:
696: Namely, consider triples $\left( E,E_0,\alpha \right)$ defined by vector bundles
697: $$E\in \limfunc{Vect}\left( V\right) ,\;\; E_0\in
698: \limfunc{Vect}\left( Y\right)
699: $$
700: (here we fix a neighbourhood $V\subset M$ of $M\backslash U$ such that if a
701: point lies in $U\cap V$ then the entire fiber containing this point also lies
702: in $U\cap V$) and a vector bundle isomorphism
703: $$
704: \pi _{!}\left. E\right|_{U\cap V}\stackrel{\alpha}\simeq E_0|_{\pi\left(U\cap V\right)} $$ on
705: $\pi\left(U\cap V\right)$.
706:
707: Let ${\rm Vect}(M,\pi)$ be the set of isomorphism classes of such triples. Here
708: two triples $(E,E_0,\alpha),(F,F_0,\gamma)$ are \emph{isomorphic} if the vector
709: bundles are pairwise isomorphic,
710: $E\stackrel{a}{\simeq}F,E_0\stackrel{b}{\simeq}F_0$, and the isomorphisms are
711: compatible: $\gamma(\pi_!a)=b\alpha$.
712:
713:
714: The linear \emph{space of sections} corresponding to the triple
715: ${\cal E}=\left(E,E_0,\alpha\right)$ is defined as
716: \[
717: C^\infty \left( M,{\cal E}\right) =\left\{ \left( u,v\right) \;\left|
718: \begin{array}{c}
719: u\in C^\infty \left( V,E\right) ,v\in C^\infty \left( Y,E_0\right) , \\
720: \alpha \beta _E\left( \left. u\right| _{U\cap V}\right) =\left. v\right|%
721: _{\pi \left( U\cap V\right) }%
722: \end{array}
723: \right. \right\} \subset C^\infty \left( V,E\right) \oplus C^\infty \left( Y,E_0\right) .
724: \]
725: For the identity covering, ${\cal E} $ defines a vector bundle on $M$ obtained
726: by clutching $E$ with $E_0$ by the transition function $\alpha$, and $C^\infty
727: \left( M,{\cal E}\right) $ is just the space of sections of $\mathcal{E}$.
728:
729: The space $C^\infty \left( M,{\cal E}\right)$ is generated by the subspaces
730: \[
731: C_0^\infty \left( V,E\right) ,C_0^\infty \left( Y,E_0\right) \subset C^\infty \left( M,{\cal
732: E}\right)
733: \]
734: of compactly supported sections. More precisely, the first embedding takes $u$
735: to the pair $(u,\widetilde{\beta_E \left. u\right| _{U\cap V}})$, where the
736: tilde stands for the extension of a function by zero at the points where the
737: function was not originally defined. Similarly, the second embedding takes $v$
738: to the pair $(\widetilde{\beta _E^{-1}\left. v\right| _{\pi(U\cap V)}},v).$ Now
739: non-local operators acting in spaces $C^\infty \left( M,{\cal E}\right) $ can
740: readily be defined by analogy with the scalar case. Namely, an \emph{admissible
741: operator of order $m$} is an operator
742: \[
743: D:C^\infty \left( M,{\cal E}\right) \longrightarrow C^\infty \left( M,{\cal
744: F}\right)
745: \]
746: that is equal, modulo operators with smooth kernel, to
747: \begin{equation}
748: D=D_1\varphi _1+D_2\varphi _2, \label{omo1}
749: \end{equation}
750: where
751: \[
752: D_1:C_0^\infty \left( V,E\right) \rightarrow C_0^\infty \left( V,F\right) ,\quad D_2:C_0^\infty
753: \left( Y,E_0\right) \rightarrow C_0^\infty \left( Y,F_0\right)
754: \]
755: are $m$th-order pseudodifferential operators with compactly supported kernels.
756: Here we assume that the cutoff function $\varphi _1$ is zero in some
757: neighbourhood of $M\backslash V$ and $\varphi _2$ is zero in a neighbourhood of
758: $M\backslash U.$
759:
760: The \emph{symbol of an admissible operator} is a pair
761: $\left(\sigma_M,\sigma_Y\right)$ of usual elliptic symbols
762: \[
763: \sigma _M:p_M^{*}\left. E\right| _{M\backslash U}\longrightarrow p_M^{*}\left. F\right|
764: _{M\backslash U},\quad \sigma _Y:p_Y^{*}\left. E_0\right| _{\overline{Y}}\longrightarrow
765: p_Y^{*}\left. F_0\right| _{\overline{Y}},
766: \]
767: where $p_M:S^{*}M\longrightarrow M$ and $p_Y:S^{*}Y\longrightarrow Y,$ are
768: compatible in the sense that
769: \[
770: \gamma (\left( \pi_0) _{!}\left. \sigma _M\right| _{\partial \overline{U}}\right) \alpha
771: ^{-1}=\sigma _Y|_{\partial \overline{Y}}.
772: \]
773:
774: Let $\limfunc{Ell}^k\left( M,\pi \right) $ be the group of stable homotopy
775: classes of admissible elliptic operators of order $k$ on $M$, modulo elliptic
776: operators with principal symbols independent of the cotangent variables.
777:
778:
779: \begin{remark}\label{rem3}
780: \emph{On manifolds with boundary, one can also consider a similar class of
781: elliptic operators and boundary value problems. More precisely, let $M$ be a
782: manifold with boundary, with a projection $\pi$ defined on a closed subset
783: $\overline{U}\subset M$ as above. We assume that $\overline{U}$ is a
784: codimension zero submanifold in the interior $M\setminus
785: \partial M$ and is the Cartesian product $[0,\varepsilon)\times
786: \overline{U_0}$ in some collar neighbourhood of the boundary for some
787: codimension zero submanifold $\overline{U_0}$ in $\partial M$. Then on $M$ we
788: consider operators similar to (\ref{omo1}), where both $D_1$ and $D_2$ are of
789: order $m$ and are differential operators with respect to the normal variables
790: in neighbourhoods of the boundaries of the corresponding manifolds (see
791: (\ref{alef1})). Such operators are considered in the spaces $C^\infty(M,{\cal
792: E})$. One considers boundary value problems of the form
793: $$
794: (D,Bj):C^\infty(M,{\cal E})\longrightarrow C^\infty(M,{\cal F}) \oplus C^\infty(\partial M,{\cal
795: G}),
796: $$
797: where $\mathcal{E},\mathcal{F}\in{\rm Vect}(M,\pi), \mathcal{G}\in{\rm
798: Vect}(\partial M,\pi|_{\partial M}),$ $j$ is the jet operator of order $m$,
799: $j:C^\infty(M,{\cal E})\to C^\infty(\partial M,{\cal E}^m|_{\partial M})$, and
800: the boundary conditions are defined by an admissible operator $B$ on the
801: boundary. One can readily extend all results of this section, including the
802: definition of trivial problems $\cal D_\pm$, the group of stable homotopy
803: classes of boundary value problems, and order reduction, to this class of
804: boundary value problems.}
805: \end{remark}
806:
807: \begin{remark}
808: {\em Let $M$ be a manifold with covering $\pi$ on the boundary. In Subsec.~1,
809: we defined the group ${\rm Ell}^m(M,\pi)$ generated by elliptic non-local
810: problems for the usual operators. At the same time, the projection $\pi\times
811: 1:\partial M\times [0,1)\to X\times [0,1)$ is defined in a collar neighbourhood
812: of the boundary, and one can consider the corresponding group ${\rm
813: Ell}^m(M,\pi\times 1)$ generated by non-local problems for admissible operators
814: in the sense of Remark~\ref{rem3}. It turns out that these two groups are
815: isomorphic under the natural mapping
816: $$
817: {\rm Ell}^m(M,\pi)\longrightarrow {\rm Ell}^m(M,\pi\times 1).
818: $$
819: This essentially follows from the isomorphism
820: ${\rm Vect}(M)\simeq{\rm Vect}(M,\pi\times 1).$ }
821: \end{remark}
822:
823: \textbf{3. Reduction to a closed manifold.} We return to the problem of
824: computing the index of non-local operators on a manifold $M$ with a covering
825: $\pi$ defined on $\partial M$. Consider an embedding $f:M\rightarrow M^{\prime
826: }$ in a closed manifold of the same dimension as $M$ (for example, $M^{\prime
827: }$ can be the double $2M=M\cup _{\partial M}M$). Just as in the classical case
828: \cite{AtSi1}, $f$ induces the direct image mapping
829: \[
830: f_{!}:\limfunc{Ell}^0\left( M,\pi \right) \longrightarrow \limfunc{Ell}^0\left(M^{\prime },\pi
831: \times 1\right) ,
832: \]
833: where $\pi \times 1$ is the extension of $\pi $ to $\partial M\times \left[
834: -1,1\right] \subset M^{\prime }.$ This mapping takes the symbol $\sigma \left(
835: D\right) =\left(\sigma _M,\sigma _X\right) $ of an elliptic
836: operator\footnote{An arbitrary operator $D':C^\infty \left( M,E\right)
837: \rightarrow C^\infty\left( M,F\right) $ is reduced to this form by adding the
838: identity operator in the sections of the complementary bundle to $F$.}
839: \[
840: D:C^\infty \left( M,E\right) \longrightarrow C^\infty \left( M,\Bbb{C}%
841: ^k\right)
842: \]
843: to the symbol on $M^{\prime }$ that coincides on $M$ with the original symbol
844: and is the identity $id:\Bbb{C}^k\rightarrow \Bbb{C}^k$ on the complement
845: $M^{\prime }\backslash M$. The extended symbol is defined on the bundle
846: obtained by clutching $E$ with $\Bbb{C}^k$ using the isomorphism $\left. \sigma
847: _X\right| _X$ and maps this bundle to the bundle $\Bbb{C}^k$ over the ambient
848: closed manifold $M^{\prime }.$
849:
850: \begin{lemma}
851: The mapping $f_{!}:\limfunc{Ell}^0\left( M,\pi \right) \longrightarrow
852: \limfunc{Ell}^0\left( M^{\prime },\pi \times 1\right) $ is well defined and is
853: index preserving.
854: \end{lemma}
855:
856: \noindent \emph{Proof}. This is a restatement of the well-known excision
857: property of the index. The proof is standard, and hence we omit it altogether.
858: \dokaend
859:
860: \textbf{4. Embedding in a universal space}. In the index theorems of the
861: present paper, we assume that the following condition is satisfied.
862: \begin{assumption}\label{as1}
863: \emph{The covering $\pi $ is regular and there is a free action of a finite
864: group $G$ on the submanifold $\overline{U}$ such that $\pi$ is the projection
865: onto the quotient.}
866: \end{assumption}
867:
868:
869: Let $\left(M,\pi\right)$ and $\left(M^{\prime},\pi^{\prime}\right) $ be two
870: pairs (both manifolds are assumed to be closed) and let $U$ and $U'$ be the
871: domains of $\pi$ and $\pi'$, respectively.
872:
873: \begin{definition}
874: \emph{We say that $f$ is an } embedding \emph{ of }$\left( M,\pi \right)
875: $\emph{\ in }$\left( M^{\prime },\pi ^{\prime }\right) $\emph{\ if there is an
876: embedding }$f:M\rightarrow M^{\prime },$\emph{\ }$f\left( \overline{U}\right)
877: \subset \overline{U}^{\prime },$\emph{\ that is equivariant on the domain of
878: $\pi$.}
879: \end{definition}
880:
881:
882:
883: Denote by $\pi _N:EG_N\longrightarrow BG_N$ the $N$-universal bundle for $G.$
884: We assume that $EG_N$ and $BG_N$ are closed manifolds. There is an explicit
885: construction for such a model (e.g., see \cite{LuMi1}). For example, consider
886: the embedding
887: \[
888: G\subset S_{\left| G\right| }\subset \Bbb{U}\left( \left| G\right| \right)
889: \]
890: in the unitary group. (Here $\left| G\right|$ is the order of $G$.) Consider
891: the bundle $V_{k,\left| G\right| }\rightarrow V_{k,\left| G\right| }/G$, where
892: $V_{k,n}$ is the Stiefel manifold of $n$-frames in $\Bbb{C}^k$. For
893: sufficiently large $k$, this bundle is $N$-universal.
894:
895: \begin{proposition}\label{pre1}
896: For $\left( M,\pi \right) $ satisfying Assumption \emph{\ref{as1}}, there
897: exists an embedding in $\left( EG_N,\pi _N\right)$ provided that $N$ is
898: sufficiently large.
899: \end{proposition}
900:
901: \noindent \emph{Proof}. By $N$-universality of $\pi _N$, there exists an
902: equivariant mapping $\overline{U}\rightarrow EG_N.$ We can assume that this
903: mapping is a smooth embedding. This can be achieved by a small deformation
904: provided that the dimension of $EG_N$ is sufficiently large.
905:
906: This embedding can be extended to a smooth mapping $M\rightarrow EG_N$ owing to
907: the $N$-connectedness of $EG_N$. Finally, a small deformation outside a
908: neighbourhood of $U$ makes it a global embedding. \dokaend
909:
910:
911: \textbf{5. The Euler operator on the disc}. Consider the Neumann problem
912: \[
913: \left\{
914: \begin{array}{c}
915: \left( d+\delta \right) u=f, \\
916: \left. \left( \ast u\right) \right| _{\Bbb{S}^{n-1}}=g,
917: \end{array}
918: \right. \quad u\in \Lambda ^e\left( \mathbb{D}^n\right) ,f\in \Lambda ^o\left(
919: \mathbb{D}^n\right),
920: \]
921: for the Euler operator in the unit disc $\mathbb{D}^n\subset \Bbb{R}^n$ with
922: the Euclidean metric. Here $g\in \Lambda ^{e+n}\left( \Bbb{S}^{n-1}\right) .$
923: This boundary value problem is elliptic, and Hodge theory shows that the
924: cokernel is trivial and the one-dimensional kernel consists of constant
925: functions.
926:
927: The same is true for the homogeneous boundary value problem, which we rewrite in the operator
928: form
929: \[
930: D_{dR}=d+\delta :\Lambda _0^e\left( \mathbb{D}^n\right) \longrightarrow \Lambda ^o\left(
931: \mathbb{D}^n\right).
932: \]
933: (Here $\Lambda _0^e\left( \mathbb{D}^n\right) $ is the space of forms
934: satisfying the homogeneous boundary condition.) This operator is $O\left(
935: n\right) $-equivariant with respect to the natural action of the orthogonal
936: group on $\mathbb{D}^n.$\vspace{1mm}
937:
938: \textbf{6. Embeddings and the index of elliptic operators}. Let $f:\left( M,\pi
939: \right)\rightarrow \left( M^{\prime },\pi ^{\prime }\right)$ be an embedding of
940: positive codimension. We choose a Riemannian metric on $M^{\prime }$ that is
941: $G$-invariant over $\overline{U}^{\prime }\subset M^{\prime }$. Denote the
942: normal bundle to $M$ by $NM.$ Then a closed tubular neighbourhood $W$ of $M$
943: in $M^{\prime } $ is diffeomorphic to the unit ball subbundle $DM\subset NM.$
944: Additionally, we can assume this diffeomorphism to be $G$-equivariant over
945: $\overline{U}\subset \overline{U}^{\prime }$.
946:
947: Consider an admissible elliptic operator
948: \[
949: D:C^\infty \left( M,{\cal E}\right) \longrightarrow C ^\infty \left( M,{\cal F}\right) .
950: \]
951:
952: We define a boundary value problem on $DM$ as the exterior tensor product of
953: $D$ by a family of boundary value problems for the Euler operator in the
954: fibers. The definition of this product is the same as in \cite{AtSi1}.
955:
956: More precisely, the exterior tensor product gives the operator
957: \[
958: \mathcal{D}=\left(
959: \begin{array}{cc}
960: \widetilde{D}\otimes 1_{\Lambda ^e} & -1_F\otimes \widetilde{D}_{dR}^{*} \\
961: 1_E\otimes \widetilde{D}_{dR} & \widetilde{D}^{*}\otimes 1_{\Lambda ^o}
962: \end{array}
963: \right),
964: \]
965: where the pullback of $D$ to the bundle $DM$ with coefficients in even forms on
966: the fibers is denoted by
967: $$
968: \widetilde{D}\otimes 1_{\Lambda ^e}: C^\infty \left( DM,p^{*}{\cal E}\otimes \Lambda _0^e\left(
969: DM\right) \right) \longrightarrow C^\infty \left( DM,p^{*}{\cal F}\otimes \Lambda _0^e\left(
970: DM\right) \right).
971: $$
972: Here $\widetilde{D}^{*}\otimes 1_{\Lambda ^o}$ is the pullback of the adjoint
973: operator with coefficients in odd forms. The family of Neumann problems for the
974: Euler operator $\widetilde{D}_{dR}^M$ with coefficients in the triple ${\cal
975: E}=\left(E,E_0,\alpha \right)$ is denoted by
976: $$
977: 1_E\otimes \widetilde{D}_{dR}:C_{\alpha \otimes 1}^\infty \left( DM,p^{*}E\otimes \Lambda
978: _0^e\left( DM\right) \right) \longrightarrow C_{\alpha \otimes 1}^\infty \left( DM,p^{*}E\otimes
979: \Lambda ^o\left( DM\right) \right).
980: $$
981: The off-diagonal entries of $\mathcal{D}$ commute with entries on the diagonal
982: by construction. As in ordinary Atiyah--Singer theory, this leads to the
983: following result.
984:
985: \begin{lemma} One has
986: $\limfunc{ind}D=\limfunc{ind}\mathcal{D}.$
987: \end{lemma}
988:
989: \noindent The \emph{proof} is similar to \cite{AtSi1}. \dokaend
990:
991: Thus an elliptic operator on the submanifold $M\subset M^{\prime }$ induces an
992: elliptic boundary value problem with the same index on the tubular
993: neighbourhood $DM\simeq W\subset M^{\prime }$. Further, we can apply the order
994: reduction procedure to this problem (see Remark~\ref{remk}) and extend the
995: resulting zero-order operator from $W$ to the entire manifold $M^{\prime} $ as
996: in Subsec.~2.
997:
998: Summarizing, we see that the embedding
999: $f$ of $\left(M,\pi\right)$ in
1000: $\left(M^{\prime},\pi^{\prime}\right) $ induces the direct image mapping
1001: \[
1002: f_{!}:\limfunc{Ell}^1\left( M,\pi \right) \longrightarrow \limfunc{Ell}^0\left( M^{\prime },\pi
1003: ^{\prime }\right),
1004: \]
1005: which preserves the index.
1006:
1007: \begin{remark}
1008: \emph{A straightforward computation shows that the linear homotopy of order
1009: reduction for boundary value problems (defined in Remark~\ref{remk}) which
1010: extends the symbol }$\sigma \left( d+\delta \right) $\emph{\ from
1011: }$T^{*}\mathbb{D}^n$\emph{\ to }$T^{*}\Bbb{R}^n$\emph{\ as an invertible
1012: element outside a compact set defines an element of the equivariant
1013: }$K$-\emph{group equal to the element}
1014: \[
1015: j_{!}\left( 1\right) \in K_{O\left( n\right) }\left( T^{*}\Bbb{R}^n\right) ,\quad
1016: j:pt\longrightarrow \Bbb{R}^n,
1017: \]
1018: \emph{\ which is used in the standard proof of the Atiyah--Singer theorem.}
1019: \end{remark}
1020:
1021: \textbf{7. The Index theorem}. Let $f$ be an embedding of $(M,\pi)$ in the
1022: universal space defined in Proposition~\ref{pre1}. For the universal space
1023: $EG_N$, the projection $\pi _N$ is defined globally. Therefore, the direct
1024: image of a non-local operator can be treated as a usual elliptic operator on
1025: the base $BG_N$; i.e., we have a natural mapping
1026: \[
1027: \left( \pi _N\right) _{!}:\limfunc{Ell}\left( EG_N,\pi _N\right) \longrightarrow
1028: \limfunc{Ell}\left( BG_N\right) \simeq K\left( T^{*}BG_N\right) .
1029: \]
1030:
1031: \begin{theorem}\label{thind1}
1032: For a pair $(M,\pi)$ satisfying Assumption~\emph{\ref{as1}}, the diagram
1033: \[
1034: \begin{array}{ccc}
1035: \limfunc{Ell}^1\left( M,\pi \right) & \stackrel{f_{!}}{\longrightarrow } &
1036: \limfunc{Ell}^0\left( EG_N,\pi _N\right) \\
1037: \limfunc{ind}\downarrow \quad & & \downarrow \left( \pi _N\right) _{!} \\
1038: \quad\Bbb{Z} & \stackrel{\limfunc{ind}_t}{\longleftarrow } & K\left( T^{*}BG_N\right) ,
1039: \end{array}
1040: \]
1041: commutes. Here $\limfunc{ind}_t$ is the usual topological index on a closed manifold.
1042: \end{theorem}
1043:
1044: \noindent \emph{Proof}. Indeed, we have
1045: \[
1046: \limfunc{ind}D=\limfunc{ind}f_{!}\left[ D\right] =\limfunc{ind}\left( \pi _N\right)
1047: _{!}f_{!}\left[ D\right] =\limfunc{ind}_t\left( \left( \pi _N\right) _{!}f_{!}\left[ D\right]
1048: \right).
1049: \]
1050: The first equality here follows from the invariance of the index for
1051: embeddings, the second from the fact that $\left( \pi_N\right) _{!}$ does not
1052: change the operator, and the last equality is just the Atiyah--Singer formula
1053: on $BG_N.$ \dokaend
1054:
1055: \textbf{8. Example}. \emph{Manifolds with reflecting boundary} \cite{Hsi2}.\
1056: Let $M$\ be a $4k$-dimensional compact oriented Riemannian manifold with
1057: boundary $\partial M$. Suppose that $\partial M$ is equipped with an
1058: orientation-reversing smooth involution $G$ without fixed points. The
1059: involution defines a free action of the group $\Bbb{Z}_2$\ and the
1060: corresponding double covering $\pi :\partial M\longrightarrow \partial
1061: M/\Bbb{Z}_2.$ Consider the Hirzebruch operator \cite{Pal1}
1062: \[
1063: d+d^{*}:\Lambda ^{+}\left( M\right) \longrightarrow \Lambda ^{-}\left( M\right).
1064: \]
1065: In a neighbourhood of the boundary, let us take a metric lifted from $\left[
1066: 0,1\right] \times
1067: \partial M/\Bbb{Z}_2.$ Then the Hirzebruch operator can be decomposed near the boundary as (see \cite{APS1})
1068: \[
1069: \frac \partial {\partial t}+A
1070: \]
1071: (up to a bundle isomorphism), where $A$ is an elliptic self-adjoint operator on
1072: the boundary and is given by the formula
1073: \[
1074: A:\Lambda ^{*}\left( \partial M\right) \longrightarrow \Lambda ^{*}\left(
1075: \partial M\right) ,\qquad
1076: A\omega =\left( -1\right) ^{k+p}\left( d*-\varepsilon *d\right) \omega;
1077: \]
1078: here for an even degree form $\omega \in \Lambda ^{2p}\left(
1079: \partial M\right) $\ we set $\varepsilon =1,$\ and $\varepsilon
1080: =-1$ otherwise. Since $G$ reverses the orientation, it follows that $A$ and
1081: $G^*$ anticommute:
1082: \[
1083: G^{*}A=-AG^{*}.
1084: \]
1085: It is known that the Hirzebruch operator has no well-posed classical boundary
1086: conditions. However, it admits the non-local boundary value problem
1087: \begin{equation}
1088: \left\{
1089: \begin{array}{ll}
1090: \left( d+d^{*}\right) \omega =f,\vspace{2mm} & \\
1091: \frac{\left( 1+G^{*}\right) }2\left. \omega \right| _{\partial M}=g, & \quad g\in \Lambda
1092: ^{*}\left( \partial M\right) ^{\Bbb{Z}_2}\simeq \Lambda ^{*}\left( \partial M/\Bbb{Z}_2\right)
1093: \end{array}
1094: \right. \label{nel1}
1095: \end{equation}
1096: on the manifold with reflecting boundary. Here $\Lambda ^{*}\left( \partial
1097: M\right) ^{\Bbb{Z}_2}$ is the subspace of $G$-invariant forms on the boundary.
1098:
1099: \begin{proposition}
1100: The non-local boundary value problem \emph{(\ref{nel1})} is elliptic.
1101: \end{proposition}
1102:
1103: \emph{Proof}. Consider an arbitrary point $x\in \partial M/\Bbb{Z}_2$. An
1104: explicit computation shows that near this point the equivalent classical
1105: boundary value problem is
1106: \[
1107: \left\{
1108: \begin{array}{c}
1109: \left( \frac \partial {\partial t}+A\right) \omega _1=f_1,\quad \left( \frac
1110: \partial {\partial t}-A\right) \omega _2=f_2,\vspace{2mm} \\
1111: \left. \omega _1\right| _{\partial M/\Bbb{Z}_2}+\left. \omega _2\right| _{\partial
1112: M/\Bbb{Z}_2}=g.
1113: \end{array}
1114: \right.
1115: \]
1116: It is elliptic (satisfies the Shapiro--Lopatinskii condition), since the symbol
1117: of the operator of boundary conditions defines an isomorphism
1118: \[
1119: \func{Im}\sigma \left( \Pi _{+}\right) \left( x,\xi \right) \oplus \func{Im}%
1120: \sigma \left( \Pi _{-}\right) \left( x,\xi \right) \simeq \Lambda ^{*}\left(
1121: \partial M\right) _x
1122: \]
1123: at an arbitrary point $\left( x,\xi \right) \in S^{*}(\partial M/\Bbb{Z}_2)$,
1124: where
1125: \[
1126: \Pi _{+}=\frac{A+\left| A\right| }{2\left| A\right| }
1127: \]
1128: is the non-negative spectral projection of $A$ and $\Pi _{-}=1-\Pi _{+}$ is the
1129: negative projection. The ellipticity of the classical boundary value problem
1130: proves the desired statement. \dokaend
1131:
1132: \begin{proposition}
1133: One has
1134: \[
1135: \limfunc{ind}\left( d+d^{*},{\left( 1+G^{*}\right) }\right) =\limfunc{%
1136: sign}M,
1137: \]
1138: where $\limfunc{sign}M$ is the signature of $M$.
1139: \end{proposition}
1140:
1141: \emph{Proof}. The symbol of (\ref{nel1}) coincides with that of the composition
1142: of the spectral Atiyah--Patodi--Singer boundary value problem
1143: \[
1144: \left\{
1145: \begin{array}{ll}
1146: \left( d+d^{*}\right) \omega =f & \\
1147: \Pi _{+}\left. \omega \right| _{\partial M}=\omega ^{\prime }, & \quad \omega ^{\prime }\in
1148: \func{Im}\Pi _{+}\subset \Lambda ^{*}\left( \partial M\right) ,
1149: \end{array}
1150: \right.
1151: \]
1152: and the Fredholm operator
1153: \begin{equation}
1154: {\left( 1+G^{*}\right) }:\func{Im}\Pi _{+}\longrightarrow \Lambda ^{*}\left( \partial M\right)
1155: ^{\Bbb{Z}_2}. \label{subs4}
1156: \end{equation}
1157: Let us compute both indices.
1158:
1159: 1) For the index of the spectral boundary value problem, one has \cite{APS1}
1160: \[
1161: \limfunc{ind}\left( d+d^{*},\Pi _{+}\right) =\limfunc{sign}M-\frac{\dim \ker A}2.
1162: \]
1163: In addition, by the Hodge--de Rham theory we obtain $ {\dim \ker A}={\dim
1164: H^{*}\left(
1165: \partial M\right) } $.
1166:
1167: 2) On the other hand, one can readily verify that the operator in
1168: Eq.~(\ref{subs4}) is surjective and its kernel coincides with the space of
1169: $G$-antiinvariant harmonic forms. The Hodge operator $*$ interchanges the
1170: antiinvariant and invariant subspaces. Thus we obtain
1171: \[
1172: \dim \ker \left. \left( 1+G^{*}\right) \right| _{\func{Im}\Pi _{+}\left( A\right) }=\frac{\dim
1173: \ker A}2.
1174: \]
1175: Adding the index of the spectral problem to the index of $\left( 1+G^{*}\right)
1176: ,$ we obtain
1177: \[
1178: \limfunc{ind}\left( d+d^{*},{\left( 1+G^{*}\right) }\right) =\limfunc{%
1179: sign}M-\frac{\dim \ker A}2+\frac{\dim \ker A}2=\limfunc{sign}M.
1180: \]
1181:
1182: The proof of the theorem is complete. \dokaend
1183:
1184:
1185:
1186:
1187:
1188:
1189: \section{The homotopy classification of non-local operators}\label{fama1}
1190:
1191: Let us cut $M$ into two parts
1192: \[
1193: M^{\prime }=M\backslash \left\{ \partial M\times \left[ 0,1\right) \right\}
1194: \simeq M\;\qquad \text{and\qquad }\partial M\times \left[ 0,1\right] .
1195: \]
1196: Then the symbol $\sigma \left(D\right) $ of an admissible elliptic operator $D$
1197: of order zero is naturally represented as a pair of usual symbols
1198: \begin{equation}
1199: \left. \sigma \left( D\right) \right| _{M^{\prime }}\text{ and }\left( \pi
1200: \times 1\right) _{!}\left. \sigma \left( D\right) \right| _{\partial M\times
1201: \left[ 0,1\right] }. \label{smb1}
1202: \end{equation}
1203: Both symbols define difference elements
1204: \[
1205: \left[ \sigma _{M}\right] \in K\left( T^{*}M^{\prime }\right) ,\qquad \left[ \sigma_X\right] \in
1206: K\left( T^{*}\left( X\times \left( 0,1\right] \right) \right).
1207: \]
1208: Here and in what follows, we use $K$-groups with compact supports. In the
1209: latter case, the elliptic symbol
1210: $\sigma_X $ of order zero is invertible over $X\times \left\{
1211: 0\right\} $ (this follows from ellipticity and the decomposition in
1212: Eq.~(\ref{alef1})) and hence defines element in the above-mentioned $K$-group
1213: with compact supports.
1214:
1215: However, it is impossible to define an element of a single topological
1216: $K$-group; indeed, the manifolds $T^{*}M^{\prime }$ and $T^{*}\left( X\times
1217: \left( 0,1\right] \right) $ can not be glued together, for their boundaries are
1218: not diffeomorphic. Nonetheless, we can glue the algebras of functions on these
1219: spaces instead of the original manifolds.
1220:
1221: \textbf{1. The $C^*$-algebra of a manifold with a covering on the boundary.} To
1222: each space, we assign an algebra of continuous functions vanishing at infinity:
1223: \[
1224: C_0\left( T^{*}M^{\prime }\right),\quad C_0\left( T^{*}\left( X\times \left( 0,1\right] \right)
1225: ,\limfunc{End}p^{*}\pi _{!}1\right).
1226: \]
1227: More precisely, on the space $T^*(X\times (0,1])$ we consider functions ranging
1228: in the set of endomorphisms of the bundle $\pi _{!}1\in \limfunc{Vect}\left(
1229: X\right) $, where $p:T^{*}\left( X\times \left( 0,1\right] \right) \rightarrow
1230: X$ is the natural projection. In the direct sum of these algebras, consider the
1231: subalgebra determined by the compatibility condition
1232: \begin{equation}
1233: \label{inty5} \mathcal{A}_{T^{*}M,\pi }=\left\{ (u,v)\;\left|
1234: \begin{array}{c}
1235: u\in C_0\left( T^{*}M^{\prime }\right), v\in C_0\left( T^{*}\left( X\times \left( 0,1\right]
1236: \right) , \limfunc{End}p^{*}\pi _{!}1\right)
1237: \vspace{1mm}\\
1238: \;\beta \left. u\right| _{\partial M^{\prime }}\beta ^{-1}=\left. v\right| _{t=1}
1239: \end{array}
1240: \right.\right\}.
1241: \end{equation}
1242: Here $t$ is the coordinate on $\left( 0,1\right] $.
1243:
1244: {For the trivial covering }$\partial M\rightarrow \partial M=X${, this algebra
1245: is just the commutative algebra of continuous functions on }$T^{*}\left(
1246: M\backslash
1247: \partial M\right)$ {\ vanishing at infinity. Let us also mention that this algebra can be also
1248: viewed as the groupoid $C^*$-algebra \cite{BrNi1} of the equivalence relation
1249: defined by $\pi$ ($x\sim y$ if either $x=y$ or $x,y\in \partial M$ and
1250: $\pi(x)=\pi(y)$). For a trivial covering, this algebra was used in
1251: \cite{Ros2}.}
1252:
1253: \textbf{2. The difference construction.} Let us define the \emph{difference
1254: construction} for non-local operators. This will be a mapping
1255: \begin{equation}
1256: \chi :\limfunc{Ell}^0\left( M,\pi \right)
1257: \longrightarrow K_0\left( %
1258: \mathcal{A}_{T^{*}M,\pi }\right) \label{stk}
1259: \end{equation}
1260: into the $K_0$ group of the $C^{*}$-algebra $\mathcal{A}_{T^{*}M,\pi }.$ To
1261: this end, we take an elliptic operator
1262: \[
1263: D:C^\infty \left( M,E\right) \longrightarrow C^\infty \left(
1264: M,{\Bbb{C}^k}\right),
1265: \]
1266: fix some embeddings of $E$ and ${\Bbb{C}^k}$ in trivial bundles of
1267: sufficiently large dimension, and denote by $P_E$ and $P_{\Bbb{C}^k}$ the
1268: projections that define the corresponding subbundles:
1269: \[
1270: E\simeq \func{Im}P_E\subset \Bbb{C}^N\oplus 0,\qquad {\Bbb{C}^k}\simeq \func{Im}%
1271: P_{\Bbb{C}^k}\subset 0\oplus \Bbb{C}^L.
1272: \]
1273: We denote the direct images of these projections near the boundary by $P_{\pi
1274: _{!}E}$ and $P_{\pi _{!}{\Bbb{C}^k}}$. The \emph{difference element of $D$}
1275: is, by definition, the difference
1276: \[
1277: \chi \left[ D\right] =\left[ P_1\oplus P_2\right] -\left[ P_{\Bbb{C}^k}\oplus P_{\pi%
1278: _{!}{\Bbb{C}^k}}\right]\in K_0\left( %
1279: \mathcal{A}_{T^{*}M,\pi }\right) ,
1280: \]
1281: where the projection $P_1$ over $M^{\prime }$ is given by
1282: \begin{equation}
1283: \left\{
1284: \begin{array}{cc}
1285: P_E\cos ^2\left| \xi \right| + P_{\Bbb{C}^k}\sin ^2\left| \xi \right| + \left(\sigma^{-1}_M
1286: \left( x,\xi \right)P_{\Bbb{C}^k} +\sigma_M \left( x,\xi \right) P_E \right) \sin \left| \xi
1287: \right| \cos \left| \xi \right|
1288: , & \left| \xi
1289: \right| \leq \pi /2,\vspace{2mm} \\
1290: P_{\Bbb{C}^k}, & \left| \xi \right| >\pi /2.
1291: \end{array}
1292: \right. \label{str}
1293: \end{equation}
1294: (We assume that the principal symbol is zero-order homogeneous in $\xi$.) The
1295: projection $P_2$ over $X\times \left[ 0,1\right] $ is defined by the formula
1296: \[
1297: P_2=
1298: \left\{
1299: \begin{array}{cc}
1300: P_{\pi _{!}E}\cos ^2\left| \xi \right| +P_{\pi _{!}{\Bbb{C}^k}}\sin ^2\left| \xi%
1301: \right| +1/2 \left(\widetilde{\sigma }^{-1}\left( x^{\prime },\xi \right) P_{\pi
1302: _{!}{\Bbb{C}^k}}+ \widetilde{\sigma }\left( x^{\prime },\xi \right) P_{\pi _{!}E} \right)\sin
1303: 2\left| \xi \right|
1304: , & \vspace{2mm} \\
1305: P_{\pi _{!}E}\cos ^2\varphi +P_{\pi _{!}{\Bbb{C}^k}}\sin ^2\varphi + 1/2\left(\widetilde{\sigma
1306: }^{-1}\left( x^{\prime },0\right) P_{\pi _{!}{\Bbb{C}^k}}+ \widetilde{\sigma }\left( x^{\prime
1307: },0\right) P_{\pi _{!}E} \right) \sin 2\varphi, &
1308: \vspace{2mm} \\
1309: P_{\pi _{!}{\Bbb{C}^k}}, &
1310: \end{array}
1311: \right.
1312: \]
1313: where the first case is used for $x^{\prime }\in X\times \left[ 1/2,1\right]
1314: ,\left| \xi \right| \leq \pi /2 $, the second for $x^{\prime }\in X\times
1315: \left[ 0,1/2\right] ,\left| \xi \right| <\pi t$, and the third otherwise. Here
1316: we write
1317: \[
1318: \varphi =\left| \xi \right| +\pi /2\left( 1-2t\right) ,\quad \widetilde{%
1319: \sigma }\left( x^{\prime },\xi \right) =\sigma_X \left( x^{\prime },\xi
1320: \right)
1321: \]
1322: for brevity. Geometrically, these projections define a subbundle that coincides
1323: with $E\subset \Bbb{C}^{N+L}$ over the zero section (for $\xi$=0); coincides
1324: with the orthogonal bundle ${\Bbb{C}^k}\subset \Bbb{C}^{N+L}$ for $\left| \xi
1325: \right| \geq \pi /2$; and is obtained by the rotation of the first bundle
1326: towards the second bundle with the use of $\sigma \left( D\right) $ at the
1327: intermediate points. (The symbol is treated as an isomorphism of the two
1328: bundles.) By construction, $P_1$ and $P_{\Bbb{C}^k}$ coincide outside a compact
1329: set in $T^{*}M^{\prime },$ and $P_2$ and $P_{\pi _{!}{\Bbb{C}^k}}$ coincide
1330: outside a compact set in $T^{*}\left( X\times \left( 0,1\right] \right).$
1331: Therefore, the difference $\left[ P_1\oplus P_2\right] -\left[
1332: P_{\Bbb{C}^k}\oplus P_{\pi _{!}{\Bbb{C}^k}}\right] $ is indeed in $K_0\left(
1333: \mathcal{A}_{T^{*}M,\pi }\right) .$
1334:
1335:
1336:
1337:
1338:
1339: \begin{remark}
1340: \emph{This element of the }$K$\emph{-group can be equivalently defined by
1341: different expressions
1342: (cf. \cite{Hig2}).}%
1343: \end{remark}
1344:
1345: \begin{theorem}
1346: \label{sem1}The difference construction $\emph{(\ref{stk})}$ is a
1347: well-defined group isomorphism.
1348: \end{theorem}
1349:
1350: \emph{Proof.} The mapping $\chi $ preserves the equivalence relations in
1351: $\limfunc{Ell}^0\left(M,\pi \right) $ and $K_0\left( \mathcal{A}_{T^{*}M,\pi
1352: }\right) .$ Indeed, under an operator homotopy the symbols vary continuously.
1353: Therefore, the corresponding projections $P_{1,2}$ are joined by a continuous
1354: homotopy. Furthermore, $\chi \left[ D\right] $ is independent of the choice of
1355: an embedding in a trivial bundle, since all such embeddings are homotopic, and
1356: for a trivial $D$ (i.e., one induced by a vector bundle isomorphism) $\chi[D]$
1357: is equal to zero. This shows that $\chi$ is well defined. The proof that this
1358: mapping is one-to-one presents no essential difficulties and is left to the
1359: reader.
1360:
1361: \dokaend\vspace{1mm}
1362:
1363:
1364:
1365:
1366: \textbf{3. Index theorem for families}. Later on in Section \ref{defe3}, we use
1367: a families index formula. Let us briefly state the corresponding results.
1368:
1369: Let $P$ be a compact space. Denote by $\limfunc{Ell}_P\left( M,\pi \right) $ the group of stable
1370: homotopy classes of elliptic families on $M$ parametrized by $P.$
1371:
1372:
1373: \begin{theorem}[\rm the index of families of non-local operators]\label{indfam} Let $(M,\pi)$ satisfy
1374: Assumption~{\rm \ref{as1}}. Then for an embedding $f:M\to EG_N$ the direct
1375: image mapping $f_!$ for families is well defined and the following diagram
1376: commutes:
1377: $$
1378: \xymatrix{ \limfunc{Ell}_P\left( M,\pi \right)\ar[d]_{\rm ind} \ar[rr]^{f_!}
1379: & & \limfunc{Ell}_P\left( EG_N,\pi_N \right) \ar[d]^{(\pi_N)_!}\\
1380: K^0(P) & \ar[l]_{{\rm ind}_t \qquad} K^0\left(P\times T^{*} BG_N\right) \ar@{=}[r]^{\;} &
1381: \limfunc{Ell}_P\left( BG_N \right) . }
1382: $$
1383: \end{theorem}
1384: The proof is similar to that of Theorem~\ref{thind1} in the previous section
1385: (cf. \cite{AtSi4}) and therefore is omitted.
1386:
1387:
1388: Let us finally note that the difference construction can also be defined in
1389: this case as a mapping
1390: \[
1391: \chi _P:\limfunc{Ell}_P\left( M,\pi \right) \longrightarrow%
1392: K_0\left( C\left( P,\mathcal{A}_{T^{*}M,\pi }\right) \right) ,
1393: \]
1394: where $C\left( P,\mathcal{A}_{T^{*}M,\pi }\right) $ is the algebra of
1395: continuous functions on $P$ ranging in the $C^{*}$-algebra
1396: $\mathcal{A}_{T^{*}M,\pi }.$
1397:
1398:
1399:
1400:
1401:
1402:
1403:
1404:
1405:
1406:
1407:
1408:
1409: \section{A homotopy invariant for manifolds with covering on the boundary
1410: \label{pr1}}
1411:
1412: \textbf{1. The class of operators}. On a manifold $M$ with covering $\pi $ on
1413: the boundary, we consider elliptic differential operators
1414: \[
1415: D:C^\infty \left( M,E\right) \longrightarrow C^\infty \left( M,F\right)
1416: \]
1417: that are lifted from the base of the covering in a neighbourhood of the
1418: boundary . Technically, we suppose that the following condition is satisfied.
1419:
1420: \begin{assumption}\label{as2}{\em
1421: The restrictions of the bundles $E$ and $F$ to the boundary are lifted from the
1422: base of the covering; moreover, we fix some isomorphisms
1423: \[
1424: \left. E\right| _{\partial M}\simeq \pi ^{*}E_0,\quad \left. F\right| _{\partial M}\simeq \pi
1425: ^{*}F_0,\quad \quad E_0,F_0\in \limfunc{Vect}\left( X\right) ,
1426: \]
1427: and for some operator $ D_0:C^\infty \left( X\times \left[ 0,1\right)
1428: ,E_0\right) \longrightarrow C^\infty \left( X\times \left[ 0,1\right)
1429: ,F_0\right)$ on the cylinder with base $X$ the direct image of $D$ in a collar
1430: neighbourhood of the boundary satisfies the commutative diagram
1431: \begin{equation}
1432: \xymatrix{ C^\infty \left( X\times \left[ 0,1\right) ,\pi _{!}E\right) \ar[d]_\simeq
1433: \ar[r]^{\left( \pi \times 1\right) _{!}D}& C^\infty \left(X\times \left[ 0,1\right) ,\pi
1434: _{!}F\right) \ar[d]^\simeq
1435: \\
1436: C^\infty \left( X\times \left[ 0,1\right) ,E_0\otimes \pi _{!}1\right) \ar[r]^{D_0\otimes 1} &
1437: C^\infty \left( X\times \left[0,1\right) ,F_0\otimes \pi _{!}1\right) . } \label{pdown}
1438: \end{equation}
1439: Here $D_0\otimes 1$ stands for the operator $D_0$ with coefficients in the flat
1440: bundle $\pi _{!}1$ (e.g., see \cite{APS3}).}
1441: \end{assumption}
1442:
1443: We also suppose that $D$ is first-order operator and the following assumption
1444: is satisfied.
1445:
1446: \begin{assumption}\label{as3}{\em
1447: In the neighbourhood $X\times\left[ 0,\varepsilon \right) $ of the boundary,
1448: the operator has the form
1449: \[
1450: \left. D_0\right| _{X\times \left[ 0,\varepsilon \right) }=\Gamma \left( \frac \partial {\partial
1451: t}+A_0\right)
1452: \]
1453: for a bundle isomorphism $\Gamma $, where $A_0$ is an elliptic self-adjoint first-order operator
1454: on $X.$ This operator is called the \emph{tangential operator} of $D_0.$}
1455: \end{assumption}
1456:
1457: If $D$ satisfies Assumptions \ref{as2} and \ref{as3}, then near the boundary it has the form
1458: \[
1459: D=\frac \partial {\partial t}+\pi ^{!}\left( A_0\otimes 1\right)
1460: \]
1461: up to a vector bundle isomorphism. For brevity, the self-adjoint operator $\pi
1462: ^{!}\left( A_0\otimes 1\right) $ will be denoted by $A.$\vspace{2mm}
1463:
1464: \textbf{2. The homotopy invariant}. For an operator $D$ satisfying
1465: Assumptions~\ref{as2} and \ref{as3}, consider the \emph{spectral
1466: Atiyah--Patodi--Singer boundary value problem} \cite{APS1}
1467: \[
1468: \left\{
1469: \begin{array}{llll}
1470: Du & = & f, & \\
1471: \Pi _{+}\left. u\right| _{\partial M} & = & g, & \qquad g\in \func{Im}\Pi
1472: _{+}\subset H^{s-1/2}\left( \partial M,E\right) ,
1473: \end{array}
1474: \right.
1475: \]
1476: where $ \Pi _{+}=(A+\left| A\right| )/{2\left| A\right| }$ is the non-negative
1477: spectral projection of the self-adjoint operator $A$. (If $A$ is not
1478: invertible, then in this formula one should replace $A$ by $A+\varepsilon$ for
1479: some $\varepsilon$ less then the absolute value of the greatest negative
1480: eigenvalue of $A$.) The spectral problem is always Fredholm. However, its index
1481: $\limfunc{ind}\left( D,\Pi _{+}\right) $ is not invariant under homotopies of
1482: $D$ and is not determined by its principal symbol. Here by definition a
1483: continuous homotopy of $D$ is a continuous homotopy in the interior of $M$ that
1484: can be covered in a neighbourhood of the boundary by a homotopy of the diagram
1485: \eqref{pdown} and a continuous homotopy of tangential operators.
1486:
1487:
1488:
1489:
1490:
1491: \begin{proposition}
1492: The sum
1493: \begin{equation}
1494: \widetilde{\limfunc{ind}}D\stackrel{def}{=}\func{mod}n\text{-}\left(
1495: \limfunc{ind}\left( D,\Pi _{+}\right) +\eta \left( A\right) -n\eta
1496: \left( A_0\right) \right) \in \Bbb{R}/n\Bbb{Z}, \label{moda1}
1497: \end{equation}
1498: is a homotopy invariant of $D.$ Here $n$ is the number of sheets of the
1499: covering and $\eta \left( A\right)$ and $\eta \left( A_0\right) $ are the
1500: spectral Atiyah--Patodi--Singer $\eta $-invariants of the tangential operators
1501: $A$ and $A_0$.
1502: \end{proposition}
1503:
1504: \emph{Proof}. Consider the non-reduced invariant
1505: \begin{equation}
1506: \limfunc{ind}\left( D,\Pi _{+}\right) +\eta \left( A\right) -n\eta \left(
1507: A_0\right) . \label{bet3}
1508: \end{equation}
1509: The results of \cite{APS3} imply that for a smooth operator family $D_t$ this
1510: expression is a piecewise smooth function of the parameter $t$ (the
1511: corresponding families of tangential operators are denoted by $A_t$ and
1512: $A_{0,t}$).
1513:
1514: 1) We claim that (\ref{bet3}) is a piecewise constant function. Indeed, the derivative of the
1515: $\eta $-invariant
1516: \[
1517: \frac d{dt}\eta \left( A_t\right)
1518: \]
1519: with respect to $t$ is local, i.e., is equal to an integral over the manifold
1520: of an expression determined by the complete symbol of the tangential family
1521: $A_t.$ However, the complete symbols of $A_t$ and $A_{0,t}$ coincide locally by
1522: Assumption \ref{as2}. Thus we have
1523: \[
1524: \frac d{dt}\eta \left( A_t\right) =n\frac d{dt}\eta \left( A_{0,t}\right) .
1525: \]
1526: Therefore, (\ref{bet3}) is a piecewise constant function.
1527:
1528: 2) Let us show that the jumps of this function are multiples of the number of
1529: sheets of the covering. Indeed, for a homotopy $D_t$ the index, as well as the
1530: $\eta $-invariants, changes by the spectral flow of the corresponding families
1531: of tangential operators. Hence
1532: \[
1533: \left. \left[ \limfunc{ind}\left( D_t,\Pi _{+,t}\right) +\eta \left(
1534: A_t\right) -n\eta \left( A_{0.t}\right) \right] \right| _{t=0,1}=\left(
1535: -1+1\right) \limfunc{sf}\left( A_t\right) _{t\in \left[ 0,1\right] }-n%
1536: \limfunc{sf}\left( A_{0,t}\right) _{t\in \left[ 0,1\right] }\in n\Bbb{Z},
1537: \]
1538: as desired. \dokaend
1539:
1540: \begin{remark}
1541: \emph{For a trivial covering, our invariant is none other than the
1542: }$\func{mod}n$-index
1543: \[
1544: \func{mod}n\text{-}\limfunc{ind}(D,\Pi_+)\in \Bbb{Z}_n\subset \Bbb{R}/n\Bbb{Z}
1545: \]\emph{\
1546: of Freed-Melrose} \emph{%
1547: \cite{FrMe1}}. \emph{On the other hand, the fractional part of the invariant}
1548: \emph{\ (\ref{moda1}) is the so-called} relative Atiyah--Patodi--Singer $\eta
1549: $-invariant
1550: \emph{\cite{APS2,APS3}} %
1551: \[
1552: \left\{ \eta \left( A_0\otimes 1_{\pi _{!}1}\right) -n\eta \left( A_0\right)
1553: \right\} \in \Bbb{R}/\Bbb{Z}
1554: \]
1555: \emph{of }$A_0$\emph{\ with coefficients in the flat bundle }$\pi _{!}1\in \limfunc{%
1556: Vect}\left( X\right) .$
1557: \end{remark}
1558:
1559:
1560: The invariant $\widetilde{\limfunc{ind}}$ has an interesting interpretation as an obstruction.
1561: Namely, suppose that $M$ is the total space of a covering $\widetilde{\pi }$ with base $Y$ that
1562: induces the covering $\pi $ over the boundary
1563: \[
1564: \begin{array}{ccc}
1565: \partial M & \subset & M \\
1566: \pi \downarrow \;\; & & \;\;\downarrow \widetilde{\pi } \\
1567: X & \subset & Y.
1568: \end{array}
1569: \]
1570:
1571: \begin{proposition}
1572: If a differential operator $D:C^\infty \left( M,E\right) \rightarrow C^\infty
1573: \left( M,F\right) $ is the pullback of an elliptic operator $D_0$ on $Y,$ then
1574: \[
1575: \widetilde{\limfunc{ind}}D=0.
1576: \]
1577: \end{proposition}
1578:
1579: \emph{Proof}. According to the Atiyah--Patodi--Singer formula (see \cite{APS1}), the sum
1580: \[
1581: \limfunc{ind}\left( D,\Pi _{+}\right) +\eta \left( A\right)
1582: \]
1583: is equal to the integral over the manifold of a local expression defined by the
1584: complete symbol of $D$. Since $D$ and $D_0$ coincide locally, one has
1585: \[
1586: \limfunc{ind}\left( D,\Pi _{+}\right) +\eta \left( A\right) =n\left(
1587: \limfunc{ind}\left( D_0,\Pi _{+,0}\right) +\eta \left( A_0\right) \right) .
1588: \]
1589: We obtain the desired formula by transposing the term $n\eta \left( A_0\right)
1590: $ to the left-hand side. \dokaend
1591:
1592:
1593: \section{The index defect formula \label{defe3}}
1594:
1595: The aim of this section is to find a topological formula for the invariant
1596: $\widetilde{\rm ind}$.\vspace{1mm}
1597:
1598:
1599: \textbf{1. The difference construction}. The pair $\left( M,\pi \right) $
1600: defines the singular space \fgr{311_2.eps}{fii2}{Singular space}
1601: \[
1602: \overline{M}^\pi =M\left/ \left\{ x\sim x^{\prime },\text{
1603: if }x,x^{\prime }\in \partial M\text{ and }\pi \left(
1604: x\right) =\pi \left( x^{\prime }\right) \right\} \right. ,
1605: \]
1606: obtained by identification of points in the fibers of $\pi$ (see Figure
1607: \ref{fii2} in the case of a trivial covering). Likewise, the boundary of the
1608: non-compact manifold $T^{*}M$ is a covering over the product $T^{*}X\times
1609: \Bbb{R}$, and the corresponding singular space will be denoted by
1610: $\overline{T^{*}M}^\pi$.
1611:
1612: Consider an elliptic operator $D$ satisfying Assumption \ref{as2}. The diagram
1613: \eqref{pdown} implies that the principal symbol defines a $K$-theory element
1614: \[
1615: \left[ \sigma \left( D\right) \right] \in K\left( \overline{T^{*}M}^\pi
1616: \right) .
1617: \]
1618: Thus we have a homomorphism
1619: \[
1620: \chi : \limfunc{Ell}\left( \overline{M}^\pi \right)\longrightarrow K\left( \overline{T^{*}M}^\pi
1621: \right) ,\quad
1622: \chi \left[ D\right] = \left[ \sigma \left( D\right) \right] .
1623: \]
1624: Here $\limfunc{Ell}\left( \overline{M}^\pi \right) $ is the Grothendieck group
1625: of homotopy classes of elliptic operators $D$ on $M$ that satisfy Assumptions
1626: \ref{as2} and \ref{as3}.
1627:
1628: The topological formula for the invariant $\widetilde{\limfunc{ind%
1629: }}$ uses the Poincar\'{e} pairing on the manifold
1630: $\overline{T^{*}M}^\pi $ with singularities. Let us define this
1631: pairing. \vspace{2mm}
1632:
1633: \textbf{2. A pairing in }$K$\textbf{-theory of a singular manifold}. By analogy
1634: with the algebra $\mathcal{A}_{T^{*}M,\pi }$ of the cotangent bundle, one can
1635: define an algebra for $M$ itself:
1636: \[
1637: \mathcal{A}_{M,\pi }=\left\{ (u,v)\; \left|\;
1638: \begin{array}{c}
1639: u\in C_0\left( M^{\prime }\right), v\in C_0\left(
1640: X\times \left( 0,1\right] ,\limfunc{End}\pi _{!}1\right)\vspace{1mm}\\
1641: \beta \left( \left. u\right| _{\partial M^{\prime }}\right)\beta ^{-1}=\left. v\right| _{t=1}
1642: \end{array}
1643: \right. \right\} .
1644: \]
1645:
1646:
1647:
1648: \begin{lemma}
1649: \label{lemc2}The group $K_0\left( \mathcal{A}_{M,\pi}\right) $ is isomorphic
1650: to the group of stable homotopy classes of triples
1651: \[
1652: \left( E,F,\sigma \right) ,\qquad E,F\in \limfunc{Vect}\left( M\right) ,\;\sigma :\pi _{!}\left.
1653: E\right| _{\partial M}\longrightarrow \pi _{!}\left. F\right| _{\partial M}.
1654: \]
1655: Here $\sigma $ is a bundle isomorphism, and trivial triples are those with
1656: $\sigma $ induced by an isomorphism over $M$.
1657: \end{lemma}
1658:
1659: \emph{Proof.} Note that this lemma is similar to Theorem~\ref{sem1}, which can
1660: be also considered as giving a realization of the group $K_0\left(
1661: \mathcal{A}_{T^{*}M,\pi }\right)$ in topological terms. Along the same lines, a
1662: triple $\left( E,\mathbb{C}^k,\sigma \right) $ defines the element
1663: \[
1664: \left[ P_E\oplus P_2\right] -\left[ P_{\mathbb{C}^k}\oplus P_{\pi _{!}\mathbb{C}^k}\right] \in
1665: K_0\left( \mathcal{A}_{M,\pi }\right) ,
1666: \]
1667: where the projection $P_2$ over $X\times \left[ 0,1\right] $ is
1668: \[
1669: P_2=P_{\pi _{!}E}\cos ^2\varphi +P_{\pi _{!}{\mathbb{C}^k}}\sin ^2\varphi +P_{\pi%
1670: _{!}{\mathbb{C}^k}}\sigma \left( x\right) P_{\pi _{!}E}\sin 2\varphi ,\quad \varphi =\frac \pi
1671: 2\left( 1-t\right) .
1672: \]
1673: Here $P_E,P_{\mathbb{C}^k}$ are projections on subbundles isomorphic to $E$ and
1674: $F$. We also suppose that the subbundles are orthogonal to each other.
1675:
1676: The proof of the fact that this mapping induces an isomorphism with the group
1677: $K_0\left( \mathcal{A}_{M,\pi }\right) $ is similar to the previous proof and
1678: is omitted. \dokaend
1679:
1680: This realization permits one to define a product
1681: \[
1682: K^0\left( \overline{T^{*}M}^\pi \right) \times K_0\left( \mathcal{A}_{M,\pi
1683: }\right) \longrightarrow K_0\left( \mathcal{A}_{T^{*}M,\pi }\right)
1684: \]
1685: using the construction of a symbol with coefficients in a vector bundle. More precisely, for
1686: elements
1687: \[
1688: \left[ \sigma \right] \in K\left( \overline{T^{*}M}^\pi \right) ,\qquad \left[
1689: E,F,\sigma ^{\prime }\right] \in K_0\left( \mathcal{A}_{M,\pi }\right),
1690: \]
1691: consider the symbol
1692: \begin{equation}
1693: \sigma \otimes 1_E\oplus \sigma ^{-1}\otimes 1_F \label{st2}
1694: \end{equation}
1695: on $M.$ The direct image of its restriction to the boundary can be written as
1696: \[
1697: \begin{array}{ccl}
1698: \pi _{!}\left( \left. \sigma \otimes 1_E\oplus \sigma ^{-1}\otimes
1699: 1_F\right| _{\partial M}\right) & = & \pi _{!}\sigma \otimes 1_{\pi
1700: _{!}\left. E\right| _{\partial M}}\oplus \pi _{!}\sigma ^{-1}\otimes 1_{\pi
1701: _{!}\left. F\right| _{\partial M}} \vspace{3mm} \\
1702: & & \simeq \left( \pi _{!}\sigma \oplus \pi _{!}\sigma ^{-1}\right) \otimes
1703: 1_{\pi _{!}\left. E\right| _{\partial M}}.
1704: \end{array}
1705: \]
1706: The latter isomorphism is induced by a vector bundle isomorphism
1707: \[
1708: \pi _{!}\left. E\right| _{\partial M}\stackrel{\sigma ^{\prime }}{\simeq }%
1709: \pi _{!}\left. F\right| _{\partial M}.
1710: \]
1711: Now $\left( \pi _{!}\sigma \oplus \pi _{!}\sigma ^{-1}\right) \otimes 1_{\pi
1712: _{!}\left. E\right| _{\partial M}}$ is trivially homotopic to the identity. The
1713: homotopy is
1714: \[
1715: \left(
1716: \begin{array}{cc}
1717: \pi _{!}\sigma & 0 \\
1718: 0 & 1
1719: \end{array}
1720: \right) \left(
1721: \begin{array}{cc}
1722: \cos \tau & \sin \tau \\ %
1723: -\sin \tau & \cos \tau%
1724: \end{array}
1725: \right) \left(
1726: \begin{array}{cc}
1727: 1 & 0 \\
1728: 0 & \pi _{!}\sigma ^{-1}
1729: \end{array}
1730: \right) \left(
1731: \begin{array}{cc}
1732: \cos \tau & -\sin \tau \\ %
1733: \sin \tau & \cos \tau%
1734: \end{array}
1735: \right) ,\quad \tau \in \left[ 0,\pi /2\right] .
1736: \]
1737: Thus we have extended the symbol (\ref{st2}) to a non-local elliptic symbol on
1738: $M.$ Now the desired product is defined as the difference construction of the
1739: latter symbol, which we denote by
1740: \[
1741: \left[ \sigma \right] \times \left[ E,F,\sigma ^{\prime }\right] \in
1742: K_0\left( \mathcal{A}_{T^{*}M,\pi }\right) .
1743: \]
1744: Using the homotopy classification $K_0\left( \mathcal{A}_{T^{*}M,\pi
1745: }\right)\simeq {\rm Ell}(M,\pi)$, we can apply the index mapping to this
1746: product and define the pairing of groups as the composition
1747: \[
1748: \left\langle ,\right\rangle :K^0\left( \overline{T^{*}M}^\pi \right) \times
1749: K_0\left( \mathcal{A}_{M,\pi }\right) \longrightarrow K_0\left( \mathcal{A}%
1750: _{T^{*}M,\pi }\right) \stackrel{\limfunc{ind}}{\longrightarrow }\Bbb{Z}.
1751: \]
1752: This pairing is an analogue of Poincar\'e duality on the singular manifold
1753: $\overline{T^{*}M}^\pi $
1754: (see Section~\ref{dvoj6}). Let us note that for a regular covering the index
1755: can be computed topologically (by the index theorem) and hence the pairing is
1756: also topologically computable. \vspace{1mm}
1757:
1758:
1759: \textbf{3}. \textbf{The element of }$K$\textbf{-theory with coefficients
1760: defined by a manifold with a covering on the boundary.} We start from a
1761: universal example. Denote the half-infinite cylinder $EG_N\times [0,+\infty)$
1762: by ${\cal M}_N$, and the projective limit of the groups
1763: $K_0({\cal{A}}_{{\cal{M}}_N,\pi_N})$ as $N\to \infty$ by
1764: $K_0({\cal{A}}_{{\cal{M}}_\infty})$. Let us show that the universal bundle
1765: $\gamma=\pi_!1 \in \limfunc{Vect}\left( BG\right), $ where $1\in{\rm
1766: Vect}(EG)$, defines an element
1767: \begin{equation}\label{volna}
1768: [\widetilde{\gamma}]\in K_0(\mathcal{A}_{\mathcal{M}_\infty},\mathbb{Q}/n\mathbb{Z}),\quad n=|G|.
1769: \end{equation}
1770: To this end, note that the universal bundle gives the element\footnote{ Here
1771: and below, the $K$-groups of classifying spaces are defined as the projective
1772: limits over their finite-dimensional approximations.}
1773: \[
1774: \left[ \gamma \right] -n\in K^0(BG).
1775: \]
1776: We use the following lemma to show that this difference defines the desired element.
1777: \begin{lemma}
1778: \label{lu2} There is an isomorphism $\widetilde{K}^0\left( BG\right) \simeq K^1\left( BG,\Bbb{Q}/n\Bbb{Z}%
1779: \right),$ which is defined as the coboundary mapping $\partial$ in the exact
1780: sequence
1781: \[
1782: \rightarrow K^1\left( BG _N\right) \otimes \Bbb{Q}%
1783: \longrightarrow K^1\left( BG _N,\Bbb{Q}/n\Bbb{Z}\right) \stackrel{\partial}\longrightarrow
1784: K^0\left( BG _N\right)
1785: \stackrel{\times n}{%
1786: \longrightarrow }K^0\left( BG _N\right) \otimes \Bbb{Q}
1787: \]
1788: induced by the inclusion of the coefficient groups $n\Bbb{Z\subset Q}$.
1789: \end{lemma}
1790:
1791:
1792: \noindent\emph{Proof}. Let us rewrite the sequence
1793: \[
1794: \rightarrow K^1\left( BG _N\right) \otimes \Bbb{Q}%
1795: \longrightarrow K^1\left( BG_N,\Bbb{Q}/n\Bbb{Z}\right)
1796: \longrightarrow K^0\left( BG_N\right) \stackrel{\times n}{%
1797: \longrightarrow }K^0\left( BG _N\right) \otimes \Bbb{Q}
1798: \]
1799: as the short exact sequence
1800: \[
1801: 0\rightarrow K^1\left( BG _N\right) \otimes \Bbb{Q}/n\Bbb{Z}%
1802: \longrightarrow K^1\left( BG _N,\Bbb{Q}/n\Bbb{Z}\right) \longrightarrow \limfunc{Tor}K^0\left(
1803: BG _N\right) \rightarrow 0.
1804: \]
1805: Then we obtain the following sequence of projective limits as $N\rightarrow
1806: \infty $ (this sequence may not be exact):
1807: \begin{equation}
1808: 0\rightarrow \stackunder{\longleftarrow }{\lim }K^1\left( BG
1809: _N\right) \otimes \Bbb{Q}/n\Bbb{Z}\longrightarrow K^1\left( BG,\Bbb{Q}/n%
1810: \Bbb{Z}\right) \longrightarrow \,\stackunder{\longleftarrow }{\lim }\limfunc{%
1811: Tor}K^0\left( BG _N\right) \rightarrow 0. \label{injlim1}
1812: \end{equation}
1813: As $N$ increases, the sequence $\widetilde{K}^*\left( BG _N\right) $ has the
1814: following property (e.g., see \cite{APS2,Ati6}). For an arbitrary $N$, there
1815: exists an $L>0$ such that the range of the mapping
1816: \begin{equation}
1817: \widetilde{K}^*\left( BG _{N+L}\right) \longrightarrow \widetilde{K}^*\left( BG _{N}\right)
1818: \label{svo1}
1819: \end{equation}
1820: is in the torsion subgroup. Using this, we obtain the following expressions
1821: for the limits:
1822: \[
1823: \stackunder{\longleftarrow }{\lim }K^1\left( BG _N\right)
1824: \otimes \Bbb{Q}/n\Bbb{Z}=0,\quad\stackunder{\longleftarrow }{\lim }%
1825: \limfunc{Tor}K^0\left( BG _N\right) =\,\stackunder{%
1826: \longleftarrow }{\lim }\widetilde{K}^0\left( BG _N\right) =%
1827: \widetilde{K}^0\left( BG\right) .
1828: \]
1829: One can also use (\ref{svo1}) to prove the exactness of (\ref{injlim1}). The
1830: proof is based on commutative diagrams of the form
1831: \begin{equation}
1832: \begin{array}{ccccc}
1833: 0\rightarrow K^1\left( BG _N\right) \otimes \Bbb{Q}/n\Bbb{Z}
1834: & \!\!\longrightarrow \!\! & K^1\left( BG _N,\Bbb{Q}/n\Bbb{Z}%
1835: \right) & \!\!\longrightarrow \!\! & \limfunc{Tor}K^0\left(
1836: BG _N\right) \rightarrow 0 \\
1837: 0\uparrow & & \uparrow & & \uparrow \\
1838: 0\rightarrow \!K^1\!\left( BG _{N+L}\right) \otimes \Bbb{Q}/n%
1839: \Bbb{Z} & \!\!\longrightarrow \!\! & \!K^1\!\left( BG
1840: _{N+L}\!,\Bbb{Q}/n\Bbb{Z}\right) \! & \!\!\longrightarrow \!\! & \!\limfunc{%
1841: Tor}\!K^0\!\left( BG _{N+L}\right) \!\rightarrow 0,
1842: \end{array}
1843: \label{coma1}
1844: \end{equation}
1845: where $L$ is chosen as in (\ref{svo1}).
1846:
1847: Thus we obtain the desired isomorphism
1848: \[
1849: \stackunder{\longleftarrow }{\lim }K^1\left( BG _N,\Bbb{Q}/n%
1850: \Bbb{Z}\right) \simeq \stackunder{\longleftarrow }{\lim }\widetilde{K}%
1851: ^0\left( BG _N\right) .
1852: \]
1853: \dokaend
1854:
1855:
1856:
1857:
1858: Finally, the desired element
1859: \eqref{volna} is obtained from the isomorphism
1860: $$
1861: K^{*+1}(BG_N,\Bbb{Q}/n\Bbb{Z})\simeq K_*({\cal{A}}_{{\cal{M}}_N,\pi_N}, \Bbb{Q}/n\Bbb{Z})
1862: $$
1863: induced by the inclusion of the ideal
1864: $C_0(BG_N\times (0,1),{\rm End}(\pi_N)_!1)$. Thus
1865: $$ K^1\left( BG,\Bbb{Q}/n\Bbb{Z}\right) \simeq
1866: K_0({\cal{A}}_{{\cal{M}}_\infty},\Bbb{Q}/n\Bbb{Z}),
1867: $$
1868: and $\left[ \widetilde{\gamma }\right]$ can be viewed as an element of both
1869: groups.
1870:
1871:
1872:
1873: Suppose that we are now given a pair $\left( M,\pi \right).$ In the remaining
1874: part of the section, we assume that $\pi$ is a principal $G$-covering for a
1875: finite group $G$. There exists a mapping $f:M\to{\cal M}_N$ that takes the
1876: boundary to the base $EG_N\times\{0\}$ and is equivariant on the boundary (see
1877: Proposition~\ref{pre1}). The inverse image of $[\widetilde{\gamma}]$ is denoted
1878: by
1879: \begin{equation}
1880: [\widetilde{\pi_!1}]\stackrel{\rm def}={f}^{*}[ \widetilde{\gamma }] \in K_0\left( \mathcal{A}%
1881: _{M,\pi },\Bbb{Q}/n\Bbb{Z}\right). \label{navor1}
1882: \end{equation}
1883: This element does not depend on the choice of $f$, since a map into the
1884: universal space is unique up to homotopy. Let us obtain a geometric realization
1885: of this element. We do this in two steps.
1886:
1887:
1888:
1889: \textbf{4. A geometric realization of $K_0\left( \mathcal{A}%
1890: _{M,\pi },\Bbb{Q}/n\Bbb{Z}\right)$.} The group $\Bbb{Q}/n\Bbb{Z}$ is the direct
1891: limit of the finite groups
1892: \[
1893: \Bbb{Z}_{nN} \subset \Bbb{Q}/n\Bbb{Z}, \quad x \mapsto x/N.
1894: \]
1895: Therefore, the $K$-group with coefficients in $\Bbb{Q}/n\Bbb{Z}$ is defined as
1896: the direct limit
1897: \begin{equation}\label{alp1}
1898: K_0\left( \mathcal{A}_{M,\pi },\Bbb{Q}/n\Bbb{Z}\right) =\stackunder{%
1899: \longrightarrow }{\lim }K_0\left( \mathcal{A}_{M,\pi },\Bbb{Z}_{nN}\right).
1900: \end{equation}
1901:
1902: Further, the elements of the groups with finite coefficients can be constructed
1903: by using the following proposition (cf. \cite{SaScS3}).
1904: \begin{proposition}\label{pre5}
1905: A triple $\left( E,F,\sigma \right) ,$ where
1906: \begin{equation}
1907: E\in \limfunc{Vect}\left( M\right) ,\;F\in \limfunc{Vect}\left( X\right) ,\qquad \pi _{!}\left(
1908: \left. E\right| _{\partial M}\right) \stackrel{\sigma }{\simeq }kF, \label{modnn1}
1909: \end{equation}
1910: and $\sigma $ is an isomorphism on $X$, defines an element in $K_0\left(
1911: \mathcal{A}_{M,\pi },\Bbb{Z}_k\right) .$
1912: \end{proposition}
1913:
1914: \noindent\emph{Proof.} By analogy with the topological case (e.g., see
1915: \cite{APS2}), the theory with coefficients in $\Bbb{Z}_k$ is defined in terms
1916: of the \emph{Moore space} $\mathbb{M}_k$ of this group by the formula
1917: \begin{equation}
1918: K_0\left( \mathcal{A}_{M,\pi },\Bbb{Z}_k\right) =K_0\left( \widetilde{C}%
1919: _0\left( \mathbb{M}_k,\mathcal{A}_{M,\pi }\right) \right) , \label{nesht1}
1920: \end{equation}
1921: where $\widetilde{C}_0\left( \mathbb{M}_k,\mathcal{A}_{M,\pi }\right) $ is the algebra of
1922: $\mathcal{A}_{M,\pi }$-valued functions on the Moore space vanishing at the fixed point.
1923:
1924:
1925: One can readily generalize Lemma~\ref{lemc2} to the case of families. More
1926: precisely, the same method shows that the group $K_0\left(
1927: \widetilde{C}_0\left( \mathbb{M}_k,\mathcal{A}_{M,\pi }\right) \right) $ is
1928: isomorphic to the group of stable homotopy classes of triples $\left( E^{\prime
1929: },F^{\prime },\sigma ^{\prime }\right) ,$ where $E^{\prime },F^{\prime }\in
1930: \limfunc{Vect}\left( M\times \mathbb{M}_k\right) $ and the isomorphism
1931: \[
1932: \sigma ^{\prime }:\pi _{!}\left(\left. E^{\prime }\right| _{\partial M}\right)\longrightarrow \pi
1933: _{!}\left(\left. F^{\prime }\right| _{\partial M}\right)
1934: \]
1935: is defined over $X\times \mathbb{M}_k.$
1936:
1937:
1938: Let $\varepsilon$ be the line bundle over the Moore space representing the
1939: generator
1940: $\left[\varepsilon\right]-1\in\widetilde{K}\left(\mathbb{M}_k\right)\simeq\Bbb{Z}_k$.
1941: (Further information about the Moore spaces can be found, e.g., in
1942: \cite{APS2,SaScS3}.) Let us also fix a trivialization $\rho :k\varepsilon
1943: \rightarrow \Bbb{C}^k.$
1944:
1945: To the triple $\left( E,F,\sigma \right)$ in \eqref{modnn1}, we assign the
1946: element
1947: \[
1948: \left[ E\otimes \varepsilon ,E,\sigma ^{\prime }\right] \in K_0\left( \widetilde{C}_0\left(
1949: \mathbb{M}_k,\mathcal{A}_{M,\pi }\right) \right) ,
1950: \]
1951: where the isomorphism $\sigma ^{\prime }$ is defined as the composition (see \cite{SaScS3})
1952: \begin{equation}
1953: \pi _{!}\left(\left. E\right| _{\partial M}\right)\otimes \varepsilon \stackrel{\sigma \otimes
1954: 1}{\rightarrow }kF\otimes \varepsilon \simeq F\otimes k\varepsilon
1955: \stackrel{1\otimes \rho }{\rightarrow }F\otimes \Bbb{C}^k\simeq kF\stackrel{%
1956: \sigma ^{-1}\otimes 1}{\rightarrow }\pi _{!}\left(\left. E\right| _{\partial M}\right).
1957: \label{longo1}
1958: \end{equation}
1959: \dokaend
1960:
1961: \textbf{5. A geometric realization of $[\widetilde{\pi_! 1}]$
1962: (see~\eqref{navor1}).} For sufficiently large $N$, consider a triple $\left(
1963: N,1,\alpha \right) ,$ where
1964: \[
1965: N\in \limfunc{Vect}\left( M\right) ,\qquad 1\in \limfunc{Vect}\left( X\right) ,
1966: \]
1967: are trivial vector bundles of the corresponding dimensions and $ \pi
1968: _{!}N\stackrel{\alpha }{\simeq }\Bbb{C}^{nN}$ is some trivialization. By
1969: Proposition~\ref{pre5}, this triple defines an element
1970: \[
1971: \left[ N,1,\alpha \right] \in K_0\left( \mathcal{A}_{M,\pi },\Bbb{Q}/n\Bbb{Z}%
1972: \right).
1973: \]
1974: Now, using the diagram (\ref{coma1}), the reader can verify that this element
1975: coincides with $[ \widetilde{\pi_{!}1}]$ if the number $N$ and the
1976: trivialization are chosen as follows.
1977:
1978: Suppose that the range of the classifying mapping
1979: \[
1980: f:M\longrightarrow \mathcal{M}_\infty
1981: \]
1982: is contained in the skeleton $\mathcal{M}_{N^{\prime }},$ then for $N^{\prime
1983: }$ there exists an $L^{\prime }$ such that property~(\ref{svo1}) is valid. Now
1984: we can choose an $N$ such that the restriction of the direct sum $N\gamma $ of
1985: the universal bundle to $BG _{N^{\prime }+L^{\prime }}$ is trivial with some
1986: trivialization
1987: \[
1988: N\gamma \stackrel{\alpha ^{\prime }}{\simeq }\Bbb{C}^{nN}.
1989: \]
1990: Finally, over $M$ we choose the induced trivialization
1991: $$
1992: \alpha=f^*\alpha'.
1993: $$
1994:
1995: \textbf{6. The index defect theorem}.
1996:
1997: \begin{theorem}
1998: \label{defect2}Let $(M,\pi)$ be a manifold with a covering on the boundary
1999: corresponding to a free action of a finite group $G$. Then the diagram
2000: $$
2001: \xymatrix{\limfunc{Ell}\left( \overline{M}^\pi \right) \ar[d]_{\widetilde{\limfunc{ind}}}
2002: \ar[r]^\chi & K\left( \overline{T^{*}M}^\pi \right) \ar[ld]^{\left\langle
2003: \cdot , [\widetilde{\pi _{!}1}] \right\rangle} \\
2004: \Bbb{R}/n\Bbb{Z}\text{,} }
2005: $$
2006: commutes. Here $\left\langle ,\right\rangle $ is the Poincar\'e pairing with
2007: coefficients,
2008: \begin{equation}
2009: \left\langle ,\right\rangle :K\left( \overline{T^{*}M}^\pi \right) \times K_0\left(
2010: \mathcal{A}_{M,\pi },\Bbb{Q}/n\Bbb{Z}\right) \longrightarrow
2011: K_0\left( \mathcal{A}_{T^{*}M,\pi },\Bbb{Q}/n\Bbb{Z}\right) \stackrel{%
2012: \limfunc{ind}}{\rightarrow }\Bbb{Q}/n\Bbb{Z}. \label{proda2}
2013: \end{equation}
2014: \end{theorem}
2015:
2016: \begin{remark}
2017: \emph{Theorem \ref{defect2} expresses $\widetilde{\rm ind}D$ in topological terms via the
2018: principal symbol. Indeed, by \eqref{alp1} and \eqref{nesht1}, the index mapping in
2019: \eqref{proda2} can be expressed topologically using the index theorem for families (Theorem
2020: \ref{indfam}).}
2021: \end{remark}
2022:
2023:
2024: \noindent\emph{Proof}. The proof of the theorem is essentially analytic in nature. The main idea
2025: is to reduce the analytic invariant $\widetilde{\rm ind}$ to the index with values in
2026: $\mathbb{Q}/n\mathbb{Z}$. Thus, we start by defining the corresponding operators.
2027:
2028:
2029: 1. First, we define elliptic theory $\limfunc{Ell}\left( M,\pi ,\Bbb{Q}/n\Bbb{Z%
2030: }\right) $ with coefficients $\Bbb{Q}/n\Bbb{Z}$. The definition can be given using the direct
2031: limit
2032: \[
2033: \limfunc{Ell}\left( M,\pi ,\Bbb{Q}/n\Bbb{Z}\right) =\stackunder{%
2034: \longrightarrow }{\lim }\limfunc{Ell}\left( M,\pi ,\Bbb{Z}_{nN}\right) ,\quad \Bbb{Z}_{nN}\subset
2035: \Bbb{Z}_{nNM}\subset \Bbb{Q}/n\Bbb{Z},
2036: \]
2037: of theories with finite coefficients. More precisely, elliptic theory with
2038: coefficients in $\Bbb{Z}_k$ is defined by families of non-local elliptic
2039: operators of order one parametrized by the Moore space $\mathbb{M}_k$ of
2040: $\Bbb{Z}_k$:
2041: \[
2042: \limfunc{Ell}\left( M,\pi ,\Bbb{Z}_k\right) =\limfunc{Ell}%
2043: _{\mathbb{M}_k}\left( M,\pi \right).
2044: \]
2045: For elliptic theory with coefficients, we refer the reader to \cite{SaScS3}.
2046:
2047:
2048: 2. Consider the mapping
2049: \begin{equation}\label{phibig}
2050: \begin{array}{ccc}
2051: \limfunc{Ell}\left( \overline{M}^\pi \right) & \stackrel{\Phi }{%
2052: \longrightarrow } & \limfunc{Ell}\left( M,\pi ,\Bbb{Q}/n\Bbb{Z}\right)
2053: \end{array}
2054: \end{equation}
2055: that takes an operator $D$ to the family
2056: \[
2057: D^{*}\oplus \left( D\otimes 1_\varepsilon \right) :C^\infty \left( M,F\oplus E\otimes \varepsilon
2058: \right) \longrightarrow C^\infty \left( M,E\oplus F\otimes \varepsilon \right)
2059: \]
2060: of first-order elliptic operators on $M$ parametrized by $\mathbb{M}_{nN}$ (the
2061: number $N$ will be chosen below). Here $D^{*}$ is the adjoint operator, and the
2062: operator family obtained by twisting $D$ with the bundle $\varepsilon $ is
2063: denoted by $D\otimes 1_\varepsilon $. Consider the direct sum of $N$ copies of
2064: this family. It turns out that if $N$ is sufficiently large, then this family
2065: admits an elliptic boundary condition. Indeed, for $N$ sufficiently large there
2066: exists a trivialization
2067: \begin{equation}
2068: N\pi _{!}1\stackrel{\alpha }{\simeq }\Bbb{C}^{nN}, \label{triva1}
2069: \end{equation}
2070: since $\pi_!1$ is flat. Hence on the base of the covering we have the vector
2071: bundle isomorphism
2072: \[
2073: \pi _{!}\left( N\left. E\right| _{\partial M}\right) \simeq \pi _{!}N\otimes E_0\stackrel{\alpha
2074: \otimes 1}{\longrightarrow }\Bbb{C}^{nN}\otimes E_0
2075: \]
2076: and the similar isomorphism
2077: \[
2078: \pi _{!}\left( N\left. E\right| _{\partial M}\right) \otimes \varepsilon
2079: \simeq \pi _{!}N\otimes E_0\otimes \varepsilon \stackrel{\alpha \otimes 1}{%
2080: \longrightarrow }\Bbb{C}^{nN}\otimes E_0\otimes \varepsilon \simeq
2081: nN\varepsilon \otimes E_0\stackrel{\rho \otimes 1}{\longrightarrow }\Bbb{C}%
2082: ^{nN}\otimes E_0.
2083: \]
2084: We denote the induced isomorphisms on sections by
2085: \[
2086: B_1=\alpha \otimes 1:C^\infty \left( X,\pi _{!} \left(N\left. E\right| _{\partial
2087: M}\right)\right) \stackrel{}{\longrightarrow }C^\infty \left( X,\Bbb{C}^{nN}\otimes E_0\right) ,
2088: \]
2089: \[
2090: B_2=\rho \otimes 1\left( \alpha \otimes 1\right) :C^\infty \left( X,\pi _{!}
2091: \left(N\left. E\right| _{\partial M}\right)\otimes \varepsilon\right) \stackrel{}{%
2092: \longrightarrow }C^\infty \left( X,\Bbb{C}^{nN}\otimes E_0\right) .
2093: \]
2094: Now we can define the family of non-local boundary value problems (in the sense
2095: of Section~\ref{para1})
2096: \begin{equation}
2097: \left\{
2098: \begin{array}{l}
2099: \begin{array}{cc}
2100: ND^{*}u=f_1,\qquad & N\left( D\otimes 1_\varepsilon \right) v=f_2,
2101: \end{array}
2102: \vspace{2mm} \\
2103: B_1\beta _E\left. u\right| _{\partial M}+B_2\beta _E\left. v\right| _{\partial M}=g,\quad g\in
2104: C^\infty \left( X,\Bbb{C}^{nN}\otimes E_0\right) ,
2105: \end{array}
2106: \right. \label{nelin1}
2107: \end{equation}
2108: which consists of elliptic elements. We define the mapping \eqref{phibig} as
2109: follows: it takes $D$ to the family of non-local problems \eqref{nelin1}. Note
2110: that $\Phi$ depends on the choice of the trivialization \eqref{triva1}.
2111:
2112:
2113: 3. There is a natural index mapping
2114: \[
2115: \limfunc{ind}:\limfunc{Ell}\left( M,\pi ,\Bbb{Q}/n\Bbb{Z}\right) \longrightarrow \Bbb{Q}/n\Bbb{Z}
2116: \]
2117: that takes an element $\left[ \mathcal{D}\right] $ represented by a family $\mathcal{D}$ of
2118: elliptic operators parametrized by $\mathbb{M}_{nN}$ to the (reduced) index of the family
2119: \[
2120: \limfunc{ind}\left[ \mathcal{D}\right] \stackrel{\rm def}=\limfunc{ind}\mathcal{D}\in
2121: \widetilde{K}\left( \mathbb{M}_{nN}\right) \simeq \Bbb{Z}_{nN}\subset \Bbb{Q}/n\Bbb{Z}.
2122: \]
2123:
2124: \begin{lemma}
2125: \label{lem3} The diagram
2126: $$
2127: \xymatrix{ \qquad\limfunc{Ell}\left( \overline{M}^\pi \right) \ar[rr]^{\Phi}
2128: \ar[rd]_{\widetilde{\limfunc{ind}}} & & \limfunc{Ell}\left( M,\pi ,\Bbb{Q}/n\Bbb{Z}\right)
2129: \ar[ld]^{\rm ind} \\
2130: & \quad \Bbb{R}/n\Bbb{Z}, }
2131: $$
2132: where $N$ and the trivialization $\alpha$ in \eqref{triva1} are chosen as in
2133: Subsec.~\emph{5}, commutes.
2134: \end{lemma}
2135:
2136: \noindent\emph{Proof of the lemma}. The boundary value problem (\ref{nelin1})
2137: is linearly homotopic to the problem
2138: \[
2139: \left\{
2140: \begin{array}{l}
2141: ND^{*}u=f_1,\qquad N\left( D\otimes 1_\varepsilon \right) v=f_2,\vspace{2mm}
2142: \\
2143: B_1\beta _E\Pi _{-}\left. u\right| _{\partial M}+B_2\beta _E\Pi _{+}\left.
2144: v\right| _{\partial M}=g,\quad g\in C^\infty \left( X,\Bbb{C}^{nN}\otimes
2145: E_0\right),
2146: \end{array}
2147: \right.
2148: \]
2149: within the class of elliptic problems. The last formula shows that the index of $\Phi \left[
2150: D\right] $ is equal to the sum of the index of the family of spectral problems for $ND^{*}$ and
2151: $N\left( D\otimes 1_\varepsilon \right) $ and the index of the operator family
2152: \begin{equation}
2153: N\func{Im}\Pi _{-}\left( A\right) \oplus N\func{Im}\Pi _{+}\left( A\right)
2154: \otimes \varepsilon \stackrel{B_1+B_2}{\longrightarrow }C^\infty \left( X,%
2155: \Bbb{C}^{nN}\otimes E_0\right) \label{opcl1}
2156: \end{equation}
2157: on the boundary. Note that we specify the self-adjoint operators in the notation of spectral
2158: projections. Let us compute the index of the former family on $X$.
2159:
2160:
2161: 1) There is a decomposition
2162: \[
2163: C^\infty \left( X,\Bbb{C}^{nN}\otimes E_0\right) \simeq nN\func{Im}\Pi _{-}\left( A_0\right)
2164: \oplus nN\varepsilon \otimes \func{Im}\Pi _{+}\left( A_0\right)
2165: \]
2166: of the target space for the family (\ref{opcl1}). This decomposition is defined as
2167: \[
2168: nN\func{Im}\Pi _{-}\left( A_0\right) \oplus nN\varepsilon \otimes \func{Im}%
2169: \Pi _{+}\left( A_0\right) \stackrel{1+\left( \rho \otimes 1\right) }{%
2170: \longrightarrow }C^\infty \left( X,\Bbb{C}^{nN}\otimes E_0\right) .
2171: \]
2172: Using this isomorphism, we represent the index of (\ref{opcl1}) in the form
2173: \[
2174: =\limfunc{ind}\left( N\func{Im}\Pi _{+}\left( A\right) \stackrel{\Pi _{+}\left( A_0\right) \beta
2175: _E}{\longrightarrow }nN\func{Im}\Pi _{+}\left( A_0\right) \right) \left( \left[ \varepsilon
2176: \right] -1\right) \in \widetilde{K}\left( \mathbb{M}_{nN}\right) .
2177: \]
2178: Finally, we rewrite the index by pushing forward the space $\func{Im} \Pi
2179: _{+}\left( A\right) $ to the base of the covering:
2180: \[
2181: =\limfunc{ind}\left( N\func{Im}\Pi _{+}\left( \pi _{!}A\right) \stackrel{\Pi _{+}\left(
2182: A_0\right) }{\longrightarrow }nN\func{Im}\Pi _{+}\left( A_0\right) \right) \left( \left[
2183: \varepsilon \right] -1\right) .
2184: \]
2185: The index of the elliptic operator (not a family!) in the last formula can be expressed by the
2186: Atiyah--Patodi--Singer formula \cite{APS3}
2187: \begin{equation}
2188: \limfunc{ind}\left( N\func{Im}\Pi _{+}\left( \pi _{!}A\right) \stackrel{\Pi _{+}\left( A_0\right)
2189: }{\longrightarrow }nN\func{Im}\Pi _{+}\left( A_0\right) \right) =N\eta \left( A\right) -nN\eta
2190: \left( A_0\right) +\left\langle \left[ \sigma \left( A_0\right) \right] ,\left[ \pi _{!}1\right]
2191: \right\rangle , \label{apps}
2192: \end{equation}
2193: where the brackets $\left\langle,\right\rangle$ denote the pairing
2194: \begin{equation}
2195: \left\langle ,\right\rangle :K^1\left( T^{*}X\right) \times K^1\left( X,\Bbb{%
2196: Q}\right) \longrightarrow \Bbb{Q} \label{pai2}
2197: \end{equation}
2198: of the difference element $ \left[ \sigma \left( A_0\right) \right] \in
2199: K^1\left( T^{*}X\right) $ of an elliptic self-adjoint operator $A_0$ with the
2200: element $\left[ \pi _{!}1\right] \in K^1\left( X,\Bbb{Q}\right) $ defined by
2201: the trivialized flat bundle $N\pi _{!}1$
2202: (more about this formula can be found in the book \cite{Gil1}).%
2203:
2204: 2) It turns out that for our choice of the trivialization (\ref{triva1}) the
2205: last term in (\ref{apps}) is equal to zero. Indeed, consider the classifying
2206: mapping $f:X\rightarrow BG _{N^{\prime }}.$ We can evaluate
2207: (\ref{pai2}) on the classifying space:
2208: \begin{equation}
2209: \left\langle \left[ \sigma \left( A_0\right) \right] ,\left[ \pi _{!}1\right] \right\rangle
2210: =\left\langle f_{!}\left[ \sigma \left( A_0\right) \right] ,\left[ \gamma \right] \right\rangle
2211: ,\quad \left[ \pi _{!}1\right] =f^{*}\left[ \gamma \right] \in K^1\left( X,\Bbb{Q}\right),
2212: \label{al1}
2213: \end{equation}
2214: where $[\gamma]\in K^1(BG_{N'})\otimes\Bbb{Q}$ is the element defined by the trivialized flat
2215: bundle $N\gamma$. The inclusion $ BG _{N^{\prime }}\subset BG _{N^{\prime }+L^{\prime }}$
2216: induces the commutative diagram
2217: \[
2218: \begin{array}{ccccc}
2219: K^1\left( T^{*} BG _{N^{\prime }}\right) & \times & K^1\left( BG _{N^{\prime }}\right) \otimes
2220: \Bbb{Q} & \longrightarrow &
2221: \Bbb{Q} \\
2222: \downarrow & & \uparrow & & \parallel \\
2223: K^1\left( T^{*} BG _{N^{\prime }+L^{\prime }}\right) & \times & K^1\left( BG _{N^{\prime
2224: }+L^{\prime }}\right) \otimes \Bbb{Q} & \longrightarrow & \Bbb{Q}
2225: \end{array}
2226: \]
2227: Using this diagram and (\ref{svo1}), one can prove the triviality of the pairing (\ref{al1}) by a
2228: diagram chase argument.
2229:
2230: Thus we have reduced $\limfunc{ind}\Phi \left[ D\right] $ to the desired form
2231: \[
2232: \limfunc{ind}\Phi \left[ D\right] =\widetilde{\limfunc{ind}}\left[ D\right] .
2233: \]
2234: \dokaend
2235:
2236: 4. To complete the proof of the theorem, it suffices to show that the value of the Poincar\'e
2237: pairing $\langle [\sigma(D)],[\widetilde{\pi_!1}]\rangle$ coincides with the index of $\Phi(D)$.
2238:
2239: We denote the product by $[ \widetilde{\pi _{!}1}] $ by $\varphi$:
2240: \[
2241: \varphi :K\left( \overline{T^{*}M}^\pi \right) \longrightarrow K_0\left( \mathcal{A}_{T^{*}M,\pi
2242: },\Bbb{Q}/n\Bbb{Z}\right).
2243: \]
2244: \begin{lemma}
2245: \label{lem3aa}Under the assumptions of Lemma \emph{\ref{lem3}}, the diagram
2246: \begin{equation}
2247: \begin{array}{ccc}
2248: \limfunc{Ell}\left( \overline{M}^\pi \right) & \stackrel{\Phi }{%
2249: \longrightarrow } & \limfunc{Ell}\left( M,\pi ,\Bbb{Q}/n\Bbb{Z}\right) \\
2250: {\chi \downarrow } & & \downarrow {\chi ^{\prime }} \\ %
2251: K\left( \overline{T^{*}M}^\pi \right) & \stackrel{\varphi }{\longrightarrow } & K\left(
2252: \mathcal{A}_{T^{*}M,\pi },\Bbb{Q}/n\Bbb{Z}\right),
2253: \end{array}
2254: \label{inty3}
2255: \end{equation}
2256: where $\chi'$ is induced by the difference constructions for families
2257: \emph{(}see Subsec.~{\rm 3.3}{\em )}.
2258: \end{lemma}
2259:
2260: \noindent\emph{Proof}. Substituting the definitions of $\left[ \sigma
2261: \left(D\right) \right] $ and $[\widetilde{\pi _{!}1}] $ (according to
2262: Subsecs.~4.4 and 4.5) into Eq.~(\ref{st2}), defining the product, one can show
2263: that the desired product $\varphi[\sigma(D)]$ is determined by the family of
2264: elliptic symbols that are equal to
2265: \begin{equation}
2266: N\sigma \left( D\right) \otimes 1_\varepsilon \oplus N\sigma \left( D\right)
2267: ^{-1}\otimes 1 \label{aa1}
2268: \end{equation}
2269: far from the boundary. The direct image of the restriction of this symbol to
2270: the boundary is equal to
2271: \begin{eqnarray}
2272: N\pi _{!}\left( \sigma \left( D\right) \otimes 1_\varepsilon \oplus \sigma \left( D\right)
2273: ^{-1}\otimes 1\right) &=&\left( \sigma \left( D_0\right) \otimes 1_{N\varepsilon \otimes \pi
2274: _{!}1}\oplus \sigma \left( D_0\right)
2275: ^{-1}\otimes 1_{N\pi _{!}1}\right) \simeq \label{aa2} \\
2276: &\simeq &\left( \sigma \left( D_0\right) \oplus \sigma \left( D_0\right)
2277: ^{-1}\right) \otimes 1_{\Bbb{C}^{nN}}. \nonumber
2278: \end{eqnarray}
2279: In the last equality, we use the isomorphisms
2280: $N\pi _{!}1\stackrel{\alpha }{\simeq }\Bbb{C}^{nN},\quad \Bbb{C}%
2281: ^{nN}\varepsilon \stackrel{\rho }{\simeq }\Bbb{C}^{nN}.$ The symbol is extended
2282: to a neighbourhood of the boundary using the homotopy of the direct sum
2283: $\sigma \left( D_0\right)\oplus \sigma \left( D_0\right) ^{-1}$ to the
2284: identity.
2285:
2286:
2287: It remains to prove that the difference elements for the principal symbol of
2288: the family of boundary value problems (\ref {nelin1}) and the symbol defined by
2289: (\ref{aa1}), (\ref{aa2}) coincide. Indeed, this equality is obvious far from
2290: the boundary, since the only difference here is in the components $\sigma
2291: \left( D^{*}\right) $ and $\sigma \left( D\right) ^{-1}.$ These components are
2292: joined by the standard homotopy
2293: \[
2294: \sigma \left( D^{*}\right) \left[ \sigma \left( D\right) \sigma \left( D^{*}\right) \right]
2295: ^{-s},\qquad s\in \left[ 0,1\right] .
2296: \]
2297: The reader can also prove the equality near the boundary using the formulae for
2298: order reduction given in Remark~\ref{remk}. \dokaend
2299:
2300:
2301: By combining Lemmata~\ref{lem3aa} and \ref{lem3}, we complete the proof of the
2302: theorem. \dokaend
2303:
2304:
2305:
2306:
2307:
2308: \section{Applications\label{exa4}}
2309:
2310: \textbf{1}. Theorem \ref{defect2} enables one to express the fractional part of the $\eta
2311: $-invariant in the following situation.
2312:
2313: Let $M$ be an even-dimensional spin manifold with boundary represented as the
2314: total space of a covering such that the spin structure on the boundary is the
2315: pullback of a spin structure on the base. Let us also fix an $E\in
2316: \limfunc{Vect}\left( M\right) $ that is also pulled back from the base near
2317: the boundary: $\left. E\right| _{\partial M}\simeq\pi^* E_0$. We choose a
2318: metric on $M$ that is a product metric induced by a metric on the base near the
2319: boundary. Finally, we choose a similar connection in $E$.
2320:
2321: \begin{proposition}
2322: The Dirac operator $D_M$ on $M$ with coefficients in $E$ satisfies the
2323: assumptions of Theorem \emph{\ref{defect2}}, and the fractional part of the
2324: $\eta $-invariant is equal to
2325: \[
2326: \left\{ \eta \left( D_X\right) \right\} =\frac 1n\left( \int\limits_M%
2327: \widehat{A}\left( M\right) \limfunc{ch}E-\left\langle \left[ \sigma \left( D_M\right) \right] ,[
2328: \widetilde{\pi _{!}1}] \right\rangle \right) \in \Bbb{R}/\Bbb{Z},
2329: \]
2330: where $D_X$ is the self-adjoint Dirac operator on $X$ with coefficients in $E_0$.
2331: \end{proposition}
2332:
2333: \emph{Proof}. The formula follows from Theorem \ref{defect2} if we decompose
2334: the index of the spectral problem using the Atiyah--Patodi--Singer formula
2335: \[
2336: \limfunc{ind}\left( D_M,\Pi _{+}\right) =\int\limits_M\widehat{A}\left(
2337: M\right) \limfunc{ch}E-\eta \left( D_{\partial M}\right) .
2338: \]
2339: \dokaend
2340:
2341: \vspace{2mm}
2342:
2343: \textbf{2}. The invariant $\widetilde{\limfunc{ind}}$ can be effectively computed via Lefschetz
2344: theory. Suppose that $\pi $ is regular, i.e., the boundary is a principal $G$-bundle for a finite
2345: group $G$. Let $D$ be a $G$-invariant elliptic differential operator of order one on $M.$ For
2346: $g\in G$, let $L\left( D,g\right) \in \Bbb{C}$ be the usual contribution to the Lefschetz formula
2347: (see \cite{Don3}) of the fixed point set of the diffeomorphism $g:M\rightarrow M.$
2348:
2349:
2350: \begin{proposition}
2351: One has
2352: \begin{equation}
2353: \label{lefa1}\widetilde{\limfunc{ind}}D\equiv -\sum_{g\neq e}L\left( D,g\right) \text{ }%
2354: \left( \func{mod}n\right) .
2355: \end{equation}
2356: \end{proposition}
2357:
2358:
2359: \emph{Proof}. Consider the equivariant index $\limfunc{ind}_g\left( D,\Pi _{+}\right) $ of the
2360: Atiyah--Patodi--Singer problem and the equivariant $\eta $-function (see \cite{Don3}) of the
2361: tangential operator $A$ on the boundary.
2362:
2363: Denote by $\left( D,\Pi _{+}\right) ^G$ and $A^G$ the restrictions of the
2364: corresponding operators to the subspaces of $G$-invariant sections. Clearly,
2365: $A^G$ is isomorphic to $A_0$ on $X$. On the other hand, one can express the
2366: usual invariants in terms of their equivariant counterparts:
2367: \[
2368: \limfunc{ind}\left( D,\Pi _{+}\right) ^G=\frac 1{\left| G\right| }\sum_{g\in
2369: G}\limfunc{ind}_g\left( D,\Pi _{+}\right) \qquad \eta \left( A^G\right) =\frac 1{\left| G\right|
2370: }\sum_{g\in G}\eta \left( A,g\right).
2371: \]
2372: These expression follow from elementary character theory. Using them, we write
2373: \[
2374: \widetilde{\limfunc{ind}}D=\limfunc{ind}_e\left( D,\Pi _{+}\right)
2375: -\sum_{g\neq e}\eta \left( A,g\right) .
2376: \]
2377: Let us substitute the expression for the $\eta $-invariant given by the
2378: equivariant Atiyah--Patodi--Singer formula
2379: (see \cite{Don3}) %
2380: \[
2381: -\eta \left( A,g\right) =\limfunc{ind}_g\left( D,\Pi _{+}\right) -L\left(
2382: D,g\right)
2383: \]
2384: into this formula. This gives the desired congruence \eqref{lefa1}:
2385: \[
2386: \widetilde{\limfunc{ind}}D=\left| G\right| \limfunc{ind}\left( D,\Pi
2387: _{+}\right) ^G-\sum_{g\neq e}L\left( D,g\right) .
2388: \]
2389: \dokaend
2390:
2391:
2392:
2393: \section{Poincar\'{e} isomorphisms}
2394:
2395:
2396:
2397: {\bf 1. A closed smooth manifold.} It is well known (see \cite{Ati4,Kas3} or
2398: the monograph \cite{HiRo1}) that elliptic operators of order zero on a compact
2399: closed manifold define elements in $K$-theory:
2400: \[
2401: \left[ \sigma \left( D\right) \right] \in K^{*}\left( T^{*}M \right) ,\quad \left[ D\right] \in
2402: K^{*}\left( C\left( M\right) \right) \equiv K_{*}\left( M\right).
2403: \]
2404: The latter group is the analytic $K$-homology group, and the grading is odd for
2405: self-adjoint operators and even otherwise. The first element is the difference
2406: element of the operator. To define the second element, we recall that an
2407: elliptic operator $D$ of order zero is a Fredholm operator
2408: \[
2409: D:L^2\left( M,E\right) \longrightarrow L^2\left( M,F\right) ,
2410: \]
2411: where both $L^2$-spaces are modules over $C\left( M\right) $ (the module structure is given by
2412: the pointwise product of functions). In addition, $D$ commutes with the module structure up to
2413: compact operators. Thus, for a self-adjoint $D$ (of course in this case the bundles coincide)
2414: the pair $\left( L^2\left( M,E\right) ,D\right) $ is an element
2415: \[
2416: \left[ D\right] \in K^1\left( C\left( M\right) \right) .
2417: \]
2418: For a nonself-adjoint $D$, we consider a self-adjoint matrix
2419: operator
2420: \begin{equation}\label{toper}
2421: T=\left(
2422: \begin{array}{cc}
2423: 0 & D^{*} \\
2424: D & 0
2425: \end{array}
2426: \right)
2427: \end{equation}
2428: in the naturally $\Bbb{Z}_2$-graded $C\left( M\right) $-module $L^2\left(
2429: M,E\right) \oplus L^2\left( M,F\right) $. The operator $T$ is odd with respect
2430: to the grading. Hence it defines a $K$-theory element, denoted by
2431: \[
2432: \left[ D\right] \in K^0\left( C\left( M\right) \right) .
2433: \]
2434:
2435: {\bf 2. Manifold with boundary.} On the other hand, elliptic operators of order
2436: one on a manifold with non-empty boundary define similar elements
2437: \[
2438: \left[ \sigma \left( D\right) \right] \in K^i\left( T^{*}M\right) ,\quad
2439: \left[ D\right] \in K_i\left( M\backslash \partial M\right) .
2440: \]
2441: The former is the Atiyah--Singer difference element, and the latter is defined
2442: as follows. Consider an embedding
2443: \[
2444: M\subset \widetilde{M}
2445: \]
2446: of $M$ in some closed manifold $\widetilde{M}$ of the same dimension (e.g., the
2447: double $2M$). Let $\widetilde{D}$ be an arbitrary extension of $D$ to
2448: $\widetilde{M}$. On $\widetilde{M}$, we consider the zero-order operator
2449: \[
2450: \widetilde{F}=\left( 1+\widetilde{D}^{*}\widetilde{D}\right) ^{-1/2}%
2451: \widetilde{D}.
2452: \]
2453: We define the restriction of this operator to $M$ as the bounded operator
2454: \begin{equation}
2455: F=i^{*}\widetilde{F}i_{*}:L^2\left( M,E\right) \longrightarrow L^2\left(
2456: M,F\right) , \label{relo}
2457: \end{equation}
2458: where $i_{*}:L^2\left( M\right) \rightarrow L^2( \widetilde{M}%
2459: ) $ is the extension by zero and $i^{*}:L^2( \widetilde{M}) \rightarrow
2460: L^2\left( M\right) $ is the restriction operator.
2461:
2462: For a symmetric $D$, we find that $F$ satisfies
2463: \[
2464: \begin{array}{c}
2465: F-F^{*}\in \mathcal{K},\quad f\left( F^2-1\right) \in \mathcal{K}, \quad \left[
2466: F,f\right] \in \mathcal{K}
2467: \end{array}
2468: \]
2469: for functions $f\in C_0\left(
2470: M\backslash \partial M\right)$ vanishing on the boundary. Here
2471: $\mathcal{K}$ is the ideal of compact operators. These relations show that
2472: $F$ defines an element of $K^1\left( C_0\left( M\backslash
2473: \partial M\right) \right) $ (see
2474: \cite{SoTr1}).%
2475:
2476: If $D$ is nonself-adjoint, then one considers the matrix as in
2477: Eq.~\eqref{toper}. This defines an element in $K^0\left( C_0\left( M\backslash
2478: \partial M\right) \right) .$
2479:
2480: The mappings
2481: \[
2482: \begin{array}{ccc}
2483: K^{*}\left( T^{*}\left( M\backslash \partial M\right) \right) &
2484: \longrightarrow & K_{*}\left( M\right) ,\vspace{1mm} \\
2485: K^{*}\left( T^{*}M\right) & \longrightarrow & K_{*}\left( M\backslash
2486: \partial M\right) ,\vspace{3mm} \\
2487: \left[ \sigma \left( D\right) \right] & \mapsto & \left[ D\right] ,
2488: \end{array}
2489: \]
2490: which take symbols to operators, define \emph{Poincar\`e isomorphisms} on a smooth manifold $M$
2491: with boundary (e.g., see \cite{Kas3}). The top mapping is defined in terms of elliptic operators
2492: of order zero on $M$ that are induced by vector bundle isomorphisms near the
2493: boundary.\vspace{1mm}
2494:
2495:
2496: {\bf 3. Manifolds with singularities.} An elliptic non-local zero-order
2497: operator $D$ defines elements
2498: \[
2499: \left[ \sigma \left( D\right) \right] \in K_{*}\left( \mathcal{A}%
2500: _{T^{*}M,\pi }\right) ,\left[ D\right] \in K^{*}\left( C\left( \overline{M}%
2501: ^\pi \right) \right) \simeq K_{*}\left( \overline{M}^\pi \right).
2502: \]
2503: The first is the difference element defined in Section \ref{hom2}. To define
2504: the second element, we note that a non-local elliptic operator $D$ of order
2505: zero does not almost commute with the entire algebra $C\left( M\right)$ but
2506: only with the functions pulled back from the quotient space $\overline{M}^\pi
2507: .$ This leads to a smaller algebra.
2508:
2509: On the other hand, the operators of Sections \ref{pr1} and \ref{defe3} define
2510: similar elements
2511: \[
2512: \left[ \sigma \left( D\right) \right] \in K^{*}\left(
2513: \overline{T^{*}M}^\pi \right) ,\left[ D\right] \in K_{*}\left(
2514: \mathcal{A}_{M,\pi }\right) .
2515: \]
2516: In this case, the corresponding operators (\ref{relo}), on the contrary, almost
2517: commute with functions $C_0\left( M\backslash
2518: \partial M\right)$ as well as with the elements of the algebra
2519: $\mathcal{A}_{M,\pi }.$
2520:
2521: \begin{theorem}
2522: For an arbitrary manifold with a covering on the boundary $(M,\pi)$, the
2523: following Poincar\'{e} isomorphisms are valid:
2524: \[
2525: \begin{array}{ccc}
2526: K_{*}\left( \mathcal{A}_{T^{*}M,\pi }\right) & \longrightarrow &
2527: K_{*}\left( \overline{M}^\pi \right) ,\vspace{1mm} \\
2528: K^{*}\left( \overline{T^{*}M}^\pi \right) & \longrightarrow & K^{*}\left(
2529: \mathcal{A}_{M,\pi }\right) ,\vspace{3mm} \\
2530: \left[ \sigma \left( D\right) \right] & \mapsto & \left[ D\right] .
2531: \end{array}
2532: \]
2533: \end{theorem}
2534:
2535: \emph{Proof}. 1) Let us prove the latter isomorphism. Consider the ideal
2536:
2537: \[
2538: I=C_0\left( T^{*}\left( X\times \left( 0,1\right) \right) ,\limfunc{End}%
2539: p^{*}\pi _{!}1\right)\subset \mathcal{A}_{T^{*}M,\pi }
2540: \]
2541: with the quotient $\mathcal{A}_{T^{*}M,\pi }/I\simeq C_0\left( T^{*}M\right).$ The long exact
2542: sequence of the pair can be written as
2543: \begin{equation}
2544: \rightarrow K\left( T^{*}X\right) \stackrel{\alpha }{\rightarrow }K_0\left(
2545: \mathcal{A}_{T^{*}M,\pi }\right) \rightarrow K\left( T^{*}M\right) \rightarrow K^1\left(
2546: T^{*}X\right) \rightarrow \ldots \label{exa}
2547: \end{equation}
2548: Here we have taken into account the isomorphism $K_{*}\left( C_0\left(
2549: Y,\limfunc{End}G\right) \right) \simeq K_{*}\left( C_0\left( Y\right) \right)
2550: \simeq K^{*}\left( Y\right)$ for a vector bundle
2551: $G\in\limfunc{Vect}\left(Y\right)$.
2552:
2553:
2554:
2555: Consider the commutative diagram
2556: \[
2557: \begin{array}{ccccccccc}
2558: \rightarrow & K^0\left( T^{*}X\right) & \rightarrow & K_0\left( \mathcal{A}%
2559: _{T^{*}M,\pi }\right) & \rightarrow & K^0\left( T^{*}M\right) & \rightarrow
2560: & K^1\left( T^{*}X\right) & \ldots \\
2561: & \downarrow & & \downarrow & & \downarrow & & \downarrow & \\
2562:
2563: \rightarrow & K_0\left( X\right) & \rightarrow & K_0\left( \overline{M}^\pi
2564: \right) & \rightarrow & K_0\left( M,\partial M\right) & \rightarrow &
2565: K_1\left( X\right) & \ldots
2566: \end{array}
2567: \]
2568: Here the lower sequence is the exact sequence of the pair $X\subset
2569: \overline{M}^\pi $ in $K$-homology. The vertical mappings of the diagram
2570: (except for the second one) are isomorphisms (see \cite{BaDo1,Kas3}). Thus,
2571: using the 5-lemma, we find that the middle mapping
2572: \[
2573: K_{*}\left( \mathcal{A}_{T^{*}M,\pi }\right) \longrightarrow K_{*}\left(
2574: \overline{M}^\pi \right)
2575: \]
2576: is also an isomorphism.
2577:
2578: 2) In the second case, the proof follows the same scheme, but one uses the diagram
2579: \[
2580: \begin{array}{ccccccccc}
2581: \!\!\leftarrow\!\! & K^1\left( T^{*}X\right) & \!\!\leftarrow\!\! & K^0\left( \overline{%
2582: T^{*}M}^\pi \right) & \!\!\leftarrow\!\! & K^0\left( T^{*}\left( M\backslash
2583: \partial M\right) \right) & \!\!\leftarrow\!\! & K^0\left( T^{*}X\right) & \ldots \\
2584: & \downarrow & & \downarrow & & \downarrow & & \downarrow & \\
2585: \!\!\leftarrow\!\! & K_1\left( X\right) & \!\!\leftarrow\!\! & K^0\left( \mathcal{A}_{M,\pi
2586: }\right) & \!\!\leftarrow\!\! & K_0\left( M\right) & \!\!\leftarrow\!\! & K_0\left( X\right)
2587: & \ldots
2588: \end{array}
2589: \]
2590: The upper row corresponds to the pair $\Bbb{R}\times T^{*}X\subset \overline{T^{*}M}%
2591: ^\pi $.
2592:
2593: The proof of the theorem is complete. \dokaend
2594:
2595: \section{Poincar\'{e} duality\label{dvoj6}}
2596:
2597: An analogue of the pairing for the groups $K^0\left( \overline{T^{*}M}^\pi
2598: \right) $ and $K_0\left( \mathcal{A}_{M,\pi }\right) $ in Section \ref{defe3}
2599: is also valid for the odd groups. The definition is left to the reader.
2600:
2601:
2602: \begin{theorem}
2603: On a manifold $M$ with covering $\pi$ on the boundary, the pairings
2604: \begin{equation}
2605: K^i\left( \overline{T^{*}M}^\pi \right) \times K_i\left( \mathcal{A}_{M,\pi }\right)
2606: \longrightarrow \Bbb{Z},\qquad i=1,2, \label{aa}
2607: \end{equation}
2608: are non-degenerate on the free parts of the groups.
2609: \end{theorem}
2610:
2611: \emph{Proof}. Fixing the first argument of the pairing, we obtain
2612: a mapping
2613: \[
2614: K^i\left( \overline{T^{*}M}^\pi \right) \otimes \Bbb{Q}\longrightarrow
2615: K_i^{\prime }\left( \mathcal{A}_{M,\pi }\right) ,
2616: \]
2617: where for brevity we write $G^{\prime }=\limfunc{Hom}\left( G,\Bbb{Q}\right) .$
2618: This mapping is part of the commutative diagram
2619: \[
2620: \begin{array}{ccccccccc}
2621: & K^1\left( T^{*}X\right) \otimes \Bbb{Q} & \!\!\leftarrow\!\! & K^0\left(
2622: \overline{T^{*}M}^\pi \right) \otimes \Bbb{Q} & \!\!\leftarrow\!\! & K^0\left(
2623: T^{*}\left( M\backslash \partial M\right) \right) \otimes \Bbb{Q} &
2624: \!\!\leftarrow\!\! & K^0\left( T^{*}X\right) \otimes \Bbb{Q} & \\
2625: & \downarrow & & \downarrow & & \downarrow & & \downarrow & \\
2626: & K^{1\prime }\left( X\right) & \!\!\leftarrow\!\! & K_0^{\prime }\left( \mathcal{A%
2627: }_{M,\pi }\right) & \!\!\leftarrow\!\! & K^{0\prime }\left(
2628: M\right) & \!\!\leftarrow\!\! & K^{0\prime }\left( X\right). &
2629: \end{array}
2630: \]
2631: Here the vertical mappings, except for the second one, are isomorphisms (by
2632: virtue of Poincar\'{e} duality on a closed manifold and on a manifold with
2633: boundary). Thus, by the 5-lemma, the second mapping is an isomorphism. Hence
2634: the pairing (\ref{aa}) is non-degenerate in the second variable.
2635:
2636: The non-degeneracy with respect to the first argument can be proved in a
2637: similar way. \dokaend
2638:
2639: By way of example, consider $M$ with a $spin^c$-structure that on the boundary
2640: is induced by a $spin^c$-structure on the base $X$ of the covering $\pi .$ Then
2641: the group $K^{*}\left( \overline{T^{*}M}^\pi \right) $ is a free $K^{*+n}\left(
2642: \overline{M}^\pi \right) $-module with one generator (where $n=\dim M$); as a
2643: generator one can take the difference construction
2644: \[
2645: \left[ \sigma \left( D\right) \right] \in K^n\left( \overline{T^{*}M}^\pi
2646: \right)
2647: \]
2648: of the principal symbol of the Dirac operator on $M$ (this can be proved by
2649: analogy with the usual case of closed manifolds; e.g., see \cite{LaMi1}).
2650: Consequently, one can define the Poincar\'{e} duality pairing
2651: \[
2652: K^{*+n}\left( \overline{M}^\pi \right) \times K_{*}\left( \mathcal{A}_{M,\pi
2653: }\right) \longrightarrow \Bbb{Z}
2654: \]
2655: as the composition with $K^{*+n}\left( \overline{M}^\pi \right) \rightarrow
2656: K\left( \overline{T^{*}M}^\pi \right) .$ The above theorem shows that this
2657: pairing is non-degenerate on the free parts of the groups.
2658:
2659:
2660:
2661: \begin{thebibliography}{10}
2662:
2663: \bibitem{APS1}
2664: M.~Atiyah, V.~Patodi, and I.~Singer.
2665: \newblock Spectral asymmetry and {Riemannian} geometry {I}.
2666: \newblock {\em Math. Proc. Cambridge Philos. Soc.}, {\bf 77}, 1975, 43--69.
2667:
2668: \bibitem{FrMe1}
2669: D.~Freed and R.~Melrose.
2670: \newblock A mod $k$ index theorem.
2671: \newblock {\em Invent. Math.}, {\bf 107}, No. 2, 1992, 283--299.
2672:
2673: \bibitem{Hig1}
2674: N.~Higson.
2675: \newblock An approach to $\mathbb{Z}/k$-index theory.
2676: \newblock {\em Int. J. Math.}, {\bf 1}, No. 2, 1990, 189--210.
2677:
2678: \bibitem{Zha1}
2679: W.~Zhang.
2680: \newblock {On the mod\,$k$ index theorem of {F}reed and {M}elrose}.
2681: \newblock {\em J. Differ. Geom.}, {\bf 43}, No. 1, 1996, 198--206.
2682:
2683: \bibitem{Bot1}
2684: B.~Botvinnik.
2685: \newblock Manifolds with singularities accepting a metric of positive scalar
2686: curvature.
2687: \newblock {\em Geom. Topol.}, {\bf 5}, 2001, 683--718.
2688:
2689: \bibitem{Ros2}
2690: J.~Rosenberg.
2691: \newblock Groupoid {$C^ *$}-algebras and index theory on manifolds with
2692: singularities.
2693: \newblock {\em Geom. Dedicata}, {\bf 100}, 2003, 65--84.
2694:
2695: \bibitem{Con1}
2696: A.~Connes.
2697: \newblock {\em Noncommutative geometry}.
2698: \newblock Academic Press Inc., San Diego, CA, 1994.
2699:
2700: \bibitem{Kas3}
2701: G.~Kasparov.
2702: \newblock Equivariant ${KK}$-theory and the {N}ovikov conjecture.
2703: \newblock {\em Inv. Math.}, {\bf 91}, No. 1, 1988, 147--201.
2704:
2705: \bibitem{MePi2}
2706: R.~Melrose and P.~Piazza.
2707: \newblock Analytic {$K$}-theory on manifolds with corners.
2708: \newblock {\em Adv. in Math.}, {\bf 92}, No. 1, 1992, 1--26.
2709:
2710: \bibitem{Kaw1}
2711: T.~Kawasaki.
2712: \newblock {{T}he index of elliptic operators over ${V}$-manifolds}.
2713: \newblock {\em Nagoya Math. J.}, {\bf 84}, 1981, 135--157.
2714:
2715: \bibitem{Far1}
2716: C.~Farsi.
2717: \newblock {${K}$-theoretical index theorems for orbifolds}.
2718: \newblock {\em Quart. J. Math. Oxford}, {\bf 43}, 1992, 183--200.
2719:
2720: \bibitem{Ant1}
2721: A.~B. Antonevich.
2722: \newblock {\em Lineinye funktsionalnye uravneniya. {O}peratornyi podkhod}.
2723: \newblock ``Universitetskoe'', Minsk, 1988.
2724:
2725: \bibitem{Hor3}
2726: L.~H{\"o}rmander.
2727: \newblock {\em The Analysis of Linear Partial Differential Operators. {III}}.
2728: \newblock Springer--Verlag, Berlin Heidelberg New York Tokyo, 1985.
2729:
2730: \bibitem{SaScS4}
2731: A.~Savin, B.-W. Schulze, and B.~Sternin.
2732: \newblock {\em The Homotopy Classification and the Index of Boundary Value
2733: Problems for General Elliptic Operators}.
2734: \newblock Univ. Potsdam, Institut f{\"u}r Mathematik, Oktober 1999.
2735: \newblock Preprint N 99/20, arXiv: math/9911055.
2736:
2737: \bibitem{AtSi1}
2738: M.~F. Atiyah and I.~M. Singer.
2739: \newblock The index of elliptic operators {I}.
2740: \newblock {\em Ann. of Math.}, {\bf 87}, 1968, 484--530.
2741:
2742: \bibitem{KoNi1}
2743: J.~Kohn and L.~Nirenberg.
2744: \newblock An algebra of pseudo-differential operators.
2745: \newblock {\em Comm. Pure Appl. Math.}, {\bf 18}, 1965, 269--305.
2746:
2747: \bibitem{LuMi1}
2748: G.~Luke and A.~S. Mishchenko.
2749: \newblock {\em Vector bundles and their applications}, volume 447 of {\em
2750: Mathematics and its Applications}.
2751: \newblock Kluwer Academic Publishers, Dordrecht, 1998.
2752:
2753: \bibitem{Hsi2}
2754: Ch.-Ch. Hsiung.
2755: \newblock The signature and {$G$}-signature of manifolds with boundary.
2756: \newblock {\em J. Diff. Geometry}, {\bf 6}, 1972, 595--598.
2757:
2758: \bibitem{Pal1}
2759: R.~S. Palais.
2760: \newblock {\em Seminar on the Atiyah--Singer index theorem}.
2761: \newblock Princeton Univ. Press, Princeton, NJ, 1965.
2762:
2763: \bibitem{BrNi1}
2764: J.-L. Brylinski and V.~Nistor.
2765: \newblock {C}yclic cohomology of etale groupoids.
2766: \newblock {\em K-theory}, {\bf 8}, 1994, 341--365.
2767:
2768: \bibitem{Hig2}
2769: N.~Higson.
2770: \newblock On the {$K$-theory} proof of the index theorem.
2771: \newblock In {\em Index Theory and Operator Algebras (Boulder, CO, 1991)},
2772: volume 148 of {\em Contemp. Math.}, 1993, pages 67--86, Providence, RI. AMS.
2773:
2774: \bibitem{AtSi4}
2775: M.~F. Atiyah and I.~M. Singer.
2776: \newblock The index of elliptic operators {IV}.
2777: \newblock {\em Ann. Math.}, {\bf 93}, 1971, 119--138.
2778:
2779: \bibitem{APS3}
2780: M.~Atiyah, V.~Patodi, and I.~Singer.
2781: \newblock Spectral asymmetry and {Riemannian} geometry {III}.
2782: \newblock {\em Math. Proc. Cambridge Philos. Soc.}, {\bf 79}, 1976, 71--99.
2783:
2784: \bibitem{APS2}
2785: M.~Atiyah, V.~Patodi, and I.~Singer.
2786: \newblock Spectral asymmetry and {Riemannian} geometry {II}.
2787: \newblock {\em Math. Proc. Cambridge Philos. Soc.}, {\bf 78}, 1976, 405--432.
2788:
2789: \bibitem{Ati6}
2790: M.~F. Atiyah.
2791: \newblock Characters and cohomology of finite groups.
2792: \newblock {\em Publ. Math. IHES}, {\bf 9}, 1961, 23--64.
2793:
2794: \bibitem{SaScS3}
2795: A.~Savin, B.-W. Schulze, and B.~Sternin.
2796: \newblock Elliptic {O}perators in {S}ubspaces and the {E}ta {I}nvariant.
2797: \newblock {\em {K}-theory}, {\bf 27}, No. 3, 2002, 253--272.
2798:
2799: \bibitem{Gil1}
2800: P.~B. Gilkey.
2801: \newblock {\em {I}nvariance theory, the heat equation, and the
2802: {A}tiyah-{S}inger index theorem}.
2803: \newblock Studies in Advanced Mathematics. CRC Press, Boca Raton, FL, second
2804: edition, 1995.
2805:
2806: \bibitem{Don3}
2807: H.~Donnelly.
2808: \newblock Eta-invariants for ${G}$-spaces.
2809: \newblock {\em Indiana Univ. Math. J.}, {\bf 27}, 1978, 889--918.
2810:
2811: \bibitem{Ati4}
2812: M.~F. Atiyah.
2813: \newblock Global theory of elliptic operators.
2814: \newblock In {\em Proc. of the Int. Symposium on Functional Analysis}, 1969,
2815: pages 21--30, Tokyo. University of Tokyo Press.
2816:
2817: \bibitem{HiRo1}
2818: N.~Higson and J.~Roe.
2819: \newblock {\em Analytic {$K$}-homology}.
2820: \newblock Oxford University Press, Oxford, 2000.
2821:
2822: \bibitem{SoTr1}
2823: Yu.~P. Solovyov and E.~V. Troitsky.
2824: \newblock {\em {$C\sp *$}-algebras and elliptic operators in differential
2825: topology}, volume 192 of {\em Translations of Mathematical Monographs}.
2826: \newblock American Mathematical Society, Providence, RI, 2001.
2827:
2828: \bibitem{BaDo1}
2829: P.~Baum and R.~G. Douglas.
2830: \newblock ${K}$-homology and index theory.
2831: \newblock In R.~Kadison, editor, {\em Operator Algebras and Applications},
2832: number~38 in Proc. Symp. Pure Math, 1982, pages 117--173. American
2833: Mathematical Society.
2834:
2835: \bibitem{LaMi1}
2836: H.~B. Lawson and M.~L. Michelsohn.
2837: \newblock {\em Spin geometry}.
2838: \newblock Princeton Univ. Press, Princeton, 1989.
2839:
2840: \end{thebibliography}
2841:
2842:
2843: \end{document}
2844: