1: \documentclass[11pt]{article}
2: \usepackage{amsmath, amssymb, amscd, epsfig}
3:
4: %\setlength{\textwidth}{13cm}
5: %\setlength{\textheight}{20cm}
6:
7:
8: \newtheorem{thm}{Theorem}
9: \newtheorem{prop}[thm]{Proposition}
10: \newtheorem{lm}[thm]{Lemma}
11: \newtheorem{claim}[thm]{Claim}
12: \newtheorem{cor}[thm]{Corollary}
13: \newtheorem{rem}[thm]{Remark}
14: \newtheorem{df}[thm]{Definition}
15: %\newtheorem{ex}[thm]{Example}
16: %\newtheorem{conj}{Conjecture}
17:
18: \renewcommand{\phi}{\varphi}
19: \renewcommand{\epsilon}{\varepsilon}
20:
21: \newcommand{\BB}{\mathbb}
22: \newcommand{\g}{\mathfrak}
23: \newcommand{\pf}{\noindent {\it Proof. }}
24: \newcommand{\qed}{\nopagebreak $\qquad$ $\square$ \vskip5pt}
25: \newcommand{\separate}{\vskip5pt}
26: %\renewcommand{\O}{{\cal O}}
27: %\renewcommand{\S}{{\cal S}}
28: %\newcommand{\N}{{\cal N}}
29: \newcommand{\supp}{\operatorname{supp}}
30: %\newcommand{\parag}{{\cal x}}
31:
32:
33: \begin{document}
34:
35: \title{\bf A Localization Argument for Characters of Reductive Lie Groups:
36: An Introduction and Examples}
37: \author{Matvei Libine}
38: \maketitle
39:
40:
41: \begin{abstract}
42: In this article I describe my recent geometric localization argument
43: dealing with actions of NONcompact groups which
44: provides a geometric bridge between two
45: entirely different character formulas for reductive Lie
46: groups and answers the question posed in \cite{Sch}.
47:
48: A corresponding problem in the compact group setting was
49: solved by N.~Berline, E.~Getzler and M.~Vergne in \cite{BGV}
50: by an application of the theory of equivariant forms
51: and, particularly, the fixed point integral localization formula.
52:
53: This localization argument seems to be the first successful attempt
54: in the direction of building a similar theory for integrals of differential
55: forms, equivariant with respect to actions of noncompact groups.
56:
57: I will explain how the argument works in the $SL(2,\BB R)$ case,
58: where the key ideas are not obstructed by technical details and
59: where it becomes clear how it extends to the general case.
60: The general argument appears in \cite{L}.
61:
62: I put every effort to make this article broadly accessible.
63: Also, although this article mentions characteristic cycles of sheaves,
64: I do not assume that the reader is familiar with this notion.
65: \end{abstract}
66:
67:
68: \tableofcontents
69:
70:
71: \begin{section}
72: {Introduction}
73: \end{section}
74:
75: For motivation, let us start with the case of a compact group.
76: Thus we consider a connected compact group $K$ and a
77: maximal torus $T \subset K$.
78: Let $\g k_{\BB R}$ and $\g t_{\BB R}$ denote the Lie algebras
79: of $K$ and $T$ respectively, and $\g k$, $\g t$ be their
80: complexified Lie algebras.
81: Let $\pi$ be a finite-dimensional representation of $K$,
82: that is $\pi$ is a group homomorphism $K \to Aut(V)$.
83: One of the most important invariants associated to a
84: representation is its {\em character}. It is a function
85: on $K$ defined by
86: $\Theta_{\pi}(x) =_{\text{def}} \operatorname{tr} (\pi(x))$,
87: $x \in K$.
88: Any finite-dimensional representation is completely determined
89: (up to isomorphism) by its character.
90: Recall the exponential map $\exp: \g k_{\BB R} \to K$.
91: If $K$ is a matrix subgroup of some $GL(n)$, then
92: $\exp(A) = e^A$. We use this exponential map to define the
93: {\em character on the Lie algebra} of the representation
94: $(\pi,V)$:
95: $$
96: \theta_{\pi}= (\det \exp_*)^{1/2} \exp^* \Theta_{\pi}.
97: $$
98: It is a smooth bounded function on $\g k_{\BB R}$.
99: Because $K$ is connected and compact, the exponential map is
100: surjective and generically non-singular.
101: Thus $\theta_{\pi}$ still determines the representation.
102:
103: Now let us assume that the representation $\pi$ is irreducible,
104: that is there are no proper $K$-invariant subspaces of $V$.
105: One reason for introducing this factor $(\det \exp_*)^{1/2}$
106: in the definition of $\theta_{\pi}$ is that for irreducible
107: representations $\pi$ the character $\theta_{\pi}$ becomes
108: an eigenfunction of the algebra of conjugation-invariant
109: constant coefficient differential operators on
110: the Lie algebra $\g k_{\BB R}$.
111:
112: There are two entirely different character formulas
113: for $\theta_{\pi}$ -- the {\em Weyl character formula}
114: and {\em Kirillov's character formula}.
115: Recall that the irreducible representations of $K$
116: can be enumerated by their {\em highest weights} which are
117: elements of the {\em weight lattice} $\Lambda$ in $i \g t_{\BB R}^*$
118: intersected with a chosen {\em Weyl chamber}.
119: Let $\lambda=\lambda(\pi) \in i \g t_{\BB R}^*$
120: denote the highest weight corresponding to $\pi$.
121: The group $K$ acts on its own Lie algebra
122: $\g k_{\BB R}$ by adjoint representation $Ad$.
123: Let $W=N_K(\g t_{\BB R})/T$,
124: where $N_K(\g t_{\BB R})$ is the normalizer
125: of $\g t_{\BB R}$ in $K$.
126: The set $W$ is a finite group called the {\em Weyl group};
127: it acts on $\g t_{\BB R}$ and hence on $i \g t_{\BB R}^*$.
128: We can choose a positive definite inner product
129: $\langle \cdot,\cdot\rangle$ on $i \g t_{\BB R}^*$ invariant under $W$.
130: Then the Weyl character formula can be stated as follows:
131: $$
132: \theta_{\pi}|_{\g t_{\BB R}}(t)=\sum_{w \in W} \frac {e^{w(\lambda+\rho)(t)}}
133: {\prod_{\alpha \in \Phi, \langle w(\lambda+\rho),\alpha\rangle>0} \alpha(t)},
134: $$
135: where $\Phi \subset i \g t_{\BB R}^*$ is the {\em root system}
136: of $\g k_{\BB R}$ and
137: $\rho \in i \g t_{\BB R}^*$ is a certain small vector independent
138: of $\pi$. Because $\theta_{\pi}$ is $Ad(K)$-invariant and
139: every $Ad(K)$-orbit in $\g k_{\BB R}$ meets $\g t_{\BB R}$,
140: this formula completely determines $\theta_{\pi}$.
141:
142: Kirillov's character formula provides a totally different expression
143: for the irreducible characters on $\g k_{\BB R}$.
144: The splitting
145: $\g k_{\BB R}=\g t_{\BB R} \oplus [\g t_{\BB R},\g k_{\BB R}]$
146: (Cartan algebra $\oplus$ root spaces)
147: induces a dual splitting of the vector space $i \g k_{\BB R}^*$,
148: which allows us to think of $\lambda$ and $\rho$ as lying in
149: $i \g k_{\BB R}^*$. The adjoint action of $K$ on
150: $\g k_{\BB R}$ has a dual action on $i \g k_{\BB R}^*$
151: called coadjoint representation. We define
152: $$
153: \Omega_{\lambda+\rho} = \text{ $K$-orbit of $\lambda+\rho$
154: in $i \g k_{\BB R}^*$}.
155: $$
156: It will be convenient to define the Fourier transform $\hat \phi$
157: of a test function $\phi \in {\cal C}^{\infty}_c (\g k_{\BB R})$
158: without the customary factor of $i=\sqrt{-1}$ in the exponent,
159: as a function on $i \g k_{\BB R}^*$:
160: $$
161: \hat \phi(\zeta) = \int_{\g k_{\BB R}}
162: \phi(x)e^{\langle \zeta, x \rangle} dx.
163: $$
164: Then Kirillov's character formula describes $\theta_{\pi}$ as a
165: distribution on $\g k_{\BB R}$:
166: $$
167: \int_{\g k_{\BB R}} \theta_{\pi} \phi dx =
168: \int_{\Omega_{\lambda+\rho}} \hat \phi d\beta,
169: $$
170: where $d\beta$ is the measure induced by the canonical symplectic
171: structure of $\Omega_{\lambda+\rho}$.
