1: \documentstyle{article}
2: \input amssymb.sty
3: %\input amssym.def
4: %\input amssym
5: \input epsf
6:
7: \newtheorem{theorem}{Theorem}[section]
8: \newtheorem{lemma}[theorem]{Lemma}
9: \newtheorem{proposition}[theorem]{Proposition}
10: \newtheorem{corollary}[theorem]{Corollary}
11: \newtheorem{conjecture}[theorem]{Conjecture}
12: \newtheorem{definition}[theorem]{Definition}
13: \newtheorem{remark}[theorem]{Remark}
14: \newtheorem{theorem-construction}[theorem]{Theorem--Construction}
15: \newtheorem{lemma-construction}[theorem]{Lemma--Construction}
16:
17: \begin{document}
18:
19: \newcommand{\Z}{{\Bbb Z}}
20: \newcommand{\R}{{\Bbb R}}
21: \newcommand{\Q}{{\Bbb Q}}
22: \newcommand{\C}{{\Bbb C}}
23: \newcommand{\lra}{\longrightarrow}
24: \newcommand{\lms}{\longmapsto}
25: \newcommand{\AAA}{{\Bbb A}}
26: \newcommand{\Alt}{{\rm Alt}}
27: \newcommand{\wg}{\wedge}
28: \newcommand{\ol}{\overline}
29: \newcommand{\CP}{{\Bbb C}P}
30: \newcommand{\bwg}{\bigwedge}
31: \newcommand{\caL}{{\cal L}}
32: \newcommand{\PP}{{\Bbb P}}
33: \newcommand{\HH}{{\Bbb H}}
34: \newcommand{\LL}{{\Bbb L}}
35:
36:
37: \begin{titlepage}
38: \title{Polylogarithms, regulators, and Arakelov motivic complexes}
39: \author{A. B. Goncharov}
40: \date{}
41: \end{titlepage}
42: \stepcounter{page}
43: \maketitle
44: \tableofcontents
45:
46: \section {Introduction}
47:
48: {\bf Abstract}. We construct an explicit regulator map
49: from the weight $n$ Bloch Higher Chow group complex to the weight $n$
50: Deligne complex
51: of a regular projective complex algebraic variety $X$.
52: We define the weight $n$
53: Arakelov motivic complex
54: as the cone of this map
55: shifted by one.
56: Its last cohomology group is (a version of) the
57: Arakelov Chow group defined by H. Gillet and C. Soul\'e ([GS]).
58:
59: We
60: relate the
61: Grassmannian $n$--logarithms (defined as in [G5])
62: to the geometry of the symmetric space
63: $SL_n(\C)/SU(n)$. For $n=2$ we recover
64: Lobachevsky's formula expressing the volume of an ideal geodesic
65: simplex in the hyperbolic space via the dilogarithm.
66: Using the relationship with symmetric spaces
67: we construct the Borel regulator
68: on $K_{2n-1}(\C)$ via the
69: Grassmannian $n$--logarithms.
70:
71: We study the Chow dilogarithm and prove a
72: reciprocity law
73: which strengthens Suslin's reciprocity law for Milnor's group $K^M_3$ on curves.
74:
75: Our note [G5] can serve as an introduction to this paper.
76:
77: {\bf 1. Beilinson's conjectures on special values of L-functions}.
78: Let $X$ be a regular scheme.
79: A.A. Beilinson [B1] defined the rational motivic cohomology
80: of $X$ via Quillen's
81: algebraic $K$-theory of $X$ by the following formula:
82: \begin{equation} \label{4.29.02.1}
83: H^i_{\cal M}(X, \Q(n)) = K^{(n)}_{2n-i}(X)_{\Q}
84: \end{equation}
85: where on the right stays the weight $n$ eigenspace for the Adams operations
86: acting on $K_{2n-i}(X)\otimes \Q$.
87: For a regular complex algebraic variety $X$ Beilinson defined the regulator map
88: to the weight $n$ Deligne cohomology of
89: $X(\C)$:
90: $$
91: r_B: H_{{\cal M}}^{i}(X, \Q(n)) \lra H_{{\cal D}}^{i}(X(\C), \R(n))
92: $$
93:
94:
95: Let $X$ be a regular projective scheme over
96: $\Q$. Let $L(h^i(X),s)$ be the $L$-function related to its i-dimensional
97: cohomology.
98: Beilinson conjectured that for any integer $n > 1+i/2$
99: its special value at $s=n$ is described, up to a nonzero rational factor,
100: by the regulator map to the weight $n$ {\it real} Deligne cohomology of
101: $X$
102: $$
103: r_B: H_{{\cal M}}^{i+1}(X, \Q(n))_{\Z} \lra H_{{\cal D}}^{i+1}(X\otimes_{\Q}\R_{/\R}, \R(n))
104: $$
105: Here the subscript $\Z$ on the left indicates the subspace of $H_{{\cal M}}^{i+1}(X, \Q(n))$
106: coming from a regular model of the scheme $X$ over $\Z$, see [B1] and [RSS] for details.
107:
108:
109:
110: This conjecture is fully established only when $X= {\rm Spec}(F)$ where
111: $F$ is a number field. In this case the regulator map $r_B$ coincides, up to a
112: non-zero rational factor,
113: with
114: the Borel regulator ([B1]),
115: and the relation with special values of the Dedekind zeta-function of $F$ was given
116: by the Borel theorem [Bo].
117:
118: Although Beilinson's conjectures are far from being
119: proved, it is interesting to see what kind of information about the
120: special values of $L$-functions they
121: suggest. So we come to the problem of {\it explicit} calculation of
122: Beilinson's regulator. This problem is already very interesting for the
123: Borel regulator.
124:
125:
126:
127: {\bf 2. Regulator maps on motivic complexes and Arakelov motivic cohomology}.
128: Beilinson [B2] and S. Lichtenbaum [L1] conjectured that the weight $n$
129: integral
130: motivic cohomology
131: of $X$ should appear as the cohomology of certain complexes of abelian groups
132: $\Gamma(X;n)$, called the weight $n$ motivic complexes:
133: $$
134: H^i_{\cal M}(X, \Z(n)):= H^i(\Gamma(X;n))
135: $$
136: These complexes are well defined as objects of the
137: derived category. They
138: should appear as the hypercohomology of certain complexes of Zarisky
139: sheaves.
140:
141: The first motivic complexes satisfying
142: Beilinson's formula (\ref{4.29.02.1}) were Bloch's Higher Chow group complexes
143: ${\cal Z}^{\bullet}(X; n)$ [Bl1]. Later on A.A. Suslin and V.A. Voevodsky
144: defined several important versions of these complexes.
145: For another candidates for motivic complexes, called the polylogarithmic motivic
146: complexes, see [G1-2]. They are very explicit and the smallest among
147: possible candidates, however Beilinson's formula (\ref{4.29.02.1})
148: is far from being established for them.
149:
150:
151: The real Deligne cohomology arises also as the cohomology
152: of certain complexes.
153: It was suggested in [G5] and [G7] that the regulator map should be {\it explicitly}
154: defined on the level of complexes.
155:
156: Let $X$ be a regular projective variety over $\C$.
157: In Chapter 2 we construct
158: a homomorphism of complexes
159: \begin{equation} \label{4.29.02.2}
160: \mbox{Bloch's weight $n$ Higher Chow group complex ${\cal Z}^ {\bullet}(X; n)$ of $X$}
161: \quad \lra
162: \end{equation}
163: $$
164: \mbox{the weight $n$ real Deligne complex
165: ${\cal C}^{\bullet}_{\cal D}(X(\C); n)$ of $X$}
166: $$
167: This construction is a version of the one given in [G5].
168: The complex ${\cal C}^{\bullet}_{\cal D}(X(\C); n)$ is the
169: truncation $\tau_{\leq 2n}$ of the complex
170: proposed by
171: Deligne [Del]. The
172: weight $n$ Arakelov motivic complex $\Gamma_{\cal A}^{\bullet}(X; n)$ is the cone of
173: the map (\ref{4.29.02.2}), shifted by $-1$:
174: \begin{equation} \label{6.11.02.14}
175: \Gamma_{\cal A}^{\bullet}(X; n):= {\rm Cone}\Bigl({\cal Z}^ {\bullet}(X; n)
176: \stackrel{(\ref{4.29.02.2})}{\lra}
177: {\cal C}^{\bullet}_{\cal D}(X(\C); n)\Bigr)[-1]
178: \end{equation}
179: For a regular projective variety $X$ over $\R$ the image of map
180: (\ref{4.29.02.2}) lies in the subcomplex
181: $$
182: {\cal C}^{\bullet}_{\cal D}(X_{/\R}; n) :=
183: {\cal C}^{\bullet}_{\cal D}(X(\C); n)^{\overline F_{\infty}}
184: $$
185: where $\overline F_{\infty}$ is the De Rham involution provided by the action of
186: complex conjugation.
187: The weight $n$
188: {\it real} Arakelov motivic complex
189: is defined as
190: \begin{equation} \label{6.11.02.4}
191: \Gamma_{\cal A}^{\bullet}(X_{/\R}; n):= {\rm Cone}\Bigl({\cal Z}^ {\bullet}(X; n)
192: \stackrel{(\ref{4.29.02.2})}{\lra}
193: {\cal C}^{\bullet}_{\cal D}(X_{/\R}; n)\Bigr)[-1]
194: \end{equation}
195: Let $X$ be a regular projective variety $X$ over a number field $F$. We view $X$ over $\Q$:
196: $X \lra {\rm Spec}(F) \lra {\rm Spec}(\Q)$, and set
197: \begin{equation} \label{6.11.02.4q}
198: \Gamma_{\cal A}^{\bullet}(X_{/F}; n):= {\rm Cone}\Bigl({\cal Z}^ {\bullet}(X; n)
199: \stackrel{}{\lra}
200: {\cal C}^{\bullet}_{\cal D}(X\otimes_{\Q} \R_{/\R}; n)\Bigr)[-1]
201: \end{equation}
202: The weight $n$ Arakelov motivic cohomology is the cohomology of this complex.
203: Our construction
204: works equally well for the Suslin-Voevodsky versions of the
205: motivic complexes.
206:
207: Taking the cohomology we get a construction of the
208: regulator map on motivic cohomology.
209: For a different construction see [Bl3].
210:
211:
212: The regulator map on the polylogarithmic motivic complexes
213: was defined in [G7] explicitly via the classical polylogarithms.
214: The Arakelov motivic complexes constructed using regulator maps
215: on different motivic complexes are supposed to lead to
216: the same object of the derived category.
217: However a precise relationship between the construction
218: given in [G7] and the one
219: in Chapter 2 is not clear.
220:
221:
222:
223: {\it Higher Arakelov Chow groups}. The last group of the
224: complex ${\cal C}^{\bullet}_{\cal D}(X(\C); n)$ consists of
225: closed distributions of a certain type on $X(\C)$. Replacing it
226: by the quotient modulo smooth closed
227: forms of the same type we get the quotient complex
228: $\widetilde {\cal C}^{\bullet}_{\cal D}(X(\C); n)$.
229: Changing ${\cal C}$ to
230: $\widetilde {\cal C}$ in (\ref{6.11.02.14})
231: we define the weight $n$
232: Higher Arakelov Chow group complex. Its last cohomology group is isomorphic to
233: the Arakelov Chow group $\widehat {CH}^n(X(\C))$ as defined by Gillet and Soul\'e
234: [GS], [S].
235:
236: {\bf Problems}. a) Show that taking
237: cohomology of
238: the map (\ref{4.29.02.2}) and using the isomorphism between
239: the rational Bloch's Higher Chow groups of $X$ and the corresponding part of the rational
240: $K$-theory of $X$ ([Bl2], [Lev])
241: we get a non-zero rational multiple of the Beilinson's regulator map.
242:
243: b) To generalize the arithmetic Riemann--Roch theorem
244: proved by Gillet and Soul\'e to the case of Higher Arakelov Chow groups.
245:
246: {\bf Remark}. The weight $n$ Arakelov motivic complex should be considered as an ingrediant
247: of a definition of the weight $n$ {\it arithmetic motivic complex}. The latter
248: is related to the regulator maps on
249: $H_{{\cal M}}^{\bullet}(X, \Q(n))_{\Z}$, while the former is related
250: to the ones on $H_{{\cal M}}^{\bullet}(X, \Q(n))$.
251: Ideally one should
252: have for every place $p$ of $\Q$ a map from the left hand side of (\ref{4.29.02.2})
253: to a certain complex, which for the Archimedian place should be given by our map.
254: Then one should take the shifted by $-1$ cone of the sum of these maps.
255:
256: {\bf 3. The Chow $n$--logarithm function}.
257: Let us describe the regulator map (\ref{4.29.02.2})
258: in the simplest case when $X = {\rm Spec}(\C)$ is a point.
259:
260: Let us choose in $\PP^m$ homogeneous coordinates $(z_0: ... : z_m)$.
261: The union of the coordinate hyperplanes is a simplex $L$.
262: Let $\AAA^m$ be the complement to
263: the hyperplane $z_1 + ... + z_m = z_0$ in $\PP^m$. The abelian group
264: ${\cal Z}_m({\rm Spec}(\C); n)$ is freely generated by
265: the codimension $n$ irreducible algebraic cycles in $\AAA^{m}$ intersecting properly
266: the faces of the simplex $L$. The intersection with codimension one faces $L_j$ of $L$ provide homomorphisms
267: $$
268: \partial_j: {\cal Z}_m({\rm Spec}(\C); n) \lra {\cal Z}_{m-1}({\rm Spec}(\C); n); \quad
269: \partial:= \sum_{j=0}^{m} (-1)^j\partial_j
270: $$
271: The weight $n$ Higher Chow group complex over ${\rm Spec}(\C)$, where $n>0$,
272: written as a homological complex, looks as follows:
273: $$
274: ... \stackrel{\partial}{\lra} {\cal Z}_2({\rm Spec}(\C); n) \stackrel{\partial}{\lra}
275: {\cal Z}_1({\rm Spec}(\C); n) \stackrel{\partial}{\lra} {\cal Z}_0({\rm Spec}(\C); n)
276: $$
277:
278: The Deligne complex of a point is the complex $(2\pi i)^{n}\R \lra \C$, with $\C$ is
279: in the degree $+1$. So it is quasiisomorphic to the group $\R(n-1):= (2\pi i)^{n-1}\R$ placed in degree $+1$.
280: The
281: regulator map (\ref{4.29.02.2})
282: boils down to a construction of a homomorphism
283: $$
284: {\cal Z}_{2n-1}({\rm Spec}(\C); n) \stackrel{{\cal P}_n}{\lra} \R(n-1), \quad
285: \mbox{such that
286: ${\cal P}_n \circ \partial = 0$}
287: $$
288: It is provided by
289: a function
290: ${\cal P}_n$ on the space of
291: codimension $n$ cycles in $\C\PP^{2n-1}$ intersecting properly
292: faces of a simplex $L$. This function, called
293: the {\it Chow $n$--logarithm function}, was constructed in [G5].
294: To recall its construction, observe that
295: a codimension $n$ cycle given by an irreducible
296: subvariety $X$ in $\PP^{2n-1} -L$ provides the
297: $(n-1)$-dimensional variety $X$ with $2n-1$ rational functions
298: $f_1, ..., f_{2n-1}$: These functions are obtained by
299: restriction of the coordinate functions
300: $z_i/z_0$ to the cycle $X$. We define a natural $(2n-2)$-form
301: $r_{2n-2}(f_1, ..., f_{2n-1})$ on $X(\C)$ and set
302: \begin{equation} \label{6.17.02.1}
303: {\cal P}_n(X; f_1, ..., f_{2n-1}):= (2\pi i)^{1-n}\int_{X(\C)}r_{2n-2}(f_1, ..., f_{2n-1})
304: \end{equation}
305:
306:
307: {\bf 4. An example: the Chow dilogarithm}.
308: Let $f_1,f_2,f_3$ be three arbitrary
309: rational functions on a complex curve $X$.
310: %$$
311: %d r_2(f_1,f_2,f_3) = {\rm Re}\Bigl(d \log f_1 \wedge d \log f_2 \wedge d \log f_3\Bigr)
312: %$$
313: Set
314: $$
315: r_2(f_1,f_2,f_3):=
316: $$
317: $$
318: {\rm Alt}_3 \Bigl(\frac{1}{6}\log|f_1| d \log|f_2|
319: \wedge d
320: \log|f_3|
321: -\frac{1}{2}
322: \log|f_1| d\arg f_2 \wedge d\arg f_3 \Bigr)
323: $$
324: where
325: ${\rm Alt}_3$ is the alternation of $f_{1}, f_{2}, f_{3}$.
326: Consider the space of
327: quadruples $(X;f_1,f_2,f_3)$. It is a union of finite dimensional algebraic varieties.
328: The Chow dilogarithm is a real function
329: on its complex points defined by the formula
330: $$
331: {\cal P}_2(X;f_1,f_2,f_3):= \frac{1}{2\pi i}\int_{X(\C)}r_2(f_1,f_2,f_3)
332: $$
333: The integral converges.
334: The Chow dilogarithm provides a homomorphism
335: \begin{equation} \label{chowh}
336: \Lambda^3 \C(X)^* \to \R,\quad f_1\wedge f_2\wedge f_3 \lms {\cal P}_2(X;f_1,f_2,f_3)
337: \end{equation}
338:
339: Why does the dilogarithm appear in the name of the function ${\cal P}_2$?
340: Recall the classical dilogarithm
341: $$
342: Li_2(z):= - \int_0^z\log(1-z)d\log z
343: $$
344: It has a single-valued cousin, the Bloch-Wigner function:
345: $$
346: {\cal L}_2(z) := {\rm Im} Li_2(z) + \arg (1-z) \log \vert z \vert
347: $$
348:
349: The Chow dilogarithm
350: is defined by a two-dimensional integral
351: over $X(\C)$, while ${\cal L}_2(z)$
352: is given by an
353: integral over a path in ${\C}{\Bbb P}^1$. In Chapter 6 we
354: show that nevertheless the Chow dilogarithm can be expressed by the function ${\cal L}_2(z)$. Here is how it
355: works when $X = \C\PP^1$.
356: For $f \in \C(X)$ let $v_x(f)$ be the order of zero of $f$ at $x \in X(\C)$.
357: Choose a point $\infty$ on ${\Bbb P}^1$. Then
358: \begin{equation} \label{6.14.02.2}
359: {\cal P}_2(\C\PP^1; f_1, f_2, f_3) = \sum_{x_i \in {\Bbb P}^1(\C)}
360: v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3){\cal L}_2(r(x_1, x_2, x_3, \infty))
361: \end{equation}where $r(...)$ denotes the cross-ratio of four points on $\PP^1$.
362: A formula for the Chow dilogarithm on elliptic curves is given
363: in Chapter 6.
364:
365: The function ${\cal L}_2$ satisfies Abel's five term functional equation:
366: \begin{equation} \label{6.14.02.1}
367: \sum_{i=1}^{5}(-1)^i {{\cal L}}_2(r(x_1,...,\widehat x_i,...,x_{5})) =0
368: \end{equation}
369: The Chow dilogarithm also satisfies functional equations. They appear as a reformulation
370: of
371: the fact that the composition
372: $$
373: {\cal Z}_4(Spec(\C); 2) \stackrel{\partial}{\lra} {\cal Z}_3(Spec(\C); 2)
374: \stackrel{{\cal P}_2}{\lra} \R(1)
375: $$
376: is zero. Namely,
377: let $Y$ be an algebraic surface with four rational functions
378: $g_1, ..., g_4$ on it corresponding to an element of ${\cal Z}_4(Spec(\C); 2)$.
379: To evaluate the composition on this element
380: we do the following.
381: Take the divisor ${\rm div} (g_i)$ and
382: restrict the other functions $g_i$
383: to it. Then applying the Chow dilogarithm to the obtained data and
384: taking the alternating sum over $1 \leq i \leq 4$ we get zero.
385: In the special case when $Y = \C\PP^2$ and ${\rm div}g_i = (l_i) - (l_{5})$,
386: where $l_1, ..., l_{5}$ are five lines in the plane, this
387: functional equation plus (\ref{6.14.02.2}) is equivalent to Abel's equation
388: (\ref{6.14.02.1}).
389:
390:
391:
392: {\bf 5. The Grassmannian $n$--logarithm and symmetric space
393: $SL_n(\C)/SU(n)$}.
394: Restricting the Chow $n$-logarithm function to the subvariety of $(n-1)$--planes in
395: $\C\PP^{2n-1}$ in general position with respect to the simplex $L$
396: we get the Grassmannian $n$--logarithm function ${\cal L}^G_n$.
397:
398: Let $G$ be a group and $X$ a $G$-set. {\it Configurations} of $n$ points in $X$ are
399: by definition the points of the quotient $X^n/G$. There is a natural bijection
400: $$
401: \mbox{ $\{ (n-1)$--planes in
402: $\PP^{2n-1}$ in generic position with respect to a simplex $L$\}}/({\Bbb G}_m^*)^{2n-1} \quad
403: $$
404: $$ <--> \quad \left \{ \mbox{Configurations of} \quad 2n \quad\mbox {generic hyperplanes in } \PP^{n-1}\right \}
405: $$
406: given by intersecting of an $(n-1)$--plane $h$ with the codimension one faces of $L$.
407:
408: \begin{figure}[ht]
409: \centerline{\epsfbox{gpol6.eps}}
410: \caption{Toric quotients of Grassmannians and configurations of hyperplanes}
411: \label{gpol6}
412: \end{figure}
413:
414:
415: Using it
416: we can view ${\cal L}^G_n$ as a function on the configurations of $2n$ hyperplanes in $\C \PP^{n-1}$. Applying the projective duality we can consider it as a function
417: on configurations of $2n$ points in $\C \PP^{n-1}$.
418:
419: In fact one can define the Grassmannian $n$-logarithm
420: ${{\cal L}}^G_n(x_1,...,x_{2n})$ as a
421: function
422: on configurations of {\it arbitrary} $2n$ points in $\C\PP^{n-1}$, see
423: Chapter 4.
424: It is a measurable function which is
425: real analytic on generic configurations.
426: It satisfies the two functional equations
427: \begin{equation} \label{6.13.02.100}
428: \sum_{i=0}^{2n}(-1)^i {{\cal L}}^G_n(x_0,...,\widehat x_i,...,x_{2n}) =0, \quad \sum_{j=0}^{2n}(-1)^j {{\cal L}}^G_n(y_j|y_0,...,\widehat y_j,...,y_{2n}) = 0
429: \end{equation}
430: In the second formula $(y_0,...,y_{2n})$ is a configuration
431: of $2n+1$ points in $\C\PP^{n}$ and
432: $(y_j|y_0,...,\widehat y_j,...,y_{2n})$
433: is a configuration of $2n$ points in $\C\PP^{n-1}$
434: obtained by projection from $y_j$.
435:
436: \begin{figure}[ht]
437: \centerline{\epsfbox{gpol20.eps}}
438: \caption{The configuration $(y_1|y_2, y_3, y_4, y_5)$ on $\PP^1$}
439: \label{gpol20}
440: \end{figure}
441:
442:
443:
444: It follows from (\ref{6.14.02.2}) that
445: the Grassmannian dilogarithm is given by the Bloch-Wigner function:
446: \begin{equation} \label{6.13.02.101}
447: {\cal L}^G_2(z_1, ..., z_4) = {\cal L}_2(r(z_1, ..., z_4))
448: \end{equation}
449: Abel's five term equation coincides with
450: (\ref{6.13.02.100}). (The two functional equations (\ref{6.13.02.100})
451: are equivalent when $n=2$).
452:
453: Lobachevsky discovered that the dilogarithm
454: appears in the computation of volumes of geodesic
455: simplices in the three dimensional hyperbolic space ${\cal H}_3$.
456: Let $I(z_1,...,z_4)$ be the ideal geodesic simplex with vertices at the
457: points $z_1,...,z_4$ on the absolute of ${\cal H}_3$.
458: The absolute is naturally identified with
459: $\C\PP^1$.
460: \begin{figure}[ht]
461: \centerline{\epsfbox{gpol21.eps}}
462: \caption{An ideal simplex in the hyperbolic $3$-space}
463: \label{gpol21}
464: \end{figure}
465: Lobachevsky's formula
466: relates its volume to the Bloch-Wigner function:
467: $$
468: {\rm vol}\Bigl(I(z_1,...,z_4)\Bigr) = {\cal L}_2(r(z_1,...,z_4))
469: $$
470: The volume function ${\rm vol}I(z_1,...,z_4)$ is invariant under the group
471: $SL_2(\C)$ of isometries of ${\cal H}_3$. So it
472: depends only on the cross ratio
473: of the points $z_1,...,z_4$. It satisfies the five term equation (\ref{6.14.02.1}).
474: Indeed, $\sum(-1)^iI(z_1,...,
475: \widehat z_i, ..., z_5) = \emptyset$.
476: By Bloch's theorem [Bl2] any measurable function $f(z)$ on $\C$
477: satisfying the five term equation is proportional to ${\cal L}_2(z)$.
478: So we get the formula up to a constant.
479:
480: We generalize this picture as follows.
481: $\C\PP^{n-1}$ is realized as
482: the smallest boundary
483: stratum of the symmetric space ${\Bbb H}_n := SL_n(\C)/SU(n)$. We define a function
484: $\psi_n(x_1, ..., x_{2n})$ on configurations of $2n$ points
485: of the symmetric space. The function $\psi_n$
486: is defined by an integral over $\C\PP^{n-1}$ similar to (\ref{6.17.02.1}).
487: We show that it can be naturally extended to a function $\overline \psi_n$
488: on configurations of $2n$ points in a compactification $\overline {\Bbb H}_n$ of
489: the symmetric space. The Grassmannian $n$--logarithm function
490: turns out to be the value of the function $\overline \psi_n$
491: on configurations of $2n$ points at the smallest boundary strata,
492: which is identified with $\C\PP^{n-1}$.
493:
494:
495: Now let $n=2$. Then
496: $SL_2(\C)/SU(2)$ is identified with the hyperbolic $3$-space.
497: We prove in Chapter 7 that $\psi_2(x_1, x_2, x_3, x_{4})$ is
498: the volume of the geodesic simplex with vertices at the points
499: $x_1, ..., x_4$. Restricting to the ideal geodesic simplices
500: and using the relation to the Grassmannian dilogarithm
501: plus (\ref{6.13.02.101}) we
502: get a
503: new proof of Lobachevsky's formula.
504:
505:
506: {\bf 6. The Grassmannian $n$--logarithms and the Borel regulator}.
507: For any point $x \in \C\PP^{n-1}$ the function
508: \begin{equation} \label{6.17.02.2}
509: c_{2n-1}^n(g_1,...,g_{2n}):= { {\cal L}}^G_n(g_1x,...,g_{2n}x)
510: \end{equation}
511: is a measurable $(2n-1)$-cocycle of the Lie group $GL_n(\C)$.
512: Indeed, it is invariant
513: under the diagonal action of $GL_n(\C)$ and the cocycle
514: condition is just the first functional equation for the function
515: $ {\cal L}^G_n$. Different points $x$ give canonically cohomologous cocycles.
516: However {\it a priori} it is not clear that the corresponding cohomology class
517: is non-zero.
518:
519:
520: Let $H^{2n-1}_{m}(GL_n(\C), \R)$ be the space of measurable
521: cohomology of the Lie group $GL_n(\C)$. It is known that
522: $$
523: H^{\ast}_{m}(GL_n(\C), \R) = \Lambda_{\Bbb
524: R}^{\ast}(b_1,b_3,...,b_{2n-1})
525: $$
526: where $b_{2k-1} \in H^{2k-1}_{m}(GL_n(\C), \R)$ are certain
527: canonical generators called the Borel classes ([Bo1]).
528: \begin {theorem}
529: \label {0.4}
530: The cohomology class of the Grassmannian cocycle (\ref{6.17.02.2}) is a non
531: zero rational multiple of the Borel class $b_{2n-1}$.
532: \end{theorem}
533:
534: For normalization of the Borel classes and precise relationship between the Grassmannian polylogarithms and the Borel regulator
535: see Chapter 5, especially Sections 5.4 and 5.5.
536:
537: The essential role in the proof is played by
538: the fact that the Grassmannian $n$--logarithm function
539: ${\cal L}^G_n$ is a boundary value of the function
540: $\overline \psi_n$. The function
541: $\overline \psi_n(x_1,...,x_{2n})$ is not continuous at certain boundary points, but
542: always satisfies the cocycle condition.
543: So taking any point $x \in \overline {\Bbb H}_n$ we get a cocycle
544: $$
545: c_x(g_1, ..., g_{2n-1}):= \overline \psi_n(g_1x,..., g_{2n}x)
546: $$
547: of the group $GL_n(\C)$. Its cohomology class
548: does not depend on $x$. If $x \in {\Bbb H}_n$
549: the corresponding cocycle is smooth. We can differentiate it, getting
550: a cohomology class of the Lie algebra $gl_n$, and relate it to the
551: Borel class. On the other hand taking $x$ to be a point on the boundary stratum
552: $\C\PP^{n-1}$ we recover the Grassmannian cocycle (\ref{6.17.02.2}).
553: So we get the theorem.
554:
555: Combining it with the technique developed in [G1-2] we
556: get a simple explicit construction of the Borel regulator
557: $$
558: K_{2n-1}(\C) \longrightarrow \R
559: $$
560: in terms of the Grassmannian $n$-logarithms.
561: The second functional equation for ${\cal L}^G_n$ plays an important role in the proof.
562: Therefore, thanks to the Borel theorem [Bo2], this allows
563: to express the special values of Dedekind $\zeta$--functions at $s=n$
564: via the Grassmannian $n$--logarithms.
565:
566: The definition of the Higher Chow groups of a variety $X$
567: is much simpler than
568: the definition of algebraic K-groups of $X$. The situation with
569: the regulator maps is similar. However relating the
570: special values of the Dedekind $\zeta$-functions to motivic
571: cohomology of the corresponding number fields
572: we need to work with the algebraic K-theory (or homology of $GL_n(F)$)
573: of number fields.
574:
575: Chapter 2 is the main core of the paper. In Chapters 3, 4 and 6
576: the main construction of Chapter 2 is investigated from different points of view.
577: Chapters 4 and 5 are rather
578: independent from the other Chapters.
579:
580: {\bf Acknowledgement}.
581: I am grateful to Spencer Bloch for discussions of Arakelov
582: motivic complexes.
583: The results of this paper were discussed in my course at Brown (Spring 1998),
584: and at
585: the Newton Institute (Cambridge, March 1998).
586: I am grateful to participants for
587: their interest and useful comments. I am very grateful to the referees
588: for many comments which greatly improved the exposition.
589:
590:
591: During the work on this paper I enjoyed the hospitality and support of the
592: MPI(Bonn) and
593: IHES (Bures-sur-Yvette). I am grateful to these
594: institutions.
595: This work was supported by the NSF grants DMS-9800998
596: and DMS-0099390.
597:
598:
599: \section { Arakelov motivic complexes}
600:
601: {\bf 1. The Higher Chow group complex}. A (non-degenerate) simplex in
602: $\PP^m$ is an ordered
603: collection of hyperplanes $L_0,...,L_m$ in generic position, i.e. with empty
604: intersection.
605: Let us choose in $\PP^m$ a simplex $L$ and a generic hyperplane $H$.
606: We might think about this data as of a simplex in the $m$-dimensional
607: affine space $\AAA^m:= \PP^m - H$. For any two non-degenerate
608: simplices in $\AAA^m$ there is a unique affine transformation sending one
609: simplex to the other.
610:
611: Let $I=(i_1,...,i_k)$ and $L_I:= L_{i_1} \cap ... \cap L_{i_k}$.
612: Let $X$ be a regular projective variety over a field $F$. Let
613: ${\cal Z}_{m}(X; n)$ be
614: the free abelian group generated by irreducible codimension $n$
615: algebraic
616: subvarieties in $X \times \AAA^m$ which intersect properly (i.e. with
617: the right codimension) all faces
618: $X \times L_I$.
619:
620: {\bf Warning}. We use the notation ${\cal Z}_{m}(X; n)$ for the group
621: denoted ${\cal Z}^{n}(X; m)$ by Bloch. This allows us to use upper and lower
622: indices to distinguish between the
623: homological and cohomological notations, see below.
624:
625: For a given codimension 1 face $L_i$ of a simplex $L$ in $\AAA^m$ the other
626: faces $L_j$ cut a simplex $\widehat L_i:= \{L_i \cap L_j\}$, in $L_i$.
627: So the intersection with codimension 1 faces $X \times L_i$ provides
628: group homomorphisms
629: $$
630: \partial_i: {\cal Z}_{m}(X; n) \longrightarrow
631: {\cal Z}_{m-1}(X; n); \qquad \partial:= \sum_{i=0}^m (-1)^i \partial_i
632: $$
633: Then $\partial^2 =0$, so
634: $({\cal Z}_{\bullet}(X; n); \partial)$
635: is a homological complex. Its homology groups
636: are Bloch's Higher
637: Chow groups.
638: By the fundamental theorem of Bloch ([Bl1-2], [Lev])
639: $$
640: H_i({\cal Z}_{\bullet}(X; n)\otimes \Q)
641: = K^{(n)}_{i}(X)\otimes \Q
642: $$
643: Let us cook up a cohomological complex setting
644: $$
645: {\cal Z}^{\bullet}(X; n):= {\cal Z}_{2n-\bullet}(X; n)
646: $$
647: Its cohomology provides a definition of the integral motivic cohomology of $X$:
648: $$
649: H_{{\cal M}}^i(X, \Z(n)) := H^i({\cal Z}^{\bullet}(X; n))
650: $$
651: Bloch's theorem guarantees Beilinson's formula (\ref{4.29.02.1})
652: for the rational motivic cohomology.
653:
654:
655:
656: {\bf 2. The Beilinson--Deligne complex}.
657: Recall that an $n$-distribution, sometimes also called an $n$-form with generalized function coefficients, or an $n$-current, on a smooth oriented manifold $X$ is
658: a continuous linear functional on the space of
659: $(\dim_{\R}X-n)$--forms with compact support. Denote by
660: ${\cal D}^{n}_{X}$ the space of all {\it real} $n$--distributions on $X$.
661: The
662: De Rham complex of distributions
663: $({\cal D}^{\bullet}_{X},d)$ is a resolution of the
664: constant sheaf $\R$. The space ${\cal A}^{n}_{X}$ of all smooth $n$--forms
665: on $X$ is a subspace of
666: ${\cal D}^{n}_{X}$.
667:
668:
669: Let $X$ be a regular projective variety over $\C$. The
670: standard weight $n$ Beilinson-Deligne complex
671: ${\R}^{\bullet}(X; n)_{{\cal D}}$
672: is the total complex associated with the following
673: bicomplex:
674: $$
675: \begin{array}{ccccccccccc} \label{del}
676: \Bigl({\cal D}_{X}^{0}&\stackrel{d}{\longrightarrow}&{\cal
677: D}_{X}^{1}&\stackrel{d}{\longrightarrow}&\ldots&\stackrel{d}{\longrightarrow}&{\cal
678: D}^{n}_{X}&\stackrel{d}{\longrightarrow}&{\cal
679: D}_{X}^{n+1}&\stackrel{d}{\longrightarrow}&\ldots\Bigr) \otimes \R(n-1)\\
680: &&&&&&&&&&\\
681: &&&&&&\uparrow\pi_{n} &&\uparrow\pi_{n}&&\\
682: &&&&&&&&&&\\
683: &&&&&&\Omega^{n}_{X}
684: &\stackrel{\partial}{\longrightarrow}&\Omega_{X}^{n+1}&\stackrel
685: {\partial}{\longrightarrow}&
686: \end{array}
687: $$
688: Here $\R(n):= (2\pi i)^n\R$ and
689: $$
690: \pi_n: {\cal D}_{X}^{p}\otimes \C \longrightarrow
691: {\cal D}_{X}^{p}\otimes \R(n-1)
692: $$ is the projection induced by the one $\C = \R(n-1) \oplus \R(n) \longrightarrow
693: \R(n-1)$. Further, ${\cal D}^{0}_{X}$ placed in
694: degree 1 and
695: $(\Omega^{\bullet}_{X}, \partial)$ is the De Rham complex of
696: holomorphic forms.
697:
698: The Beilinson-Deligne complex ${\R}^{\bullet}(X; n)_{{\cal D}}$
699: is quasiisomorphic to the complex
700: $$
701: \R(n) \lra {\cal O}_X \lra \Omega^{1}_{X}
702: \lra \Omega^{2}_{X}\lra ... \lra \Omega^{n-1}_{X}
703: $$
704:
705: %If $X$ is not projective we should take a
706: %compactification $\overline X$ of $X$ such that $Y:= \overline X - X$
707: %is a normal crossing divisor, and use
708: %the complex $F^n\Omega^{\bullet}_{X}<Y>$ of forms
709: %{\it with logarithmic singularities}
710: % at $Y$ in the definition above.
711:
712:
713: {\bf 3. The truncated Deligne complex}.
714: Let ${\cal D}_X^{ p,q} = {\cal D}^{ p,q}$ be the abelian group of complex valued distributions of type $(p,q)$ on $X(\C)$. Consider the following cohomological
715: ``bicomplex'', where ${\cal D}^{n,n}_{\rm cl} $
716: is the subspace of the space ${\cal D}^{n,n}$ of closed currents, and
717: ${\cal D}^{0,0}$ is in degree $1$:
718: $$
719: \begin{array}{ccccccccc}
720: &&&&&&&&{\cal D}_{cl}^{n,n}\\
721: &&&&&&&&\\
722: &&&&&&&2 \overline \partial \partial\nearrow&\\
723: &&&&&&&&\\
724: {\cal D}^{0,n-1}&\stackrel{\partial}{\longrightarrow}&{\cal D}^{1,n-1}&\stackrel{\partial}{\longrightarrow}&
725: ...&\stackrel{\partial}{\longrightarrow}&{\cal D}^{n-1,n-1}&&\\
726: &&&&&&&&\\
727: \overline \partial \uparrow &&\overline \partial \uparrow &&&&\overline \partial \uparrow&&\\
728: ...&...&...&...&...&...&...&&\\
729: \overline \partial \uparrow&&\overline \partial \uparrow&&&&\overline \partial \uparrow&&\\
730: &&&&&&&&\\
731: {\cal D}^{0,1}&\stackrel{\partial}{\longrightarrow}& {\cal D}^{1,1}&\stackrel{\partial}{\longrightarrow}&... &\stackrel{\partial}{\longrightarrow}& {\cal D}^{n-1,1}&&\\
732: &&&&&&&&\\
733: \overline \partial \uparrow&&\overline \partial \uparrow&&&&\overline \partial \uparrow&&\\
734: &&&&&&&&\\
735: {\cal D}^{0,0}&\stackrel{\partial}{\longrightarrow} &{\cal D}^{1,0}& \stackrel{\partial}{\longrightarrow} &...&\stackrel{\partial}{\longrightarrow} &{\cal D}^{n-1,0}&&
736: \end{array}
737: $$
738: Properly speaking, it is not a bicomplex due to the presence of the operator
739: $2\overline \partial \partial$, but we can handle it the same way we handle the bicomplexes. Namely, we define its total complex
740: $Tot^{\bullet}$. It is concentrated
741: in degrees $[1,2n]$. The complex $C^{\bullet}_{{\cal D}}(X(\C); n)
742: = C^{\bullet}_{{\cal D}}(n)$ is a subcomplex
743: of the complex
744: $Tot^{\bullet}$ defined as follows.
745: Take the intersection of the part of the complex $Tot^{\bullet}$
746: coming from the $n \times n$ square in the diagram (and concentrated in degrees
747: $[1,2n-1]$) with the complex of distributions with values in $\R(n-1)$. Consider
748: the subgroup ${\cal D}_{\R, cl}^{n, n}(n) \subset {\cal D}_{cl}^{n, n}$ of
749: the $\R(n)$-valued distributions of type $(n, n)$. They form a subcomplex in
750: $Tot^{\bullet}$ because
751: $\overline \partial \partial$ sends $\R(n-1)$-valued distributions to
752: $\R(n)$--valued distributions.
753: This is the complex $C^{\bullet}_{{\cal D}}(n)$. It is a truncation of the
754: complex considered by Deligne ([Del]). Its cohomology is the absolute Hodge cohomology defined by Beilinson [B3].
755:
756:
757:
758: \begin{proposition} \label{dcom} Let $X$ be a regular complex projective
759: variety.
760: Then the complex $C^{\bullet}_{{\cal D}}(X; n)$ is quasiisomorphic to the
761: truncated Beilinson-Deligne complex $\tau_{\leq 2n}\R^{\bullet}(X; n)_{{\cal D}}$.
762: \end{proposition}
763:
764: {\bf Proof}. We need the following general construction.
765: Let $f^{\bullet}: X^{\bullet} \lra Y^{\bullet}$ be a morphism of complexes such that
766: the map $f^i $ is injective for $i \leq p$ and surjective for $i \geq p$ (and hence is an isomorphism for $i=p$). Consider a complex
767: $$
768: Z^{\bullet}:= \quad {\rm Coker}f^{<p}[-1] \stackrel{D}{\lra} {\rm Ker}f^{>p}
769: $$
770: where the differential $D: {\rm Coker} f^{p-1} \lra {\rm Ker} f^{p+1}[1]$ is defined via the following diagram (the vertical sequences are exact):
771: $$
772: \begin{array}{ccccccccc}
773: &&&&&& 0 &&0\\
774: &&&&&&\downarrow &&\downarrow \\
775: 0&&0 &&&&{\rm Ker}f^{p+1}&\lra &{\rm Ker}f^{p+2}\\
776: \downarrow &&\downarrow &&&&\downarrow &&\downarrow \\
777: X^{p-2} &\lra &X^{p-1} &\lra &X^{p}&\lra & X^{p+1}&\lra &X^{p+2}\\
778: \downarrow &&\downarrow &&f^p \downarrow = && \downarrow &&\downarrow \\
779: Y^{p-2}&\lra &Y^{p-1} &\lra &Y^{p} &\lra & Y^{p+1}&\lra &Y^{p+2}\\
780: \downarrow &&\downarrow &&&&\downarrow &&\downarrow \\
781: {\rm Coker} f^{p-2}&\lra &{\rm Coker} f^{p-1}&&&& 0&&0\\
782: \downarrow &&\downarrow &&&&&&\\
783: 0&&0 &&&&&&
784: \end{array}
785: $$
786:
787: \begin{lemma} \label{decom}
788: The complex $Z^{\bullet}$ is canonically quasiisomorphic to ${\rm Cone} (X^{\bullet} \stackrel{f^{\bullet}}{\lra} Y^{\bullet})$.
789: \end{lemma}
790:
791: {\bf Proof}. Let
792: $$
793: \widetilde \tau_{<p}X^{\bullet}:= \quad ... \stackrel{d_X}{\lra} X^{p-2} \stackrel{d_X}{\lra} X^{p-1} \stackrel{d_X}{\lra} {\rm Im} d_X
794: $$
795: $$
796: \widetilde \tau_{\geq p}Y^{\bullet}:= \quad Y^{p}/{\rm Im} d_Y \stackrel{d_Y}{\lra} Y^{p+1}
797: \stackrel{d_Y}{\lra} Y^{p+2} \stackrel{d_Y}{\lra} ...
798: $$
799: Then there is an exact sequence of complexes $0 \lra \widetilde \tau_{<p}X^{\bullet} \lra X^{\bullet} \lra \widetilde \tau_{\geq p}X^{\bullet} \lra 0$. The conditions on the maps $f^{\bullet}$ imply that
800: $$
801: \widetilde \tau_{<p}f^{\bullet}: \quad \widetilde \tau_{<p}X^{\bullet} \lra \widetilde \tau_{<p}Y^{\bullet} \quad \mbox{is injective}
802: $$
803: $$
804: \widetilde \tau_{\geq p}f^{\bullet}: \quad \widetilde \tau_{\geq p}Y^{\bullet} \lra \widetilde \tau_{\geq p}Y^{\bullet} \quad \mbox{is surjective}
805: $$
806: We get maps of complexes
807: $$
808: {\rm Cone}(\widetilde \tau_{<p}X^{\bullet} \lra f^{\bullet}(\widetilde \tau_{<p}X^{\bullet})) \stackrel{\alpha}{\hookrightarrow}
809: {\rm Cone}(X^{\bullet} \lra Y^{\bullet}) \stackrel{\beta}{\longrightarrow}
810: {\rm Cone}(\widetilde \tau_{\geq p}Y^{\bullet} \lra \widetilde \tau_{\geq p}Y^{\bullet})
811: $$
812: where $\alpha$ is injective and $\beta$ is surjective.
813: The complex ${\rm Ker} (\beta)/{\rm Im}(\alpha)$ looks as follows:
814: $$
815: \begin{array}{ccccccccc}
816: 0&\lra &0 &\lra & {(f^p)}^{-1}{\rm Im}(d_Y)/{\rm Im}(d_X)&\lra &{\rm Ker} f^{p+1}&\lra &{\rm Ker} f^{p+2}\\
817: \downarrow &&\downarrow &&\downarrow &&\downarrow &&\downarrow \\
818: {\rm Coker} f^{p-2}&\lra &{\rm Coker} f^{p-1}&\lra & {\rm Im}(d_Y)/{f^p}{\rm Im}(d_X) &\lra &0&\lra &0
819: \end{array}
820: $$
821: Since the map $f^p : {\rm Im}(d_Y)/{\rm Im}(d_X) \to {\rm Im}(d_Y)/{f^p}{\rm Im}(d_X)$ is an isomorphism it is quasiisomorphic to $Z^{\bullet}$. The lemma is proved.
822:
823:
824:
825: Applying the lemma to the morphism of complexes
826: $$
827: {\rm Tot}({\cal D}^{\geq n, \bullet}) \stackrel{\pi_n}{\longrightarrow} {\cal D}^{n+\bullet}\otimes_{\R}\R(n-1)
828: $$
829: we see that the complex $\R(n)_{{\cal D}}$ is canonically quasiisomorphic to the following complex:
830:
831: \begin{figure}[ht]
832: \centerline{\epsfbox{gpol1.eps}}
833: \caption{The weight $n$ real Deligne complex }
834: \label{gpol1}
835: \end{figure}
836: To compute the differential ${\cal D}_{\R}^{n-1, n-1}(n-1) \lra {\cal D}_{\R}^{n, n}(n)$ we proceed as follows. Take $\alpha \in {\cal D}_{\R}^{n-1, n-1}(n-1)$, so $\alpha = (-1)^{n-1}\overline \alpha$. Then
837: $$
838: d \alpha \quad = \quad \partial \alpha + \overline \partial \alpha \quad = \quad \partial \alpha + (-1)^{n-1}\overline {\partial \alpha } \quad = \quad 2 \pi_n(\partial \alpha )
839: $$
840: Applying $d = \partial + \overline \partial$ again and taking the $(n,n)$-component we get $2\overline \partial \partial (\alpha )$.
841: Truncating this complex we obtain the proof of the proposition.
842: (Note that $d^{C}:=
843: (4\pi i)^{-1}(\partial - \overline \partial)$,
844: so $dd^{C} = (2\pi i)^{-1}\overline \partial\partial$.)
845:
846: Now if $X$ is a variety over $\R$, then we set
847: $$
848: C^{\bullet}_{{\cal D}}(X_{/{\R}}; n):= C^{\bullet}_{{\cal D}}(X; n)^{\overline F_{{\infty}}}; \qquad H_{{\cal D}}^i(X_{/{\R}}; \R(n)) :=
849: H^i\Bigl( C^{\bullet}_{{\cal D}}(X_{/{\R}}; n)\Bigr)
850: $$
851: where $\overline F_{{\infty}}$ is the De Rham involution, i.e. the composition of the involution $F_{{\infty}}$ on $X(\C)$ induced by the complex conjugation with the complex conjugation of coefficients.
852:
853:
854: \begin{theorem-construction} \label{6.11.02.1}
855: Let $X$ be a regular complex projective variety. Then
856: there exists a {\rm canonical} homomorphism of complexes
857: $$
858: {\cal P}^{\bullet}(n): {\cal Z}^{\bullet}(X; n)
859: \longrightarrow C^{\bullet}_{{\cal D}}(X; n)
860: $$
861: If $X$ is defined over $\R$ then the image of the map
862: ${\cal P}^{\bullet}(n)$ lies in
863: the subcomplex $C^{\bullet}_{{\cal D}}(X_{/\R}; n)$.
864: \end{theorem-construction}
865:
866: %The complex $C^{\bullet}_{{\cal D}}(X; n)$
867: %is quasiisomorphic to Deligne's complex when $X(\C)$ is compact.
868: %When $X(\C)$ is non compact
869: %we define the Arakelov motivic complexes using
870: %the standard Beilinson--Deligne complexes.
871:
872: %\begin{theorem-construction} \label{6.11.02.1s}
873: %If $X$ be an open regular complex variety then the same construction
874: %provides a homomorphism of complexes
875: %$$
876: %{\cal P}^{\bullet}(n): {\cal Z}^{\bullet}(X; n)
877: %\longrightarrow \tau_{\leq 2n}\R^{\bullet}(X; n)_{{\cal D}}
878: %$$
879: %\end{theorem-construction}
880:
881: To construct this homomorphism we need to define certain
882: homomorphism $r_{n-1}$ ([G5]). In the next section we
883: recall its definition and establish its basic properties.
884: Using it we define an $(m-1)$-form
885: $r_{m-1}(L; H)$ canonically attached to the pair $(\AAA^m; L) =
886: (\PP^m - H, L)$, and then define the homomorphism ${\cal P}^{\bullet}(n)$.
887:
888:
889: {\bf 4. The homomorphism $r_{m-1}$}.
890: Let $X$ be a variety over $\C$. Let $f_1,...,f_m$ be $m$ rational functions on $X$.
891: We attach to them the $(m-1)$-form
892: \begin {equation} \label{1wq}
893: r_{m-1}(f_1,..., f_m) :=
894: \end {equation}
895: $$
896: {\rm Alt}_m \sum_{j\geq 0, 2j+1\leq 2m+1} c_{j,m}\log|f_1|d\log|f_2|
897: \wedge ... \wedge d\log|f_{2j+1}|\wedge di\arg f_{2j+2}\wedge ... \wedge
898: di\arg f_{m}
899: $$
900: Here $c_{j,m}:= \frac{1}{(2j+1)!(m-2j-1)!} $ and ${\rm Alt}_m$ is the operation of alternation:
901: $$
902: {\rm
903: Alt}_m F(x_1,...,x_m):= \sum_{\sigma \in S_m}(-1)^{|\sigma|}F(x_{\sigma
904: (1)},...,x_{\sigma (m)})
905: $$
906:
907:
908:
909: So $r_{m-1}(f_1,..., f_m)$ is an $\R(m-1)$-valued $(m-1)$-form and it
910: is easy to check that
911: \begin{equation} \label{6.16.04.3}
912: d r_{m-1}(f_1,..., f_m) = \pi_m \Bigl(
913: d \log f_1 \wedge ... \wedge d \log f_m\Bigr) \quad
914: \end{equation}
915:
916: The form (\ref{1wq}) is a part of a cocycle representing the product in real Deligne cohomology of
917: 1-cocycles $(\log|f_i|, d\log f_i)$.
918:
919: Here is a yet another, a bit more general
920: way to look at the homomorphism $r_{m-1}$. Let
921: ${\cal A}^i(M)$ be the space of smooth $i$-forms on a real smooth manifold $M$.
922: Consider the following map
923: \begin{equation} \label{6.16.04.1}
924: \omega_{m-1}: \Lambda^{m}{\cal A}^0(M) \lra {\cal A}^{m-1}(M)
925: \end{equation}
926: $$
927: \omega_{m-1}(\varphi_1 \wedge ... \wedge \varphi_{m}) :=
928: $$
929: $$
930: \frac{1}{m!}{\rm Alt}_{m}
931: \Bigl(\sum_{k=1}^{m}
932: (-1)^{k-1}\varphi_1\partial \varphi_2 \wedge ... \partial \varphi_k \wedge \overline
933: \partial \varphi_{k+1} \wedge ... \wedge \partial \varphi_{m}\Bigr)
934: $$
935: For example
936: $$
937: \omega_0(\varphi_1) = \varphi_1;\qquad
938: \omega_1(\varphi_1 \wedge \varphi_{2}) = \frac{1}{2}\Bigl(\varphi_1\partial \varphi_2 -
939: \varphi_2\partial \varphi_1 - \varphi_1\overline \partial \varphi_2 +
940: \varphi_2\overline \partial \varphi_1 \Bigr)
941: $$
942: Then one easily checks that
943: \begin{equation} \label{6.16.04.2}
944: d\omega_{m-1}(\varphi_1 \wedge ... \wedge \varphi_{m}) =
945: \partial \varphi_1 \wedge ... \wedge \partial \varphi_{m}
946: +(-1)^m \overline \partial \varphi_1 \wedge ... \wedge \overline \partial \varphi_{m} +
947: \end{equation}
948: $$
949: \sum_{i=1}^{m}(-1)^{i}\overline \partial \partial \varphi_i \wedge
950: \omega_{m-2}(\varphi_1 \wedge ... \wedge \widehat \varphi_i \wedge ... \wedge \varphi_{m})
951: $$
952:
953:
954:
955:
956: Now let $f_i$ be rational functions on
957: a complex algebraic variety $X$. Set $M:= X^0(\C)$, where $X^0$ is the open part
958: of $X$ where the functions $f_i$ are regular. Then $\varphi_i := \log|f_i|$
959: are smooth functions on $M$, and we have an identity
960: $$
961: \omega_{m-1}(\log |f_1| \wedge ... \wedge \log |f_{m}|) = r_{m-1}(f_1 \wedge ... \wedge f_{m})
962: $$
963:
964: Observe that $\overline \partial \partial \log|f| = 0$
965: on $X^0(\C)$. Therefore the second term in the formula (\ref{6.16.04.2}) is zero, and so this formula
966: is consistent with the one (\ref{6.16.04.3}). Notice however that
967: if we understood
968: $\overline \partial \partial \log|f| $ as a distribution on $X(\C)$, then
969: by the Poincar\'e-Lelong formula one has
970: \begin{equation} \label{6.16.04.4}
971: 2\overline \partial \partial \log|f| = 2\pi i \delta(f)
972: \end{equation}
973:
974: Our next goal is to interpret the form $r_{m-1}(f_1 \wedge ... \wedge f_{m})$ as a distribution
975: on $X(\C)$ and calculate the differential of this distribution, taking into account
976: formula (\ref{6.16.04.4}).
977:
978:
979: {\bf 5. The distribution $r_{m-1}(f_1 \wedge ... \wedge f_{m})$}.
980: Recall (see for instance [S])
981: that for a subvariety $Y$ of a smooth complex variety $X$
982: we define the $\delta$-distribution $\delta_{Y}$ by setting
983: $$
984: <\delta_{Y}, \omega>:= \int_{Y^0(\C)} \omega
985: $$
986: where $Y^0$ is the nonsingular part of $Y$.
987:
988: \begin{theorem} \label{8.6.02.2}
989: Let $Y$ be an arbitrary irreducible subvariety of a smooth complex variety
990: $X$ and $f_1, ..., f_m \in {\C}^*(Y)$. Then for any smooth differential
991: form $\omega$
992: with compact support on $X(\C)$ the following integral is convergent:
993: $$
994: \int_{Y^0(\C)}r_{m-1}(f_1, ..., f_m) \wedge i_Y^0\omega
995: $$
996: Here $Y^0$ is the nonsingular part of $Y$ and $i_Y^0\omega$ is the
997: restriction of the form $\omega$ to $Y^0(\C)$.
998: Thus the form $r_{m-1}(f_1,..., f_m)$ defines a distribution
999: $r_{m-1}(f_1, ...,f_m)\delta_Y$ on
1000: $X(\C)$ given by
1001: $$
1002: <r_{m-1}(f_1, ...,f_m)\delta_Y, \omega>:=
1003: \int_{Y^0(\C)}r_{m-1}(f_1, ..., f_m) \wedge i_Y^0\omega
1004: $$
1005: It provides a group homomorphism
1006: \begin{equation} \label{2}
1007: r_{m-1}: \Lambda^m\C(Y)^{\ast} \longrightarrow {\cal D}^{m-1}_{X(\C)}(m-1)
1008: \end{equation}
1009: \end{theorem}
1010:
1011: {\bf Proof}. We need the following lemma
1012:
1013: \begin{lemma} \label{8.6.02.112}
1014: Let $Y$ be a smooth complex projective variety. Then
1015: for any non zero rational functions $f_1,...,f_m$ on $X$ and
1016: for any smooth form $\omega$
1017: with compact support on $Y(\C)$ the integral
1018: $$
1019: \int_{Y(\C)}r_{m-1}(f_1, ..., f_m) \wedge \omega
1020: $$
1021: is convergent. So the form
1022: $r_{m-1}(f_1, ..., f_m)$ defines a distribution
1023: on $Y(\C)$.
1024: \end{lemma}
1025:
1026:
1027:
1028: {\bf Basic example}. The integral $\int_{\C}\log|z|d\log(z-a) \wedge
1029: d\log \overline {(z-b)}$ is
1030: divergent at infinity, where all the functions $z, z-a, z-b$ have a
1031: simple pole, since $\int_{\C}\log|z|\frac{dz\wedge d\overline z}{|z|^2}$
1032: is divergent (both near zero and infinity). However
1033: $$
1034: 4\cdot \int_{\C}\log|z|d\log|z-a| \wedge
1035: d\log |z-b| =
1036: $$
1037: $$
1038: \int_{\C}\log|z|\Bigl( d\log(z-a) \wedge d\log \overline {(z-b)} \quad + \quad d\log \overline {(z-a)} \wedge d\log (z-b)\Bigr)
1039: $$
1040: is convergent: the divergent parts cancel each other.
1041: In $r_{m-1}(f_1,...,f_{m})$ such divergences
1042: cancel because of multiplicativity and skew-symmetry of $r_{m-1}$.
1043:
1044: {\bf Proof of Lemma \ref{8.6.02.112}}.
1045: Resolving singularities we reduce the statement of the lemma to the case when divisors
1046: ${\rm div}f_i$
1047: have normal crossing. Using the fact that $r_{m-1}$ is a homomorphism
1048: to differential forms
1049: we may suppose that these divisors are
1050: different. Our statement is local, so we can assume that in
1051: local coordinates $z_1,...,z_m$ one has $f_1 = z_1,...,f_k = z_k$ and
1052: ${\rm div}f_j$ for $j>k$ does not intersect the origin.
1053: After this the statement of lemma is obvious: each term in (\ref{1wq})
1054: defines
1055: a distribution near the origin. For instance the worst possible singularities have
1056: the term $\log|z_1|d\log|z_2|
1057: \wedge ... \wedge d\log|z_{k}|\wedge \omega$ where $\omega$ is smooth
1058: near the origin. It is clearly integrable with a smooth test form. The lemma is proved.
1059:
1060: {\bf Remark}. In particular if ${\rm dim}_{\C}X=n$ the integral
1061: \begin{equation} \label{a5}
1062: \int_{X(\C)}r_{2n}(f_1,...,f_{2n+1})
1063: \end{equation}
1064: is convergent.
1065:
1066:
1067:
1068:
1069: Below we use the following form of the resolution of singularities
1070: theorem. Recall that the proper preimage $\widetilde Y$
1071: is the closure in $\widetilde X$
1072: of the preimage of
1073: the generic part of $Y$.
1074:
1075: \begin{theorem} \label{8.9.02.4}
1076: Let $Y$ be an arbitrary
1077: subvariety of a regular variety $X$ over a characteristic
1078: zero field, and $Z$
1079: a divisor of $Y$. Then there exists a sequence of blow ups
1080: $\widetilde X \lra X_1 \lra ... \lra X$ providing a projection
1081: $\pi: \widetilde X \to
1082: X$ such that the proper preimage $\widetilde Y$ of $Y$
1083: is non singular,
1084: $\widetilde Z:= p^*Z$, where $p= \pi|_{\widetilde Y}:\widetilde Y \to Y$,
1085: is a normal crossing divisor in $\widetilde Y$,
1086: and restriction of $\pi$ to the nonsingular part
1087: of $\widetilde Y- \widetilde Z$
1088: is an isomorphism.
1089: \end{theorem}
1090:
1091:
1092:
1093:
1094: {\bf Proof of Theorem \ref{8.6.02.2}}. By Theorem \ref{8.9.02.4}
1095: there exists a sequence of blow ups providing a projection
1096: $\pi: \widetilde X \lra X$ such that
1097: the proper preimage $\widetilde Y$ of $Y \subset X$ is
1098: smooth.
1099: %Indeed, the sequence of blow ups of $Y$, resolving the singularities
1100: %of $Y$, reinterpreted as a sequence of blow ups of $X$
1101: %would do it.
1102: By the above lemma the integral
1103: \begin{equation} \label{8.6.02.1}
1104: \int_{\widetilde Y(\C)}r_{m-1}(\pi^*f_1, ..., \pi^*f_m) \wedge \pi^*\omega
1105: \end{equation}
1106: is convergent. Therefore the similar integral
1107: over any Zariski dense subset of $\widetilde Y(\C)$ is also convergent and
1108: coincides with (\ref{8.6.02.1}). Since
1109: $\pi$ is an isomorphism on the nonsingular part of
1110: $\widetilde Y- \widetilde Z$, we are done.
1111: The Theorem \ref{8.6.02.2} is proved.
1112:
1113:
1114: Below we employ notation $r_{m-1}(f_1 \wedge ... \wedge f_m)$
1115: for the distribution given by (\ref{2}).
1116:
1117:
1118:
1119: {\bf 6. Differential of the distribution
1120: $r_{n-1}(f_1 \wedge ... \wedge f_n)$}.
1121: Let $X$ be a normal variety. Then there is the residue homomorphism
1122: $$
1123: {\rm Res}: \Lambda^n\C(X)^* \lra \oplus_{Y \subset X^{(1)}}\Lambda^{n-1}\C(Y)^*, \qquad
1124: $$
1125: where the sum is over all irreducible divisors of $X$.
1126:
1127: Here is its definition. Let $K$ be a field with
1128: a discrete valuation $v$ and the residue field $
1129: k_v$. The group of units $U$ has a natural
1130: homomorphism $U\longrightarrow k_v^{\ast}\; , \; u\mapsto
1131: \overline u$. An element $\pi \in K^{\ast}$ is prime if
1132: ${\rm ord}_{v}\pi = 1$. There is a homomorphism ${\rm res}_v:\Lambda^{n}K^{\ast}
1133: \longrightarrow\Lambda^{n-1} k_v^{\ast}$ uniquely defined
1134: by the properties $(u_{i}\in U)$:
1135: $$
1136: {\rm res}_v (\pi\wedge u_{1}\wedge \cdots\wedge u_{n-1})
1137: = \overline u_{1}\wedge\cdots \wedge \overline u_{n-1}\quad \mbox{and} \quad
1138: {\rm res}_v
1139: (u_{1}\wedge \cdots \wedge u_{n}) = 0
1140: $$
1141: It does not depend on the choice of $\pi$.
1142:
1143:
1144:
1145: Observe that if $X$ is normal then
1146: the local ring of any irreducible divisor of
1147: $X$ is a discrete valuation ring, so we can apply the above construction.
1148: We set ${\rm Res}:= \sum {\rm res}_{v}$ where the sum is over
1149: all valuations of the field $\C(X)$
1150: corresponding to the codimension one points of $X$.
1151:
1152: {\bf Remark}. If for any $i$ the restrictions of
1153: the functions $f_j$ for $j \not = i$
1154: to the generic points of all irreducible components of
1155: the divisor ${\rm div}f_i$ are non zero, then
1156: \begin{equation} \label{8.12.02.100}
1157: {\rm Res}(f_1 \wedge ... \wedge f_n) = \sum_{Y \in X^{(1)}}\sum_{i=1}^n(-1)^{i-1}v_Y(f_i)
1158: \cdot {f_1}_{|_Y} \wedge ... \wedge \widehat {f_i}_{|_Y}\wedge ...
1159: \wedge {f_n}_{|_Y}
1160: \end{equation}
1161: where $v_Y(f)$ is the order of zero of $f$ at the generic point of $Y$.
1162:
1163: Let $f_1 \wedge ...\wedge f_n \in \Lambda^n \C(Y)^*$ where $Y$ is a
1164: subvariety of a regular complex variety $X$. We define
1165: a distribution
1166: \begin{equation} \label{8.9.02.1}
1167: (r_{n-2}\circ {\rm Res})(f_1 \wedge ...\wedge f_n)
1168: \end{equation}
1169: on $X(\C)$ as follows.
1170:
1171:
1172: i) If $Y$ is a normal then we have defined
1173: the residue map ${\rm Res}$ on $\Lambda^n \C(Y)^*$.
1174: So we define
1175: (\ref{8.9.02.1}) as
1176: $$
1177: \sum_{Z \in Y^{(1)}}r_{n-2}{\rm res}_Z(f_1 \wedge ...\wedge f_n)\delta_Y
1178: $$
1179: i.e. for any smooth form $\omega$ on $X(\C)$
1180: $$
1181: <(r_{n-2}\circ {\rm Res})(f_1 \wedge ...\wedge f_n), \omega> :=
1182: \sum_Z \int_{Z^0(\C)}r_{n-2}{\rm res}_Z(f_1\wedge ... \wedge f_n)\wedge \omega
1183: $$
1184:
1185: ii) For an arbitrary $Y$ we take the normalization
1186: $\pi: Y^{\nu} \to Y$ and define (\ref{8.9.02.1}) as
1187: $\pi_*((r_{n-2}\circ {\rm Res}) (\pi^*f_1 \wedge ...\wedge \pi^* f_n))$
1188:
1189:
1190:
1191:
1192: \begin{lemma} \label{Lemma 2.7} Let $Y$ be a subvariety of a regular
1193: complex variety $X$, $f_1 \wedge ...\wedge f_n \in \Lambda^n
1194: \C(Y)^*$, and $Z:= \cup_i {\rm div}f_i$.
1195: Let $\pi: \widetilde X \to X$ is a blow up of $X$
1196: as in Theorem \ref{8.9.02.4}. Then
1197: \begin{equation} \label{8.9.02.3}
1198: \pi_*\Bigl((r_{n-2}\circ {\rm Res})(\pi^* f_1 \wedge
1199: ... \wedge \pi^* f_n) \Bigr) =
1200: (r_{n-2}\circ {\rm Res})(f_1 \wedge
1201: ... \wedge f_n)
1202: \end{equation}
1203: \end{lemma}
1204:
1205: {\bf Proof}. Thanks to the definition ii)
1206: of distribution (\ref{8.9.02.1}) for non normal varieties
1207: we may assume without loss
1208: of generality
1209: that $Y$ is normal.
1210:
1211:
1212: Let $Z = \cup_{i \in I}Z_i$ be the decomposition of the divisor $Z$ into
1213: irreducible components parametrised by a set $I$.
1214: By the very definition the right hand side of (\ref{8.9.02.3})
1215: is a sum over $i \in
1216: I$ of
1217: distributions $\psi_i= r_{n-2}(F_i)$ on
1218: $Z_i$ corresponding to certain elements
1219: $F_i \in \Lambda^{n-2}\C(Z_i)^*$.
1220:
1221: Let $\widetilde Z = \cup_{j \in J}\widetilde Z_j$
1222: be the decomposition of $\widetilde Z$ into
1223: irreducible components parametrised by a set $J$.
1224: The left hand side of (\ref{8.9.02.3})
1225: is a sum over $j \in J$ of distributions $\widetilde \psi_j
1226: = r_{n-2}(\widetilde F_j)$ corresponding to certain elements
1227: $\widetilde F_j \in \Lambda^{n-2}\C(\widetilde Z_j)^*$.
1228: One has $I \subset
1229: J$ since the proper preimage $\widetilde Z_i$
1230: of $Z_i$ is an irreducible component
1231: of
1232: $\widetilde Z$.
1233:
1234: The lemma follows from the following two claims:
1235: $$
1236: \pi_*(\widetilde \psi_i) = \psi_i, \quad i\in I; \qquad
1237: \pi_*(\widetilde \psi_j) = 0, \quad j\in J - I
1238: $$
1239: The first one is obvious since both distributions
1240: $\widetilde \psi_i$ and $\psi_i$ are defined
1241: by their restriction
1242: to any nonsingular Zariski dense open part of the corresponding
1243: divisor, and $\pi$, being restricted to such a sufficiently small part
1244: of $\widetilde Z_i$, is an isomorphism. Let us prove the second claim.
1245: Observe that for $j \in J-I$ the subvariety $\pi(\widetilde Z_j)$ is
1246: of
1247: codimension
1248: $2$ in $Y$, and restriction of $\pi$ to a Zariski open part of
1249: $\widetilde Z_j$ is a
1250: fibration with fibers of positive dimension. We need to show that for
1251: any smooth form $\omega$ on $X(\C)$
1252: $$
1253: \int_{\widetilde Z_j(\C)} \widetilde \psi_j \wedge \pi^*\omega
1254: := \int_{\widetilde Z'_j(\C)} \widetilde \psi_j\wedge \pi^*\omega =0
1255: $$
1256: where $Z_j'$ is a (sufficiently small)
1257: nonsingular Zariski dense open part of $Z_j$.
1258: Observe that $\widetilde \psi_j \wedge \pi^*\omega$
1259: is a smooth form on $\widetilde
1260: Z'_j(\C)$. For any vector field
1261: $v$ on $\widetilde Z_j'$ tangent to the fibers of $\pi$ we have
1262: $i_v\pi^*\omega =0$. So it suffices to show that
1263: $i_v \widetilde \psi_j =0$. Since the statement is local,
1264: we can choose a local equation of an open part of $\widetilde Z'_j$
1265: in the form $\pi^*z =0$. Using it as a local
1266: parameter at the definition of ${\rm res}_{\pi^*z =0}$, we see that
1267: ${\rm res}_{\pi^*z =0}(\pi^*f_1 \wedge ... \wedge \pi^* f_n)$ is
1268: lifted from $Y$, and hence so is
1269: $$
1270: \widetilde \psi_j = r_{n-2}
1271: {\rm
1272: res}_{\pi^*z =0}
1273: (\pi^*f_1 \wedge ... \wedge \pi^* f_n)
1274: $$
1275: Thus $i_v \widetilde \psi_j =0$. The lemma is proved.
1276:
1277:
1278: \begin{proposition} \label{8.9.02.6} Let $Y$ be an arbitrary
1279: subvariety of a regular
1280: complex variety $X$ and $f_1 \wedge ...\wedge f_n \in \Lambda^n
1281: \C(Y)^*$. Then
1282: $$
1283: d r_{n-1}(f_1\wedge ... \wedge f_n) =
1284: $$
1285: \begin{equation} \label{p106}
1286: \pi_n \Bigl(d \log f_1
1287: \wedge ... \wedge d \log f_n\Bigr)
1288: + 2 \pi i \cdot (r_{n-2}\circ {\rm Res})
1289: (f_1\wedge ... \wedge f_n)
1290: \end{equation}
1291: \end{proposition}
1292:
1293: {\bf Proof}. Let us resolve singularities as in Lemma \ref{Lemma 2.7}.
1294: Since $\pi$ is a birational isomorphism,
1295: and the distribution
1296: $r_{n-1}(f_1 \wedge ... \wedge f_n)$ is determined by its
1297: restriction to the generic point,
1298: one has
1299: $$
1300: \pi_*r_{n-1}( \pi^* f_1 \wedge ... \wedge \pi^* f_n) =
1301: r_{n-1}( f_1 \wedge ... \wedge f_n)
1302: $$
1303: So this and
1304: Lemma \ref{Lemma 2.7} imply that we may assume that $\cup_i {\rm div}f_i$
1305: is a normal crossing divisor. The proposition
1306: follows immediately from the Poincar\'e-Lelong formula (\ref{6.16.04.4})
1307: $$
1308: d(d i
1309: \arg f) = 2 \pi i \delta(f) := 2 \pi i \delta_{{\rm div}(f)}
1310: $$
1311: The Proposition \ref{8.9.02.6} is proved.
1312:
1313: {\bf Remark}. To prove the Poincar\'e-Lelong formula one may
1314: resolve the singularities and argue just as in the proof of Lemma
1315: \ref{Lemma 2.7} that
1316: it is sufficient to prove this formula on a blow up. When the divisor of $f$
1317: is a normal
1318: crossing divisor the formula follows from
1319: $$
1320: d(d i
1321: \arg z) = 2 \pi i \delta(z):= 2 \pi i (\delta_0 - \delta_{\infty}
1322: )$$
1323:
1324:
1325:
1326:
1327:
1328: {\bf 7. The distribution $r_{m-1}(L;H)$}. Let $\Omega_L$ be the canonical
1329: $m$-form in ${\Bbb P}^m - L$ with logarithmic singularities
1330: at $L$. It represents a generator of $H^m_{{\rm DR}}({\Bbb P}^m - L)$ defined over $\Z$. Let us give its coordinate description. Choose homogeneous coordinates
1331: $(z_0:...:z_m)$ in $\PP^m$ such that
1332: $L_i $ is given by equation $\{z_i = 0\}$. Then
1333: $$
1334: \Omega_L = d\log z_1/z_0 \wedge ... \wedge d\log
1335: z_m/z_0
1336: $$
1337: The form $\Omega_L$ has periods in $\Z(m)$.
1338: So
1339: $\pi_m(\Omega_L)$ is exact. However
1340: there is no
1341: canonical choice of a primitive $(m-1)$-form for it: the group
1342: $(\C^{\ast})^m$ acting on $\C \PP^m - L$ leaves the form
1343: invariant and acts non trivially on the
1344: primitives.
1345: But if we consider a simplex $L$ in the affine complex space $\AAA^m$
1346: (or, what is the same, choose an additional hyperplane $H$ in $\C \PP^m$, which should be thought of as the infinite hyperplane) then there
1347: is a {\it canonical} primitive.
1348:
1349:
1350:
1351: Choose a coordinate system $(z_0:...:z_m)$ in $\PP^m$ as above
1352: such that $H$ is given by $\{\sum_{i=1}^{m}
1353: z_i =z_0\}$. Set
1354: \begin{equation} \label{mint4}
1355: r_{m-1}(L;H) := r_{m-1}(z_1/z_0 \wedge ... \wedge z_m/z_0)
1356: \end{equation}
1357: Here is a more invariant definition. Choose one of the faces
1358: of the simplex $L$, say
1359: $L_0$. Consider the simplex $(H, L_1, ... , L_m)$.
1360: Let $f_i$ be the rational function on $\C \PP^m$ such that
1361: $(f_i) = L_i - L_0$ normalized by
1362: $f_i(l_i)= 1$, where $l_i$ is the vertex of the simplex
1363: $(H, L_1, ... , L_m)$ opposite to the face
1364: $L_i$. Then $f_i = \frac{z_i}{z_0}$ and
1365: $$
1366: r_{m-1}(L;H) := r_{m-1}(f_1 \wedge ... \wedge f_m)
1367: $$
1368:
1369: \begin{figure}[ht]
1370: \centerline{\epsfbox{gpol5.eps}}
1371: \caption{A simplex $L$ and an infinite hyperplane $H$}
1372: \label{gpol5}
1373: \end{figure}
1374: This form
1375: is skewsymmetric with respect to the permutation of the
1376: hyperplane faces of the simplex $L$.
1377: One has
1378: $$
1379: d r_{m-1}(L;H) = \pi_m (\Omega_L) \quad \mbox{in} \quad \C \PP^m - L
1380: $$
1381: So a choice of an ``infinite'' hyperplane $H \subset \C
1382: \PP^{m}$ provides the form
1383: $r_{m-1}(L;H)$.
1384:
1385: {\bf Example}. If $m=1$ then
1386: $$
1387: \AAA^1 = \PP^1 - \{1\}, \quad L = \{0\} \cup \{\infty \}, \quad \Omega_L = d\log z, \quad \pi_1 (d\log z) = d\log |z|, \quad
1388: $$
1389: $$
1390: r_0(\{0\} \cup \{\infty \}; \{1\}) = \log |z|
1391: $$
1392:
1393:
1394:
1395:
1396: The $(n-1)$-form $ r_{n-1}(L;H)$ provides
1397: an $(n-1)$-distribution on $\C \PP^n$.
1398: Recall the simplex $\widehat L_i$ which is cuted out
1399: by $L$ in the hyperplane $L_i$, and put
1400: $H_i := L_i \cap H$. Consider the $(n-2)$-form
1401: $r_{n-2}(\widehat L_i; H_i )$ on the hyperplane
1402: $L_i$ as $n$-distribution in $\C \PP^n$. We denote
1403: it as $r_{n-2}(\widehat L_i; H_i ) \cdot \delta_{L_i}$.
1404:
1405:
1406: \begin {corollary} \label{dif}
1407: One has
1408: \begin{equation} \label{1p1}
1409: d r_{n-1}(L; H) = \pi_n ( \Omega_L ) + 2 \pi i \cdot
1410: \sum_{i=0}^n (-1)^ir_{n-2}(\widehat L_i; H_i ) \delta_{L_i}
1411: \end{equation}
1412: \end {corollary}
1413:
1414: {\bf Proof}. Follows immediately from Proposition \ref{8.9.02.6}.
1415:
1416:
1417: {\bf 8. A coordinate free description of the form $r_{m-1}(L;H)$}.
1418: Let $V_m$ be an $m$-dimensional vector space over a field $F$.
1419: Choose a volume form ${\rm vol}_m \in {\rm det}V_m^{\ast}$. Set
1420: $\Delta(v_1,...,v_m):= \langle{\rm vol}_m, v_1\wedge ...\wedge v_m\rangle \in F^{\ast}$.
1421:
1422: \begin {lemma} For a configuration $(l_0,...,l_m)$ of $m+1$ vectors in
1423: generic position
1424: \begin{equation} \label{4p}
1425: f_m(l_0,...,l_m):= \sum_{i=0}^m(-1)^i \Lambda_{j\not = i}\Delta(l_0,...,\widehat l_j,...,
1426: l_i,...,l_m) \in \Lambda^m F^{\ast}
1427: \end{equation}
1428: does not depend on the choice of the volume form ${\rm vol}_m$.
1429: \end {lemma}
1430:
1431: {\bf Proof}. See proof of Lemma 3.1 in [G3].
1432:
1433: For a point $z\in \AAA^m- L$ let $l_i(z)$ be the
1434: vector from
1435: $z$ to the vertex $l_i$, see Figure \ref{gpol4}.
1436:
1437:
1438: \begin{figure}[ht]
1439: \centerline{\epsfbox{gpol4.eps}}
1440: \caption{The vectors $l_i(z)$}
1441: \label{gpol4}
1442: \end{figure}
1443: We get a canonical element
1444: $$
1445: f_m(l_0(z),...,l_m(z)) \in \Lambda^m\Q(\AAA^m - L)^{\ast}
1446: $$
1447: If $F=\C$, applying the homomorphism $r_m$ to this
1448: element we get a canonical $(m-1)$-form in $\C \PP^{m} - L$.
1449: It coincides with $ r_{m-1}(L;H)$.
1450:
1451: {\bf Example}. If $m=1$ and $(L_0, L_1, H)= (0, \infty, 1)$ then $t = \frac{z}{z-1}$
1452: is an affine coordinate on $\PP^1 - \{1\}$ and $$
1453: l_0(t) = \frac{z}{z-1},
1454: l_1(t) = \frac{1}{z-1}, \quad \mbox{so} \quad
1455: f_1(l_0(t), l_1(t)) = \frac{l_0(t)}{l_1(t)} = z
1456: $$
1457:
1458: {\bf Remark}. The map $z\in \AAA^m - L \lms f_m(l_0(z),...,l_m(z))$
1459: provides an isomorphism
1460: $F_m: CH^m({\rm Spec}(F),0) \lra K_m^M(F)$;
1461: see [NS] where the isomorphism $F_m$ was presented in a bit different way.
1462:
1463:
1464:
1465:
1466:
1467:
1468:
1469:
1470:
1471:
1472:
1473: {\bf 9. The main construction}. We have to construct a morphism of complexes
1474: $$
1475: \begin{array}{ccccccccc}
1476: ... &\longrightarrow&{\cal Z}^{1}(X; n)&\longrightarrow &...&\longrightarrow&{\cal Z}^{2n-1}(X; n)&\longrightarrow&{\cal Z}^{2n}(X; n)\\
1477: &&&&&&&&\\
1478: &&\downarrow {\cal P}^{1}(n)&&... &&\downarrow {\cal P}^{2n-1}(n)&&\downarrow
1479: {\cal P}^{2n}(n)\\
1480: &&&&&&&&\\
1481: 0&\lra &{\cal D}^{0,0}_{\R}(n-1)& \stackrel{}{\longrightarrow}&...&\stackrel{}{\longrightarrow}&{\cal D}_{\R}^{n-1,n-1}(n-1)&\stackrel{2 \overline \partial \partial}{\longrightarrow}&{\cal D}_{\R}^{n,n}(n)
1482: \end{array}
1483: $$
1484: Let $Y \in {\cal Z}^{2n}(X; n)$
1485: be a codimension $n$ cycle in $X$. By definition
1486: $$
1487: {\cal P}^{2n}(n)(Y):= (2\pi i)^n\delta_Y%, \quad \mbox{where} \quad <\delta_Y,\omega>:= \int_{Y(\C)} \omega
1488: $$
1489:
1490: Let us construct homomorphisms
1491: $$
1492: {\cal P}^{2n-i}(n): {\cal Z}^{2n-i}(X; n)
1493: \longrightarrow {\cal D}^{2n-i-1}_{X(\C)}(n-1), \quad i>0
1494: $$
1495: Denote by $\pi_{{\AAA}^i}$ (resp. $\pi_{X}$) the projection of
1496: $X \times \AAA^i$ to $\AAA^i$ (resp. $X$), and by
1497: $\overline \pi_{{\AAA}^i}$ (resp. $\overline \pi_{X}$) the projection of
1498: $X \times \C{\Bbb P}^i$ to $\C{\Bbb P}^i$ (resp. $X$).
1499:
1500: Recall the element
1501: \begin{equation} \label{mint3}
1502: \frac{z_1}{z_0} \wedge... \wedge \frac{z_i }{z_0}
1503: \in \Lambda^{i}\C(\AAA^i)^*
1504: \end{equation}
1505: defining the form $r_{i-1}(L;H)$, see (\ref{mint4}). Let
1506: \begin{equation} \label{mint2}
1507: g_1 \wedge... \wedge g_i \in \Lambda^{i}\C(Y)^*
1508: \end{equation}
1509: be the restriction
1510: to $Y$ of the inverse image of element
1511: (\ref{mint3}) by the projection $\pi^*_{\AAA^i}$.
1512: The element (\ref{mint2}) provides, by
1513: Theorem \ref{8.6.02.2}, a distribution on $X(\C) \times \C {\Bbb P}^i$.
1514: Pushing this distribution down by $(2\pi i)^{n-i}\cdot\pi_{\overline X}$ we get
1515: the distribution $ {\cal P}^{2n-i}(n)(Y)$:
1516:
1517: \begin{definition} \label{mint6}
1518: $$
1519: {\cal P}^{2n-i}(n)(Y) := (2\pi i)^{n-i}\cdot
1520: \pi_{\overline X *}r_{i-1}(g_1 \wedge... \wedge g_i) :=
1521: $$
1522: $$
1523: (2\pi i)^{n-i}\cdot \pi_{\overline X *}r_{i-1}\Bigl( i_Y^*\pi^*_{\AAA^i}
1524: (\frac{z_1}{z_0} \wedge... \wedge \frac{z_i }{z_0})\Bigr)
1525: $$
1526: \end{definition}
1527: Here $i_{Y}: Y
1528: \hookrightarrow X \times {\Bbb P}^i$
1529:
1530: {\bf Remark}. This definition works if and only if
1531: the cycle $Y$ has proper intersection with all codimension one faces
1532: of $X \times L$. Indeed, if $Y$ does not have proper intersection
1533: with one of the faces, then the equation of this face restricts to zero
1534: to $Y$, and so (\ref{mint2}) does not make sence.
1535: As soon as all equations of the
1536: codimension one faces restrict to non zero functions on $Y$,
1537: (\ref{mint2}) makes sense, and we can apply Theorem \ref{8.6.02.2}.
1538:
1539:
1540:
1541:
1542:
1543: {\bf Remark}. We just proved that the product of distributions $\delta_Y \wedge \pi_{\AAA}^{\ast}r_{i-1}(L;H)$ makes sense and
1544: $$
1545: {\cal P}^{2n-i}(n)(Y)= \quad (2\pi i)^{n-i}
1546: {\pi_X}_*(\delta_Y \wedge \pi_{\AAA}^{\ast}r_{i-1}(L;H))
1547: $$
1548:
1549: It is handy to rewrite Definition \ref{mint6} more explicitly
1550: as an integral over $Y(\C)$. Namely,
1551: %let $k= 2 {\rm dim}Y - i +1$.
1552: %Observe that
1553: %$$
1554: %2 {\rm dim}X - k = 2({\rm dim}X +i - {\rm dim}Y) - i - 1 = 2n-i-1
1555: %$$
1556: %So $(2n-i-1)$-distributions are functionals on smooth $k$-forms with
1557: %compact support on $X(\C)$.
1558: %Let $\omega$ be such a $k$-form and
1559: let $\omega$ be a smooth form on $X(\C)$ and
1560: $Y \in {\cal Z}^{2n-i}(X; n)$. Then
1561: \begin{equation} \label{mint}
1562: <{\cal P}^{2n-i}(n)(Y),\omega>= \quad (2\pi i)^{n-i}
1563: \int_{Y(\C)}\pi_{\AAA}^{\ast}r_{i-1}(L;H) \wedge \pi_X^{\ast}\omega =
1564: \end{equation}
1565: \begin{equation} \label{8.9.02.10}
1566: (2\pi i)^{n-i}\int_{Y^0(\C)}
1567: r_{i-1}(g_1\wedge ... \wedge g_i)\wedge i_{Y(\C)}^*\pi^*_X\omega
1568: \end{equation}
1569: where $Y^0$ is the nonsingular part of $Y$
1570:
1571:
1572:
1573:
1574:
1575:
1576: Since the form $r_{i-1}(L;H)$ is $\R(i-1)$-valued, for $i>0$ the
1577: distribution
1578: ${\cal P}^{2n-i}(n)(Y)$ takes values in $\R(n-1)$.
1579: Further, ${\cal P}^{2n}(n)(Y)$ is obviously an $\R(n)$-valued distribution.
1580:
1581: Let us show that for $i>0$ the distribution ${\cal P}^{2n-i}(n)(Y)$ lies precisely
1582: in the left bottom $(n-1) \times (n-1)$ square of the Dolbeault
1583: bicomplex. The integral (\ref{mint})
1584: is non zero only if
1585: $\pi^*_{X}\omega \wedge \pi^*_{{\Bbb A}}r_{i-1}(L;H)$ is of type
1586: $$
1587: ({\rm dim}Y, {\rm dim}Y) = ({\rm dim}X+i-n, {\rm dim}X+i-n)
1588: $$
1589: Since $r_{i-1}(L;H)$ is an
1590: $(i-1)$-form we see that the integral vanish if $\omega$ is
1591: a form of type $(p,q)$ where
1592: $p$ or $q$ is smaller then ${\rm dim} X+1-n$. This just means that
1593: the distribution lies in the left bottom $(n-1)\times (n-1)$ square of
1594: the Dolbeault bicomplex.
1595: The proposition is proved.
1596:
1597: Therefore we have constructed the maps ${\cal P}^{i}(n)$.
1598:
1599: \begin{theorem} \label{hhmmoo}
1600: ${\cal P}^{\bullet}(n)$ is a homomorphism of complexes.
1601: \end{theorem}
1602:
1603: {\bf Proof}. One has
1604: \begin{equation} \label{98}
1605: <d {\cal P}^{2n-i}(n)(Y), \omega> =
1606: %= - <{\cal P}^{2n-i}(n)(Y), d\omega> =
1607: %$$
1608: %$$- \int_{\overline Y(\C)}i_{\overline Y}(\overline \pi_X^{\ast}d\omega \wedge
1609: %\overline \pi_{\AAA}^{\ast}r_{i-1}(L;H)) = - \int_{\overline Y(\C)}i_{\overline Y}^*\p%i_X^*d\omega \wedge
1610: %r_{i-1}(g_1\wedge... \wedge g_i) =
1611: %$$
1612: %
1613: \int_{\overline Y(\C)}dr_{i-1}(g_1\wedge
1614: ... \wedge g_i)\wedge
1615: i_{\overline Y}^*\pi_X^*\omega
1616: \end{equation}
1617: where $g_1\wedge
1618: ... \wedge g_i$ is as in (\ref{mint2}).
1619: We use Proposition \ref{8.9.02.6} to calculate
1620: $dr_{i-1}(g_1\wedge... \wedge
1621: g_i)$.
1622:
1623:
1624: To handle the first term in (\ref{p106}) observe that
1625: by the very definition of the complex
1626: ${\cal C}_{\cal D}(X;n)$ we need to investigate integral (\ref{98})
1627: only for smooth forms $\omega$ of type $(p,q)$ where $|p-q| \leq i-1$.
1628: Since $\pi_i( \Omega_L ) = \Omega_L \pm \overline
1629: \Omega_L$ is a sum of forms of type $(i,0)$
1630: and $(0,i)$
1631: the form $\pi_{\AAA}^{\ast}\pi_i(\Omega_L)
1632: \wedge \pi_X^{\ast}\omega$
1633: can not be of type $(k,k)$.
1634: Therefore only the second term in (\ref{p106}) contributes.
1635:
1636:
1637:
1638:
1639: Since $r_0(f) = \log|f|$, the commutativity of the last square
1640: follows from the Poincar\'e-Lelong formula
1641: $
1642: 2 \overline \partial \partial \log |f| = 2 \pi i \cdot \delta_{{\rm div}(f)}
1643: $.
1644:
1645:
1646:
1647: {\it Commutativity of the $i$-th square of the diagram,
1648: $i>1$, counting from the right}.
1649: If $\overline Y$ is normal this
1650: follows from Proposition \ref{8.9.02.6}. Indeed,
1651: since $Y$ meets the codimension two faces properly all equations of the codimension one
1652: faces of $X \times L$ but $z_j$ have non zero restriction
1653: to the generic point of ${\rm div}z_j$. Therefore we may use formula
1654: (\ref{8.12.02.100}) to calculate ${\rm Res}$, and then the claim is obvious.
1655:
1656: In the case when $\overline Y$ is not normal
1657: we face the following subtle problem.
1658:
1659: Calculating
1660: $dr_{i-1}(g_1\wedge... \wedge
1661: g_i)$ and hence $d \circ {\cal P}^{2n-i}(n)(Y)$ when $Y$ is not
1662: normal we need to take
1663: ${\rm Res}_{\widetilde Z}(\widetilde g_1\wedge... \wedge
1664: \widetilde g_i)$ for all irreducible divisors $\widetilde Z$ in the normalization
1665: $\pi: \widetilde Y \to Y$, where $\widetilde g:= \pi^*g$,
1666: and then take
1667: \begin{equation} \label{8.9.02.7}
1668: \sum_{\widetilde Z} \pi_*r_{i-2}\circ {\rm Res}_{\widetilde Z}
1669: (\widetilde g_1\wedge... \wedge
1670: \widetilde g_i)
1671: \end{equation}
1672: Computation of ${\cal P}^{2n-i+1}(n)(Z)\circ d $
1673: does not involve the normalization of $Y$: we intersect
1674: $Y$ with all codimension one faces of $X \times L$.
1675: So we need to compute (\ref{8.9.02.7}) using the intersection data of $Y$ and
1676: $X \times L$.
1677:
1678:
1679:
1680: To handle this
1681: we use the condition that $Y$ meets the codimension two faces of
1682: the simplex $X \times L$ properly. It implies that
1683: \begin{equation} \label{8.9.02.20}
1684: {\rm Res}_{\widetilde Z}(\widetilde g_1\wedge... \wedge
1685: \widetilde g_i) = \pi^*G_{i-1}; \qquad
1686: G_{i-1}\in \Lambda^{i-1}\C(\pi(\widetilde Z))^*
1687: \end{equation}
1688: Indeed, since all equations of the codimension one
1689: faces of $X \times L$ but $z_j$ have non zero restriction
1690: to the generic point of ${\rm div}z_j$, the wedge
1691: product of these restrictions can be taken as $G_{i-1}$.
1692:
1693: Having (\ref{8.9.02.20}) the statement is obvious since
1694: $$
1695: \pi_* r_{i-2} \pi^*G_{i-1} = [\C(\widetilde Z): \C(\pi (\widetilde Z))]
1696: \cdot r_{i-2}(G_{i-1})
1697: $$
1698: Indeed, recall (see [F], page 9) that if $\widetilde Y \to Y$ is the
1699: normalization of $Y$ and $g \in \C(Y)^* = \C(\widetilde Y)^*$ then
1700: $$
1701: {\rm ord}_Z(g) = \sum_{\widetilde Z}
1702: {\rm ord}_{\widetilde Z}(g)[\C(\widetilde Z): \C(Z)]
1703: $$
1704: where the sum is over all irreducible
1705: divisors projecting onto $Z$.
1706:
1707: Theorem \ref{hhmmoo} is proved. Therefore we finished the proof of
1708: Theorem-Construction
1709: \ref{6.11.02.1}.
1710:
1711:
1712:
1713: {\bf 10. The Higher Arakelov Chow groups}. Let $X$ be a regular complex variety.
1714: Denote by $\widetilde C^{\bullet}_{{\cal D}}(n)$ the quotient of the complex $C^{\bullet}_{{\cal D}}(n)$ along the subgroup ${\cal A}_{cl}^{n,n}(n) \subset
1715: {\cal D}_{cl}^{n,n}(n) $ of closed smooth form of type $(n,n)$ with values in $\R(n)$.
1716:
1717: Consider the cone of the homomorphism ${\cal P}^{\bullet}(n)$ shifted by $-1$:
1718: $$
1719: \widehat {\cal Z}^\bullet(X; n):= {\rm Cone}
1720: \Bigl( {\cal Z}^\bullet(X; n)
1721: \stackrel{ }{\longrightarrow} \widetilde C^{\bullet}_{{\cal D}}(X(\C); n)
1722: \Bigr)[-1]
1723: $$
1724:
1725: \begin{definition} \label{arackd} The Higher Arakelov Chow groups are
1726: \begin{equation} \label{6.11.02.11}
1727: \widehat {CH}^n(X; i):= H^{2n-i}(\widehat {\cal Z}^{\bullet}(X; n))
1728: \end{equation}
1729: \end{definition}
1730:
1731:
1732:
1733: Recall the arithmetic Chow groups defined by Gillet-Soule [GS] as follows:
1734: $$
1735: \widehat {CH}^n(X):=
1736: $$
1737: \begin{equation} \label{6.11.02.10}
1738: \frac{\{(Z,g); \frac { \overline \partial \partial}{\pi i} g+\delta_Z
1739: \in {\cal A}^{n,n}\}}{\{(0,\partial u + \overline \partial v); ({\rm div}
1740: f, -\log|f|), f \in \C(Y),
1741: {\rm codim} (Y) = n-1\}}
1742: \end{equation}
1743: Here $Z$ is a divisor in $X$, $f$ is a rational function on a divisor $Y$ in $X$,
1744: $$
1745: g \in {\cal D}_{\R}^{n-1,n-1}(n-1), \quad (u,v) \in ({\cal D}^{n-2,n-1}
1746: \oplus {\cal D}^{n-1,n-2})_{\R}(n-1)
1747: $$
1748:
1749:
1750:
1751: \begin{proposition} \label{ch}
1752: $\widehat {CH}^n(X; 0) = \widehat {CH}^n(X)$.
1753: \end{proposition}
1754:
1755:
1756: {\bf Proof}. Let us look at the very right part of the complex
1757: $\widehat {\cal Z}^{\bullet}(X; n)$:
1758: $$
1759: \begin{array}{ccccc}
1760: ...&\longrightarrow&{\cal Z}^{2n-1}(X; n)&\longrightarrow&{\cal Z}^{2n}(X; n) \\
1761: &&&&\\
1762: &&\downarrow {\cal P}^{2n-1}(n)&& \downarrow {\cal P}^{2n}(n)\\
1763: &&&&\\
1764: ({\cal D}^{n-2,n-1} \oplus {\cal D}^{n-1,n-2})_{\R}(n-1)&\stackrel{(\partial , \overline \partial)}{\longrightarrow}&{\cal D}_{\R}^{n-1,n-1}(n-1)&\stackrel{2\overline \partial \partial}{\longrightarrow}&{\cal D}_{\R}^{n,n}(n)/{\cal A}_{\R}^{n,n}(n)
1765: \end{array}
1766: $$
1767: Consider the very end of the Gersten complex on $X$:
1768: $$
1769: \prod_{Y \in X_{n-2}}\Lambda^2\C(Y)^* \stackrel{\partial}{\longrightarrow} \prod_{Y \in X_{n-1}}\C(Y)^* \longrightarrow {\cal Z}_0(X; n)
1770: $$
1771: where $\partial$ is the tame symbol. It maps to the complex
1772: $\widehat {\cal Z}^{\bullet}(X; n)$, i.e. to the top row of the bicomplex above,
1773: as follows. Recall that ${\cal Z}_0(X; n) = {\cal Z}^{2n}(X; n)$, so the very right component of our map is provided by this identification.
1774: Further, a pair $(Y;f)$ where $Y$ is an irreducible codimension $n-1$ subvariety of $X$ maps to the cycle $\{(y,f(y))| y \in Y\} \subset X \times \AAA^1$.
1775: Similarly any element in $\Lambda^2\C(Y)^*$ can be represented as a linear combination
1776: of elements
1777: $\sum_i (Y;f_i\wedge g_i)$ where $Y$ is an irreducible codimension $n-2$ subvariety of $X$ and $f_i, g_i$ are rational functions on $Y$ such that ${\rm div} f_i$ and
1778: ${\rm div} g_i$ share no irreducible divisors. Then we send $(Y;f_i\wedge g_i)$ to
1779: the cycle $(y,f_i(y),g_i(y)) \subset X \times \AAA^2$.
1780: It is well known that in this way we get an isomorphism on the
1781: last two cohomology groups.
1782: Computing the composition of this map with the homomorphism
1783: ${\cal P}^{\bullet}(n)$ we end up precisely with the denominator
1784: in (\ref{6.11.02.10}). The proposition is proved.
1785:
1786:
1787:
1788:
1789: \section { The Chow polylogarithms}
1790:
1791: Suppose $X = {\rm Spec}(\C)$. Then ${\cal P}_n(Y): = {\cal P}^{1}(n)(Y)$ is a function on the space of all codimension $n$ cycles in $\PP^{2n-1}$ intersecting properly faces of the simplex $L$. It is called
1792: the {\it Chow polylogarithm function}. For $i>1$ all the distributions ${\cal P}^{i}(n)(Y)$ are zero. However modifying
1793: the construction of the previous chapter we get a very interesting object, the Chow polylogarithm, even when $X$ is a point. The Chow polylogarithm function is the
1794: first component of the Chow polylogarithm. One can define the Chow
1795: polylogarithm for an arbitrary variety $X$, but we spell out the details
1796: in the most
1797: interesting case when $X$ is a point.
1798:
1799:
1800: {\bf 1 Chow polylogarithms [G5]}. Let $L = (L_0, ..., L_{p+q})$
1801: be a simplex in $\PP^{p+q}$,
1802: $H$ is a hyperplane in generic position to $L$, and $H_i := H \cap L_i$.
1803:
1804: Let $ {\cal Z}^{q}_{p}(L)$ be the variety of all codimension $q$
1805: effective algebraic cycles in
1806: $\C \PP^{p+q}$ which intersect properly, i.e. each irreducible component in the right codimension,
1807: all faces of the simplex $L$.
1808: It is a union of an infinite number of finite
1809: dimensional complex algebraic varieties.
1810:
1811: {\bf Example}.
1812: $ \C \PP^n - L = (\C^{\ast})^n$ is an irreducible component of
1813: $ {\cal Z}^{n}_{0}(L)$ parametrizing the irreducible subvarieties, i.e. points.
1814:
1815: Let $\widehat L_i$ be the simplex in the projective space $L_i$ cut by the hyperplanes $L_j$, $j \not = i$.
1816: The intersection of a cycle with a codimension
1817: 1 face $L_i$
1818: of the simplex $L$ provides a map
1819: $$
1820: a_i: {\cal Z}^{q}_{p}(L) \longrightarrow {\cal Z}^{q}_{p-1}(\widehat L_i),
1821: \qquad 0 \leq i \leq p+q
1822: $$
1823:
1824: Let $l_j$ be the vertex opposite to the face $L_j$.
1825: Consider an open part ${\cal Z}^{q}_{p}(L)^0$ of ${\cal Z}^{q}_{p}(L)$
1826: parametrising the cycles $C$ such that projection with the center $l_j$ sends $C$ to a codimension $q-1$ cycle. Then
1827: projection with the center at the vertex $l_j$ of $L$
1828: defines a map
1829: $$
1830: b_j: {\cal Z}^{q}_{p}(L)^0 \longrightarrow
1831: {\cal Z}^{q-1}_{p}(\widehat L_j), \qquad 0 \leq i \leq p+q
1832: $$
1833:
1834:
1835:
1836: \begin{theorem-construction} \label{chw}
1837: For given $q \geq 0$ there is an
1838: explicitly constructed chain of $(q-p-1)$-distributions $\omega_{p}^q =
1839: \omega_{p}^q(L; H)$ on ${\cal Z}^{q}_{p}(L)$ such that
1840:
1841: \begin{equation} \label{i}
1842: i)\qquad d \omega^q_{0}(L,H) = \pi_q( \Omega_L)
1843: \end{equation}
1844: %r({\Bbb A}^{q-1};\Delta_{q-1})
1845: \begin{equation} \label{ii}
1846: ii)\quad d\omega_{p}^q(L; H) = \sum_{i=0}^{p+q}(-1)^ia_i^{\ast}\omega_{p-1}^q(L; H_i)
1847: \end{equation}
1848: \begin{equation} \label{iii}
1849: iii)\qquad \sum_{j=0}^{p+q+1}(-1)^j b_j^{\ast}\omega_{p}^q(L; H) = 0
1850: \end{equation}
1851: The restriction of $\omega_{p}^q$ to the subvariety $\widehat
1852: {\cal Z}^{q}_{p}(L)$ of smooth cycles
1853: in generic position with respect to the simplex $L$ is a real-analytic
1854: differential $(q-p-1)$ form.
1855: \end{theorem-construction}
1856:
1857: For a given positive integer
1858: $q$ the collection $\{\omega_{p}^q\}$ is called the {\it $q$-th Chow
1859: polylogarithm}.
1860:
1861: The varieties $ {\cal Z}^{q}_{p}(L)$ for $ p\geq 0$ form a truncated simplicial variety
1862: $ {\cal Z}^{q}_{\bullet}(L)$. The conditions i) and ii) just mean that
1863: the sequence of forms $\omega_{p}^q$ is
1864: a $2q$-cocycle in
1865: the complex computing the Deligne cohomology $H^{2q}(
1866: {\cal Z}^{q}_{\bullet}(L), \R_{\cal D}(q))$.
1867:
1868:
1869:
1870:
1871: {\bf Proof}.
1872: We define
1873: $\omega^q_p$ as the Radon transform of the distribution
1874: $r_{p+q-1}(L;H)$ in $\C \PP^{p+q}$ over the family of cycles $Y_{\xi}$
1875: parametrized by
1876: $ {\cal Z}^q_p(L)$. This means the following. Consider the incidence variety:
1877: $$
1878: \Gamma_p := \{(x,\xi) \in \C\PP^{p+q} \times {\cal Z}^q_p(L)(\C)
1879: \quad \mbox{such that} \quad x \in Y_{\xi}\}
1880: $$
1881: where $Y_{\xi}$ is the cycle in $\C \PP^{p+q}$ corresponding to $\xi
1882: \in {\cal Z}^q_p(L)$. We get a double bundle
1883: $$
1884: \begin{array}{ccccc}
1885: && \Gamma_p \subset \C \PP^{p+q} \times {\cal Z}^q_p(L)(\C)&&\\
1886: &&&&\\
1887: &\pi_1\swarrow&&\searrow \pi_2&\\
1888: &&&&\\
1889: \C \PP^{p+q}&&&& {\cal Z}^q_p(L)(\C)
1890: \end{array}
1891: $$
1892: Then
1893: $$
1894: \omega_p^q:= {\pi_2}_{\ast} {\rm Res}_{\Gamma_p} \pi_1^{\ast} (2\pi i)^{-q}r_{p+q}(L;H)
1895: $$
1896: Observe that $r_{p+q}(L;H)$ is a distribution on $\C \PP^{p+q}$,
1897: and hence
1898: $\pi_1^{\ast} r_{p+q}(L;H)$ is a distribution on
1899: $\C \PP^{p+q} \times {\cal Z}^q_p(L)(\C)$. The fact that this
1900: distribution can be restricted to
1901: $\Gamma_p$ is a version of Theorem \ref{8.6.02.2}, and is proved in the same way.
1902: The push forward ${\pi_2}_{\ast}$ of this
1903: distribution is well defined since $\pi_2$ is a proper map.
1904:
1905: The property i) is true by the very definition.
1906:
1907: \begin {lemma} \label{2.4}
1908: $\sum_{j=0}^{n+1}(-1)^j b_j^{\ast}\omega^n_0 = 0$.
1909: \end {lemma}
1910:
1911: {\bf Proof}. Let $s(z_0,...,z_n):= z_1/z_0 \wedge ...
1912: \wedge z_{n}/z_0$. The lemma follows from the identity
1913: $$
1914: \sum_{j=0}^{n+1}(-1)^j s(z_0,...,\widehat {z_j},..., z_{n+1}) = 0
1915: $$
1916: So we have iii).
1917: To check ii) observe that the push forward ${\pi_2}_{\ast}$ of
1918: distributions commutes with the De Rham differential.
1919: The theorem is proved.
1920:
1921:
1922:
1923:
1924:
1925: {\bf 2. Properties of the Chow polylogarithm function}.
1926: The function
1927: ${\cal P}_q:= \omega_{q-1}^q$ on $ {\cal Z}^{q}_{q-1}(L)$
1928: is called the {\it Chow $q$--logarithm function}.
1929: It satisfies
1930: two functional
1931: equations:
1932: $$
1933: \sum_{i=0}^{2q}(-1)^ia_i^{\ast}{\cal P}_q =0 \qquad \sum_{j=0}^{2q}(-1)^j
1934: b_j^{\ast}{\cal P}_q = 0
1935: $$
1936:
1937: \begin{theorem} \label{6.10.04.1q}
1938: The Chow polylogarithm function is invariant under the natural
1939: action of the torus
1940: $(\C^*)^{p+q}$ on ${\cal Z}^q_p(\C)$. In particular it does not
1941: depend on the choice of the hyperplane $H$.
1942: \end{theorem}
1943:
1944: {\bf Remark}. The statements of Theorem \ref{6.10.04.1q} are no longer true
1945: for the forms $\omega_{p}^q$ for
1946: $p<q-1$.
1947:
1948: Here is a reformulation of Theorem \ref{6.10.04.1q}.
1949: \begin {theorem}
1950: \label {0.2}
1951: Suppose that $dimX=n$ and $f_1,...,f_{2n+1
1952: }$ are rational functions on $X$. Then
1953: the integral
1954: $$
1955: (2\pi i)^{1-n}\int_{X(\C)}r_{2n}(f_1,...,f_{2n+1})
1956: $$
1957: does not change if we multiply one of the
1958: functions $f_i$ by a non zero constant.
1959: \end {theorem}
1960:
1961: {\bf Proof}. Multiplying, say, $f_1$ by $\lambda$ we see that the difference
1962: between the two integrals is
1963: \begin{equation} \label{a6}
1964: \log|\lambda| \sum_k a_k \int_{X(\C)} {\rm Alt}_{2n} d\log|f_2|\wedge ... \wedge
1965: d\log|f_{2k-1}| \wedge d\arg f_{2k} \wedge ... \wedge d\arg f_{2n+1}
1966: \end{equation}
1967: where the $a_k$ are some rational constants (easily computable from
1968: (\ref{1wq})). We will prove that for each $k$ the corresponding integral in this sum
1969: is already zero. Using the identity
1970: $$
1971: (d\log|f_2| + i d \arg f_2) \wedge ... \wedge (d\log|f_{2n+1}| + i d \arg
1972: f_{2n+1}) = 0
1973: $$
1974: we can rewrite the integral
1975: $$
1976: \int_{X(\C)} d\arg f_2\wedge ... \wedge d\arg f_{2n+1}
1977: $$
1978: as a sum of similar integrals containing $d\log |f_i|$. Our statement
1979: follows from
1980: \begin {proposition} \label{5.2}
1981: \label {0.2b} Suppose that $dimX=n$. Then
1982: $$
1983: d {\rm Alt}_{2n}\Bigl(\log|f_2|d\log|f_3|\wedge ... \wedge
1984: d\log|f_{2k-1}| \wedge d\arg f_{2k} \wedge ... \wedge d\arg f_{2n+1}\Bigr) =
1985: $$
1986: \begin{equation} \label{d6}
1987: {\rm Alt}_{2n} \Bigl( d\log|f_2| \wedge ... \wedge
1988: d\log|f_{2k-1}| \wedge d\arg f_{2k} \wedge ... \wedge d\arg f_{2n+1}\Bigr)
1989: \end{equation}
1990: in the sense of distributions.
1991: \end {proposition}
1992:
1993: {\bf Proof}. Since $d di \arg f = 2\pi i \delta(f) $, the left hand
1994: side is equal to the right hand side {\it
1995: plus} the following terms concentrated on the divisors $f_{2k+j} = 0$:
1996: $$
1997: 4\pi\cdot2(n-k+1) {\rm Alt}_{2n} \Bigl( \delta(f_{2n+1})\log|f_2|d\log|f_3| \wedge ... \wedge
1998: d\log|f_{2k-1}| \wedge d\arg f_{2k} \wedge
1999: ... \wedge d\arg f_{2n}\Bigr)
2000: $$
2001: However all these additional terms vanish thanks to the following
2002: proposition.
2003:
2004:
2005: \begin {proposition} \label{5.3}
2006: \label {0.2a}
2007: Suppose that $dim X = n$. Then for each $0 \leq j \leq n-1$ one has
2008: \begin{equation} \label{b6}
2009: {\rm Alt}_{2n}\Bigl(d \log|f_1|
2010: \wedge ... \wedge d\log|f_{2j+1}| \wedge d\arg f_{2j+2}\wedge ... \wedge
2011: d\arg f_{2n} \Bigr) =0
2012: \end{equation}
2013: \begin{equation} \label{c6}
2014: {\rm Alt}_{2n}\Bigl(d \log|f_1|
2015: \wedge ... \wedge d\log|f_{2j}| \wedge d\arg f_{2j+1}\wedge ... \wedge
2016: d\arg f_{2n} \Bigr) =
2017: \end{equation}
2018: $$
2019: \frac{(2n)! {n \choose j}}{{2n \choose 2j}}\cdot d \log|f_1|
2020: \wedge ... \wedge d\log|f_{2n}|
2021: $$
2022: \end {proposition}
2023:
2024:
2025: {\bf Proof}. The idea is this.
2026: One has $n$ equations
2027: $$
2028: d\log f_{1} \wedge ... \wedge d\log f_{2n} =0
2029: $$
2030: $$
2031: d\log |f_{1}| \wedge d\log f_{2} \wedge ... \wedge d\log f_{2n} =0
2032: $$
2033: \begin{equation}
2034: ............
2035: \end{equation}
2036: $$
2037: d\log |f_{1}|\wedge ... \wedge d\log |f_{n-1}| \wedge d\log f_{n}
2038: \wedge ... \wedge d\log f_{2n} = 0
2039: $$
2040: Taking imaginary part of each of them and alternating $f_1,...,f_{2n}$ we
2041: get $n$ linear equations. Solving them we get the proposition.
2042: See the details in the Appendix in [GZ].
2043:
2044:
2045:
2046: \section{The Grassmannian polylogarithms}
2047:
2048:
2049:
2050: {\bf 1. Configurations of vectors and Grassmannians: a dictionary}.
2051: Let $G$ be a group. Let $X$ be a $G$-set.
2052: We define {\it configurations of $m$ points of} $X$ as $G$-orbits in $X^m$.
2053:
2054: {\bf Example 1}. If $X:= V$ is a vector space and $G:= GL(V)$ we get
2055: configurations of vectors
2056: in $V$.
2057: A configuration of vectors $(l_1,...,l_m)$ is in generic position
2058: if each $k\leq {\rm dim}V$ of
2059: the vectors are linearly independent.
2060:
2061: {\bf Example 2}. If $X:= \PP(V)$ is a projective space and $G:= PGL(V)$ we get
2062: configurations $(x_1,...,x_m)$ of $m$ points in $\PP(V)$.
2063: A configuration of points is in generic position if each $k\leq {\rm dim}V$ of
2064: them generate a plane of dimension $k-1$.
2065:
2066: Let $T_{p+q}$ be the quotient of the torus ${\Bbb G}_m^{p+q+1}$
2067: by the diagonal subgroup ${\Bbb G}_m = (t, ..., t)$. Below $V_n$ denotes a vector space of dimension $n$.
2068:
2069: \begin{lemma-construction} \label{6.12.02.5}
2070: i) There are canonical isomorphisms between the following sets of geometric
2071: objects:
2072:
2073: (a) Configurations of $p+q+1$ vectors in generic position in $V_q$.
2074:
2075: (b) Isomorphism classes of triples
2076: $\{$a projective space ${\PP}^{p+q}$ together with a
2077: simplex $L$, an ``infinite'' hyperplane $H$ in generic position to $L$, and a $p$-dimensional plane in generic position to $L$ (but not necessarily to $H$) $\}$.
2078:
2079: (c) Isomorphism classes of triples
2080: $\{$a vector space $V_{p+q+1}$, a basis $(e_0,...,e_{p+q})$ of
2081: $V_{p+q+1}$, and a $p+1$-dimensional subspace of $V_{p+q+1}$ in generic position with respect to the coordinate hyperplanes$\}$.
2082:
2083: ii) The torus $T_{p+q}$ acts naturally, and without fixed points,
2084: on each of the objects
2085: a), b), c), and the isomorphisms above are compatible with this action.
2086: \end{lemma-construction}
2087:
2088: {\bf Proof}. i) (a) $ \rightarrow$ (c).
2089: For the $(p+1)$-dimensional subspace, take the kernel of the linear map from
2090: $V_{p+q+1}$ to $V_q$ sending $e_i$ to $l_i$.
2091:
2092: (c) $ \rightarrow$ (a). Take the quotient of $V_{p+q+1}/h$ along the given subspace $h$ and consider the images of the vectors $(e_0,...,e_{p+q})$ there.
2093:
2094: (c) $ \rightarrow$ (b). Let $\PP^{p+q}:= \PP(V_{p+q+1})$. Let
2095: ${\AAA}^{p+q}$ be the affine hyperplane in $V_{p+q+1}$
2096: passing through the ends of the basis vectors $e_i$.
2097: Then ${\AAA}^{p+q} \subset \PP^{p+q}$.
2098: The coordinate hyperplanes in $V_{p+q+1}$ provide a simplex
2099: $L_{p+q} \subset {\PP}^{p+q}$.
2100: The projectivization of a generic $(p+1)$-dimensional
2101: subspace $h$ in $V_{p+q+1}$ gives a
2102: $p$-plane $\overline h$ in generic position with respect to this simplex.
2103: (Notice that we do not impose any condition on the mutual
2104: location of $H$ and $\overline h$. For instance $\overline h$ may be inside of $H$.)
2105:
2106: (b) $\rightarrow$ (c). The triple $(\PP^{p+q}, H, L)$ provides a unique up to an isomorphism data
2107: $(V_{p+q+1}, (e_0,...,e_{p+q}))$. Namely, the partial data $(\PP^{p+q}, H)$
2108: provides us with $(V_{p+q+1}, \widetilde H)$ where $\widetilde H$ is the subspace of
2109: $V_{p+q+1}$ projecting to $H$. Now the vertices $l_i$ of the simplex $L$ provide coordinate lines $\widetilde l_i$ in
2110: $V_{p+q+1}$. Intersecting these coordinate lines with a parallel shift of the subspace $\widetilde H$ we get a point on each of the coordinate lines. By definition the endpoints of the basis vectors $e_i$ are these points. Taking the subspace $\widetilde h$ in $V_{p+q+1}$ projecting to a given $p$-plane $h$ in
2111: $\PP^{p+q}$ we get the desired correspondence.
2112:
2113:
2114:
2115:
2116: ii) The torus $T_{p+q}$ acts on the
2117: configurations of vectors in a)
2118: as $$
2119: (t_1, ..., t_{p+q+1}):
2120: (l_1, ..., l_{p+q+1}) \lms (t_1l_1, ..., t_{p+q+1} l_{p+q+1})
2121: $$
2122: The torus $T_{p+q}$ is identified with $\PP^{p+q} - L$ in b),
2123: and so acts naturally on the data in b). The action on the data in c) is similar.
2124: The lemma is proved.
2125:
2126: If we use the description c) for the Grassmannians then
2127: $b_j$ is obtained by factorization along the coordinate
2128: axis $\langle e_j \rangle$.
2129:
2130:
2131:
2132: {\bf 2. The Grassmannian and bi-Grassmannian polylogarithms}.
2133: Let us fix a positive integer $q$.
2134: The operations $a_i$ and $b_j$ from Section 3.1 transform planes to planes. So we get
2135: the following diagram of varieties called the
2136: bi-Grassmannian $\widehat G(q)$:
2137: $$
2138: \begin{array}{cccccccccccc}
2139: &&&&&&\downarrow...\downarrow&&\downarrow...\downarrow\\
2140: &&&&&&&&&\\
2141: \widehat G(q):=&&&&&\stackrel{\rightarrow}{\stackrel{...}{\rightarrow}}&\widehat G_1^{q+1}&\stackrel{\rightarrow}{\stackrel{...}{\rightarrow}}&\widehat G_0^{q+1}\\
2142: &&&&&&&&&\\
2143: &&&&\downarrow...\downarrow&&\downarrow...\downarrow&&\downarrow...\downarrow\\
2144: &&&&&&&&&\\
2145: &&...&\stackrel{\rightarrow}{\stackrel{...}{\rightarrow}}&\widehat G_2^{q}&\stackrel{\rightarrow}{\stackrel{...}{\rightarrow}}&\widehat G_1^{q}&\stackrel{\rightarrow}{\stackrel{...}{\rightarrow}}&\widehat G_0^{q}\\
2146: \end{array}
2147: $$
2148: Here the horizontal arrows are the maps $a_i$ and the vertical ones
2149: are $b_j$.
2150:
2151: {\bf Remark}. The bi-Grassmannian $\widehat G(n)$ is not a (semi)bisimplicial
2152: scheme. (It is a truncated semi{\it hypersimplicial} scheme.
2153: See s.2.6 in [G4]).
2154:
2155: {\it Configurations of hyperplanes and torus quotients of Grassmannians}.
2156: Let $\widehat G_p^q$ be the
2157: Grassmannian of $p$-planes in ${\Bbb P}^{p+q}$
2158: in generic position with respect to a given simplex $L$.
2159:
2160: Taking the $T_{p+q}$--orbits of the objects
2161: a) and b) in the lemma we arrive at
2162:
2163:
2164: \begin{corollary} \label{6.13.02.1}
2165: There is a
2166: bijective correspondence
2167: $$
2168: \widehat G^q_p/T_{p+q}
2169: <-> \left \{ \mbox{Configurations of $ p+q+1$ generic hyperplanes in } \PP^{p}\right \}
2170: $$
2171: sending a $p$--plane $h$ to the configuration $
2172: (h \cap L_0,...,h \cap L_{p+q})$ in $h$.
2173: \end{corollary}
2174:
2175: Let $\psi^q_{p}(q)$ be the restriction of the differential form
2176: $\omega_{p}^q$ to $\widehat G^{q}_p$. The properties i), ii) from Theorem \ref{chw}
2177: are exactly the defining conditions for the single--valued
2178: Grassmannian polylogarithm whose existence was conjectured in
2179: [HM], [BMS], see also [GGL].
2180:
2181: Let us extend these forms by zero
2182: to the other rows of the bi-Grassmannian $\widehat G(q)$, i.e.
2183: set $\psi^{q+i}_{p}(q) =0$ if $i>0$. Then the
2184: property iii) from Theorem \ref{chw} guarantees that the forms $\psi^{q+i}_{p}(q)$
2185: form a $2q$-cocycle in
2186: the bicomplex computing the Deligne cohomology $H^{2q}(\widehat
2187: G(q)_{\bullet}, \R(q)_{\cal D})$. It is called {\it the
2188: bi-Grassmannian $q$-logarithm}.
2189: ( [G5]).
2190:
2191:
2192:
2193: A sequence of {\it
2194: multivalued analytic } forms on Grassmannians satisfying conditions
2195: similar to i), ii) was defined in [HaM1], [HaM2].
2196: Another construction of the multivalued analytic Grassmannian
2197: polylogarithms was suggested in [G5] in the more general setting
2198: of the multivalued Chow polylogarithm.
2199:
2200:
2201:
2202:
2203:
2204: {\bf 3. The Grassmannian $n$-logarithm function}.
2205: By Theorem \ref{0.2} the Chow polylogarithm function is invariant under the action
2206: of the torus $(\C^*)^{2n-1}$. So restricting it to the open Grassmannian $\widehat G^n_{n-1} \subset
2207: \widehat {\cal Z}^n_{n-1}$ and using the bijection
2208: $$
2209: \mbox{ $\{ (n-1)$--planes in
2210: $\PP^{2n-1}$ in generic position with respect to a simplex $L$\}}/({\Bbb G}_m^*)^{2n-1} \quad
2211: $$
2212: $$ <--> \quad \left \{ \mbox{Configurations of} \quad 2n \quad\mbox {generic
2213: hyperplanes in } \PP^{n-1}\right \}
2214: $$
2215: we get a function on the configurations of $2n$ hyperplanes in $\C \PP^{n-1}$, called
2216: the Grassmannian polylogarithm function ${\cal L}^G_n$.
2217:
2218: The Grassmannian polylogarithm function has the following simple description on
2219: the language of configurations of hyperplanes. It is intresting that in this description we can
2220: work with {\it any} configuration of $2n$
2221: hyperplanes, assuming nothing about their mutual location.
2222:
2223: Let $h_1,...,h_{2n}$ be $2n$ arbitrary hyperplanes in $\C \PP^{n-1}$.
2224: Choose an additional hyperplane $h_0$. Let $f_i$ be a
2225: rational function on $\C \PP^{n-1}$ with divisor $(h_i) - (h_0)$. It is
2226: defined up to a scalar
2227: factor.
2228: Set
2229: $$
2230: {\cal L}^G_n(h_1,...,h_{2n}):= (2\pi i)^{1-n}\int_{\C \PP^{n-1}}r_{2n-2}
2231: (\sum_{j=1}^{2n}(-1)^j f_1\wedge ...
2232: \wedge \widehat f_j \wedge ... \wedge f_{2n})
2233: $$
2234: It is skewsymmetric by definition.
2235: Notice that
2236: $$
2237: \sum_{j=1}^{2n}(-1)^j f_1\wedge ...
2238: \wedge \widehat f_j \wedge ... \wedge f_{2n} = \frac{f_1}{f_{2n}} \wedge
2239: \frac{f_2}{f_{2n}} \wedge ... \wedge \frac{f_{2n-1}}{f_{2n}}
2240: $$
2241: So we can define ${\cal L}^G_n(h_1,...,h_{2n})$ as follows: choose rational
2242: functions $g_1,...,g_{2n-1}$ such that ${\rm div} g_i = (h_i) - (h_{2n})$ and put
2243: $$
2244: {\cal L}^G_n(h_1,...,h_{2n}) = (2\pi i)^{1-n}
2245: \int_{\C \PP^{n-1}}r_{2n-2}(g_1,...,g_{2n-1})
2246: $$
2247:
2248: {\bf Remark}. The function ${\cal L}^G_n$ is defined on the set of all configurations of $2n$ hyperplanes in $\C\PP^{n-1}$. However it is
2249: not even continuous on this set. It is real analytic on the submanifold of generic
2250: configurations. Since we put no restrictions on the hyperplanes $h_i$
2251: the following theorem is stronger than Theorem \ref{chw} in the case of
2252: linear subvarieties.
2253:
2254: \begin {theorem}
2255: \label {0.3a} The function ${\cal L}^G_n$ has the following properties:
2256:
2257: a) It does not depend on the choice of hyperplane $h_0$.
2258:
2259: b) For any $2n+1$ hyperplanes $(h_1, ..., h_{2n+1})$ in $\C \PP^{n}$ one has
2260: \begin{equation} \label{4}
2261: \sum_{j=1}^{2n+1}(-1)^j {\cal L}^G_n(h_j \cap h_1,..., \widehat {h_j \cap h_j}, ..., h_j \cap h_{2n+1}) =0
2262: \end{equation}
2263:
2264: c) For any $2n+1$ hyperplanes $(h_1, ..., h_{2n+1})$ in $\C \PP^{n-1}$ one has
2265: \begin{equation} \label{5}
2266: \sum_{j=1}^{2n+1}(-1)^j {\cal L}^G_n(h_1,...,\widehat h_j,...,h_{2n+1})=0
2267: \end{equation}
2268: \end {theorem}
2269:
2270:
2271:
2272: {\bf Proof}. a) Choose another hyperplane $ h_0'$. Take a rational
2273: function $f_0$ with divisor $(h_0') - (h_0)$. Set $f_i' = \frac{f_i}{f_0}$. Then
2274: $$
2275: \sum_{j=1}^{2n+1}(-1)^j f_1\wedge ...
2276: \wedge \widehat f_j \wedge ... \wedge f_{2n+1} - \sum_{j=1}^{2n+1}(-1)^j
2277: f_1' \wedge ...
2278: \wedge \widehat { f_j'} \wedge ..., \wedge f_{2n+1}' = 0
2279: $$
2280:
2281: Indeed, substituting $f_i' = \frac{f_i}{f_0}$ in this formula we find
2282: that the only possible
2283: nontrivial term $f_0 \wedge f_1 \wedge ... \wedge \widehat f_i \wedge ...
2284: \wedge \widehat f_j \wedge ... \wedge f_{2n}$ vanishes because it is symmetric
2285: in $i,j$.
2286:
2287:
2288: b) Let $g_1,...,g_{2n+1}$ be rational functions on $\C \PP^{n}$ with
2289: ${\rm div} g_i = (h_i) - (h_0)$.
2290: Then
2291: \begin{equation} \label{8}
2292: dr_{2n-1}\Bigl(\sum_{j=1}^{2n+1}(-1)^j g_1 \wedge ...
2293: \wedge \widehat g_j \wedge ... \wedge g_{2n+1}\Bigr) =
2294: \end{equation}
2295: $$
2296: \sum_{j \not = i}(-1)^{j+i-1} 2 \pi i \delta(f_i) \wedge r_{2n-2}\Bigl(
2297: g_1\wedge ... \wedge \widehat g_i \wedge ...
2298: \wedge \widehat g_j \wedge ... \wedge g_{2n+1}\Bigr)
2299: $$
2300: (Notice that $d\log g_1 \wedge ... \wedge \widehat {d\log g_j} \wedge ...
2301: \wedge d\log g_{2n+1} =0$ on $\C \PP^{n}$).
2302: Integrating (\ref{8}) over $\C \PP^{n}$ we see that the left hand side equals
2303: zero, while the right hand side equals to the sum of the expressions staying
2304: on the left of (\ref{4}). So we get b).
2305:
2306: c) is obvious: we apply $r_{2n-2}$ to the zero element.
2307: Theorem is proved.
2308:
2309:
2310: {\bf 4. ${\Bbb P}^1 - \{0, \infty\}$ as a
2311: special stratum in the configuration space of $2n$ points in $\PP^{n-1}$}.
2312: A {\it special configuration} is
2313: a configuration of $2n$ points
2314: \begin{equation} \label{--00}
2315: (l_0,...,l_{n-1}, m_0,...,m_{n-1})
2316: \end{equation}
2317: in $\PP^{n-1}$ such that $l_0,...,l_{n-1}$ are vertices of a
2318: simplex in $\PP^{n-1}$ and $ m_i$ is a point
2319: on the edge $l_il_{i+1}$ of the simplex different from $l_i$ and
2320: $l_{i+1}$, as on the picture.
2321:
2322: \begin{figure}[ht]
2323: \centerline{\epsfbox{fig1coc.eps}}
2324: \caption{A special configuration of $8$ points in $\PP^{3}$.}
2325: \label{fig1coc}
2326: \end{figure}
2327:
2328:
2329:
2330: \begin{proposition} \label{6.12.02.1}
2331: The set of special configurations of $2n$ points in $\PP^{n-1}$
2332: is canonically identified with $\PP^1 \backslash \{0, \infty \}$.
2333: \end{proposition}
2334:
2335: \begin{figure}[ht]
2336: \centerline{\epsfbox{gpol12.eps}}
2337: \caption{$\PP^1 - \{0, \infty\}$ is a stratum in the configuration space of $2n$ points in $\PP^{n-1}$.}
2338: \label{gpol12}
2339: \end{figure}
2340:
2341:
2342:
2343: {\bf Proof}.
2344: We define {\it generalized cross-ratio}
2345: $$
2346: r(l_0,...,l_{n-1},m_0,...,m_{n-1}) \in F^*,
2347: $$
2348: where $F$ is the common field of definition of the points $l_i, m_j$, as follows.
2349: Consider the one-dimensional subspaces $L_i, M_j$ in the $n$-dimensional vector space $V$ projecting to the points $l_i,m_j$ in $\PP^{n-1}$ respectively. The
2350: subspaces $L_i, M_i, L_{i+1}$ generate a two dimensional subspace. Its quotient
2351: along $M_i$ can be identified with $L_i$ as well as with $L_{i+1}$.
2352: So we get a canonical linear map
2353: $
2354: \overline M_i: L_i \longrightarrow L_{i+1}
2355: $.
2356: The composition of these maps (the ``linear monodromy'')
2357: $$
2358: \overline M_0 \circ ... \circ \overline M_{n-1} : L_0 \longrightarrow L_0
2359: $$
2360: is multiplication by an element of $F^*$
2361: called the generalized cross-ratio of the special configuration (\ref{--00}).
2362:
2363: \begin{figure}[ht]
2364: \centerline{\epsfbox{gpol7.eps}}
2365: \caption{The generalized cross-ratio of a special configuration}
2366: \label{gpol7}
2367: \end{figure}
2368:
2369:
2370: It is clearly invariant under the cyclic permutation
2371: $$
2372: l_0 \to l_0 \to ... \to l_{n-1}\to l_0; \quad m_0 \to m_1 \to ... \to m_{n-1}\to m_0
2373: $$
2374: Notice that $r(l_0,...,l_{n-1},m_0,...,m_{n-1}) =1$
2375: if and only if the points $m_0,...,m_{n-1}$ belong to a hyperplane.
2376:
2377:
2378:
2379:
2380: Let $\widehat m_i$
2381: be the point of intersection of the line $l_il_{i+1}$ with the
2382: hyperplane passing through all the points $m_j$ except $m_i$.
2383: Let $r(x_1, ..., x_4)$ be the cross-ratio of the four points $x_i$ on $\PP^1$. Then
2384: $$
2385: r(l_0,...,l_{n-1},m_0,...,m_{n-1}) =
2386: r(l_i,l_{i+1},m_i,\widehat m_{i+1})
2387: $$
2388:
2389:
2390: {\it The special configurations and classical polylogarithms}.
2391: Consider the configuration of $2n$ hyperplanes in $\PP^{n-1}$ given by the following
2392: equations in homogeneous coordinates $z_0: ... :z_{n-1}$
2393: $$
2394: z_0 = 0, \quad ..., \quad z_{n-1} = 0, \quad z_0 = z_1, \quad z_1 + z_2 = z_0,
2395: $$
2396: \begin{equation} \label{ewa99}
2397: z_2-z_3=0,\quad ..., \quad z_{n-2} - z_{n-1} = 0, \quad z_{n-1} = az_0
2398: \end{equation}
2399: It admits the following interpretation. Recall that the classical
2400: polylogarithm function
2401: $Li_{n-1}(z)$ can be defined by an iterated integral:
2402: $$
2403: Li_{n-1}(a) =
2404: \int_0^a\frac{dt}{1-t} \circ \frac{dt}{t} \circ ... \circ \frac{dt}{t} =
2405: \int_{\Delta_a}\frac{dz_1}{z_1} \wedge ... \wedge \frac{dz_{n-1}}{z_{n-1}}
2406: $$
2407: If $a \in (0,1]$, then the simplex $\Delta_a$ is defined by the equations
2408: $$
2409: \Delta_a:= \quad \{(z_1,...,z_{n-1}) \in \R^{n-1}| \quad
2410: 0 \leq 1-z_1 \leq z_2 \leq z_3 \leq ... \leq z_{n-1} \leq a\}
2411: $$
2412: The faces of the simplex $\Delta_a$ can be defined for arbitrary $a$.
2413: Then the codimension one faces $\{z_i =0\}$ of the coordinate
2414: simplex
2415: and the codimension one faces of the simplex $\Delta_a$
2416: form the configuration (\ref{ewa99}).
2417:
2418:
2419:
2420:
2421: We can
2422: reorder hyperplanes of this configuration as follows:
2423: $$
2424: z_0 =0, \quad z_1 =0, \quad z_1 =z_0, \quad z_1+z_2 = z_0, \quad z_2 =0,
2425: $$
2426: $$
2427: z_2 = z_3, \quad z_3 =0, \quad ...\quad , \quad z_{n-2} =0, \quad
2428: z_{n-2} = z_{n-1}, \quad z_{n-1} = a z_0
2429: $$
2430: Applying the projective duality to this configuration of hyperplanes
2431: we get the special configuration of $2n$ points in $\PP^{n-1}$
2432: with the generalized cross ratio $a$.
2433:
2434: The correspondence between the configuration (\ref{ewa99}) and the
2435: special configuration of points is illustrated in the case $n=3$
2436: on Figure \ref{gpol11.fig}.
2437:
2438: \begin{figure}[ht]
2439: \centerline{\epsfbox{gpol11.fig.eps}}
2440: \caption{Classical polylogarithm configurations and special configurations}
2441: \label{gpol11.fig}
2442: \end{figure}
2443:
2444: {\bf Remark}. It is amusing that the special configuration of
2445: $2n$ points in $\PP^{n-1}$,
2446: which is related to the classical $n$-logarithm by Theorem \ref{clpoly} below,
2447: is constructed using the geometry of the mixed motive corresponding to $Li_{n-1}(a)$.
2448:
2449:
2450:
2451:
2452:
2453: {\bf 5. Restriction of the Grassmannian $n$--logarithm to the special stratum}.
2454: The function $Li_n(z)$ has a remarkable single-valued
2455: version ([Z1], [BD]):
2456: \begin{eqnarray*}
2457: {\cal L}_{n}(z) &:=& \begin{array}{ll}
2458: {\rm Re} & (n:\ {\rm odd}) \\
2459: {\rm Im} & (n: \ {\rm even}) \end{array}
2460: \left( \sum^{n-1}_{k=0} \beta_k
2461: \log^{k}\vert z\vert \cdot Li_{n-k}(z)\right)\; , \quad n\geq
2462: 2 \\
2463: \end{eqnarray*}
2464: It is continuous on $\C \PP^1$. Here $\frac{2x}{e^{2x} -1} =
2465: \sum_{k=0}^{\infty}\beta_k x^k $, so $\beta_k = \frac{2^kB_k}{k!}$
2466: where the $B_k$ are the Bernoulli numbers. For example ${\cal L}_2(z)$
2467: is the Bloch - Wigner function.
2468:
2469: Let us consider the following modification of the
2470: function ${\cal L}_n(z)$ proposed by A. M. Levin in [Le]:
2471: $$
2472: \widetilde {\cal L}_{n}(x):=
2473: $$
2474: $$
2475: \frac{(2n-3)}{(2n-2)}
2476: \sum_{\mbox{$k$ even; $0 \leq k \leq n-2$}}
2477: \frac{2^k (n-2)!(2n-k-3)!}{(2n-3)!(k+1)!(n-k-2)!} {\cal L}_{n-k}(x)\log^k|x|
2478: $$
2479: For example $\widetilde {\cal L}_{n}(x) = {\cal L}_{n}(x)$ for $n \leq 3$,
2480: but already $\widetilde {\cal L}_{4}(x)$ is different from ${\cal L}_{4}(x)$.
2481: A direct integration carried out in Proposition 4.4.1 of [Le]
2482: shows that
2483: $$
2484: -(2\pi i)^{n-1}(-1)^{(n-1)(n-2)/2}\widetilde {\cal L}_{n}(x) =
2485: $$
2486: $$
2487: \int_{\C \PP^{n-1}}\log|1-z_1| \prod_{i=1}^{n-1}d\log|z_i| \wedge \prod_{i=1}^{n-2}d\log|z_{i} - z_{i+1}|
2488: \wedge d\log|z_{n-1} -a|
2489: $$
2490: This combined with Proposition \ref{0.3} below implies
2491:
2492: \begin {theorem} \label{clpoly}
2493: The value of the function ${\cal L}^G_n$ on the special configuration
2494: (\ref{--00}) is equal to $$
2495: -(-1)^{n(n-1)/2}4^{n-1}{2n-2\choose n-1}^{-1}\widetilde {\cal L}_{n}(a)
2496: $$ where $a = r(l_0,...,l_{n-1},m_0,...,m_{n-1}) $.
2497: \end {theorem}
2498:
2499: Another proof in the case $n=2$ is given in Proposition 6.8.
2500:
2501:
2502: \begin {conjecture} \label{conj}
2503: The Chow $n$-logarithm function can be expressed by the
2504: Grassmannian $n$-logarithm function.
2505: \end {conjecture}
2506:
2507:
2508:
2509:
2510:
2511: {\bf Remark}. Suppose that an element $\sum_k\{f^{(k)}_1,...,f^{(k)}_{2n+1}\}
2512: \in K^M_{2n+1}(\C(X))$ has zero residues at all the divisors on an
2513: $n$-dimensional variety $X$ over $\C$. Then it defines an element
2514: $$
2515: \alpha \in gr^{\gamma}_{2n+1}K_{2n+1}(X) = Ext^{2n+1}_{{\cal M}}(\Q(0)_X,\Q(2n+1)_X)
2516: $$
2517: Its direct image to the point is
2518: an element
2519: $$
2520: \pi_{\ast}(\alpha) \in gr^{\gamma}_{n}K_{2n+1}(Spec \C) = Ext^{1}_{{\cal M}}(\Q(0),\Q(n))
2521: $$
2522: Applying the regulators we see that the
2523: integral
2524: $\sum_k \int_{X(\C)}r_{2n}(f^{(k)}_1,...,f^{(k)}_{2n+1})$ coincides, up to a factor,
2525: with
2526: the value of the Borel regulator map on $\pi_{\ast}(\alpha)$ and so by
2527: results of the next chapter is
2528: expressible by the Grassmannian $n$-logarithms. Conjecture \ref{conj}
2529: tells us that this should be true {\it for any} element in
2530: $K^M_{2n+1}(\C(X))$.
2531:
2532:
2533:
2534:
2535:
2536: \section{Grassmannian polylogarithms, symmetric
2537: spaces and Borel regulators}
2538:
2539: {\bf 1. The function $\psi_n$.} Let $V_n$ be a complex
2540: vector space of dimension $n$.
2541: Let
2542: $$
2543: \HH_n: = \left \{ \mbox{ positive definite Hermitian forms in
2544: $V_n$ }\right \}/\R ^*_+ = SL_n(\C)/SU(n)
2545: $$
2546: $$
2547: = \left \{ \mbox{ positive definite Hermitian forms in
2548: $V_n$ with determinant } =1 \right \}
2549: $$
2550: It is a symmetric space of rank $n-1$. For example
2551: $\HH_2 = {\cal H}_3$ is the hyperbolic 3-space.
2552: Replacing positive definite by non-negative definite Hermitian forms
2553: we get a compactification $\overline \HH_n$ of the symmetric space $\HH_n$.
2554:
2555:
2556:
2557: Let $G_x$ be the subgroup of $SL_N(\C)$ stabilizing the point $x \in \HH_n$ .
2558: A point $x$ defines a one-dimensional vector space $M_x$:
2559: $$
2560: x \in \HH_n \longmapsto M_x:= \left \{ \mbox{measures on } \C \PP^{n-1} %\subset \partial \HH_n
2561: \mbox{ invariant under } G_x\right \}
2562: $$
2563: Namely, a point $x$ corresponds to a hermitian metric in $V_n$.
2564: This metric provides the Fubini-Studi metric on $\C \PP^{n-1} =P(V_n)$.
2565: Moreover there is the Fubini-Studi K\"ahler form on $\C \PP^{n-1} = P(V_n)$;
2566: its imaginary part is a symplectic form. Raising it to
2567: $(n-1)$-th power we get the Fubini-Studi volume form.
2568: The elements of $M_x$ are the multiples of the Fubini-Studi volume form.
2569:
2570:
2571:
2572: So $\HH_n$ embeds
2573: into the projectivization of the space of all measures in $\C \PP^{n-1}$.
2574: Taking its closure we get a compactification of $\HH_n$.
2575:
2576:
2577:
2578: Let us choose for any point $x \in \HH_n$ an invariant measure $\mu_x \in M_x$.
2579: Then, for any $y \in {\Bbb H}_n$, the ratio $ \mu_x/\mu_y$ is a real function on $\C \PP^{n-1}$.
2580:
2581: Let $x_0,...,x_{2n-1}$ be points of the symmetric space $SL_n(\C)/SU(n)$.
2582: Consider the following function
2583: \begin{equation} \label{1221q}
2584: \psi_n(x_0,...,x_{2n-1}) := \int_{\C \PP^{n-1}}
2585: \log | \frac{\mu_{x_1}}{\mu_{x_0}}| d\log|\frac{\mu_{x_2}}{\mu_{x_0}} | \wedge ... \wedge d\log| \frac{\mu_{x_{2n-1}}}{\mu_{x_0}} |
2586: \end{equation}
2587:
2588:
2589:
2590:
2591:
2592:
2593: {\bf 2. General properties of the function $\psi_n$}. Let us study the properties of
2594: integral (\ref{1221q}) in a more general situation. Let $X$ be an $m$-dimensional manifold. For any $m+2$ measures $\mu_0,...,\mu_{m+1}$
2595: on $ X$ such that $\frac{\mu_{i}}{\mu_j}$ are smooth functions we can construct a differential $m$-form on $X$:
2596: $$
2597: \overline r_{ m}(\mu_0:...:\mu_{ m+1}) := \log | \frac{\mu_1}{\mu_0}| d\log|\frac{\mu_2}{\mu_0} | \wedge ... \wedge d\log| \frac{\mu_{ m+1}}{\mu_0} |
2598: $$
2599:
2600: \begin{proposition} \label{1.2.}
2601: The integral
2602: \begin{equation} \label{i2}
2603: \int_{ X}\overline r_{ m}(\mu_0:...:\mu_{ m+1})
2604: \end{equation}
2605: satisfies the following properties:
2606:
2607: 1) Skew symmetry with respect to the permutations of $\mu_i$.
2608:
2609: 2) Homogeneity:
2610: $$
2611: \int_{ X}\overline r_{ m}(\lambda_0 \mu_0: ... :\lambda_{ m+1} \mu_{ m+1}) = \int_{X }\overline r_{ m}(\mu_0:...:\mu_{ m+1})
2612: $$
2613:
2614: 3)Additivity: for any $ m+3$ measures $\mu_i$ on $X $ one has
2615: $$
2616: \sum_{i=0}^{m+2} (-1)^i \int_{X }\overline r_{ m}(\mu_0: ... :\widehat \mu_i: ... :\mu_{ m+2}) =0
2617: $$
2618:
2619: 4) Let $g$ be a diffeomorphism of $X$. Then
2620: $$
2621: \int_{X }\overline r_{ m}(g^*\mu_0: ... :g^*\mu_{ m+1}) = \int_{X }\overline r_{ m}( \mu_0: ... : \mu_{ m+1})
2622: $$
2623: \end{proposition}
2624:
2625: {\bf Proof}. 1). Follows from $\log f \cdot d\log g + \log g \cdot d\log f =
2626: d (\log f \cdot \log g)$.
2627:
2628: 2) Using 1) we may assume $\lambda_i =1$ for $i>0$. Then
2629: $$
2630: \int_{ X} (\overline r_{ m}(\lambda_0\mu_0:\mu_1: ...:\mu_{ m+1}) - \overline r_{ m}(\mu_0:\mu_1 :...:\mu_{ m+1})) =
2631: $$
2632: $$
2633: - \log |\lambda| \cdot \int_{ X}
2634: d(\log| \frac{\mu_2}{\mu_0}| d \log| \frac{\mu_3}{\mu_0}| \wedge ... \wedge d\log| \frac{\mu_{m+1}}{\mu_0}|) =0
2635: $$
2636:
2637: 3) Taking into account the skewsymmetry of the integral we have to prove that
2638: \begin{equation} \label{12321}
2639: {\rm Alt}_{(0,..., m+2)} \left \{
2640: \log | \frac{\mu_2}{\mu_1}| d\log| \frac{\mu_3}{\mu_1}| \wedge ... \wedge d\log| \frac{\mu_{m+2}}{\mu_1} | \right \} = 0
2641: \end{equation}
2642: Let us write $ \frac{\mu_i}{\mu_j} = \frac{\mu_i}{\mu_0}/ \frac{\mu_j}{\mu_0}$ and substitute it into (\ref{12321}). Then the terms in (\ref{12321}) where
2643: $$
2644: \log | \frac{\mu_2}{\mu_0}| d\log| \frac{\mu_3}{\mu_0}| \wedge ... \wedge d\log| \frac{\mu_{m+2}}{\mu_0} |
2645: $$
2646: will appear look as follows:
2647: $$
2648: \log | \frac{\mu_2}{\mu_1}| d\log| \frac{\mu_3}{\mu_1}| \wedge ... \wedge d\log| \frac{\mu_{m+2}}{\mu_1} | - \log | \frac{\mu_1}{\mu_2}| d\log| \frac{\mu_3}{\mu_2}| \wedge ... \wedge d\log| \frac{\mu_{m+2}}{\mu_2} |
2649: $$
2650: $$
2651: -\log | \frac{\mu_2}{\mu_0}| d\log| \frac{\mu_3}{\mu_0}| \wedge ... \wedge d\log| \frac{\mu_{m+2}}{\mu_0} | + \log | \frac{\mu_0}{\mu_2}| d\log| \frac{\mu_3}{\mu_2}| \wedge ... \wedge d\log| \frac{\mu_{m+2}}{\mu_2} |
2652: $$
2653: (The first two terms comes from ${\rm Alt}_{(1,..., m+2)}\overline r_m(\mu_1: ... :\mu_{m+2})$ and the second two from ${\rm Alt}_{(0,2,..., m+2)}\overline r_m(\mu_0: \mu_2: ... :\mu_{m+2})$. The expression ${\rm Alt}_{ m+2}\overline r_m(\mu_0: ... :\widehat \mu_i: ... :\mu_{m+2})$ provides no such terms if $i>1$).
2654:
2655: 4) Clear. The proposition is proved.
2656:
2657:
2658: Recall the following general construction. Let $G$ be a group.
2659: Let $X$ be a $G$-set and $f$ a function on $X^n$ satisfying
2660: $$
2661: \sum_{i=1}^{n+1}(-1)^{i}f(x_1, ..., \widehat x_i, ..., x_{n+1}) =0
2662: $$
2663: Choose a point $x \in X$. Then there is an $(n-1)$-cocycle of the group $G$:
2664: $$
2665: f_x(g_1, ..., g_n):= f(g_1 x, ..., g_n x)
2666: $$
2667:
2668: \begin{lemma} \label{point}
2669: The cohomology class of the cocycle $f_x$ does not depend on $x$.
2670: \end{lemma}
2671:
2672: {\bf Proof}. The difference $f_y - f_x$ is the coboundary of the $(n-2)$-chain
2673: \begin{equation} \label{1.22.1}
2674: h_{x,y}(g_1, ..., g_{n-1}) =
2675: \sum_{i=1}^{n-1}(-1)^{k-1} f(g_1 x, g_2 x, ..., g_kx, g_ky, g_{k+1}y,
2676: ..., g_{n-1} y)
2677: \end{equation}
2678: Here is the geometric picture leading to this formula. Consider the prism
2679: $\Delta_{g_1, ..., g_{n}}^{(n-1)}\times \Delta_{x,y}^{(1)}$ given by product
2680: of the $(n-1)$-simplex with vertices
2681: $g_1, ..., g_n$ by the $1$-simplex with vertices $(x,y)$. Decomposing its side face
2682: $\Delta_{g_1, ..., g_{n-1}}^{(n-2)}\times \Delta_{x,y}^{(1)}$
2683: into simplices
2684: we come to the right hand side of (\ref{1.22.1}). Then the terms of the
2685: formula $f_y - f_x - \delta h_{x,y}$
2686: correspond to the boundary faces of the prism. Cutting the prism into
2687: simplices we see that
2688: the sum of the terms corresponding to the prism boundary is zero thanks
2689: to the cocycle relation.
2690: The lemma is proved.
2691:
2692: So, for any $x \in {\Bbb H}_n$,
2693: $$
2694: (\psi_n)_x(g_0,...,g_{2n-1}):= \psi_n(g_0x,...,g_{2n-1}x)
2695: $$
2696: is a smooth $(2n-1)$-cocycle of $GL_n(\C)$.
2697:
2698: {\bf Remark}. This cocycle is the restriction to $GL_n(\C)$
2699: of the Bott cocycle for the group of diffeomorphisms of $\C \PP^{n-1}$.
2700:
2701:
2702:
2703: Let $h_0, \dots, h_{2n-1}$ be any hyperplanes in ${\Bbb CP}^{n-1}$.
2704: Recall that the Grassmannian $n$-logarithm is defined by
2705: $$
2706: {\cal L}_n^{G}(h_0,\dots,h_{2n-1})=(2\pi i)^{1-n}
2707: \int_{{\Bbb CP}^{n-1}} r_{2n-1}(f_1,\dots, f_{2n-1})$$
2708: where $f_i$ is a rational function on ${\Bbb CP}^{n-1}$ with the divisor $(h_i) -(h_0)$.
2709:
2710:
2711: \begin{proposition} \label{0.3}
2712: One has
2713: $$
2714: {\cal L}_n^{\rm G}(h_0,\dots,h_{2n-1}) = -\frac {(-4)^{n-1}(n-1)!^2} {(2\pi i)^{n-1} (2n-2)!}
2715: \int_{{\Bbb CP}^{n-1}}\log|f_1|\bigwedge_{j=2}^{2n-1} d\log|f_j|
2716: $$
2717: \end{proposition}
2718:
2719:
2720: {\bf Proof}. See Proposition 6.2 in [GZ].
2721:
2722:
2723: {\bf 3. The Grassmannian polylogarithm ${\cal L}_n^G$ as the boundary value
2724: of the function $\psi_n$.}
2725: We start from an explicit formula for the Fubini-Studi form. Let $\widehat \PP^{n-1}$
2726: be the variety of all hyperplanes in $\PP^{n-1}$. Consider the incidence divisor
2727: $$
2728: D \subset \widehat \PP^{n-1} \times \PP^{n-1} \qquad D := \{(h, x)| x \in h\}
2729: $$
2730: where $h$ is a hyperplane and $x$ is a point in $\PP^{n-1}$.
2731:
2732: Let $(x_0:...:x_{n-1})$ be the
2733: homogeneous coordinates of a point $x$ in $\PP^{n-1}$.
2734: Let
2735: $$
2736: \sigma_n(x,dx):= \sum_{i=0}^{n-1} (-1)^i x_i dx_0 \wedge ...
2737: \wedge \widehat dx_i \wedge ... \wedge dx_{n-1} = i_E {\rm vol}_x
2738: $$
2739: be the Leray from. Here ${\rm vol}_x = dx_0 \wedge ...
2740: \wedge dx_{n-1}$ and $E =\sum x_i \partial_{x_i}$.
2741:
2742: There is a canonical differential $(n-1,n-1)$-form
2743: $\omega_D$ on $ \widehat \PP^{n-1} \times \PP^{n-1} - D$
2744: with a polar singularity at the divisor $D$. Namely,
2745: let $x \in V_n$ and $\xi \in V_n^*$. Then
2746: $$
2747: \omega_D:= \frac{1}{(2\pi i)^{n-1}}\frac{\sigma_n(\xi,d\xi)
2748: \wedge \sigma_n(x,dx)}{<\xi,x>^{n}}
2749: $$
2750: It is $PGL_{n}$-invariant.
2751: A hermitian metric $H$ in $V_n$ provides an isomorphism
2752: $H: V_n \lra \overline V_n^*$, and hence an isomorphism
2753: ${\Bbb C} \PP^{n-1} \lra \widehat {\overline {{\Bbb C}P}}^{n-1}$.
2754: The graph $\Gamma_H$ of this map does not intersect the incidence divisor $D$. Thus
2755: restricting the form $\omega_D$ to $\Gamma_H$ we
2756: get a volume form on ${{\Bbb C}P}^{n-1}$:
2757: \begin{equation} \label{FUBS}
2758: \omega_{FS}(H) := \frac{1}{(2\pi i)^{n-1}}\frac{\sigma_n(z,dz) \wedge \sigma_n(\overline z, d \overline {z})}{H(z, \overline z)^{n}}
2759: \end{equation}
2760: It is clearly invariant under the group preserving the Hermitian form $H$. Moreover,
2761: it is the Fubini-Studi volume form:
2762: a proof can be obtained by using the explicit formula for the Fubini-Studi
2763: K\"ahler form given in [Ar], complement 3.
2764:
2765:
2766:
2767: One can realize
2768: $\C \PP^{n-1}$ as the smallest stratum of the boundary of $\HH_n$ . Namely, for
2769: a hyperplane $h$ in an $n$-dimensional complex vector space $V_n$ let
2770: $$
2771: F_h:= \left \{\mbox{nonnegative definite hermitian forms in
2772: $V_n$ with kernel} h \right \}/ \R _+^*
2773: $$
2774: The set of
2775: hermitian forms in $V_n$ with the
2776: kernel $h$ is isomorphic to $\R _+^*$, so $F_h$ defines a point on
2777: the boundary of $\overline \HH_n$.
2778:
2779: For any nonzero
2780: nonnegative definite hermitian form $H$ one can define the corresponding
2781: Fubini-Studi form by formula (\ref{FUBS}). It is
2782: a differential form with singularities along the projectivization of
2783: the kernel of $H$. In particular if $h$ is a hyperplane then the degenerate hermitian
2784: form $F_{h}$ provides the Lebesgue measure on the affine space $\C\PP^{n-1} - h$.
2785: Indeed, if
2786: $h_0 = \{z_0 =0\}$ then (\ref{FUBS}) specializes to
2787: $$
2788: \frac{1}{(2\pi i)^{n-1}} d\frac{z_1}{z_0} \wedge ... \wedge d\frac{z_{n-1}}{z_0} \wedge d\frac{ \overline z_1}{\overline z_0} \wedge ... \wedge d\frac{ \overline z_{n-1}}{\overline z_0}
2789: $$
2790: Denote by $M_{h}$ the one dimensional real vector space generated by this form.
2791: For any hyperplane $h$ in $\C \PP^{n-1}$
2792: let us choose a measure $\mu_{h} \in M_{h}$.
2793:
2794:
2795:
2796:
2797:
2798: \begin{proposition} \label{1/18.1}
2799: For any $2n$ hyperplanes $h_0,...,h_{2n-1}$ in $\C \PP^{n-1}$ the integral
2800: \begin{equation} \label{1221N}
2801: \psi_n(h_0,...,h_{2n-1}) := \int_{\C \PP^{n-1}}
2802: \log | \frac{\mu_{h_1}}{\mu_{h_0}}| d\log|\frac{\mu_{h_2}}{\mu_{h_0}} | \wedge ... \wedge d\log| \frac{\mu_{h_{2n-1}}}{\mu_{h_0}} |
2803: \end{equation}
2804: is convergent and equals to
2805: $$
2806: (-4)^{-n} \cdot (2\pi i)^{n-1} (2n)^{2n-1}{2n-2\choose n-1} \cdot {\cal L}_n^G(h_0, ..., h_{2n-1})
2807: $$
2808: \end{proposition}
2809:
2810: {\bf Proof}. Let $h_1,h_2$ be hyperplanes in $\C \PP^{n-1}$ and $f$ be a rational function
2811: such that $(f) = (h_1) -(h_2)$.
2812: From the explicit description of $M_h$ given above we immediately see that
2813: \begin{equation} \label{AG362}
2814: \mu_{h_1}/\mu_{h_2} = \lambda \cdot |f|^{2n}
2815: \end{equation}
2816: Using this and Theorem 2.4 we see that
2817: integral (\ref{1221N}) is convergent. The second statement follows from
2818: Proposition \ref{0.3} and (\ref{AG362}). The proposition is proved.
2819:
2820:
2821: More generally, take any $2n$ Hermitian forms
2822: $H_0, ..., H_{2n-1}$, possibly degenerate. For each of the forms $H_i$
2823: consider the corresponding measure $\mu_{H_i}$ ( a multiple of the Fubini-Studi form related to $H_i$). Using the convergence of the integral (\ref{1221N})
2824: we can deduce that the integral
2825: \begin{equation} \label{1221M}
2826: \psi_n(H_0,...,H_{2n-1}) := \int_{\C \PP^{n-1}}
2827: \log | \frac{\mu_{H_1}}{\mu_{H_0}}|
2828: d\log|\frac{\mu_{H_2}}{\mu_{H_0}} | \wedge ... \wedge
2829: d\log| \frac{\mu_{H_{2n-1}}}{\mu_{H_0}} | =
2830: \end{equation}
2831: $$
2832: -n^{2n-1}\cdot \int_{\C \PP^{n-1}}
2833: \log | \frac{H_1(z, \overline z)}{H_0(z, \overline z)}| d\log|\frac{H_2(z, \overline z)}{H_0(z, \overline z)} | \wedge ... \wedge d\log| \frac{H_{2n-1}(z, \overline z)}{H_{0}(z, \overline z)} |
2834: $$
2835: is also convergent. This enables us to extend $\psi_n$ to the function
2836: $\overline \psi_n(x_0,...,x_{2n-1})$
2837: on the configuration space of $2n$ points in $\overline \HH_{n-1}$. The function $\overline \psi_n$ is discontinuous. For instance it is discontinuous at the point
2838: $x_1 = ... = x_{2n-1} = F_h$ for a given hyperplane $h$ in $\C \PP^{n-1}$.
2839: It is however a smooth function on an open part of any given strata. We will keep the notation
2840: $$
2841: \psi_n(h_0,..., h_{2n-1}) = \overline \psi_n(F_{h_0},...,F_{h_{2n-1}})
2842: $$
2843:
2844: Applying Lemma \ref{point} to the case when $X$ is $\overline {\Bbb H}_n$ and using only the fact that
2845: the function $\overline \psi_n(x_0,...,x_{2n-1})$ is well defined for {\it any}
2846: $2n$ points in $\overline {\Bbb H}_n$ and satisfies the cocycle condition for any $2n+1$ of them we get
2847:
2848: \begin{corollary} \label{pointx1}
2849: Let $x \in {\Bbb H}_n$ and let $h$ be a hyperplane in $\C\PP^{n-1}$.
2850: Then the cohomology classes of the following cocycles coincide:
2851: $$
2852: \psi_n(g_0 x,...,g_{2n-1}x) \quad \mbox{and} \quad \psi_n(g_0 h,...,g_{2n-1}h)
2853: $$
2854: \end{corollary}
2855:
2856: {\bf 4. A normalization of the Borel class $b_n$}. Choose a hermitian metric
2857: in $V_n$.
2858: Let $e$ be the corresponding point of the symmetric space ${\Bbb H}_n$; its
2859: stabilizer is the subgroup $SU(n)$. One has
2860: $$
2861: \Bigl(\Lambda^{\bullet}T^*_e{\Bbb H}_n\Bigr)^{SU(n)} =
2862: {\cal A}^{\bullet}(SL_n(\C)/SU(n))^{SL_n(\C)}
2863: $$
2864: There are well known
2865: canonical ring isomorphisms (see [B2] and references there):
2866: $$
2867: \Bigl(\Lambda^{\bullet}T^*_e{\Bbb H}_n\Bigr)^{SU(n)}\otimes_{\R} \C =
2868: \Lambda^{\bullet}(sl_n(\C))^{sl_n(\C)} =
2869: $$
2870: \begin{equation} \label{1.20.2}
2871: H^{\bullet} (sl_n(\C), \C) \stackrel{\alpha}{=} H_{\rm top}^{\bullet} (SU(n), \C)
2872: \stackrel{\beta}{=}
2873: H^{\bullet}_m(SL_n(\C), \C)
2874: \end{equation}
2875: where $H^{\bullet}$ is the Lie algebra cohomology, $H^{\bullet}_{\rm top}$ is the topological cohomology, and $H_m(G)$ denotes the measurable cohomology of a Lie group $G$.
2876: The first isomorphism is obvious: $T_e{\Bbb H}_n\otimes_{\R} \C = sl_n(\C)$.
2877: The map
2878: $$
2879: \alpha_{\rm DR}: \Lambda_{\Q}^{\bullet}(sl_n)^{sl_n} \stackrel{\sim}{\lra}
2880: H_{\rm DR}^{\bullet} (SL_n(\C), \Q)
2881: $$
2882: sends an $sl_n$--invariant exterior form
2883: on $sl_n$ to the right--invariant one, and hence
2884: biinvariant differential form on $SL_n(\C)$.
2885: Let us describe the map
2886: $$
2887: \beta_{\rm DR}: H_{\rm DR}^{\bullet} (SL_n(\C), \Q)
2888: \stackrel{}{\lra}
2889: H^{\bullet}_m(SL_n(\C), \C)
2890: $$
2891:
2892: Let $C$ be a biinvariant, and hence closed,
2893: differential $(2n-1)$--form on $SL_n(\C)$. Let us restrict it first
2894: to the Lie algebra, and then
2895: to the orthogonal
2896: complement $su(n)^{\perp}$ to the Lie subalgebra $su(n) \subset sl_n(\C)$.
2897: We identify the $\R$--vector spaces $T_e{\Bbb H}_n$ and $su(n)^{\perp}$.
2898: The obtained exterior form on $T_e{\Bbb H}_n$ is the restriction of
2899: an invariant differential form, denoted $\omega_{C}$,
2900: on the symmetric space
2901: ${\Bbb H}_n$. It is a closed differential form.
2902:
2903:
2904:
2905: For any ordered $2n$ points $x_1,...,x_{2n}$
2906: in ${\Bbb H}_n$ there is
2907: a geodesic
2908: simplex $I(x_1,...,x_{2n})$ in ${\Bbb H}_n$. It is constructed inductively
2909: as follows. Let
2910: $I(x_1,x_2)$ be the geodesic from $x_1$ to $x_2$. The
2911: geodesics from $x_3$ to
2912: the points of $I(x_1,x_2)$ form a geodesic triangle
2913: $I(x_1,x_2,x_3)$. All the geodesics from $x_4$ to the points
2914: of the geodesic triangle $I(x_1,x_2,x_3)$ form a geodesic simplex
2915: $I(x_1,x_2,x_3,x_4)$, and so on. When the
2916: rank of the symmetric space is greater than 1 (i.e. $n>2$)
2917: the geodesic simplex $I(x_1,...,x_k)$ depends on the
2918: ordering of the vertices $x_1,...,x_k$.
2919:
2920: The
2921: differential $(2n-1)$-form $\omega_{C}$ on
2922: $SL_n(\C)/SU(n)$ provides a volume of the geodesic simplex:
2923: $$
2924: {\rm vol}_C I(x_1,...,x_{2n}):= \int_{I(x_1,...,x_{2n})}\omega_{C}
2925: $$
2926: For every $2n+1$ points $x_1,...,x_{2n+1}$ the boundary of the simplex
2927: $I(x_1,...,x_{2n+1})$ is the alternating sum of the simplices
2928: $I(x_1,..., \widehat x_i,..., x_{2n+1})$. Since the form $\omega_{C}$ is
2929: closed, the Stokes theorem yields
2930: \begin{equation} \label{6.15.02.1}
2931: \sum_{i=1}^{2n+1}(-1)^i \int_{I(x_1,..., \widehat x_i,..., x_{2n+1})}\omega_{C} =
2932: \int_{I(x_1,..., x_{2n+1})}d\omega_{C} = 0
2933: \end{equation}
2934: This just means that for a given point $x$ the function
2935: $ {\rm vol_C} I(g_1 x,...,g_{2n} x)$ is a smooth $(2n-1)$-cocycle
2936: of the Lie group
2937: $SL_n(\C)$. It was considered by J.Dupont [D].
2938: By Lemma \ref{point} cocycles corresponding to different points $x$
2939: are canonically cohomologous.
2940: The obtained cohomology class is the class $\beta_{\rm DR}([C])$.
2941:
2942:
2943:
2944: {\bf Remark}.
2945: ${\rm vol}_CI(x_1,...,x_{2n})$ is independent up to a sign
2946: of the ordering of its vertices.
2947: Indeed,
2948: consider $2n+1$ points $(x_1, x_2, x_1, x_3,...,x_{2n})$
2949: and apply relation (\ref{6.15.02.1}).
2950:
2951:
2952: {\it The Betti cohomology of $SL_n(\C)$}. Recall that $SU(n)$ is a retract of
2953: $SL_n(\C)$.
2954: It is well known that
2955: $$
2956: H_{\rm top}^{\bullet} (SU(n), \Z) =
2957: H_{\rm top}^{\bullet} (S^3 \times S^5 \times ... \times S^{2n-1}, \Z) =
2958: \Lambda^*(B_3,
2959: B_5, ..., B_{2n-1})
2960: $$
2961: The restriction from $SU(n)$
2962: to $SU(m)$ kills the classes $B_{2k-1}$ for $k>m$. If $k\leq m$
2963: it identifies the class
2964: $B_{2k-1}$ for $SU(n)$ with the one for $SU(m)$.
2965: The class $B_{2n-1}$ for $SU(n)$
2966: is provided by the fundamental class of the sphere $S^{2n-1} \subset \C^n$.
2967: Namely, it is the pull back of the
2968: fundamental class under the map $SU(n) \lra S^{2n-1}$ provided by a choice of a point on
2969: $S^{2n-1}$. This sphere has the orientation
2970: induced by the one of $\C^n$.
2971: Thus
2972: \begin{equation} \label{1.20.1}
2973: \Z \cdot B_{n} = {\rm Ker}\Bigl(H_{\rm top}^{2n-1}(SU(n), \Z) \lra
2974: H_{\rm top}^{2n-1}(SU(n-1), \Z) \Bigr)
2975: \end{equation}
2976: The transgression in the Leray spectral sequence for the universal $SU(n)$-bundle
2977: $EU(n) \lra BU(n)$ provides an isomorphism
2978: $$
2979: \Z\cdot B_n \lra \frac{H^{2n}(BSU(n), \Z)}{\oplus_{0 < i < 2n}H^i \cdot H^{2n-i}}
2980: $$
2981: and identifies $B_{n}$ with the Chern class
2982: $c_n \in H^{2n}(BSU(n), \Z)$ of the associated vector bundle.
2983:
2984: {\it The De Rham cohomology of $SL_n(\C)$}. Consider the differential form
2985: \begin{equation} \label{7.1.02.2}
2986: \omega_{D_n}:= {\rm tr}(g^{-1}dg)^{2n-1} \in \Omega^{2n-1}(SL_N)
2987: \end{equation}
2988: Its restriction to the subgroup $SL_m$ is zero for $m<n$. It follows that
2989: the cohomology class
2990: $$
2991: [\omega_{D_n}] \in H^{2n-1}_{\rm DR}(SL_n, \C)
2992: $$
2993: is a multiple of $B_n$.
2994: The Hodge considerations
2995: show that $[\omega_{D_n}] \in (2\pi i)^n \Q \cdot
2996: B_n$.
2997:
2998: \begin{lemma} \label{7.1.02.3} The differential form $\omega_{D_n}$ is an
2999: $\R(n-1)$-valued form. In particular it provides
3000: a cohomology class
3001: $$
3002: b_n := \beta_{\rm DR}(\omega_{D_n}) \in H^{2n-1}_m(SL_n(\C), \R(n-1))
3003: $$
3004: \end{lemma}
3005:
3006: {\bf Proof}. An easy calculation
3007: shows that the value of the exterior form ${\omega_{D_n}}_{|T_e{\Bbb H}_n}$
3008: on
3009: $$
3010: (e_{1,n} + e_{n,1}) \wedge i(e_{1,n} - e_{n,1}) \wedge ... \wedge
3011: (e_{n-1, n} + e_{n, n-1}) \wedge i(e_{n-1, n} - e_{n, n-1}) \wedge e_{n,n}
3012: $$
3013: is non zero, and obviously lies in $\Q(n-1)$.
3014:
3015: On the other hand the values of the form $\omega_{D_n}$
3016: lie in a one dimensional $\R$--vector space. Indeed, the space of
3017: $su(n)$--invariant real exterior
3018: $(2n-1)$--forms on the space of all
3019: hermitian $n \times n$ matrices,
3020: which have zero restriction to the subspace of hermitian
3021: $(n-1) \times (n-1)$ matrices,
3022: is one-dimensional. The exterior form ${\omega_{D_n}}_{|T_e{\Bbb H}_n}$
3023: belongs to the complexification of this space. The lemma follows from this.
3024:
3025:
3026:
3027: We call the cohomology class provided by this lemma the Borel class, and
3028: use it below to construct the Borel regulator.
3029:
3030:
3031:
3032:
3033: {\bf 5. Comparison of the Grassmannian and Borel cohomology classes of $GL_n(\C)$}.
3034: Let $[C_n^G]$ be the cohomology class of the $(2n-1)$--cocycle
3035: of $GL_n(\C)$ provided by the Grassmannian $n$--logarithm (see Corollary 5.5).
3036: We want to compare it with the Borel class.
3037:
3038: Let us consider the following integral
3039: $$
3040: \widetilde C_n(H_1, ..., H_{2n-1}):=
3041: $$
3042: \begin{equation} \label{A362}
3043: -n^{2n-1}\cdot \int_{\C \PP^{n-1}}
3044: \frac{H_1(z, \overline z)}{(z, \overline z)} d\frac{H_2(z, \overline z)}{(z, \overline z)} \wedge ... \wedge d\frac{H_{2n-1}(z, \overline z)}{(z, \overline z)}
3045: \end{equation}
3046: where the $H_i$ are arbitrary complex matrices and $H_i(z, \overline z)$
3047: are the bilinear form in $z, \overline z$ given by the matrix $H_i$.
3048: We claim that it is a $(2n-1)$-cocycle of the Lie algebra $gl_n(\C)$, and it is
3049: obtained by differentiating the
3050: group cocycle provided by the function (\ref{1221M}).
3051: We put these facts in the following framework.
3052:
3053: If we restrict to the case when $H_i$ are hermitian matrices,
3054: integral (\ref{A362}) admits the following interpretation.
3055: Let us construct a map
3056: $$
3057: {\Bbb M}_e: \C \PP^{n-1} \lra T^*_e{\Bbb H}_n
3058: $$
3059: which is a version of the moment map.
3060: For a point $z \in \C \PP^{n-1}$ the value of the $\langle M_e(z), v\rangle$
3061: of the functional ${\Bbb M}_e(z)$ on a vector $v \in T_e{\Bbb H}_n$ is defined as follows.
3062: Let $e(t)$ be a path in
3063: ${\Bbb H}_n$ such that $e(0) =e$ and $\stackrel{{\bf .}}{e}(0) =v$.
3064: recall the measure $\mu_x$ defined in Section 5.1. Then
3065: $$
3066: <{\Bbb M}_e(z), v>:= \frac{d}{dt}\log \frac{\mu_{e(t)}(z)}{\mu_{e}(z)}|_{t=0}
3067: $$
3068: Choose coordinates $z_1, ..., z_{n}$ in $V_n$ such that
3069: $(z, \overline z):= |z_1|^2 + ... + |z_{n}|^2$ corresponds to the point $e$.
3070: Then $T_e{\Bbb H}_n$ is identified with the space of hermitian $(n\times n)$
3071: matrices $H$. It follows from
3072: (\ref{FUBS}) that
3073: \begin{equation} \label{1/18.2}
3074: <{\Bbb M}_e(z), H>:= n \frac{H(z,\overline z)}{(z,\overline z)}
3075: \end{equation}
3076: The map ${\Bbb M}_e$ is clearly $SU(n)$-invariant. Its image is an $SU(n)$-orbit
3077: in $T^*_e{\Bbb H}_n$ isomorphic to $\C \PP^{n-1}$.
3078:
3079: We need the following general construction. Let $V$ be a real vector space and $M$ a compact subset of $V^*$ which is the closure of a $k$-dimensional submanifold. Any element $\omega \in \Lambda^kV$ can be viewed as a $k$-form
3080: $\omega$ on $V^*$. Integrating it over $M$ we get
3081: an exterior form $C_M \in \Lambda^kV^*$. If $M$ is a cone over $M'$ with the vertex at the origin then $\int_M\omega = \int_{M'}i_E\omega$ where $E$ is the Euler vector field on $V$.
3082:
3083:
3084: Applying this construction to the cone over the orbit $M_e$
3085: based at the origin we get an
3086: $SU(n)$-invariant element
3087: $\widetilde C_n \in \Lambda^{2n-1}T^*_e{\Bbb H}_n$.
3088: It follows from (\ref{1/18.2}) that
3089: it is given by formula (\ref{A362}) multiplied by $2n$.
3090:
3091:
3092: Another invariant $(2n-1)$-cocycle
3093: $C_n$ of the Lie algebra $gl_n$, considered by Dynkin [Dy], is given by
3094: \begin{equation} \label{7.1.02.1}
3095: C_n(X_1, ..., X_{2n-1}) = \frac{1}{n! }{\rm Alt}_{2n-1}{\rm Tr}(X_1 X_2 ... X_{2n-1})
3096: \end{equation}
3097: Let $[C_n]$ be the cohomology class
3098: of $GL_n(\C)$
3099: corresponding to the cocycle $C_n$.
3100:
3101: \begin{theorem} \label{1.20.10} One has
3102: $$
3103: \widetilde C_n \quad = -(-1)^{\frac{(n-1)n}{2}}\frac{(2\pi i)^{n-1}n^{2n-1}(n-1)!}{(2n-1)!}
3104: \cdot C_n
3105: $$
3106: and the class $[C_n^G]$ is a non zero rational multiple of $[C_n]$:
3107: $$
3108: [C_n^G] = - (-1)^{\frac{n(n+1)}{2}}\frac{(n-1)!^3}{(2n-2)!(2n-1)!} [C_n]
3109: $$
3110: \end{theorem}
3111:
3112: {\bf Proof}. The second claim follows from the first
3113: using Proposition \ref{1/18.1}.
3114:
3115:
3116: Let us prove the first claim.
3117: The restriction of the cocycle $C_n$ to the Lie subalgebra $gl_{n-1}(\C)$ equals to zero.
3118: This follows, for instance, from the Amitsur-Levitsky theorem: for any $n \times n$
3119: matrices $A_1, ..., A_{2n}$ one has
3120: ${\rm Alt}_{2n} (A_1, ..., A_{2n})=0$.
3121:
3122:
3123: On the other hand the restriction of the cocycle $\widetilde C_n$
3124: to the Lie subalgebra of matrices
3125: $(a_{ij})$ where $a_{1j} = a_{j1} =0$
3126: is zero.
3127: Indeed, in this case the form we integrate
3128: in (\ref{A362}) is a differential $(2n-2)$-form in $dz_2, ..., dz_{n-1},
3129: d \overline z_2, ..., d\overline z_{n-1}$ and thus it is zero.
3130: So thanks to (\ref{1.20.2}) and (\ref{1.20.1}) we conclude that the
3131: cocycle $C_n $ is proportional to $\widetilde C_n$.
3132: To determine the proportionality coefficient
3133: we compute the values of the both cocycle on a special element
3134: $E_n \in \Lambda^{2n-1}gl_n$. To write it down denote by $e_{i,j}$ the
3135: elementary $n\times n$ matrix whose only non-zero entry is $1$ on the $(i,j)$ place. Then
3136: \begin{equation} \label{1/16.2}
3137: E_n:= \bigwedge_{j=1}^{n-1}(e_{j,n}\wedge e_{n,j} )\wedge e_{n,n}
3138: \end{equation}
3139: A direct computation shows that
3140: $$<C_n, E_n> = 1
3141: $$
3142: Indeed, to get a non zero trace we have to multiply $(n-1)$ blocks
3143: $e_{n,j} e_{j,n}$, as well as $e_{n,n}$, which can be inserted
3144: anywhere between these blocks.
3145: So there are $(n-1)! n = n!$ possibilities.
3146:
3147: Let us compute the value of the cocycle $\widetilde C_n$ on $E_n$.
3148: \begin{lemma}
3149: Integral (\ref{A362}) equals
3150: \begin{equation} \label{1/18.4}
3151: \frac{-n^{2n-1}}{(2n-1)!}\cdot {\rm Alt}_{2n-1}\int_{\C \PP^{n-1}}
3152: \frac{H_1(z, \overline z)d H_2(z, \overline z) \wedge ... \wedge dH_{2n-1}(z, \overline z)}{(z, \overline z)^{2n-1}}
3153: \end{equation}
3154: \end{lemma}
3155:
3156: {\bf Proof}. By Proposition \ref{1.2.} integral (\ref{A362}) equals to
3157: \begin{equation} \label{1?1}
3158: \frac{-n^{2n-1}}{(2n-1)!}{\rm Alt}_{2n-1}\cdot \int_{\C \PP^{n-1}}
3159: \frac{H_1(z, \overline z)}{(z, \overline z)} d\frac{H_2(z, \overline z)}{(z, \overline z)} \wedge ... \wedge d\frac{H_{2n-1}(z, \overline z)}{(z, \overline z)}
3160: \end{equation}
3161: One has, for $i=2, ..., 2n-1$, that
3162: $$
3163: d\frac{H_{i}(z, \overline z)}{(z, \overline z)} =
3164: \frac{(z, \overline z) d H_{i}(z, \overline z) - H_{i}(z, \overline z) d (z, \overline z)}{(z, \overline z)^2}
3165: $$
3166: Substituting
3167: $$
3168: \frac{- H_{i}(z, \overline z) d (z, \overline z)}{(z, \overline z)^2}\quad \mbox{instead of} \quad d\frac{H_i(z, \overline z)}{(z, \overline z)} \quad \mbox{in (\ref{1?1})}
3169: $$
3170: we get zero since $H_1$ and $H_i$
3171: appear in a symmetric way and thus disappear after the alternation.
3172: The lemma follows.
3173:
3174:
3175: Let us calculate integral (\ref{1?1}) in the special case
3176: $$
3177: H_{2n-1}(z, \overline z) = |z_n|^2, \quad H_{2k-1}(z, \overline z) = z_k \overline z_n, \quad H_{2k}(z, \overline z) = z_n \overline z_k
3178: $$
3179: so that $H_1 \wedge ... \wedge H_{2n-1} =E_n$.
3180: We will restrict the integrand to the affine part $\{z_n=1\}$
3181: and then perform the integration.
3182: Since $dH_{2n-1}(z, \overline z) =0$ on $\{z_n=1\}$ and
3183: $d z_k \wedge d\overline z_k = -2i dx_k \wedge dy_k$
3184: we get
3185: $$
3186: -(-1)^{\frac{(n-1)(n-2)}{2}}\frac{(-2i)^{n-1}n^{2n-1}}{2n-1}\int_{\C ^{n-1}}\frac{
3187: d^{n-1}x d^{n-1} y}
3188: {(1+ |z_1|^2 + ... + |z_{n-1}|^2)^{2n-1}} =
3189: $$
3190: $$
3191: -(-1)^{\frac{(n-1)n}{2}}\frac{(2i)^{n-1}n^{2n-1}}{2n-1} {\rm vol}(S^{2n-3})
3192: \int_0^{\infty}\frac{r^{2n-3}dr}{(1+r^2)^{2n-1}} =
3193: $$
3194: $$
3195: -(-1)^{\frac{(n-1)n}{2}}\frac{(2\pi i)^{n-1}n^{2n-1}}{(n-1)!(2n-1)}
3196: {2n-2 \choose n-1}^{-1}
3197: $$
3198: since the volume of the sphere $S^{2n-3}$ is
3199: $\frac{2\pi^{n-1}}{(n-2)!}$ and
3200: \begin{equation} \label{1/18.3}
3201: \int_0^{\infty}\frac{r^{2n-3}dr}{(1+r^2)^{2n-1}} \quad = \quad \frac{1}{2}\int_0^{\infty}\frac{r^{n-2}dr}{(1+r)^{2n-1}} \quad = \quad \frac{1}{2n-2} {2n-2 \choose n-1}^{-1}
3202: \end{equation}
3203: To get the last equality we integrate by parts:
3204: $$
3205: \int_0^{\infty}\frac{r^{n-2}dr}{(1+r)^{2n-1}} = - \frac{1}{2n-2}\int_0^{\infty}r^{n-2} \left(
3206: \frac{1}{(1+r)^{2n-2}}\right)'dr =
3207: $$
3208: $$
3209: \frac{n-2}{2n-2}\int_0^{\infty}r^{n-3}\frac{dr}{(1+r)^{2n-2}} = ... =
3210: \frac{(n-2)!n!}{(2n-2)!}\int_0^{\infty}\frac{dr}{(1+r)^{n+1}} =
3211: \frac{1}{n-1} {2n-2 \choose n-1}^{-1}
3212: $$
3213: Theorem \ref{1.20.10} is proved.
3214:
3215:
3216:
3217:
3218: {\bf 6. Construction of the Borel regulator
3219: via Grassmannian polylogarithms}.
3220: Let $G$ be a group. The diagonal map $\Delta: G \lra G \times G$ provides a homomorphism $\Delta_*: H_n(G) \lra H_n(G \times G)$.
3221: %By the Kunneth formula $H_n(G \times G)_{\Q} =
3222: %\oplus_{ 0 \leq k \leq n}H_k(G)_{\Q} \otimes H_{n-k}(G)_{\Q}$.
3223: Recall that
3224: $$
3225: {\rm Prim}H_n(G) := \{x \in H_n(G)| \Delta_*(x) = x \otimes 1 + 1 \otimes x\}
3226: $$
3227: Set $A_{\Q}:= A \otimes \Q$. One has
3228: $$
3229: K_n(F)_{\Q} = {\rm Prim}H_n(GL(F))_{\Q} = {\rm Prim}H_nGL_n(F)_{\Q}
3230: $$
3231: where the second isomorphism is provided by Suslin's stabilization theorem.
3232:
3233: The Borel regulator is a map
3234: $$
3235: r_{n}^{\rm Bo}: K_{2n-1}(\C)_{\Q} \lra \R(n-1)
3236: $$
3237: provided by pairing the class $b_n \in H^{2n-1}(GL_{2n-1}(\C), \R(n-1))$ with
3238: the subspace $K_{2n-1}(\C)_{\Q} \subset H_{2n-1}(GL_{2n-1}(\C), {\Q})$.
3239:
3240: Recall the Grassmannian complex $C_*(n)$
3241: $$
3242: ... \stackrel{d}{\lra} C_{2n-1}(n) \stackrel{d}{\lra} C_{2n-2}(n)
3243: \stackrel{d}{\lra} ... \stackrel{d}{\lra} C_{0}(n)
3244: $$
3245: where $C_k(n)$ is the free abelian group generated by configurations
3246: of $k+1$
3247: vectors $(l_0, ..., l_{k})$ in generic position in an $n$--dimensional
3248: vector space over a field $F$, and $d$ is given by the standard
3249: formula
3250: (see s. 3.1 in [G2]). The group $C_k(n)$ is in degree $k$.
3251: Since it is a homological resolution of the trivial $GL_n(F)$--module $\Z$
3252: (see Lemma 3.1 in [G2]),
3253: there is a canonical homomorphism
3254: $$
3255: \varphi_{2n-1}^n: H_{2n-1}(GL_n(F)) \lra H_{2n-1}(C_*(n))
3256: $$
3257: Thanks to Lemma \ref{7.1.02.3}
3258: the Grassmannian $n$--logarithm function provides a homomorphism
3259: \begin{equation} \label{6.24.02.3}
3260: {\cal L}_n^G: C_{2n-1}(n) \lra \R(n-1); \quad (l_0, ..., l_{2n-1}) \lms
3261: {\cal L}^G_n(l_0, ..., l_{2n-1})
3262: \end{equation}
3263: Thanks to the first $(2n+1)$--term functional equation for ${\cal L}_n^G$,
3264: see (\ref{4}),
3265: it is zero on the subgroup $dC_{2n}(n)$. So it induces a homomorphism
3266: $$
3267: {\cal L}_n^G: H_{2n-1}(C_{*}(n)) \lra \R(n-1);
3268: $$
3269: \begin{lemma} \label{6.24.02.1} The composition ${\cal L}_n^G\circ
3270: \varphi_{2n-1}^n$ coincides with the class $[C_n^G]$.
3271: \end{lemma}
3272:
3273: {\bf Proof}. Standard, see [G4].
3274:
3275:
3276: To construct the Borel regulator we extend, as in s. 3.10 of [G2],
3277: the class $[C_n^G]$
3278: to a class of $GL_{2n-1}(\C)$. Let us recall the key steps.
3279:
3280: Let $\Z[S]$be the free abelian group generated
3281: by a set $S$. Let $F$ be a field.
3282: Applying the covariant functor $\Z \lms \Z[X(F)]$
3283: to the bi-Grassmannian
3284: $\widehat G(n)$ (see s. 4.2), and taking the alternating sum of
3285: the obtained homomorphisms,
3286: we get a bicomplex.
3287: Using Lemma \ref{6.12.02.5} we see that
3288: it looks as follows ([G2], s. 3.7):
3289: $$
3290: \begin{array}{cccccccc}
3291: &&&&&... & \stackrel{d}{\lra}&C_{2n-1}(2n-1)\\
3292: &&& &&&&\downarrow \\
3293: &&&...&&...&&...\\
3294: &&& \downarrow &&&&\downarrow \\
3295: & ... & \stackrel{d}{\lra} & C_{2n-1}(n+1) & \stackrel{d}{\lra} & ... &
3296: \stackrel{d}{\lra} & C_{n+1}(n)\\
3297: &\downarrow && \downarrow &&&&\downarrow \\
3298: ... \stackrel{d}{\lra} & C_{2n-1}(n) & \stackrel{d}{\lra} & C_{2n-2}(n) &
3299: \stackrel{d}{\lra}& ... &\stackrel{d}{\lra} & C_n(n)
3300: \end{array}
3301: $$
3302: In particular the bottom row is the stupid truncation of the
3303: Grassmannian complex at
3304: the group $C_n(n)$.
3305: The total complex of this bicomplex is a homological complex,
3306: called the weight $n$ bi--Grassmannian complex $BC_*(n)$.
3307: In particular there is a homomorphism
3308: \begin{equation} \label{6.24.02.2}
3309: H_{2n-1}(C_*(n)) \lra H_{2n-1}(BC_*(n))
3310: \end{equation}
3311:
3312: In [G1-2] we proved that there are homomorphisms
3313: $$
3314: \varphi_{2n-1}^m: H_{2n-1}(GL_m(F)) \lra H_{2n-1}(BC_*(n)),
3315: \quad m \geq n
3316: $$
3317: whose restriction to the subgroup $GL_n(F)$ coincides
3318: with the composition
3319: $$
3320: H_{2n-1}(GL_n(F)) \stackrel{{\varphi_{2n-1}^n}}{\lra} H_{2n-1}(C_*(n))
3321: \stackrel{(\ref{6.24.02.2})}{\lra} H_{2n-1}(BC_*(n)),
3322: $$
3323: Let us extend homomorphism
3324: (\ref{6.24.02.3}) to a homomorphism
3325: $$
3326: {\cal L}_n^G: BC_{2n-1}(n) \lra \R(n-1)
3327: $$
3328: by setting it zero on the groups $C_{2n-1}(n+i)$ for $i>0$.
3329: The second $(2n-1)$--term functional equation for
3330: the Grassmannian $n$--logarithm function, see (\ref{5}),
3331: just means that the composition
3332: $$
3333: C_{2n}(n+1) \lra C_{2n-1}(n) \stackrel{{\cal L}_n^G}{\lra} \R(n-1),
3334: $$
3335: where the first map is a vertical arrow in $BC_*(n)$, is zero.
3336: Therefore we get a homomorphism
3337: $$
3338: {\cal L}_n^G: H_{2n-1}(BC_{*}(n)) \lra \R(n-1)
3339: $$
3340: \begin{corollary} \label{1.20.11as}
3341: One has
3342: $$
3343: [C^G_n] = -(-1)^{n(n+1)/2} \frac{(n-1)!^2}{(2n-2)!(2n-1)!}
3344: \cdot \frac{b_n}{n}
3345: $$
3346: \end{corollary}
3347:
3348: {\bf Proof}. Indeed,
3349: $$
3350: {\rm Alt}_{2n-1}{\rm Tr}(X_1 \cdot ... \cdot X_{2n-1}) =
3351: <{\rm tr}(g^{-1}dg)^{2n-1}_{|sl_n}, X_1 \wedge ... \wedge X_{2n-1}>
3352: $$
3353: So the claim follows from Theorem \ref{1.20.10} since, as it clear from comparison of formulas
3354: (\ref{7.1.02.1}) and (\ref{7.1.02.2}),
3355: $
3356: b_n = n![C_n]
3357: $.
3358: The corollary is proved.
3359:
3360:
3361: \begin{theorem} \label{6.24.02.7}
3362: The composition
3363: \begin{equation} \label{6.24.02.6}
3364: K_{2n-1}(\C) \stackrel{\sim}{\lra}
3365: {\rm Prim} H_{2n-1}(GL_{2n-1}(\C), \Q) \stackrel{\varphi_{2n-1}^{2n-1}}{\lra}
3366: \end{equation}
3367: $$
3368: H_{2n-1}(BC_{*}(n)_{\Q}) \stackrel{{\cal L}_n^G}{\lra} \R(n-1)
3369: $$
3370: equals to
3371: \begin{equation} \label{7.02.1.4}
3372: -(-1)^{n(n+1)/2} \cdot \frac{(n-1)!^2}{n(2n-2)!(2n-1)!}r_{n}^{\rm Bo}
3373: \end{equation}
3374: \end{theorem}
3375:
3376: {\bf Proof}. Recall that
3377: restriction to $GL_n$ of the map $\varphi_{2n-1}^{2n-1}$
3378: coincides with the map $\varphi_{2n-1}^n$. Therefore Lemma \ref{6.24.02.1}
3379: guarantees that restriction to $GL_n(\C)$
3380: of the composition of the last two arrows
3381: coincides with the map given by the class $[C_{2n-1}^G]$.
3382: So Corollary \ref{1.20.11as} implies the theorem.
3383:
3384:
3385:
3386: {\bf 7. Comparing $[D_n]$ and $B_n$}.
3387: The following result is not used below.
3388:
3389: \begin{theorem} \label{1.20.11}
3390: $$
3391: [D_n] \quad = (2\pi i)^{n}(2n-1)B_n
3392: $$
3393: \end{theorem}
3394:
3395: {\bf Proof}. The transgression identifies the class $B_n$ with
3396: the Chern class of the universal bundle over $BG$, where $G = GL_n(\C)$.
3397: We will compute explicitly the transgression of the $n$-th component of the
3398: Chern character of the
3399: universal vector bundle $p: E \lra BG$.
3400: Let ${\cal A}$ be a connection
3401: on $E$. Then the $n$-th Chern class is
3402: represented by the $2n$-form
3403: $$
3404: c_n(A):= \frac{tr F_{\cal A}^n}{{(2\pi i)}^n}
3405: $$
3406: where $F_A := d{\cal A} + {\cal A} \wedge {\cal A}$ is the curvature form.
3407:
3408: Let $q: EG \lra BG$ be the principal fibration associated with $E$.
3409: Then the form $q^*c_n({\cal A})$ is exact. If
3410: $d\omega = q^*c_n({\cal A})$ and
3411: $F$ is a fiber of $q$ then $\omega|_F$
3412: is closed, its cohomology class is transgressive, and goes to
3413: $[c_n({\cal A})]$. To do
3414: the computation we choose a connection ${\cal A}_0$ on $BG$ which is
3415: flat in a neighbourhood $U$
3416: of a point $x \in BG$. It provides a trivialization of the bundle $E$
3417: over $U$ as well as a trivialization $\varphi: EG|_U \lra G \times U$.
3418:
3419: The bundle
3420: $q^*E$ has a canonical trivialization. It provides a connection ${\cal A}_1$
3421: on $q^*E$. So there are two connections, $q^*{\cal A}_0$ and
3422: ${\cal A}_1$ on $q^*E$. One has ${\cal A}_1 = q^*{\cal A}_0 + g^{-1}dg$, where
3423: $(g,u) = \varphi(x)$. Let
3424: $$
3425: {\cal A}(t):= t {\cal A}_1 +(1-t) q^*{\cal A}_0 =
3426: tg^{-1}dg + q^*{\cal A}_0
3427: $$
3428: It can be thought of as a connection on the lifting of the
3429: bundle $q^*E$
3430: to $EG \times [0,1]$; here $t \in [0,1]$. The curvature $F(t)$ of this
3431: connection is
3432: $$
3433: F(t)= g^{-1}dg dt + t^2g^{-1}dg\wedge g^{-1}dg
3434: $$
3435: The push forward of the form $trF(t)^n$ down to $EG$ is a primitive for the form
3436: $trF_{{\cal A}(1)}^n - trF_{{\cal A}(0)}^n$. It is given (in $q^{-1}U$) by
3437: $$
3438: \int_0^1tr F(t)^n = \frac{1}{2n-1}tr (g^{-1}dg )^{2n-1}
3439: $$
3440: Theorem \ref{1.20.11} is proved.
3441:
3442:
3443: {\bf 8. On the
3444: motivic nature of the Grassmannian $n$--logarithm functions}.
3445: According to the results of the previous section understanding of the Borel regulator,
3446: and hence special values of the Dedekind $\zeta$--functions, is reduced to
3447: study of properties of the Grassmannian $n$--logarithm function
3448: ${\cal L}_n^{G}$.
3449:
3450: Recall that a framed mixed Hodge-Tate structure has a natural $\R$-valued invariant ([BD]), called the Lie period. Thus a variation of Hodge-Tate structures ${\Bbb L}$
3451: over a base $X$
3452: provides a period function ${\Bbb L}^{\cal M}$ on $X(\C)$.
3453:
3454:
3455:
3456: \begin{conjecture} \label{6.24.02.10} There exists a variation ${\Bbb L}^{\cal M}_n$
3457: of framed
3458: mixed Tate motives over $\widehat G_{n-1}^n$ such that:
3459: $$
3460: \sum_{i=0}^{2n}(-1)^i a_i^*{\Bbb L}_n^{\cal M} = 0; \quad
3461: \sum_{j=0}^{2n}(-1)^j b_j^*{\Bbb L}_n^{\cal M} = 0;
3462: $$
3463: and the Lie period ${\cal L}^{\cal M}_n$
3464: of its Hodge realization satisfies
3465: \begin{equation} \label{6.24.02.11}
3466: {\cal L}^{\cal M}_n - {\cal L}^{G}_n = \sum_{i=0}^{2n-1}(-1)^i
3467: a_i^*{F}_n
3468: \end{equation}
3469: where $F_n$ is a function on $\widehat G_{n-2}^n(\C)$.
3470:
3471: b) The functional equations satisfied by ${\cal L}^{\cal M}_n$
3472: essentially determine it:
3473: the space of all smooth/measurable functions satisfying these functional equations is finite dimensional.
3474: \end{conjecture}
3475:
3476: {\bf Remark}. The function $F_n$ is obviously
3477: not determined by
3478: (\ref{6.24.02.11}) -- add a function coming from $\widehat G^n_{n-3}(\C)$.
3479: Nevertheless we expect that there exists a {\it canonical} explicit
3480: choice for $F_n$. Then formula (\ref{6.24.02.11}) can be considered
3481: as an explicit formula for
3482: ${\cal L}^{\cal M}_n$ in terms of ${\cal L}^{G}$'s.
3483:
3484: Moreover we expect that there exists a canonical homotopy
3485: between the Grassmannian $n$-logarithm (understood as
3486: a cocycle in the Deligne cohomology of the bi-Grassmannian)
3487: and its ``motivic''
3488: bi-Grassmannian counterpart.
3489: Observe that the motivic bi-Grassmannian $n$-logarithm should
3490: have non-trivial components outside of the bottom line of
3491: the bi-Grassmannian, while the defined above (or in [G5])
3492: Grassmannian $n$--logarithm
3493: is concentrated entirely
3494: at the bottom line.
3495:
3496:
3497: A variation of mixed Tate motives over $\widehat G_{n-1}^n$
3498: was constructed in [HaM]. However it is not clear how to relate it to
3499: the function ${\cal L}^G_n$.
3500:
3501:
3502:
3503: Conjecture \ref{6.24.02.10} is known for $n=2$ and $n=3$.
3504:
3505: The $n=2$ case
3506: follows from (\ref{6.13.02.101}), the well known
3507: motivic realization of the dilogarithm, and
3508: Bloch's theorem characterizing the Bloch-Wigner function by
3509: Abel's $5$--term equation it satisfies.
3510:
3511: The $n=3$ case of conjecture \ref{6.24.02.10} follows from the results of
3512: [G1-2], [GZ] and the motivic realization of the trilogarithm.
3513: In particular the part b) is given by
3514: Theorem 1.10 in [G1].
3515:
3516: {\bf Examples}. 1. $n=2$. Then ${\cal L}^{\cal M}_2 = {\cal L}^{G}_2$.
3517:
3518: 2. $n= 3$. The motivic Grassmannian trilogarithm function
3519: has been constructed in [G1-2] in terms of the
3520: classical trilogarithm function. Namely, one has
3521: $$
3522: {\cal L}^{\cal M}_3(l_0, ..., l_5) = \frac{1}{90} {\rm Alt}_6
3523: {\cal L}_3\Bigl( \frac{\Delta(l_0, l_1, l_3)\Delta(l_1, l_2, l_4)
3524: \Delta( l_2, l_0, l_5)}{\Delta(l_0, l_1, l_4)
3525: \Delta(l_1, l_2, l_5)\Delta(l_2, l_0, l_3)}
3526: \Bigr)
3527: $$
3528:
3529: According to Theorem 1.3 of [GZ] ${\cal L}^{\cal M}_3$
3530: is different from ${\cal L}^{G}_3$, and
3531: $$
3532: F_3(l_0, ..., l_4) = \frac{1}{9}{\rm Alt}_5\Bigl(\log
3533: |\Delta(l_0, l_1, l_2)| \log |\Delta(l_1, l_2, l_3)| \log |\Delta(l_2, l_3, l_4)| \Bigr)
3534: $$
3535:
3536: 3. If $n>3$ then ${\cal L}^{\cal M}_n$
3537: is different from ${\cal L}^{G}_n$ since it is already so
3538: for the restriction to the special configuration,
3539: see Theorem \ref{clpoly}.
3540:
3541:
3542: The space of the functional equations
3543: for the function ${\cal L}^G_3$ is smaller than the one for
3544: ${\cal L}^{\cal M}_3$, see Chapter 1.5 of [G3].
3545: A similar situation is expected
3546: for all $n>3$.
3547:
3548:
3549: The space of the functional equations for the motivic
3550: $n$--logarithm function ${\cal L}^{\cal M}_n$ should provide an
3551: explicit construction of the weight $n$ part of the motivic Lie coalgebra of
3552: an arbitrary field $F$, as explained in s. 4.1 in [G6], taking into
3553: account the following correction.
3554:
3555: {\it Correction}. In s. 4.2 of [G6] the subgroup of the functional equations ${\cal R}_n^G$ is supposed to be defined as the subgroup of all functional equations for the function ${\cal L}^{\cal M}_n$, not ${\cal L}^{G}_n$.
3556:
3557:
3558:
3559:
3560:
3561: \section{The Chow dilogarithm and a reciprocity law}
3562:
3563:
3564: The Chow dilogarithm provides a homomorphism $\Lambda^3 \C(X)^* \to \R$
3565: given by
3566: \begin{equation} \label{chowhh}
3567: f_1\wedge f_2\wedge f_3 \lms {\cal P}_2(X;f_1,f_2,f_3):= \frac{1}{2\pi i}
3568: \int_{X(\C)}r_2(f_1,f_2,f_3)
3569: \end{equation}
3570:
3571: In this section we show that the Chow dilogarithm can be
3572: expressed by the function ${\cal L}_2(z)$.
3573: The precise versions of this claim are discussed below.
3574:
3575: %The statement "for any point $(X;f_1,f_2,f_3)$ one has
3576: %${\cal P}_2(X;f_1,f_2,f_3) = \sum_i {\cal L}_2(z_i)$ for some
3577: % complex numbers $z_i$"
3578: %is empty since any complex number can be written in this form.
3579: %However if, for example, a point
3580: %$(X;f_1,f_2,f_3) \in {\cal Z}_2^1$ is defined over ${\overline \Q}$,
3581: %then I claim that one can choose $z_i \in {\overline \Q}$.
3582:
3583: {\bf 1. The set up ([G1-2])}.
3584: For any field $F$ we defined in [G1] the groups
3585: $$
3586: {\cal B}_n(F):= \frac{\Z[F^*]}{{\cal R}_n(F)}, \quad n \geq 2
3587: $$
3588: and homomorphisms
3589: $$
3590: \delta_n : {\cal B}_n(F) \lra {\cal B}_{n-1}(F)\otimes F^*; \quad
3591: \{x\}_n \lms \{x\}_{n-1} \otimes x, \quad n\geq 3
3592: $$
3593: $$
3594: \delta_2:
3595: {\cal B}_2(F) \lra \Lambda^2F^*, \quad \{x\}_2 \lms (1-x) \wedge x
3596: $$
3597: There is a complex $\Gamma(F;n)$
3598: $$
3599: {\cal B}_{n}(F) \stackrel{\delta_n}{\longrightarrow} {\cal B}_{n-
3600: 1}(F)\otimes F^{\ast} \stackrel{\delta_n}{\rightarrow} \ldots \stackrel{\delta_n}{\rightarrow} {\cal B}_{2}(F)\otimes \Lambda^{n-2}F^{\ast}
3601: \stackrel{\delta_n}{\rightarrow}
3602: \Lambda^{n}F^{\ast}
3603: $$
3604: where $\delta_n(\{x\}_k \otimes Y) := \{x\}_{k-1} \otimes x \wedge Y$ for $k>2$, and $(1-x)\wedge x \wedge y$ for $k=2$,
3605: called the weight $n$ polylogarithmic complex.
3606:
3607: If $K$ is a field with
3608: a discrete valuation $v$ and the residue field $
3609: k_v$, then there is a homomorphism of complexes
3610: ${\rm res}_v: \Gamma(K, n) \lra \Gamma(k_v, n-1)[-1]$ (see s.\ 14 of \S
3611: 1 in [G1]). For example for $n=3$ we have
3612: \begin{equation}
3613: \begin{array}{ccccc}
3614: {\cal B}_3(K)&\stackrel{\delta_3}{\longrightarrow}&{\cal B}_2(K)\otimes
3615: K^{\ast}&
3616: \stackrel{\delta_3}{\longrightarrow}&\Lambda^3K^{\ast}\\
3617: &&\downarrow {\rm res}_v &&\downarrow {\rm res}_v \\
3618: &&{\cal B}_2(k_v)&\stackrel{\delta_2}{\longrightarrow}&\Lambda^2k_v^{\ast}
3619: \end{array}
3620: \end{equation}
3621: Here ${\rm res}_v(\{x\}_2 \otimes y)$ is zero unless $v(x) =0$. In the latter case
3622: it is ${\rm res}_v(\{x\}_2 \otimes y) = v(y) \{\overline x\}_2$, where $\overline x$
3623: denotes projection of $x$ to the residue field of $K$.
3624:
3625:
3626: Let $X$ be a regular curve over an algebraically closed field
3627: $k$ and $F:= k(X)^{\ast}$.
3628: Set ${\rm Res}:= \sum_x {\rm res}_x$ where ${\rm res}_x$ is
3629: the residue homomorphism for the valuation on $F$
3630: corresponding to a
3631: point $x$ of $X$. For instance for $n=3$ we get a morphism of complexes
3632: $$
3633: \begin{array}{ccccc}
3634: {\cal B}_3(F)&\stackrel{\delta_3}{\longrightarrow}&{\cal B}_2(F )\otimes
3635: F^{\ast}&
3636: \stackrel{\delta_3}{\longrightarrow}&\Lambda^3F^{\ast}\\
3637: &&\downarrow {\rm Res}&&\downarrow {\rm Res}\\
3638: && {\cal B}_2(k )&\stackrel{\delta_2}{\longrightarrow}& \Lambda^2k^{\ast}
3639: \end{array}
3640: $$
3641:
3642: We will also need a more explicit version $B_2(F)$
3643: of the group ${\cal B}_2(F)$.
3644: Denote by $R_2(F)$ the subgroup of $\Z[{\Bbb P}^1(F)]$
3645: generated by the elements $$
3646: \{0\}, \{\infty\}\quad \mbox{and} \quad \sum_{i=1}^{5}(-1)^i
3647: \{r(x_1,...,\widehat x_i,...,x_5)\}
3648: $$
3649: when $(x_1,...,x_5)$ runs through all $5$-tuples of distinct points in ${\Bbb P}^1(F)$,
3650: and the cross-ratio $r(...)$ is normalized by $r(\infty, 0, 1, x) =x$. Define the Bloch group $B_2(F)$ as
3651: $$
3652: B_2(F):= \frac{\Z[{\Bbb P}^1(F)]}{R_2(F)}
3653: $$
3654:
3655:
3656: One can show that $R_2(F)\subset {\cal R}_2(F)$.
3657: So there is a map
3658: \begin{equation} \label{6.18.02.10}
3659: i: B_2(k)\lra {\cal B}_2(k)
3660: \end{equation}
3661: induced by the identity map on the generators.
3662:
3663: \begin{proposition} \label{mot}
3664: Let $k$ be a number field. Then (\ref{6.18.02.10}) is an isomorphism
3665: modulo torsion.
3666: \end{proposition}
3667:
3668: {\bf Proof}. The map $i$ is clearly surjective. The diagram
3669: $$
3670: \begin{array}{ccc}
3671: B_2(k) &\lra&{\cal B}_2(k) \\
3672: &&\\
3673: \downarrow \delta_2&&\downarrow \delta_2\\
3674: &&\\
3675: \Lambda^2k^*&=&\Lambda^2k^*
3676: \end{array}
3677: $$
3678: is commutative. So we need only to show that if $0 \not = x \in B_2(k)_{\Q}$ and $\delta_2(x)=0$, then $i(x)\not =0$. This follows from the injectivity of the regulator map on $K^{ind}_3(k)_{\Q}$.
3679: Indeed, by Suslin's theorem for a field $F$ one has $K^{ind}_3(F)_{\Q} = {\rm Ker} \delta_2\otimes \Q $. Let us identify $K^{ind}_3(\C)_{\Q}$ with this subgroup of $B_2(\C)_{\Q} $. The restriction of the dilogarithm map
3680: $$
3681: B_2(k) \lra {\cal B}_2(k)\lra (\Z[Hom(k,\C)]\otimes 2 \pi i \R)^+, \qquad \{z\}_2 \lms \{2 \pi i{\cal L}_2(\sigma_i(z))\}
3682: $$
3683: to the subgroup ${\rm Ker} \delta_2\otimes \Q $ gives the Borel regulator $K^{ind}_3(k)_{\Q} \lra \R^{r_2}$ ([G1]) and thus is injective by Borel's theorem.
3684:
3685: {\bf Remark}. For any field $k$ the rigidity conjecture for $K_3^{ind}$
3686: implies that the map $i$ should be an isomorphism, see [G1].
3687:
3688: {\bf 2. The strong reciprocity law}.
3689: \begin{conjecture} \label{reh}
3690: Let $X$ be a regular projective curve over an algebraically closed field
3691: $k$ and $F:= k(X)^{\ast}$.
3692: Then there exists a canonical homomorphism of groups
3693: $
3694: h:\Lambda^3F^* \rightarrow
3695: {\cal B}_2(k)
3696: $
3697: satisfying the following two conditions:
3698:
3699: a) $h(k^* \wedge \Lambda^2F^*) =0$ and the diagram
3700: \begin{equation} \label{6.17.02.3}
3701: \begin{array}{ccccc} \label{reh1}
3702: {\cal B}_3(F)& \stackrel{\delta_3}{\longrightarrow}&{\cal B}_2(F)\otimes F^{\ast}
3703: & \stackrel{\delta_3}{\longrightarrow} & \Lambda^3F^{\ast}\\
3704: &&&&\\
3705: &&{\rm Res}\downarrow&h \swarrow&\downarrow {\rm Res}\\
3706: &&&&\\
3707: &&{\cal B}_2(k)&\stackrel{\delta_2}{\longrightarrow}&\Lambda^2 k^{\ast}
3708: \end{array}
3709: \end{equation}
3710: is commutative.
3711:
3712: b) If $X$ is a curve over $\C$
3713: then
3714: \begin{equation} \label{homotq}
3715: \frac{1}{2\pi i}\int_{X(\C)}r_2(f_1\wedge f_2 \wedge f_3) = {\cal L}_2 \Bigl(h(f_1\wedge
3716: f_2 \wedge f_3)\Bigr)
3717: \end{equation}
3718: \end{conjecture}
3719:
3720:
3721:
3722:
3723:
3724: {\bf Remarks}. 1. b) follows easily if we have a functorial map $h$ such that
3725: ${\rm Res} = \delta_2 \circ h $, see Theorem \ref{55} below.
3726:
3727: 2. According to Suslin's
3728: reciprocity law for the Milnor group $K^M_3(F)$ the projection of
3729: ${\rm Res}(\Lambda^3F^{\ast}) \subset \Lambda^2k^*$ to $K_2(k)$ is zero.
3730: Since by Matsumoto's
3731: theorem $K_2(k)={\rm Coker}(\delta_2)$, one has
3732: ${\rm Res}(\Lambda^3F^{\ast}) \subset {\rm Im}(\delta_2)$. However
3733: ${\rm Ker}(\delta_2)$ is nontrivial, so it is {\it a priory}
3734: unclear that we can lift naturally the map ${\rm Res}$ to a map $h$.
3735: One of the reasons why we can do this is provided by (\ref{homotq}).
3736:
3737:
3738:
3739:
3740:
3741:
3742:
3743:
3744: We prove this conjecture in the following cases:
3745:
3746: a) $X={\Bbb P}^1$; we construct {\it explicitly} a reciprocity homomorphism
3747: $h: \Lambda^3F^*\to B_2(k)$ in Theorem \ref{p4}.
3748:
3749: b) $X$ is an elliptic curve over an algebraically closed field; we construct {\it explicitly} a reciprocity homomorphism $h: \Lambda^3F^* \to {\cal B}_2(k)$ in Theorem \ref{ecur}.
3750:
3751: c) $k = {\overline \Q}$, $X$ is any curve; see Theorem \ref{rl}.
3752:
3753: In the cases a) and b) the homomorphism $h$ satisfies the following additional property. Let $F = k(X)$ and $k$ is not necessarily algebraically closed. Let $k'$ be the field of definition of the divisors $(f_1), (f_2), (f_3)$. Then $h(f_1\wedge f_2\wedge f_3) \in {\cal B}_2(k')$.
3754:
3755: \begin{conjecture}
3756: Let $X$ be a projective regular curve over an algebraically closed field
3757: $k$ and $F:= k(X)$. Then the homomorphism
3758: $$
3759: {\rm Res}: \Gamma(F;n) \longrightarrow \Gamma(k;n-1)[-1]
3760: $$
3761: is homotopic to zero.
3762: \end{conjecture}
3763:
3764: \begin{lemma} \label{mot1}
3765: Assume that we have a map $h$ such that $h(k^* \wedge \Lambda^2F^*) =0$ and
3766: ${\rm Res} = \delta_2 \circ h $. Then $ h \circ \delta_3 = {\rm Res}$.
3767: \end{lemma}
3768:
3769: {\bf Proof}. The image of the group ${\cal B}_2(F)\otimes F^{\ast}$ under the map $h \circ \delta_3$ belongs to the subgroup ${\rm Ker}\delta_2$. Since $h(k^* \wedge \Lambda^2F^*) =0$ one has $h \circ \delta_3 = {\rm Res}$ on ${\cal B}_2(F)\otimes k^{\ast}$. Any element of $k(X)^*$ can be connected via a curve to a constant. This together with the rigidity of ${\rm Ker}\delta_2$ (built into the definition of the group ${\cal B}_2(k)$) implies the result.
3770:
3771:
3772: {\bf 3. The $X= \PP^1$ case}.
3773: Recall that $v_x(f)$ is the order of zero of $f\in k(X)$ at $x$.
3774: Choose a point $\infty$ on ${\Bbb P}^1$.
3775:
3776: \begin{theorem} \label{p4}
3777: Assume that $k = \overline k$. Then the map $
3778: h:\Lambda^3 k({\Bbb P}^1)^* \rightarrow
3779: B_2(k)
3780: $ given by the formula
3781: $$
3782: h(f_1\wedge f_2\wedge f_3):= \sum_{x_i \in {\Bbb P}^1(k)}
3783: v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3)\{r(x_1, x_2, x_3, \infty)\}_2
3784: $$
3785: satisfies all the conditions of conjecture \ref{reh} modulo $6$-torsion.
3786: \end{theorem}
3787:
3788: {\bf Proof}. Let us show that $h$ is independent of the choice of $\infty$, i.e.
3789: $$
3790: \sum_{x_i \in {\Bbb P}^1(k)}
3791: v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3)\{r(x_1, x_2, x_3, a)\}_2
3792: \in B_2(k)
3793: $$
3794: does not depend on $a$. Indeed, the 5-term
3795: relation for the 5-tuple of points $(x_1, x_2, x_3, a, b)$ gives
3796: $$
3797: \sum_{x_i \in {\Bbb P}^1(k)}
3798: v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3)\Bigl(\{r(x_1, x_2, x_3, a)\}_2 -
3799: \{r(x_1, x_2, x_3, b)\}_2\Bigr) =
3800: $$
3801: $$
3802: -\sum_{x_i \in {\Bbb P}^1(k)}
3803: v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3)\Bigr(\{r(x_1, x_2, a, b )\}_2
3804: -\{r( x_1, x_3, a, b)\}_2 + \{r(x_2, x_3, a, b)\}_2\Bigr)
3805: $$
3806: Each of these 3 terms vanishes because $\sum_{x \in {\Bbb P}^1(k)}v_{x}(f) =0$.
3807: for any $f \in k(\PP^1)^*$.
3808:
3809: \begin{proposition} \label{not}
3810: Let $k= \overline k$. Then modulo $6$-torsion
3811: $$
3812: h((1-f) \wedge f \wedge g) = \sum_{x \in {\Bbb P}^1(k)}v_{x}(g)\{f(x)\}_2
3813: $$
3814: \end{proposition}
3815:
3816: {\bf Proof}. Using linearity with respect to $g$ and projective invariance of the cross--ratio we see that it is sufficient to prove the identity for $g=t$.
3817: Then it boils down to
3818: \begin{equation} \label{7.3.02.2}
3819: \sum_{x_i \in {\Bbb P}^1(k)}
3820: v_{x_1}(1-f)v_{x_2}(f)\{r(x_1, x_2, 0, \infty)\}_2 =
3821: \{f(0)\}_2 - \{f(\infty)\}_2
3822: \end{equation}
3823:
3824:
3825: \begin{lemma} \label{notxzc}
3826: Applying $\delta_2$ to both parts of
3827: (\ref{7.3.02.2}) we get the same result modulo $6$-torsion.
3828: \end{lemma}
3829:
3830: {\bf Proof}.
3831: Choose a coordinate $t$ on $\PP^1$ such that $f(\infty) =1$.
3832: Then
3833: \begin{equation} \label{7.3.02.4}
3834: f(t) = \frac{\prod_i (a_i -t)^{\alpha_i}}{\prod_k (c_k -t)^{\gamma_k}}; \quad
3835: 1-f(t)= \frac{B \prod_j (b_j -t)^{\beta_j}}{\prod_k (c_k -t)^{\gamma_k}}
3836: \end{equation}
3837: Observe that $\{f(\infty)\}_2 =0$ modulo $6$-torsion.
3838: The left hand side equals
3839: \begin{equation} \label{7.3.02.1}
3840: \sum_{x_i \in {\Bbb P}^1(k)}
3841: v_{x_1}(1-f)v_{x_2}(f)\{x_1/x_2\}_2 =
3842: \end{equation}
3843: $$
3844: \sum \beta_j \alpha_i \{b_j/a_i\}_2 -
3845: \sum \gamma_k \alpha_i \{c_k/a_i \}_2
3846: - \sum \beta_j\gamma_k \{b_j/c_k\}_2
3847: $$
3848: Applying $\delta_2$ to it we get
3849: $$
3850: \sum \beta_j \alpha_i \cdot \frac{a_i-b_j}{a_i} \wedge \frac{b_j}{a_i} -
3851: \sum \gamma_k \alpha_i \cdot \frac{a_i- c_k}{a_i} \wedge \frac{c_k}{a_i}
3852: - \sum \gamma_k \beta_j \cdot \frac{c_k - b_j}{c_k} \wedge \frac{b_j}{c_k} =
3853: $$
3854: $$
3855: \sum \beta_j \alpha_i \cdot b_j \wedge a_i +
3856: \sum \alpha_i\gamma_k \cdot a_i \wedge c_k
3857: + \sum \gamma_k \beta_j \cdot c_k \wedge b_j \quad +
3858: $$
3859: $$
3860: - \sum_i\frac{\prod_j (a_i-b_j)^{\beta_j }}{\prod_k (a_i-c_k)^{\gamma_k }}
3861: \wedge a_i^{\alpha_i}
3862: + \sum_j\frac{\prod_i (a_i - b_j)^{\alpha_i}}
3863: {\prod_k (c_k - b_j)^{\gamma_k}} \wedge b_j^{\beta_j}
3864: - \sum_k \frac{\prod (a_i- c_k)^{\alpha_i}}{\prod (c_k - b_j)^{\beta_j}}
3865: \wedge c_k^{\gamma_k}
3866: $$
3867: Using (\ref{7.3.02.4}) we see that the second line equals modulo $2$--torsion
3868: to
3869: $$
3870: - \sum_i (1-f(a_i)) \wedge a_i^{\alpha_i} + \sum_j f(b_j)
3871: \wedge b_j^{\beta_i} - \sum_k
3872: \frac{\prod (a_i- c_k)^{\alpha_i}}{\prod (b_j - c_k)^{\beta_j}}
3873: \wedge c_k^{\gamma_k} +
3874: B \wedge \prod a_i^{\alpha_i}
3875: $$
3876: The first two terms are zero since $f(a_i)=0$
3877: and $f(b_j) =1$. The third term equals to
3878: $-(B \wedge \prod c_k^{\gamma_k})$ since, as it follows from (\ref{7.3.02.4}),
3879: $f(c_k) = \infty$ and thus
3880: $$
3881: \frac{\prod (a_i- c_k)^{\alpha_i}}{\prod (b_j - c_k)^{\beta_j}} = -B
3882: $$
3883: On the other hand, we have
3884: $$
3885: \delta_2(\{f(0)\}_2) = (1-f(0)) \wedge f(0) =
3886: \frac{B\prod_j b_j^{\beta_j}}{\prod_k c_k^{\gamma_k}} \wedge
3887: \frac{\prod_i a_i^{\alpha_i}}{\prod_k c_k^{\gamma_k}}
3888: $$
3889: which matches the expression we got for the left hand side. The lemma is proved.
3890:
3891: To prove the proposition it remains to use a rigidity argument.
3892: Namely, we need to show that the identity is valid
3893: for some particular $f$, which is easy, or use proposition \ref{p44}
3894: plus injectivity of the regulator on $K_3(\overline \Q)_{\Q}$.
3895: The proposition is proved.
3896:
3897: Now let us prove the key fact that ${\rm Res} = \delta_2\circ h$. We need to show that
3898: for any 3 rational functions $f_1, f_2, f_3$ on
3899: ${\Bbb P}^1$
3900: $$
3901: \sum_{x \in {\Bbb P}^1(k)} {\rm res}_x (f_1\wedge f_2\wedge f_3) =
3902: $$
3903: \begin{equation} \label {1}
3904: \delta_2 \Bigl( \sum_{x_i \in {\Bbb P}^1(k)}
3905: v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3)\{r(x_1, x_2, x_3, \infty)\}_2 \Bigr)
3906: \end{equation}
3907:
3908: Both sides are obviously homomorphisms
3909: from $\Lambda^3 F^{\ast}$ to $\Lambda^2 k^{\ast}$ which are zero on
3910: $k^{\ast} \wedge \Lambda^2 F^{\ast}$. (The last property for the map
3911: $\sum {\rm res}_x$
3912: is provided by the Weil reciprocity law). We normalize the cross ratio of four points on the projective
3913: line by $r(\infty,0,1,z) =z$.
3914: So it suffices to
3915: check the formula on elements
3916: $$
3917: \frac{z-a_2}{z-a_1} \wedge \frac{z-b_2}{z-b_1} \wedge \frac{z-c_2}{z-c_1}
3918: $$
3919: In this case it follows from
3920: $$
3921: \delta_2 \{r(a_2, b_2, c_2, \infty)\}_2 = \delta_2\left\{\frac{a_2-c_2}{b_2-c_2}\right\}_2 = \frac{b_2-a_2}{b_2-c_2}\wedge \frac{a_2-c_2}{b_2-c_2}
3922: $$
3923:
3924:
3925:
3926: It remains to prove the following proposition.
3927:
3928: \begin{proposition} \label{p44}
3929: $$
3930: {\cal P}_2({\Bbb P}^1;f_1,f_2,f_3)= \sum_{x_i \in {\Bbb P}^1(\C)} v_{x_1}(f_1)v_{x_2}(f_2)v_{x_3}(f_3){\cal L}_2(r(x_1, x_2, x_3, \infty))
3931: $$
3932: \end{proposition}
3933:
3934: {\bf Proof}. We immediately reduce the statement to the situation when $f_1 = 1-z, f_2 = z, f_3 = z-a$
3935: which is a particular case of the following lemma
3936:
3937: \begin{lemma} \label{l454} Let $X$ be an arbitrary curve over $\C$. Then
3938: $$
3939: \int_{X(\C)}r_2((1-f)\wedge f\wedge g)= - \sum_{x \in X(\C)} v_{x}(g) \cdot
3940: {\cal L}_2(f(x))
3941: $$
3942: \end{lemma}
3943:
3944: {\bf Proof}. For functions $f(z)$ and $g(z)$ on $X(\C)$ set
3945: $$
3946: \alpha(f,g): = \log|f|d\log|g| - \log|g|d\log|f|
3947: $$
3948: Consider the following 1-form on $X(\C)$
3949: \begin{equation} \label{99}
3950: {\cal L}_{2}(f) d\arg g -\frac{1}{3} \alpha(1-f,f)\log|g|
3951: \end{equation}
3952: It defines a current on $X(\C)$. We claim that its derivative is equal to:
3953: \begin{equation} \label{11}
3954: 2 \pi \cdot {\cal L}_{2}(f)\delta(g) + r_2((1-
3955: f) \wedge f \wedge g)
3956: \end{equation}
3957: Using $d (d \arg g) = 2\pi \cdot \delta(g)$
3958: %where $\delta(g)$ is a $2$-current given by $<\delta(g), \varphi(x)>:=
3959: %\sum_{x\in X(\C)} v_x(g)\varphi(x)$ for any smooth test function $\varphi(x)$
3960: and
3961: \begin{equation} \label{dl2}
3962: d{\cal L}_2(z) = -\log|1-z| d\arg z + \log|z| d\arg(1-z)
3963: \end{equation}
3964: we see that the differential of the current (\ref{99}) equals to
3965: $$
3966: 2 \pi {\cal L}_{2}(f)\delta(g) \quad + \quad \Bigl( -\log|1-f| d\arg f + \log|f| d\arg(1-f)\Bigr)\wedge d\arg g +
3967: $$
3968: $$
3969: \frac{1}{3}\Bigl(\log|1-f|d\log|f| - \log|f|d\log|1-f|\Bigr)\wedge d\log|g| - \frac{2}{3}\log|g| \cdot d\log|1-f|\wedge d\log|f|
3970: $$
3971: Since $d\log(1-f)\wedge d\log f = 0$ we have $$
3972: d\log|1-f|\wedge d\log|f| = d\arg(1-f)\wedge d\arg f$$
3973: Using this and writing $r_2(f_1 \wedge f_2 \wedge f_3)$ as
3974: $$
3975: \frac{1}{3}(\log|f_1|d\log|f_2|\wedge d\log|f_3| + \quad \mbox{cyclic permutations})
3976: $$
3977: $$
3978: - (\log|f_1|d\arg(f_2)\wedge d\arg f_3 + \quad \mbox{cyclic permutations})
3979: $$
3980: we come to (\ref{11}).
3981: Integrating we get the lemma. The theorem is proved.
3982:
3983: {\bf 4. Expressing the Chow dilogarithm via the classical one}. Let $\pi:
3984: Y \rightarrow S$ be a family of curves over a base $S$ over $\C$ and
3985: $f_1,f_2,f_3 \in \C(Y)^*$.
3986: We get a function at the generic point of $S$. Its value at
3987: $s \in S$ is given by the Chow dilogarithm
3988: ${\cal P}_2(Y^s;f^s_1,f^s_2,f^s_3)$, where
3989: $Y^s$ is the fiber of
3990: $\pi$ at $s$. Denote it by
3991: ${\cal P}_2(Y \rightarrow S ;f_1,f_2,f_3)$.
3992:
3993: \begin{theorem} \label{55}
3994: a) Let $\pi:
3995: Y \rightarrow S$ be a family of curves over a base $S$ over $\C$. Then
3996: there are rational functions $\varphi_i$ on $S$ such that
3997: $$
3998: {\cal P}_2(Y \rightarrow S;f_1,f_2,f_3) = \sum_i {\cal L}_2(\varphi_i(s))
3999: $$
4000:
4001: b) Let $k = \C(S)$, $X$ is the generic fiber of $\pi$, and $F=k(X)$. Suppose that there exists a map $h:\Lambda^3F^* \rightarrow
4002: {\cal B}_2(k)$ such that ${\rm Res} = \delta_2 \circ h $. Then
4003: \begin{equation} \label{3211}
4004: d{\cal P}_2(Y \rightarrow S;f_1,f_2,f_3) \quad = \quad d{\cal L}_2(h(f_1,f_2,f_3))
4005: \end{equation}
4006: \end{theorem}
4007:
4008: {\bf Proof}.
4009: a)
4010: We use the existence of the transfer map on $K^M_3$
4011: to reduce the statement to the case $X= {\Bbb P}^1$.
4012:
4013: Choose a projection $p: X \rightarrow {\Bbb P}^1$.
4014: We may suppose without loss of generality that $p$ is a (ramified)
4015: Galois covering
4016: with the Galois group $G$.
4017: Indeed, let $p_1:Y \rightarrow X$
4018: be such a covering that its composition with $p$
4019: is a Galois covering. Indeed,
4020: $$
4021: {\cal P}_2(Y\lra S; p_1^*f_1,p_1^*f_2, p_1^*f_3) = \frac{1}{{\rm deg} p_1}{\cal P}_2(X\lra S; f_1, f_2, f_3)
4022: $$
4023:
4024: Then $\sum_{g \in G} g^*\{f_1, f_2, f_3\} \in p^*K_3^M(k({\Bbb P}^1))$.
4025: It coincides with $p^*$ of the transfer of the element $\{f_1, f_2, f_3\} \in K_3^M(F)$.
4026: This means that there exist $s^{(i)}_1,s^{(i)}_2,s^{(i)}_3 \in k({\Bbb P}^1)$
4027: and $g_j,h_j \in k(X)$ such that
4028: \begin{equation} \label{6.18.02.1}
4029: \sum_{g \in G} g^*(f_1\wedge f_2 \wedge f_3) - p^*\sum_i
4030: s^{(i)}_1\wedge s^{(i)}_2\wedge s^{(i)}_3 =
4031: \sum_j(1-g_j)\wedge g_j \wedge h_j
4032: \end{equation}
4033: Therefore
4034: $$
4035: {\cal P}_2(Y \rightarrow S;f_1,f_2,f_3)=
4036: \frac{1}{|G|}\sum_i {\cal P}_2({\Bbb P}^1 \times S \rightarrow S;
4037: s^{(i)}_1, s^{(i)}_2, s^{(i)}_3) +
4038: $$
4039: $$
4040: \sum_j {\cal P}_2({\Bbb P}^1 \times S \rightarrow S; (1-g_j), g_j, h_j)
4041: $$
4042: It remains to use Lemma \ref{l454} and Proposition \ref{p44}.
4043: The part a) of the theorem is proved.
4044:
4045: b) We need the following lemma.
4046: \begin{lemma} \label{2.14}
4047: \begin{equation} \label{dhomot}
4048: d {\cal P}_2(Y \rightarrow S;f_1,f_2,f_3) = (2 \pi)^{-1} \cdot {\rm Alt}_3 \Bigl(v_x(f_1) \log|f_2(x)| d_s \arg f_3(x) \Bigr)
4049: \end{equation}
4050: \end{lemma}
4051:
4052: {\bf Proof}. Using $dd \log f = 2\pi i \delta(f)$ we get an equality of $3$-currents on $Y$
4053: $$
4054: dr_2(f_1,f_2,f_3) =
4055: $$
4056: \begin{equation} \label{2233}
4057: \pi_3\Bigl(\frac{df_1}{f_1}\wedge\frac{df_2}{f_2}\wedge\frac{df_3}{f_3}\Bigr) + 2 \pi \cdot {\rm Alt}_3 \Bigl(\delta(f_1) \log|f_2(x)| d \arg f_3(x) \Bigr)
4058: \end{equation}
4059: The second term in (\ref{2233}) is a $1$-form on the divisor
4060: $D:= \cup_{i=1}^3{\rm div}(f_i)$ considered as a $3$-current on $Y$.
4061: This $1$-form is the composition of the residue map
4062: $$
4063: {\rm res}: \Lambda^3 \C(Y)^* \longrightarrow \coprod_{X \in Y_1}
4064: \Lambda^2\C(X)^*
4065: $$
4066: with the map
4067: $$
4068: r_1: \Lambda^2 \C(X)^* \longrightarrow {\cal A}^1({\rm Spec }(\C(X))), \qquad f \wedge g \longmapsto -2\pi (\log|f| d\arg g - \log|g| d\arg f)
4069: $$
4070:
4071:
4072: The push forward of the first term in (\ref{2233}) vanishes
4073: (since the fibers are complex curves the push down of any $(3,0)$-form
4074: to $S$ is zero). Integrating the second $3$-current in (\ref{2233})
4075: along the fibers of $Y$ we get (\ref{dhomot}). The lemma is proved.
4076:
4077:
4078:
4079: According to Lemma \ref{2.14} and formula (\ref{dl2}) for $d {\cal L}_2$, and using ${\rm Res} = \delta_2 \circ h $ we get the proof of the
4080: part b) of the theorem.
4081:
4082:
4083:
4084:
4085: {\bf Remark} The function ${\cal L}_2(z)$ is continuous on $\C {\Bbb P}^1$. Therefore part a) of the theorem
4086: implies that the function ${\cal P}_2(Y \rightarrow S ;f_1,f_2,f_3)$ can be extended to a continuous
4087: function on $S$.
4088:
4089:
4090: {\bf 5. Conjecture \ref{reh} for $\overline k = \overline \Q$}.
4091: \begin{theorem} \label{rl}
4092: Let $X$ be a regular projective curve over
4093: ${\overline \Q}$ and $F:= {\overline \Q}(X)$.
4094: Then there exists a homomorphism
4095: $
4096: h:\Lambda^3F^* \rightarrow
4097: B_2({\overline \Q})\otimes \Q
4098: $ as in conjecture \ref{reh} such that for any embedding $\sigma: {\overline \Q}
4099: \hookrightarrow \C$ one has
4100: \begin{equation} \label{homot}
4101: \frac{1}{2\pi i}\int_{X(\C)}r_2(\sigma(f_1\wedge f_2 \wedge f_3)) = {\cal L}_2 \Bigl(\sigma(h(f_1\wedge
4102: f_2 \wedge f_3))\Bigr)
4103: \end{equation}
4104: \end{theorem}
4105:
4106:
4107:
4108: {\bf Proof}. It is similar to the proof of theorem \ref{55}.
4109: Choose a projection $p:X \rightarrow {\Bbb P}^1$.
4110: We may suppose that $p$ is a Galois covering
4111: with the Galois group $G$.
4112: Indeed, let $p_1:Y \rightarrow X$
4113: be such a covering that its composition with $p$
4114: is a Galois covering.
4115: Setting
4116: $$
4117: h(f_1\wedge f_2 \wedge f_3):= h(\frac{1}{{\rm deg}p_1} \cdot p_1^*(f_1\wedge f_2 \wedge f_3))
4118: $$
4119: we may suppose that $p$ is Galois.
4120:
4121: Then $\sum_{g \in G} g^*\{f_1, f_2, f_3\} \in p^*K_3^M(k({\Bbb P}^1))$.
4122: So there exist $s^{(i)}_1,s^{(i)}_2,s^{(i)}_3, \in k({\Bbb P}^1)$
4123: and $g_j,h_j \in k(X)$ such that (\ref{6.18.02.1}) holds.
4124: Set
4125: $$
4126: |G| \cdot h(f_1\wedge f_2 \wedge f_3) := \sum_i h(s^{(i)}_1\wedge s^{(i)}_2\wedge s^{(i)}_3) + \sum_j \sum_{x \in X(\overline \Q)}\{g_j(x)\}_2 \cdot v_x(h_j)
4127: $$
4128: \begin{lemma} \label{1.8}
4129: Suppose $\sum_i (1-f_i) \wedge f_i \wedge g_i = 0$ in $\Lambda^3 {\overline \Q}(X)^*$.
4130: Then $$
4131: \sum_i \sum_{x \in X(\overline \Q)} v_x(g_i) \cdot \{f_i(x)\}_2 =0 \quad \mbox{in the group $ B_2({\overline \Q})$.}
4132: $$
4133: \end{lemma}
4134:
4135: This lemma implies that $h$ is well defined.
4136: Indeed, suppose that we have a different presentation
4137: $$
4138: \sum_{g \in G} g^*\{f_1, f_2, f_3\} = p^*\sum_k
4139: \widetilde s^{(k)}_1\wedge \widetilde s^{(k)}_2\wedge \widetilde
4140: s^{(k)}_3 + \sum_j(1-\widetilde g_j)\wedge \widetilde g_j \wedge \widetilde h_j
4141: $$
4142: We need to show that
4143: \begin{equation} \label{**}
4144: h\Bigl( \sum_k \widetilde s^{(k)}_1\wedge \widetilde s^{k)}_2\wedge
4145: \widetilde s^{(k)}_3 -
4146: \sum_i s^{(i)}_1\wedge s^{(i)}_2\wedge s^{(i)}_3\Bigr) +
4147: \end{equation}
4148: $$
4149: \sum v_x(h_j)\{g_j(x)\}_2 - \sum v_x(\widetilde h_j)\{\widetilde g_j(x)\}_2 = 0
4150: $$
4151: There exist $a_j, b_j \in k({\Bbb P}^1)$ such that modulo $k^* \wedge
4152: \Lambda^2k({\Bbb P}^1)^*$ one has
4153: $$
4154: \sum_k \widetilde s^{(k)}_1\wedge \widetilde s^{(k)}_2\wedge \widetilde s^{(k)}_3 -
4155: \sum_i s^{(i)}_1\wedge s^{(i)}_2\wedge s^{(i)}_3 - \sum_j (1-a_j)\wedge a_j
4156: \wedge b_j =0
4157: $$
4158: According to Theorem \ref{p4} the homomorphism $h$ for ${\Bbb P}^1$ annihilates
4159: the left hand side. On the other hand
4160: $$
4161: p^*\Bigl( \sum_k \widetilde s^{(k)}_1\wedge \widetilde s^{(k)}_2\wedge \widetilde
4162: s^{(k)}_3 -
4163: \sum_i s^{(i)}_1\wedge s^{(i)}_2\wedge s^{(i)}_3\Bigr) -
4164: $$
4165: $$
4166: \sum_j(1-\widetilde g_j)\wedge \widetilde g_j \wedge \widetilde h_j +\sum_j(1- g_j)\wedge g_j \wedge h_j = 0
4167: $$
4168: Using Lemma \ref{1.8} we get (\ref{**}).
4169: To get (\ref{homot}) we use Theorem \ref{55} and notice that
4170: \begin{equation} \label{***}
4171: {\cal P}_2(Y \rightarrow S;f_1,f_2,f_3) = 1/m \cdot {\cal P}_2(Z \rightarrow S;
4172: p_1^*f_1, p_1^*f_2, p_1^*f_3)
4173: \end{equation}
4174:
4175:
4176:
4177: {\bf Proof of Lemma \ref{1.8}}.
4178: For a regular curve $X$ over an algebraically closed field $k$ there is a commutative diagram
4179: $$
4180: \begin{array}{ccc} \label{reh1a}
4181: B_2(F)\otimes F^{\ast}
4182: & \stackrel{\delta_3}{\longrightarrow} & \Lambda^3F^{\ast}\\
4183: &&\\
4184: {\rm Res}\downarrow&&\downarrow {\rm Res}\\
4185: &&\\
4186: B_2(k)\otimes \Z[X(k)]&\stackrel{\delta_2}{\longrightarrow}&\Lambda^2 k^{\ast}\otimes \Z[X(k)]
4187: \end{array}
4188: $$
4189: Thus for any point $x$ of the curve $X$ the element $\sum_{i} v_x(g_i) \cdot
4190: \{f_i(x)\}_2 $ lies in ${\rm Ker }\delta_2$. Therefore it defines an
4191: element $\gamma_x \in K_3({\overline \Q})_{\Q}$.
4192:
4193: For any embedding $\sigma: {\overline \Q}
4194: \hookrightarrow \C$ the value of the
4195: Borel regulator on $\sigma(\gamma_x)$ is equal to $\sum_{x \in X} v_x(g_i) \cdot {\cal L}_2(\sigma(f_i(x)))$.
4196: So by Lemma \ref{l454} the value of the
4197: Borel regulator on $\sum_x \sigma(\gamma_x)$ is equal to $2 \pi \cdot\int_{\C {\Bbb P}^1} r_2(\sum_i (1-f_i) \wedge f_i \wedge g_i)$
4198: and hence it is zero by our assumption. So Borel's theorem implies that the element is also zero.
4199:
4200: A similar argument using Lemma \ref{1.8} shows that the homomorphism $h$ does not depend on the choice of the (finite) Galois extension of $\Q({\Bbb P}^1)$ containing the field $\Q(X)$.
4201:
4202:
4203:
4204: {\bf 6. Explicit formulas for the reciprocity homomorphism $h$ and the Chow dilogarithm in the case of an elliptic curve}.
4205: Let $E$ be an elliptic curve. We want to calculate the integral
4206: $\int_{E(\C)} r_2(f_1 \wedge f_2 \wedge f_3)$.
4207: Let us suppose that $E$ is realized as a
4208: plane curve. Then any rational function $f$ on $E$ can be written as a ratio
4209: of products of {\it linear} homogeneous functions:
4210: $$
4211: f = \frac{l_1\cdot ... \cdot l_k}{l_{k+1}\cdot ... \cdot l_{2k}}
4212: $$
4213: So it is enough to calculate the integral
4214: $\int_{E(\C)} r_2(l_1/l_0 \wedge l_2/l_0 \wedge l_3/l_0)$ where $l_i$ are
4215: linear functions in homogeneous coordinates.
4216: We will do this in a more general setting.
4217:
4218: {\it Notations}. Let $X$ be a plane algebraic curve
4219: and $l_i$ linear functions in homogeneous coordinates.
4220: Denote by $L_i$ the line $l_i =0$ in
4221: $\PP^2$.
4222: Let $D_i$ be the divisor $L_i \cap X$.
4223: Set $l_{ij}:= L_i \cap L_j$.
4224: For three points $a,b,c$ and a
4225: divisor $D =\sum n_i (x_i)$ on a
4226: line we will use the following notation (see Figure \ref{kj}):
4227: $$
4228: \{r(a,b,c,D)\}_2:= \sum_i n_i \{r(a,b,c,x_i)\}_2
4229: $$
4230:
4231:
4232:
4233: \begin{figure}[ht]
4234: \centerline{\epsfbox{kj.eps}}
4235: \caption{Defining $\{r(a,b,c,D)\}_2$ for a plane algebraic curve $X$.}
4236: \label{kj}
4237: \end{figure}
4238:
4239: \begin{theorem} \label{ecur}
4240: Let $E$ be an elliptic curve over an algebraically closed field $k$. Then there exists a homomorphism of groups $h:\Lambda^3F^* \rightarrow
4241: {\cal B}_2(k)$ such that for any linear homogeneous functions $l_0,...,l_3$ one has
4242: \begin{equation} \label{hrule}
4243: h(l_1/l_0 \wedge l_2/l_0 \wedge l_3/l_0) = -\sum_{i=0}^3(-1)^i\{r(l_{i0},..., \widehat l_{ii}, ..., \l_{i3}, D_i)\}_2
4244: \end{equation}
4245: and which satisfies all the properties of conjecture \ref{reh}. In particular, if $k =\C$ then
4246: \begin{equation} \label{homot1}
4247: {\cal P}_2(E; f_1\wedge f_2 \wedge f_3) = {\cal L}_2 \Bigl( h(f_1\wedge
4248: f_2 \wedge f_3 )\Bigr)
4249: \end{equation}
4250: \end{theorem}
4251:
4252:
4253:
4254: {\bf Proof}. Suppose we have four generic lines $L_0,L_1,L_2,L_3$ in $\PP^2$. Any two of them, say $L_0$ and $L_1$, provide a {\it canonical} rational function $(l_0/l_1)$ on $\PP^2$ with the divisor $L_0 - L_1$ normalized by the condition that its value at the point $l_{23}$ is equal to $1$.
4255:
4256:
4257: \begin{lemma} \label{3.9}
4258: a) On the line $L_3$ one has $(l_1/l_0) + (l_2/l_0) =1$.
4259:
4260: b) $\frac{(l_1/l_0)}{(l_2/l_0)} = -(l_1/l_2)$
4261: \end{lemma}
4262:
4263: {\bf Proof}. Let $m$ be a point on the line $L_3$. Then
4264: \begin{equation} \label{88}
4265: (l_1/l_0)(m) = r(l_{03},l_{13},l_{23},m); \qquad (l_2/l_0)(m) = r(l_{03},l_{23},l_{13},m)
4266: \end{equation}
4267: This gives a).
4268: It follows from this that if the point $m$ approaches the point $l_{03}$ then $\frac{(l_1/l_0)}{(l_2/l_0)}$ tends to $-1$. This implies b).
4269:
4270: \begin{lemma} \label{lrez}
4271: For any plane curve $X$ one has
4272: $$
4273: \sum_{x \in X} {\rm res}_x \Bigl((l_1/l_0) \wedge (l_2/l_0) \wedge ( l_3/l_0)\Bigr) =
4274: - \delta_2\Bigl(\sum_{i=0}^3(-1)^i \{r(l_{i0},..., \widehat l_{ii}, ..., \l_{i3}, D_i)\}_2\Bigr)
4275: $$
4276: \end{lemma}
4277:
4278:
4279: {\bf Proof}. Let us compute first the residues at the divisors $D_1,D_2,D_3$ using part a) of Lemma \ref{3.9}. For example
4280: the residue at $x \in D_1$ is equal to
4281: $$
4282: v_x((l_1/l_0))\cdot (l_2/l_0)(x) \wedge (l_3/l_0)(x) = v_x((l_1/l_0))\cdot (1-(l_3/l_0)(x) ) \wedge (l_3/l_0)(x)
4283: $$
4284: $$
4285: \stackrel{(\ref{88})}{=} v_x((l_1/l_0))\cdot \{r(l_{13}, l_{10}, l_{12}, x)\}_2 = v_x((l_1/l_0))\cdot \{r(l_{10}, l_{12}, l_{13}, x)\}_2
4286: $$
4287: It remains to compute the residues on the line $L_0$. According to part b) of Lemma \ref{3.9} one has
4288: $$
4289: (l_1/l_0) \wedge (l_2/l_0) \wedge ( l_3/l_0) = - (l_0/l_1) \wedge (l_2/l_1) \wedge
4290: (l_3/l_1)
4291: $$
4292: Using this we reduce the calculation to the previous case.
4293:
4294: \begin{proposition} \label{hrez} Let $E$ be an elliptic curve over an algebraically closed field $k$. Then formula (\ref{hrule}) provides a well defined homomorphism of groups $h:\Lambda^3F^* \rightarrow
4295: {\cal B}_2(k)$.
4296: \end{proposition}
4297:
4298:
4299:
4300:
4301: {\bf Proof}. Let $D:= \sum_i n_i(x_i)$ be the divisor of a rational function $f$ on $E$.
4302: To decompose it into a fraction of products of linear homogeneous functions $l_i$ we proceed as follows. Let $l_{x,y}$ (resp. $l_{x }$) be a linear homogeneous equation of the line in $\PP^2$ through the points $x$ and $y$ on $E$ (resp. $x $ and $-x $). The divisor of the function $l_{x,y}/l_{x}$ is $(x) +(y) - (x+y) - (0)$.
4303: If $D = (x)+(y) + D_1$, we write $f =
4304: l_{x,y}/l_{x+y}\cdot f'$, so $(f') = (0) + (x+y) + D_1$.
4305: After a finite number of such steps we get the desired decomposition.
4306:
4307:
4308:
4309: \begin{figure}[ht]
4310: \centerline{\epsfbox{kj3.eps}}
4311: \caption{}
4312: \label{kj3}
4313: \end{figure}
4314:
4315: There are the following relations
4316: $$
4317: \frac{l_{x,y}}{l_{x+y}} \cdot \frac{l_{x+y,z}}{l_{x+y+z}} \Big /\frac{l_{y,z}}{l_{y+z}} \cdot \frac{l_{x,y+z}}{l_{x+y+z}} \quad = \quad \mbox{constant}
4318: $$
4319:
4320: One can prove that they generate all the relations between the functions $l_{x,y}/l_{x+y}$.
4321: So $h$ is well defined if it annihilates the following expression:
4322: $$
4323: F(x,y,z; l_0,l_2,l_3):= \Bigl(\frac{l_{x,y}}{l_{x+y}} \cdot \frac{l_{x+y,z}}{l_{x+y+z}} \Big / \frac{l_{y,z}}{l_{y+z}} \cdot \frac{l_{x,y+z}}{l_{x+y+z}} \Bigr) \wedge (l_2/l_0) \wedge ( l_3/l_0)
4324: $$
4325: It follows from Lemma \ref{lrez} that $\delta_2(F(x,y,z; l_0,l_2,l_3)) =0$. Thus according to the definition of the group ${\cal B}_2(k)$ it is enough to check that $h(F(x',y',z'; l_0,l_2,l_3))=0$ for a certain triple of points $(x',y',z')$. It is easy to see that $h(F(a,a,a; l_0,l_2,l_3))=0$ since then the first factor in $F$ is a constant. The proposition is proved.
4326:
4327:
4328:
4329: \begin{proposition} \label{hrezza} Let $X$ be an algebraic curve in $\PP^2$ over $\C$ and $l_0,...,l_3$ linear homogeneous functions on $\C^3$. Then one has (using the notations defined above)
4330: \begin{equation} \label{elfo}
4331: \int_{X(\C)} r_2(l_1/l_0 \wedge l_2/l_0 \wedge l_3/l_0) = 2 \pi \cdot \sum_{i=0}^3(-1)^i
4332: {\cal L}_2(r(l_{i0},..., \widehat l_{ii}, ..., \l_{i3}, D_i)
4333: \end{equation}
4334: \end{proposition}
4335:
4336: {\bf Proof}. It follows from Lemma \ref{lrez} and Theorem \ref{55} that the differentials of the both sides coincide. So their difference is a constant.
4337: To show that this constant is zero
4338: we deform $X$ to a union of lines in $\PP^2$. Using Proposition \ref{p44} one sees that formula (\ref{elfo}) is valid when $X$ is a
4339: line in $\PP^2$.
4340:
4341:
4342:
4343:
4344: \section{Appendix: on volumes of simplices in symmetric spaces}
4345:
4346: {\bf 1. Volumes of hyperbolic geodesic simplices as boundary integrals}.
4347: A point $y$ of the $n$-dimensional hyperbolic space ${\cal H}_n$ defines a one
4348: dimensional space $M_y$ of volume forms on the absolute $\partial{\cal H}_n$. It
4349: consists of the volume forms invariant under the
4350: action of the isotropy group of $y$. We write them as follows.
4351: Let $x_0, ..., x_{n}$ be the coordinates in a vector space $V_{n+1}$ of dimension $n+1$,
4352: and
4353: $
4354: Q(x):= x_0^2 + ... + x_{n-1}^2 - x_{n}^2
4355: $.
4356: Then ${\cal H}_n$ can be realized as the
4357: projectivization of the cone $Q(x) < 0$, and its boundary is the projectivization of the cone
4358: $Q(x) = 0$. Choose a point $y \in {\cal H}_n$. Lifting $y$ to a vector $y' \in V_{n+1}$
4359: we have the following
4360: volume form on the boundary $\partial{\cal H}_n$:
4361: $$
4362: \frac{\delta(Q(x)) \sigma_{n+1}(x,dx)}{(x,y')^{n-1}}
4363: $$
4364: If $y$ belongs to the boundary $\partial{\cal H}_n$, this formula provides a
4365: space $M_y$ of singular volume forms on