172: In other words,
173: $$
174: \hat \theta_{\pi} = \text{ integration over $\Omega_{\lambda+\rho}$}.
175: $$
176: Kirillov calls this the ``universal formula'' for irreducible characters.
177:
178: \separate
179:
180: The geometric relationship between these two formulas is even
181: more striking. As a homogeneous space, $\Omega_{\lambda+\rho}$
182: is isomorphic to the {\em flag variety} $X$, i.e the variety
183: of Borel subalgebras
184: $\g b \subset \g k = \g k_{\BB R} \otimes_{\BB R} \BB C$.
185: The space $X$ is a smooth complex projective variety which is also
186: isomorphic to $K/T$ as a homogeneous space.
187: The Borel-Weil-Bott theorem can be regarded as an explicit
188: construction of a holomorphic $K$-equivariant line
189: bundle ${\cal L}_{\lambda} \to X$ such that the resulting
190: representation of $K$ in the cohomology groups is:
191: \begin{eqnarray*}
192: && H^p(X, {\cal O}({\cal L}_{\lambda}))=0 \text{\quad if $p \ne 0$}, \\
193: && H^0(X, {\cal O}({\cal L}_{\lambda})) \simeq \pi.
194: \end{eqnarray*}
195: Then the Weyl character formula is a consequence of
196: the Atiyah-Bott fixed point formula or the Lefschetz fixed point formula.
197: On the other hand, N.~Berline, E.~Getzler
198: and M.~Vergne proved in \cite{BGV}
199: Kirillov's character formula using the integral
200: localization formula for $K$-equivariant forms.
201: They showed that the right hand side of Kirillov's character formula
202: equals the right hand side of the Weyl character formula.
203:
204: \separate
205:
206: Equivariant forms were introduced in 1950 by Henri Cartan.
207: There are many good texts on this subject including
208: \cite{BGV} and \cite{GS}.
209:
210: Let $K$ act on a smooth manifold $M$, and let $\Omega^{\bullet}(M)$
211: denote the algebra of smooth differential forms on $M$.
212: A $K$- equivariant form is a smooth map
213: $$
214: \omega: \g k_{\BB R} \to \Omega^{\bullet}(M)
215: $$
216: whose image need not lie entirely in any single degree component of
217: $\Omega^{\bullet}(M)$ and which is $K$-invariant,
218: i.e. for all elements $k \in \g k_{\BB R}$ and $\tilde k \in K$ we have
219: $$
220: \omega(k) = (\tilde k)^{-1} \cdot \omega \bigl( Ad(\tilde k) k \bigr).
221: $$
222:
223: We say that an equivariant form $\omega$ is {\em equivariantly closed}
224: if $d_{equiv}(\omega)=0$, where
225: $$
226: \bigl( d_{equiv}(\omega) \bigr) (k) = d(\omega(k)) + \iota (k) \omega(k).
227: $$
228: Here, the first term is the ordinary deRham differential of $\omega(k)$
229: and the last term denotes the contraction of the differential form
230: $\omega(k)$ by the vector field on $M$ generated by the infinitesimal
231: action of $k$.
232:
233: If $N \subset M$ is a submanifold, then
234: $$
235: \int_N \omega =_{\text{def}}
236: \int_N \text{component of $\omega$ of degree $\dim N$};
237: $$
238: it is a function on $\g k_{\BB R}$. Then the localization formula
239: reduces integration of an equivariantly closed form to
240: summation over the zeroes of the vector field in $M$ generated
241: by $k \in \g k_{\BB R}$.
242: In other words, it expresses a global object such as integral of a
243: differential form in terms of quantities which can be calculated
244: locally at the zeroes of the vector field.
245: It is crucial for the localization formula to hold that the group $K$
246: is compact.
247:
248: \separate
249:
250: Now let $G_{\BB R}$ be a connected, linear, reductive Lie group.
251: We let $\g g_{\BB R}$ denote its Lie algebra.
252: Then most representations of interest have infinite dimension.
253: We always consider representations on complete, locally convex
254: Hausdorff topological vector spaces and require that the action
255: of $G_{\BB R}$ is continuous.
256: Let $K$ be a maximal compact subgroup of $G_{\BB R}$.
257: A reasonable category of representations consists of
258: {\em admissible} representations of {\em finite length}.
259: (A representation $\pi$ has finite length if every increasing chain
260: of closed, invariant subspaces breaks off after finitely many steps;
261: $\pi$ is admissible if its restriction to $K$ contains any irreducible
262: representation of $K$ at most finitely often.) Admissibility
263: is automatic for irreducible unitary representations.
264: Although trace of a linear operator in an infinite-dimensional
265: space cannot be defined in general, it is still possible to
266: define a character $\theta_{\pi}$ as an
267: $Ad(G_{\BB R})$-invariant distribution on $\g g_{\BB R}$.
268:
269: M.~Kashiwara and W.~Schmid in their paper \cite{KSch}
270: generalize the Borel-Weil-Bott construction.
271: (Recall that $X$ denotes the flag variety of Borel subalgebras of the
272: complexified Lie algebra $\g g = \g g_{\BB R} \otimes_{\BB R} \BB C$.)
273: Instead of line bundles on $X$ they consider
274: $G_{\BB R}$-equivariant sheaves ${\cal F}$ and, for each integer
275: $p \in \BB Z$, they define representations of $G_{\BB R}$ in
276: $\operatorname{Ext}^p({\cal F},{\cal O})$.
277: Such representations turn out to be admissible of finite length.
278: Then W.~Schmid and K.~Vilonen prove in \cite{SchV2}
279: two character formulas for these representations --
280: the fixed point character formula and the integral character formula.
281: In the case when $G_{\BB R}$ is compact, the former reduces to
282: the Weyl character formula and the latter -- to Kirillov's character
283: formula. The fixed point formula was conjectured by M.~Kashiwara
284: in \cite{K}, and its proof uses a
285: generalization of the Lefschetz fixed point formula to sheaf
286: cohomology due to M.~Goresky and R.~MacPherson in \cite{GM}.
287: On the other hand, W.~Rossmann in \cite{R} established existence
288: of an integral character formula over an unspecified
289: Borel-Moore cycle.
290: W.~Schmid and K.~Vilonen prove the integral character formula
291: where integration takes place over the {\em characteristic cycle} of
292: ${\cal F}$, $Ch({\cal F})$, and their proof
293: depends totally on representation theory.
294:
295: $Ch({\cal F})$ is a conic Lagrangian cycle in the cotangent space
296: $T^*X$ associated to the sheaf ${\cal F}$.
297: Characteristic cycles were introduced by M.~Kashiwara and their
298: definition can be found in \cite{KaScha}. On the other hand,
299: W.~Schmid and K.~Vilonen give a geometric way to understand
300: characteristic cycles in \cite{SchV1}.
301:
302: In this article we only deal with characteristic cycles when
303: considering the important cases in the situation of $SL(2,\BB R)$.
304: Each time a characteristic cycle appears here it is described
305: explicitly and, therefore, the reader is not expected to be
306: familiar with this notion.
307:
308: \separate
309:
310: The equivalence of these two formulas can be stated in terms
311: of the sheaves ${\cal F}$ alone, without any reference to
312: their representation-theoretic significance.
313: In the announcement \cite{Sch} W.~Schmid posed a question:
314: ``Can this equivalence be seen directly without a detour to
315: representation theory, just as in the compact case.''
316:
317: In this article I provide such a geometric link.
318: I introduce a localization technique which allows
319: to localize integrals to the zeroes of vector fields on $X$
320: generated by the infinitesimal action of $\g g_{\BB R}$.
321: Thus, in addition to a representation-theoretical result,
322: we obtain a whole family of examples where it is possible
323: to localize integrals to fixed points with respect to an
324: action of a noncompact group.
325: As far as I am aware, this localization argument is the first
326: successful attempt in the direction of building an integral
327: localization theory for noncompact groups.
328:
329: \separate
330:
331: The following convention will be in force throughout these notes:
332: whenever $A$ is a subset of $B$,
333: we will denote the inclusion map $A \hookrightarrow B$ by
334: $j_{A \hookrightarrow B}$.
335:
336:
337: \separate
338:
339: \begin{section}
340: {Two Character Formulas}
341: \end{section}
342:
343: In these notes we try to keep the same notations as W.~Schmid and K.~Vilonen
344: use in \cite{SchV2} as much as possible.
345: That is they fix a connected, complex algebraic,
346: reductive group $G$ which is defined over $\BB R$.
347: The representations they consider are representations of a real form
348: $G_{\BB R}$ of $G$ -- in other words, $G_{\BB R}$ is a subgroup of
349: $G$ lying between the group of real points $G(\BB R)$ and the
350: identity component $G(\BB R)^0$.
351: They regard $G_{\BB R}$ as a reductive Lie group and
352: denote by $\g g$ and $\g g_{\BB R}$ the Lie algebras of
353: $G$ and $G_{\BB R}$ respectively,
354: they also denote by $X$ the flag variety of $G$.
355:
356: If $g \in \g g$ is an element of the Lie algebra,
357: we denote by $\operatorname{VF}_g$ the vector field on $X$
358: generated by $g$: if $x \in X$ and $f \in {\cal C}^{\infty}(X)$, then
359: $$
360: \operatorname{VF}_g (x)f=
361: \frac d{d\epsilon} f(\exp(\epsilon g) \cdot x)|_{\epsilon=0}.
362: $$
363: We call a point $x \in X$ a {\em fixed point of $g$} if the vector field
364: $\operatorname{VF}_g$ on $X$ vanishes at $x$, i.e.
365: $\operatorname{VF}_g (x)= 0$.
366:
367: In this paragraph we explain the general picture, but since objects
368: mentioned here will not play any role in what follows they will not
369: be defined, rather the reader is referred to \cite{SchV2}.
370: W.~Schmid and K.~Vilonen denote by $\g h$ the universal Cartan
371: algebra. They pick an element $\lambda \in \g h^*$ and introduce
372: the ``$G_{\BB R}$-equivariant derived category on $X$ with twist
373: $(\lambda - \rho)$'' denoted by
374: $\operatorname{D}_{G_{\BB R}}(X)_{\lambda}$.
375: They also introduce ${\cal O}_X(\lambda)$,
376: the twisted sheaf of holomorphic functions on $X$,
377: with twist $(\lambda - \rho)$.
378: Then, for ${\cal F} \in \operatorname{D}_{G_{\BB R}}(X)_{-\lambda}$,
379: they define a virtual representation of $G_{\BB R}$
380: $$
381: \sum_p (-1)^p \operatorname{Ext}^p (\BB D {\cal F}, {\cal O}_X(\lambda)),
382: $$
383: where $\BB D {\cal F} \in \operatorname{D}_{G_{\BB R}}(X)_{\lambda}$
384: denotes the Verdier dual of ${\cal F}$.
385: It was shown in \cite{KSch} that each
386: $\operatorname{Ext}^p (\BB D {\cal F}, {\cal O}_X(\lambda))$
387: is admissible of finite length.
388: There are two formulas expressing the character $\theta$ of this
389: virtual representation as a distribution on $\g g_{\BB R}$.
390: We think of a character as a linear functional defined on the space of
391: smooth compactly supported differential forms $\phi$ on $\g g_{\BB R}$
392: of top degree, and write $\int_{\g g_{\BB R}} \theta \phi$ for the value
393: of $\theta$ at $\phi$.
394: In my article \cite{L} I start with the right hand side of
395: the integral character formula
396: \begin{equation} \label{intformula}
397: \int_{\g g_{\BB R}} \theta \phi =
398: \frac 1{(2\pi i)^nn!} \int_{Ch({\cal F})}
399: \mu_{\lambda}^* \hat \phi (-\sigma+\pi^* \tau_{\lambda})^n
400: \end{equation}
401: and show that it is equivalent to the right hand side of the
402: fixed point character formula
403: \begin{equation} \label{fpformula}
404: \int_{\g g_{\BB R}} \theta \phi = \int_{\g g_{\BB R}} F_{\theta} \phi,
405: \qquad
406: F_{\theta} (g) =
407: \sum_{k=1}^{|W|}
408: \frac {m_{x_k(g)} e^{\langle g,\lambda_{x_k(g)} \rangle}}
409: {\alpha_{x_k(g),1}(g) \dots \alpha_{x_k(g),n}(g)},
410: \end{equation}
411: where $F_{\theta}$ is a function in $L^1_{loc} (\g g_{\BB R})$,
412: $x_1, \dots, x_n$ are the fixed points of $g$ and
413: integers $m_{x_k(g)}$'s are the local invariants of ${\cal F}$.
414:
415: \separate
416:
417: Because both character formulas depend on
418: ${\cal F} \in \operatorname{D}_{G_{\BB R}}(X)_{-\lambda}$ only through
419: its characteristic cycle $Ch({\cal F})$,
420: as far as characteristic cycles concern, we can simply
421: replace ${\cal F}$ with a $G_{\BB R}$-equivariant sheaf on the flag
422: variety $X$ with the same characteristic cycle.
423: We will use the same notation ${\cal F}$ to denote this
424: $G_{\BB R}$-equivariant sheaf on $X$.
425: Let $n = \dim_{\BB C} X$, let $\pi: T^*X \twoheadrightarrow X$ be
426: the projection map, and equip $\g g_{\BB R}$ with some orientation.
427: Then $Ch({\cal F})$ is a $2n$-cycle in $T^*X$ which has real dimension $4n$.
428:
429: We will make an elementary calculation of the integral
430: $$
431: \frac 1{(2\pi i)^nn!} \int_{Ch({\cal F})}
432: \mu_{\lambda}^* \hat \phi (-\sigma+\pi^* \tau_{\lambda})^n
433: $$
434: where $\phi$ is a smooth compactly supported differential
435: form on $\g g_{\BB R}$ of top degree,
436: $$
437: \hat \phi(\zeta) = \int_{\g g_{\BB R}} e^{\langle g, \zeta \rangle} \phi
438: \qquad (g \in \g g_{\BB R},\,\zeta \in \g g^*)
439: $$
440: is its Fourier transform
441: (without the customary factor of $i=\sqrt{-1}$ in the exponent),
442: $\mu_{\lambda}: T^*X \to \g g^*$
443: is the twisted moment map defined in \cite{SchV1}
444: and $\tau_{\lambda}$, $\sigma$ are 2-forms on
445: $X$ and $T^*X$ respectively defined in \cite{SchV2}.
446: $\sigma$ is the complex algebraic symplectic form on $T^*X$.
447: On the other hand, the precise definition of the form $\tau_{\lambda}$
448: will not be important. What will be important, however,
449: is that, for each $g \in \g g$, the $2n$-form on $T^*X$
450: \begin{equation} \label{integrand}
451: e^{\langle g, \mu_{\lambda}(\zeta) \rangle}
452: (-\sigma+\pi^* \tau_{\lambda})^n
453: \end{equation}
454: is closed.
455:
456: \separate
457:
458: If $\lambda =0$, the twisted moment map $\mu_{\lambda}$ reduces to
459: the ordinary moment map $\mu_0=\mu$ defined by
460: $$
461: \mu(\zeta): g \mapsto \langle \zeta, \operatorname{VF}_g \rangle,
462: \qquad \zeta \in T^*X.
463: $$
464: The moment map $\mu$ takes values in the nilpotent cone
465: ${\cal N^*} \subset \g g^*$.
466: In general, $\mu_{\lambda} = \mu + \lambda_x$, where
467: $\lambda_x$ is a function on $X$ mapping $x \in X$ into
468: $\lambda_x \in \g g^*$.
469:
470: Suppose that $\lambda \in \g h^*$ is regular. For $\g g = \g{sl}(2,\BB C)$
471: this simply means $\lambda \ne 0$. Then $\mu_{\lambda}$ is a real analytic
472: diffeomorphism of $T^*X$ onto $\Omega_{\lambda} \subset \g g^*$
473: -- the orbit of $\lambda$ under the coadjoint action of $G$ on $\g g^*$.
474: Let $\sigma_{\lambda}$ denote the canonical $G$-invariant complex algebraic
475: symplectic form on $\Omega_{\lambda}$. Then
476: $e^{\langle g, \zeta \rangle} (\sigma_{\lambda})^n$ is a holomorphic
477: $2n$-form of maximal possible degree, hence closed.
478: It turns out that
479: $\mu_{\lambda}^* (\sigma_{\lambda}) = -\sigma+\pi^* \tau_{\lambda}$.
480: This shows that, for $\lambda$ regular, $g \in \g g$,
481: $$
482: e^{\langle g, \mu_{\lambda}(\zeta) \rangle}
483: (-\sigma+\pi^* \tau_{\lambda})^n =
484: \mu_{\lambda}^* \bigl( e^{\langle g, \zeta \rangle} (\sigma_{\lambda})^n \bigr)
485: $$
486: is a closed $2n$-form on $T^*X$.
487: Note that neither map $\mu_{\lambda}$ nor the form (\ref{integrand})
488: is holomorphic. Because the form
489: $e^{\langle g, \mu_{\lambda}(\zeta) \rangle}
490: (-\sigma+\pi^* \tau_{\lambda})^n$
491: depends on $\lambda$ real analytically and the set of regular elements
492: is dense in $\g h^*$, we conclude that the form in the equation
493: (\ref{integrand}) is closed.
494:
495:
496: \separate
497:
498: \begin{section}
499: {Localization Argument for $SL(2,\BB R)$}
500: \end{section}
501:
502: In this section we will deal with characteristic cycles of sheaves.
503: I remind it once again that all characteristic cycles which
504: appear here will be described explicitly and the reader is not expected
505: to be familiar with this notion.
506:
507: \separate
508:
509: There are three major obstacles to any geometric proof of the equivalence
510: of formulas (\ref{intformula}) and (\ref{fpformula}):
511: first of all, the characteristic cycle $Ch({\cal F})$ need not be smooth,
512: in fact, it may be extremely singular;
513: secondly, the integrand in (\ref{intformula}) is an equivariant form with
514: respect to some compact real form $U_{\BB R} \subset G$,
515: but $U_{\BB R}$ does not preserve $Ch({\cal F})$
516: (unless $Ch({\cal F})$ is a multiple of the zero section of $T^*X$
517: equipped with some orientation);
518: and, lastly, $Ch({\cal F})$ is not compactly supported.
519:
520: \separate
521:
522: In this section we will consider two different examples when
523: $G_{\BB R} = SL(2,\BB R)$.
524:
525: Recall that when $G_{\BB R} = SL(2,\BB R)$, the flag variety $X$ is
526: just a 2-sphere $\BB C P^1$ on which $SL(2,\BB R)$ acts by
527: projective transformations.
528: There are exactly three $SL(2,\BB R)$-orbits: two open hemispheres
529: and one circle which is their common boundary.
530: Also, $n = \dim_{\BB C} X =1$.
531:
532: \separate
533:
534: To deal with the problem of noncompactness of $Ch({\cal F})$
535: we fix a norm $\|.\|$ on the space of linear functionals on
536: $\g{sl}(2,\BB R)$, $\g{sl}(2,\BB R)^*$. Then the moment map $\mu$
537: induces a vector bundle norm on $T^*X$: for $\zeta \in T^*X$
538: its norm will be $\|\mu(\zeta)\|$. We will use the same notation
539: $\|.\|$ for this norm too.
540:
541: Let $\g{sl}(2,\BB R)'$ denote the set of regular semisimple elements
542: in $\g{sl}(2,\BB R)$: $\g{sl}(2,\BB R)'$ consists of those elements
543: in $\g{sl}(2,\BB R)$ which are diagonalizable over $\BB C$,
544: with distinct eigenvalues.
545:
546: Since the complement of $\g{sl}(2,\BB R)'$ in $\g{sl}(2,\BB R)$
547: has measure zero, we can replace integration over $\g{sl}(2,\BB R)$
548: by integration over $\g{sl}(2,\BB R)'$. Then
549: \begin{multline} \label{R}
550: \frac 1{2\pi i} \int_{Ch({\cal F})}
551: \mu_{\lambda}^* \hat \phi (-\sigma+\pi^* \tau_{\lambda}) \\
552: =\lim_{R \to \infty}
553: \frac 1{2\pi i}
554: \int_{\g{sl}(2,\BB R) \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\})}
555: e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
556: (-\sigma+\pi^* \tau_{\lambda}) \\
557: =\lim_{R \to \infty} \frac 1{2\pi i}
558: \int_{\g{sl}(2,\BB R)' \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\})}
559: e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
560: (-\sigma+\pi^* \tau_{\lambda}).
561: \end{multline}
562: (Of course, the orientation on
563: $\g{sl}(2,\BB R)' \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\})$
564: is induced by the product orientation on
565: $\g{sl}(2,\BB R) \times Ch({\cal F})$.)
566:
567: We will interchange the order of integration:
568: integrate over the characteristic cycle first and only then
569: perform integration over $\g{sl}(2,\BB R)'$.
570: As we already mentioned, the integrand
571: is an equivariant form with respect to some
572: compact real form $U_{\BB R} \subset G = SL(2,\BB C)$.
573: But $U_{\BB R}$ does not preserve $Ch({\cal F})$
574: (unless $Ch({\cal F})$ is a multiple of the zero section of $T^* \BB CP^1$
575: equipped with some orientation).
576: We regard the integral as an integral of a closed differential form
577: $e^{\langle g, \mu_{\lambda}(\zeta) \rangle}
578: \phi(-\sigma+\pi^* \tau_{\lambda})$
579: over a chain in
580: $\g{sl}(2,\BB R)' \times (T^*\BB CP^1 \cap \{\|\zeta\| \le R\})$.
581:
582: Each regular semisimple element $g \in \g{sl}(2,\BB R)'$
583: has exactly two fixed points on $X=\BB CP^1$.
584: We will use the open embedding theorem of W.~Schmid and K.~Vilonen
585: (\cite{SchV1}) to construct a deformation of
586: $Ch({\cal F})$ into a simple cycle of the following kind:
587: $$
588: m_1 T^*_{x_1}X + m_2 T^*_{x_2}X,
589: $$
590: where $m_1,m_2$ are some integers,
591: $x_1,x_2$ are the points in $\BB CP^1$ fixed by $g$,
592: and each cotangent space $T^*_{x_k}X$ is given some orientation.
593: To ensure that the integral behaves well, we will stay during the
594: process of deformation inside the set
595: $$
596: \{(g,\zeta) \in \g{sl}(2,\BB R)' \times T^* \BB CP^1;\,
597: Re( \langle g,\mu(\zeta) \rangle) \le 0\}.
598: $$
599:
600: \separate
601:
602: The most important case is the split Cartan case,
603: i.e. the case when the eigenvalues of $g$ are real and
604: the unique Cartan algebra containing $g$ is split.
605: The compact Cartan case also needs to be considered, but
606: because it can be treated using the classical theory of
607: equivariant forms it is not so interesting.
608:
609: In the split Cartan case one fixed point in $\BB CP^1$ is stable
610: (the vector field $\operatorname{VF}_g$ on $\BB CP^1$ points towards it)
611: and the other fixed point in $\BB CP^1$ is unstable (the vector field
612: $\operatorname{VF}_g$ points away from it).
613: We can position $X=\BB CP^1$ in space so that the unstable fixed point
614: is the north pole $N$ and the stable fixed point is the south pole $S$.
615: Furthermore, we can position it so that the eastern and western hemispheres
616: are stable under the $SL(2,\BB R)$-action. So, let $H$ denote one of these
617: two open hemispheres, say, the western one; and let $S^1 \subset \BB CP^1$
618: denote the circle containing the Greenwich meridian; $S^1 = \partial H$.
619:
620:
621: \begin{figure}
622: \centerline{\psfig{figure=AMSfig1.eps}}
623: \caption{Positioning of $N$, $S$, $H$, $S^1$ and $U_t$.}
624: \end{figure}
625:
626:
627: There are only two essential choices for the
628: $SL(2,\BB R)$-equivariant sheaf ${\cal F}$:
629: ${\cal F}_{\text{principal}}$ -- the sheaf which is
630: ``zero outside of $S^1$ and the constant sheaf along $S^1$'' --
631: defined by
632: $$
633: {\cal F}_{\text{principal}} = (j_{S^1 \hookrightarrow \BB CP^1})_* \BB C_{S^1},
634: $$
635: where $\BB C_{S^1}$ denotes the constant sheaf on $S^1$;
636: and ${\cal F}_{\text{discrete}}$ -- the sheaf which is
637: ``zero outside of $H$ and the constant sheaf along $H$'' --
638: defined by
639: $$
640: {\cal F}_{\text{discrete}} = (j_{H \hookrightarrow \BB CP^1})_! \BB C_H,
641: $$
642: where $\BB C_H$ denotes the constant sheaf on $H$.
643: If we take any other $SL(2,\BB R)$-equivariant sheaf ${\cal F}$,
644: its characteristic cycle $Ch({\cal F})$ will be an integral linear combination
645: of $Ch({\cal F}_{\text{principal}})$, $Ch({\cal F}_{\text{discrete}})$
646: and $Ch({\cal F}_{\text{discrete'}})$, where
647: $$
648: {\cal F}_{\text{discrete'}} = (j_{H' \hookrightarrow \BB CP^1})_! \BB C_{H'}
649: $$
650: and $H'$ denotes the open hemisphere opposite to $H$ (the eastern hemisphere).
651:
652: In the case of ${\cal F}_{\text{principal}}$, the integral character formula
653: (\ref{intformula}) calculates the character of the principal series
654: representation.
655: On the other hand,
656: in the case of ${\cal F}_{\text{discrete}}$, for certain values of $\lambda$,
657: the integral character formula (\ref{intformula}) calculates the character
658: of the discrete series representation.
659:
660: If $M$ is a submanifold of $X$, we will denote by $T^*_MX$ the
661: {\em conormal vector bundle} to $M$ in $T^*X$.
662: That is $T^*_MX$ is a subbundle of $T^*X$ restricted to $M$ consisting of those
663: covectors in $T^*_mX$ which annihilate $T_m M \subset T_m X$, $m \in M$.
664:
665: \begin{figure}
666: \centerline{\psfig{figure=AMSfig2.eps}}
667: \caption{$Ch({\cal F}_{\text{principal}})$.}
668: \end{figure}
669:
670: As promised, we give an explicit description of the characteristic cycles:
671: $$
672: Ch({\cal F}_{\text{principal}}) = T^*_{S^1} \BB CP^1,
673: $$
674: the conormal bundle to $S^1$; and the open embedding theorem of
675: W.~Schmid and K.~Vilonen (\cite{SchV1}) tells us that
676: \begin{multline*}
677: Ch({\cal F}_{\text{discrete}}) = T^*_H \BB CP^1 \bigcup
678: \{ \zeta \in T^*_{S^1} \BB CP^1;\, \langle \zeta, v \rangle \ge 0 \\
679: \text{ for all tangent vectors $v$ pointing outside of $H$} \}.
680: \end{multline*}
681: Of course, their orientations must be specified too, but they are not
682: important at this level of details.
683:
684: \begin{figure}
685: \centerline{\psfig{figure=AMSfig3.eps}}
686: \caption{$Ch({\cal F}_{\text{discrete}})$.}
687: \end{figure}
688:
689: For $t \in (-90^{\circ},90^{\circ}]$ we will denote by $U_t$ the open
690: set consisting of all points on the sphere $\BB CP^1$ whose
691: latitude is strictly less than $t$.
692:
693: Now we will describe the deformations of $Ch({\cal F}_{\text{discrete}})$
694: first and then of $Ch({\cal F}_{\text{principal}})$.
695: For $t \in (-90^{\circ},90^{\circ}]$ we can consider the sheaf
696: ${\cal F}_{\text{discrete}, t}$ which is
697: ``zero outside of $H \cap U_t$ and the constant sheaf along $H \cap U_t$''
698: defined by
699: $$
700: {\cal F}_{\text{discrete}, t} =
701: (j_{U_t \hookrightarrow \BB CP^1})_! ({\cal F}_{\text{discrete}} |_{U_t})
702: = (j_{H \cap U_t \hookrightarrow \BB CP^1})_! \BB C_{H \cap U_t}.
703: $$
704: Observe that
705: $$
706: {\cal F}_{\text{discrete}, 90^{\circ}} = {\cal F}_{\text{discrete}}.
707: $$
708: Heuristically, if one ``deforms a sheaf slightly'' (whatever that means)
709: one could expect its characteristic cycle make only slight changes too.
710:
711: Then we can consider the characteristic cycle of
712: ${\cal F}_{\text{discrete}, t}$, $Ch({\cal F}_{\text{discrete}, t})$.
713: If $t=90^{\circ}$,
714: $Ch({\cal F}_{\text{discrete}, t}) = Ch({\cal F}_{\text{discrete}})$.
715: When $t \in (-90^{\circ},90^{\circ})$,
716: we use the open embedding theorem of W.~Schmid and K.~Vilonen (\cite{SchV1})
717: once again to see that
718: \begin{multline*}
719: Ch({\cal F}_{\text{discrete}, t})
720: = Ch \bigl(
721: (j_{H \cap U_t \hookrightarrow \BB CP^1})_! \BB C_{H \cap U_t} \bigr) \\
722: = T^*_{H \cap U_t} \BB CP^1 \bigcup
723: \{ \zeta \in T^*_x \BB CP^1;\, x \in \partial (H \cap U_t), \,
724: \langle \zeta, v \rangle \ge 0 \\
725: \text{ for all tangent vectors $v$ pointing outside of $H \cap U_t$} \}.
726: \end{multline*}
727:
728: \begin{figure}
729: \centerline{\psfig{figure=AMSfig4.eps}}
730: \caption{$Ch ({\cal F}_{\text{discrete}, t})$.}
731: \end{figure}
732:
733: \begin{figure}
734: \centerline{\psfig{figure=AMSfig5.eps}}
735: \caption{$Ch ({\cal F}_{\text{discrete}})$ is deformed into $T^*_S \BB CP^1$.}
736: \end{figure}
737:
738: We see that, as we decrease $t$ from $90^{\circ}$ to $0^{\circ}$,
739: we continuously deform the cycle $Ch({\cal F}_{\text{discrete}})$.
740: The end result of this deformation
741: is the cotangent space at the south pole $S$, $T^*_S \BB CP^1$.
742: Moreover, because the vector field $\operatorname{VF}_g$ never points
743: outside of $H \cap U_t$, it is true that during the
744: process of deformation we always stayed inside the set
745: $$
746: \{\zeta \in T^* \BB CP^1;\, Re( \langle g,\mu(\zeta) \rangle) =
747: \langle \operatorname{VF}_g, \zeta \rangle \le 0\}.
748: $$
749: In other words, $Ch({\cal F}_{\text{discrete}}) - T^*_S \BB CP^1$ is the
750: boundary of a chain which lies entirely inside
751: \begin{equation} \label{negative}
752: \{\zeta \in T^* \BB CP^1;\, Re( \langle g,\mu(\zeta) \rangle) \le 0\}.
753: \end{equation}
754:
755: Notice also that although the north pole $N$ and the south pole $S$ apriori
756: seem to be entirely symmetric with respect to ${\cal F}$, it is only the
757: south pole $S$ that counts because the condition (\ref{negative}) must be
758: satisfied during a deformation. And there is no way to deform
759: $Ch({\cal F}_{\text{discrete}})$ into $T^*_N \BB CP^1$ without breaking
760: the condition (\ref{negative}).
761:
762: \separate
763:
764: Next we show how to deform $Ch({\cal F}_{\text{principal}})$.
765: First of all, we observe that the flag variety
766: $$
767: X = \BB CP^1 = U_{90^{\circ}} \coprod \{N\}.
768: $$
769: (For general real reductive Lie groups $G_{\BB R}$ we use
770: the Bruhat cell decomposition.)
771: Hence we get a distinguished triangle:
772: $$
773: (j_{U_{90^{\circ}} \hookrightarrow \BB CP^1})_!
774: ({\cal F}_{\text{principal}}|_{U_{90^{\circ}}})
775: \to {\cal F}_{\text{principal}} \to
776: (j_{\{N\} \hookrightarrow \BB CP^1})_!
777: ({\cal F}_{\text{principal}}|_{\{N\}}),
778: $$
779: which is equivalent to
780: $$
781: (j_{S^1 \cap U_{90^{\circ}} \hookrightarrow \BB CP^1})_!
782: \BB C_{U_{S^1 \cap U_{90^{\circ}}}}
783: \to {\cal F}_{\text{principal}} \to
784: (j_{\{N\} \hookrightarrow \BB CP^1})_! \BB C_N.
785: $$
786: The first term is the sheaf which is ``zero outside
787: $S^1 \cap U_{90^{\circ}} = S^1 \setminus \{N\}$ and the constant sheaf
788: along $S^1 \cap U_{90^{\circ}}$''; the last term is the sheaf which is
789: ``zero outside $\{N\}$ and the constant sheaf on $\{N\}$''.
790:
791: It is a basic property of characteristic cycles that the characteristic
792: cycle of the middle sheaf in a distinguished triangle equals the sum of the
793: characteristic cycles of the other two sheaves:
794: $$
795: Ch({\cal F}_{\text{principal}}) =
796: Ch \bigl( (j_{S^1 \cap U_{90^{\circ}} \hookrightarrow \BB CP^1})_!
797: \BB C_{S^1 \cap U_{90^{\circ}}} \bigr) +
798: Ch \bigl( (j_{\{N\} \hookrightarrow \BB CP^1})_! \BB C_N \bigr).
799: $$
800: The term
801: $$
802: Ch \bigl( (j_{\{N\} \hookrightarrow \BB CP^1})_! \BB C_N \bigr) =
803: T^*_N \BB CP^1.
804: $$
805: already has the desired form. We deform the other term
806: $$
807: Ch \bigl( (j_{S^1 \cap U_{90^{\circ}} \hookrightarrow \BB CP^1})_!
808: \BB C_{S^1 \cap U_{90^{\circ}}} \bigr) = T^*_{S^1} \BB CP^1 - T^*_N \BB CP^1
809: $$
810: only, and we do it very similarly to the case of
811: $Ch({\cal F}_{\text{discrete}})$.
812:
813: Thus, for $t \in (-90^{\circ},90^{\circ}]$, we consider the sheaf
814: ${\cal F}_{\text{principal}, t}$ which is ``zero outside
815: $S^1 \cap U_t = S^1 \setminus \{N\}$ and the constant sheaf
816: along $S^1 \cap U_t$'' defined by
817: $$
818: {\cal F}_{\text{principal}, t} =
819: (j_{U_t \hookrightarrow \BB CP^1})_! ({\cal F}_{\text{principal}} |_{U_t})
820: = (j_{S^1 \cap U_t \hookrightarrow \BB CP^1})_! \BB C_{S^1 \cap U_t}.
821: $$
822: Then we consider its characteristic cycle
823: $Ch({\cal F}_{\text{principal}, t})$.
824: When $t=90^{\circ}$,
825: $Ch({\cal F}_{\text{principal}, t}) = T^*_{S^1} \BB CP^1 - T^*_N \BB CP^1$.
826: When $t \in (-90^{\circ},90^{\circ})$,
827: we use the open embedding theorem of W.~Schmid and K.~Vilonen (\cite{SchV1})
828: once again to see that
829: \begin{multline*}
830: Ch({\cal F}_{\text{principal}, t})
831: = Ch((j_{S^1 \cap U_t \hookrightarrow \BB CP^1})_! \BB C_{S^1 \cap U_t}) \\
832: = T^*_{S^1 \cap U_t} \BB CP^1 \bigcup
833: \{ \zeta \in T^*_x \BB CP^1;\, x \in \partial (S^1 \cap U_t), \,
834: \langle \zeta, v \rangle \ge 0 \\
835: \text{ for all tangent vectors $v \in T_x S^1$ pointing outside of $U_t$} \}.
836: \end{multline*}
837:
838: \begin{figure}
839: \centerline{\psfig{figure=AMSfig6.eps}}
840: \caption{
841: $Ch \bigl( (j_{\{N\} \hookrightarrow \BB CP^1})_! \BB C_N \bigr)$
842: and $Ch({\cal F}_{\text{principal}, t})$.}
843: \end{figure}
844:
845: \begin{figure}
846: \centerline{\psfig{figure=AMSfig7.eps}}
847: \caption{$Ch ({\cal F}_{\text{principal}})$ is deformed into
848: $T^*_N \BB CP^1 + T^*_S \BB CP^1$.}
849: \end{figure}
850:
851: We see that, as we decrease $t$ from $90^{\circ}$ to $0^{\circ}$,
852: we continuously deform the cycle
853: $$
854: Ch \bigl( (j_{S^1 \cap U_{90^{\circ}} \hookrightarrow \BB CP^1})_!
855: \BB C_{S^1 \cap U_{90^{\circ}}} \bigr).
856: $$
857: The end result of this deformation
858: is the cotangent space at the south pole $S$, $T^*_S \BB CP^1$.
859: Moreover, as before, it is true that during the
860: process of deformation we always stayed inside the set
861: $$
862: \{\zeta \in T^* \BB CP^1;\, Re( \langle g,\mu(\zeta) \rangle) =
863: \langle \operatorname{VF}_g, \zeta \rangle \le 0\}.
864: $$
865: In other words, $Ch({\cal F}_{\text{principal}}) -
866: (T^*_N \BB CP^1 + T^*_S \BB CP^1)$ is the
867: boundary of a chain which lies entirely inside
868: $$
869: \{\zeta \in T^* \BB CP^1;\, Re( \langle g,\mu(\zeta) \rangle) \le 0\}.
870: $$
871:
872: \separate
873:
874: Let me remind once again what we have achieved so far.
875: We started with a cycle $\g{sl}(2,\BB R)' \times Ch({\cal F})$
876: inside $\g{sl}(2,\BB R)' \times T^* \BB CP^1$ and deformed it
877: into a new cycle. Let us call it $C_{\cal F}$.
878: The new cycle $C_{\cal F}$ is characterized by the following two properties:
879: First of all, for each regular semisimple $g \in \g{sl}(2,\BB R)'$,
880: $C_{\cal F}$ intersects $\{ g \} \times T^*\BB CP^1$ transversally and
881: $$
882: C_{\cal F} \cap (\{ g \} \times T^*\BB CP^1)
883: = m_1 T^*_{x_1} \BB CP^1 + m_2 T^*_{x_2} \BB CP^1,
884: $$
885: where $m_1,m_2$ are some integers,
886: $x_1,x_2$ are the points in $\BB CP^1$ fixed by $g$,
887: and each cotangent space $T^*_{x_k} \BB CP^1$ is given some orientation.
888: Secondly, the deformation of the cycle
889: $\g{sl}(2,\BB R)' \times Ch({\cal F})$ into $C_{\cal F}$ takes place
890: inside the set
891: $$
892: \{(g,\zeta) \in \g{sl}(2,\BB R)' \times T^* \BB CP^1;\,
893: Re( \langle g,\mu(\zeta) \rangle) \le 0\}.
894: $$
895: These two properties determine the cycle $C_{\cal F}$ uniquely.
896:
897: \separate
898:
899: The way this deformation procedure is done in general, it becomes
900: very similar to the classical Morse's lemma which says
901: that if we have a smooth real valued function $f$ on a manifold
902: $M$, then the sublevel sets $\{ m \in M;\,f(m) < a \}$ and
903: $\{ m \in M;\,f(m) < b \}$ can be deformed one into the other
904: as long as there are no critical values of $f$ in an open
905: interval containing $a$ and $b$.
906:
907: \separate
908:
909: After deforming the cycle
910: $\g{sl}(2,\BB R)' \times Ch({\cal F})$ into $C_{\cal F}$
911: we encounter the following obstacle: The integrand
912: $e^{\langle g, \mu_{\lambda}(\zeta) \rangle}
913: \phi(-\sigma+\pi^* \tau_{\lambda})$
914: when restricted to $C_{\cal F}$ becomes zero!
915: There is no contradiction to Stokes' theorem because
916: the cycles in question are not compact.
917: In this situation we are simply not allowed to interchange pointwise limits
918: and integration. See Remark \ref{zero_remark}.
919:
920: In order to deal with this problem, for each regular semisimple
921: $g \in \g{sl}(2,\BB R)'$, we define another deformation
922: $\Theta_t(g): T^*X \to T^*X$, $t \in [0,1]$.
923: Its description will be given right after the equation (\ref{RR}).
924: These $\Theta_t(g)$'s combine into a deformation
925: $$
926: \Theta_t: \g{sl}(2,\BB R)' \times T^*X \to \g{sl}(2,\BB R)' \times T^*X,
927: \qquad t \in [0,1].
928: $$
929: The idea of this deformation was inspired by the classical proof of
930: the Fourier inversion formula
931: $$
932: \phi(g)= \frac 1{(2\pi i)^{\dim_{\BB C}\g g}}
933: \int_{\zeta \in i \g g_{\BB R}^*}
934: \hat \phi(\zeta) e^{-\langle g,\zeta \rangle}
935: $$
936: where we multiply the integrand by a term like $e^{-t \|\zeta\|^2}$
937: to make it integrable over $\g g_{\BB R} \times i \g g_{\BB R}^*$,
938: and then let $t \to 0^+$.
939: The deformation $\Theta_t$ has a very similar effect --
940: it makes our integrand an $L^1$-object.
941: The following result is Lemma 17 in \cite{L}:
942:
943: \begin{lm} \label{slanting}
944: For any $t \in [0,1]$, we have:
945: \begin{multline*}
946: \lim_{R \to \infty}
947: \int_{\g{sl}(2,\BB R)' \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\})}
948: \bigl( e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
949: (-\sigma + \pi^* \tau_{\lambda}) \\
950: - \Theta_t^*(e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
951: (-\sigma + \pi^* \tau_{\lambda})) \bigr) =0.
952: \end{multline*}
953: \end{lm}
954:
955: It essentially says that it is permissible to substitute
956: $$
957: \Theta_t^*(e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
958: (-\sigma + \pi^* \tau_{\lambda}))
959: $$
960: in place of the original integrand
961: $e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
962: (-\sigma + \pi^* \tau_{\lambda})$.
963:
964: Proof of Lemma \ref{slanting} is very technical, but the idea is quite simple.
965: The difference between the original integral and the deformed
966: one is expressed by an integral of
967: $e^{\langle g, \mu_{\lambda}(\zeta) \rangle}
968: \phi(-\sigma+\pi^* \tau_{\lambda})$ over a certain cycle
969: $\tilde C(R)$ supported in
970: $\g{sl}(2,\BB R)' \times (T^*\BB CP^1 \cap \{\|\zeta\| = R\})$
971: which depends on $R$ by scaling along the fiber.
972: Recall that the Fourier transform $\hat \phi$ decays rapidly
973: in the imaginary directions which is shown by an integration
974: by parts. We modify this integration by parts argument to prove
975: a similar statement about behavior of the integrand on
976: the support of $\tilde C(R)$ as $R \to \infty$.
977: Hence the difference of integrals in question tends to zero.
978:
979: \separate
980:
981: Let $C$ be the Borel-Moore chain in $\g{sl}(2,\BB R)' \times T^*\BB CP^1$
982: of dimension 6 which lies inside the set
983: $$
984: \{(g,\zeta) \in \g{sl}(2,\BB R)' \times T^* \BB CP^1;\,
985: Re( \langle g,\mu(\zeta) \rangle) \le 0\}
986: $$
987: and such that
988: $$
989: \partial C = \g{sl}(2,\BB R)' \times Ch({\cal F}) - C_{\cal F}.
990: $$
991: Take an $R \ge 1$ and restrict all cycles to the set
992: $$
993: \{ (g,\zeta) \in \g{sl}(2,\BB R)' \times T^*\BB CP^1;\, \|\zeta\| \le R\}.
994: $$
995: Then the restricted chain $C$ has boundary
996: \begin{multline*}
997: \partial (C \cap \{\|\zeta\| \le R\}) =
998: \g{sl}(2,\BB R)' \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\}) \\
999: - (C_{\cal F} \cap \{\|\zeta\| \le R\}) - C'(R),
1000: \end{multline*}
1001: where $C'(R)$ is a 5-chain supported in the set
1002: $$
1003: \{ (g, \zeta) \in \g{sl}(2,\BB R)' \times T^*X;\, \|\zeta\|=R,\,
1004: Re( \langle g, \mu(\zeta) \rangle ) \le 0 \}.
1005: $$
1006: Because our chain $C$ is conic, the piece of boundary $C'(R)$
1007: depends on $R$ by appropriate scaling of $C'(1)$ in the fiber direction.
1008:
1009: The following result is Lemma 18 in \cite{L}:
1010:
1011: \begin{lm} \label{zero}
1012: For a fixed $t \in (0,1]$,
1013: $$
1014: \lim_{R \to \infty} \int_{C'(R)}
1015: \Theta_t^* \bigl( e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1016: (-\sigma + \pi^* \tau_{\lambda})^n \bigr) =0.
1017: $$
1018: \end{lm}
1019:
1020: \separate
1021:
1022: We start with the right hand side of the
1023: integral character formula (\ref{intformula}) and
1024: continuing calculations (\ref{R}) using Lemmas \ref{slanting} and \ref{zero}
1025: we obtain:
1026: \begin{multline} \label{RR}
1027: \frac 1{2\pi i} \int_{Ch({\cal F})}
1028: \mu_{\lambda}^* \hat \phi (-\sigma+\pi^* \tau_{\lambda}) \\
1029: =\lim_{R \to \infty} \frac 1{2\pi i}
1030: \int_{\g{sl}(2,\BB R)' \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\})}
1031: e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1032: (-\sigma+\pi^* \tau_{\lambda}) \\
1033: =\lim_{R \to \infty} \frac 1{2\pi i}
1034: \int_{\g{sl}(2,\BB R)' \times (Ch({\cal F}) \cap \{\|\zeta\| \le R\})}
1035: \Theta_t^*(e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1036: (-\sigma + \pi^* \tau_{\lambda})) \\
1037: =\lim_{R \to \infty} \frac 1{2\pi i}
1038: \int_{(C_{\cal F} \cap \{\|\zeta\| \le R\}) + C'(R)}
1039: \Theta_t^*(e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1040: (-\sigma + \pi^* \tau_{\lambda})) \\
1041: =\lim_{R \to \infty} \frac 1{2\pi i}
1042: \int_{C_{\cal F} \cap \{\|\zeta\| \le R\}}
1043: \Theta_t^*(e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1044: (-\sigma + \pi^* \tau_{\lambda})) \\
1045: = \frac 1{2\pi i} \int_{C_{\cal F}}
1046: \Theta_t^* \bigl( e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1047: (-\sigma + \pi^* \tau_{\lambda}) \bigr).
1048: \end{multline}
1049: It turns out that the last integral is pretty easy to calculate.
1050:
1051: \separate
1052: As before,
1053: $g \in \g{sl}(2,\BB R)'$ is a regular semisimple element, and
1054: $x_1,x_2 \in \BB CP^1$ are the zeroes of the vector field
1055: $\operatorname{VF}_g$ on $\BB CP^1$.
1056: For $k =1,2$, we let $\g b_k \subset \g g$ be the Borel subalgebra
1057: of $\g g$ consisting of all elements in $\g g$ fixing $x_k$.
1058: Let $\g n_k \subset \g g$ be the root
1059: spaces with respect to the unique Cartan algebra $\g t \subset \g g$
1060: containing $g$ such that $\g g = \g b_k \oplus \g n_k$ as linear spaces.
1061: Define maps $\psi_k: \g n_k \to \BB CP^1$, $n \mapsto \exp(n) \cdot x_k$.
1062: Each $\psi_k$ is a diffeomorphism onto its image and their images
1063: cover all of $\BB CP^1$, i.e. $\{\psi_1,\psi_2\}$ is an atlas of $\BB CP^1$.
1064: Let $z_k: \g n_k \tilde \to \BB C$ be a linear coordinate on $\g n_k$.
1065: Then it is not hard to show that the vector field $\operatorname{VF}_g$
1066: can be expressed in these coordinates by
1067: $$
1068: \alpha_{x_k}(g) z_k \frac{\partial}{\partial z_k},
1069: $$
1070: where $\alpha_k \in \g t^*$ denotes the root of $\g{sl}(2,\BB C)$
1071: corresponding to the root space $\g n_k$.
1072:
1073: We expand $z_k$ to a standard coordinate system $z_k, \xi_k$ on
1074: $T^* \g n_k$ so that every element of
1075: $T^* \g n_k \simeq \g n_k \times \g n_k^*$ is expressed in these
1076: coordinates as $(z_k, \xi_k dz_k)$. This gives us two charts
1077: $\tilde \psi_k : (z_k,\xi_k) \to T^*\BB CP^1$ and an atlas
1078: $\{\tilde \psi_1, \tilde \psi_2 \}$ of $T^*\BB CP^1$.
1079:
1080: The diffeomorphism $\Theta_t(g): T^*\BB CP^1 \to T^*\BB CP^1$ is defined
1081: using the atlas $\{\tilde \psi_1, \tilde \psi_2 \}$ and smooth cutoff
1082: functions. It is designed so that it approximates the maps
1083: \begin{equation} \label{Theta}
1084: (z_k, \xi_k) \mapsto
1085: \bigl( z_k - t \frac{\bar \alpha_k(g)}{|\alpha_k(g)|} \bar \xi_k,\xi_k \bigr).
1086: \end{equation}
1087: Of course, these maps cannot be defined globally on $T^*\BB CP^1$, but for
1088: the purposes of calculating the integral (\ref{RR}) we are allowed
1089: to pretend that, in each chart $\tilde \psi_k$, $\Theta_t(g)$ is given
1090: by the equation (\ref{Theta}).
1091:
1092: Next we calculate the restriction of the integrand of (\ref{RR})
1093: $$
1094: \Theta_t(g)^* \bigl( e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1095: (-\sigma + \pi^* \tau_{\lambda}) \bigr)
1096: $$
1097: to $T^*_{x_k} \BB CP^1$:
1098: $$
1099: \Theta_t(g)^*
1100: \bigl( \langle g, \mu_{\lambda}(\zeta) \rangle \bigr) |_{T^*_{x_k} \BB CP^1}
1101: = - t |\alpha_k(g)| \cdot |\xi_k|^2
1102: + \Theta_t(g)^* \langle g, \lambda_x \rangle,
1103: $$
1104: $$
1105: \Theta_t(g)^* \sigma |_{T^*_{x_k} \BB CP^1}
1106: = t \frac{\bar \alpha_k(g)}{|\alpha_k(g)|} d\bar \xi_k \wedge d\xi_k,
1107: $$
1108: and $\Theta_t(g)^* (\pi^* \tau_{\lambda})$ on $T^*_{x_k} \BB CP^1$
1109: can be expressed as $t^2$ times a bounded $2$-form.
1110: Putting it all together, we obtain
1111: \begin{multline*}
1112: \Theta_t(g)^* \bigl( e^{\langle g, \mu_{\lambda}(\zeta) \rangle} \phi
1113: (-\sigma + \pi^* \tau_{\lambda}) \bigr) |_{T^*_{x_k} \BB CP^1} \\
1114: = e^{- t |\alpha_k(g)| \cdot |\xi_k|^2
1115: + \Theta_t(g)^* \langle g, \lambda_x \rangle}
1116: \phi(g) \wedge
1117: \Bigl( t \frac{\bar \alpha_k(g)}{|\alpha_k(g)|} d\xi_k \wedge d\bar\xi_k
1118: + t^2 \cdot (\text{bounded term}) \Bigr).
1119: \end{multline*}
1120: One can argue using a coordinate change and the Lebesgue dominated
1121: convergence theorem that, as $t \to 0^+$, we can replace the integrand with
1122: \begin{equation} \label{integrand_t}
1123: t \frac{\bar \alpha_k(g)}{|\alpha_k(g)|}
1124: e^{\langle g, \lambda_{x_k} \rangle}
1125: e^{- t |\alpha_k(g)| \cdot |\xi_k|^2}
1126: \phi(g) \wedge d\xi_k \wedge d\bar\xi_k,
1127: \end{equation}
1128: which integrates over $T^*_{x_k} \BB CP^1$ to
1129: $$
1130: (2\pi i) \frac {e^{\langle g, \lambda_{x_k} \rangle}}{\alpha_k(g)} \phi(g).
1131: $$
1132: The last expression may appear to be missing a negative sign, but it is
1133: correct because we are using the convention (11.1.2) in \cite{KaScha}
1134: to identify the real cotangent bundle $T^*(X^{\BB R})$ with the
1135: holomorphic cotangent bundle $T^*X$ of the complex manifold $X$.
1136: Under this identification, the standard symplectic
1137: structure on $T^*(X^{\BB R})$ equals $2 Re (\sigma)$.
1138:
1139: Substituting into (\ref{RR}) we obtain
1140: $$
1141: \frac 1{2\pi i} \int_{Ch({\cal F})}
1142: \mu_{\lambda}^* \hat \phi (-\sigma+\pi^* \tau_{\lambda})
1143: = \int_{\g g_{\BB R}} \sum_{k=1}^2 m_k(g)
1144: \frac {e^{\langle g, \lambda_{x_k(g)} \rangle}}{\alpha_{k(g)}(g)} \phi(g)
1145: = \int_{\g g_{\BB R}} F_{\theta} \phi,
1146: $$
1147: which says that the right hand side of the integral character formula
1148: (\ref{intformula}) is equal to
1149: the right hand side of the fixed point character formula (\ref{fpformula}).
1150:
1151: \begin{rem} \label{zero_remark} {\em
1152: Notice that, as $t \to 0^+$, the integrand (\ref{integrand_t}) tends to zero
1153: pointwise, but its integral does not.
1154: We had to introduce the deformation $\Theta_t(g) : T^* X \to T^*X$ so that
1155: we could interchange pointwise limits and integrals and resolve the
1156: situation when integrand (\ref{integrand}) restricted to the cycle
1157: $C_{\cal F}$ is zero, but the integral does not tend to zero.
1158: }\end{rem}
1159:
1160: \separate
1161:
1162: The key ideas are the deformation of $Ch({\cal F})$,
1163: the definition of $\Theta_t(g): T^*X \to T^*X$ and
1164: Lemma \ref{slanting}. Because of the right definition
1165: of $\Theta_t(g)$, Lemma \ref{slanting} holds and our calculation
1166: of the integral becomes very simple.
1167: We see that, as $R \to \infty$,
1168: the integral concentrates more and more inside $T^*U$,
1169: where $U$ is a neighborhood of the set of fixed points of $g$ in $X$.
1170: In the limit, we obtain the right hand side of the
1171: fixed point character formula.
1172: This means that the integral is localized at the fixed points of $g$.
1173:
1174: \separate
1175:
1176: As was stated before, complete proofs with all technical details
1177: are available in \cite{L}.
1178:
1179:
1180:
1181: \separate
1182:
1183: %\newpage
1184:
1185: \begin{thebibliography}{[KaSchm]}
1186: %\bibitem[BB]{BB} A. Beilinson and J. Bernstein,
1187: %{\em Localisation de $\g g$-modules}, C. R. Acad. Sci. Paris {\bf 292}
1188: %(1981), 15-18.
1189: \bibitem[BGV]{BGV} N. Berline, E. Getzler, M. Vergne,
1190: {\em Heat Kernels and Dirac Operators},
1191: Springer-Verlag, 1992.
1192: %\bibitem[Bo]{Bo} A. Borel et al., {\em Algebraic {\D}-modules},
1193: %Perspectives in Mathematics, Academic Press, 1987.
1194: %\bibitem[Fo]{Fo} G. B. Folland,
1195: %{\em Real Analysis: Modern Techniques and Their Applications, Second edition},
1196: %Pure and Applied Mathematics, John Wiley \& Sons, Inc., New York, 1999.
1197: \bibitem[DM]{DM} L. van den Dries and C. Miller,
1198: {\em Geometric categories and o-minimal structures},
1199: Duke Math. Jour. {\bf 84} (1996), 497-540.
1200: \bibitem[GM]{GM} M. Goresky and R. MacPherson,
1201: {\em Local contribution to the Lefschetz fixed point formula},
1202: Inventiones Math. {\bf 111} (1993), 1-33.
1203: %\bibitem[GM]{GM} M. Goresky and R. MacPherson,
1204: %{\em Stratified Morse Theory}, Ergebnisse der Mathematik,
1205: %vol 14, Springer-Verlag, Heidelberg, 1989.
1206: %\bibitem[Ka]{Ka} M. Kashiwara, {\em Index theorem for constructible sheaves},
1207: %Ast\'erisque {\bf 130} (1985), 193-209.
1208: \bibitem[GS]{GS} V. Guillemin and S. Sternberg,
1209: {\em Supersymmetry and Equivariant de Rham Theory},
1210: Springer-Verlag, 1999.
1211: \bibitem[K]{K} M. Kashiwara, {\em Character, character cycle,
1212: fixed point theorem, and group representations},
1213: Advanced Studies in Pure Mathematics, vol. 14, Kinokuniya, Tokyo,
1214: 1988, pp. 369-378.
1215: \bibitem[KaScha]{KaScha} M. Kashiwara and P. Schapira,
1216: {\em Sheaves on Manifolds}, Springer, 1990.
1217: \bibitem[KSch]{KSch} M. Kashiwara and W. Schmid,
1218: {\em Quasi-equivariant ${\cal D}$-modules, equivariant derived category, and
1219: representations of reductive Lie groups},
1220: Lie Theory and Geometry, in Honor of Bertram Kostant,
1221: Progress in Mathematics, vol. 123, Birkh\"auser, Boston, 1994, pp. 457-488.
1222: \bibitem[L]{L} M. Libine,
1223: {\em A Localization Argument for Characters of Reductive Lie Groups},
1224: math.RT/0206019, 2002, to appear in the Jour. of Functional Analysis.
1225: \bibitem[R]{R} W. Rossmann, {\em Invariant Eigendistributions on a
1226: Semisimple Lie Algebra and Homology Classes on the Conormal Variety I, II},
1227: Jour. Func. Anal. {\bf 96} (1991), 130-193.
1228: %\bibitem[Ru]{Ru} W. Rudin, {\em Functional Analysis},
1229: %International Series in Pure and Applied Mathematics, McGraw-Hill, 1991.
1230: \bibitem[Sch]{Sch} W. Schmid,
1231: {\em Character formulas and localization of integrals},
1232: Deformation Theory and Symplectic Geometry, Mathematical Physics Studies,
1233: {\bf 20} (1997), Kluwer Academic Publishers, 259-270.
1234: \bibitem[SchV1]{SchV1} W. Schmid and K. Vilonen,
1235: {\em Characteristic cycles of constructible sheaves},
1236: Inventiones Math. {\bf 124} (1996), 451-502.
1237: \bibitem[SchV2]{SchV2} W. Schmid and K. Vilonen,
1238: {\em Two geometric character formulas for reductive Lie groups},
1239: Jour. AMS {\bf 11} (1998), 799-876.
1240: \bibitem[SchV3]{SchV3} W. Schmid and K. Vilonen,
1241: {\em Characteristic cycles and wave front cycles of representations of
1242: reductive groups}, Annals of Math. (2), {\bf 151} (2000), 1071-1118.
1243: \end{thebibliography}
1244:
1245: \noindent
1246: {matvei@math.umass.edu}
1247:
1248: \noindent
1249: {\em Department of Mathematics and Statistics, University of Massachusetts,
1250: Lederle Graduate Research Tower, 710 North Pleasant Street, Amherst,
1251: MA 01003}
1252:
1253: \enddocument
1254: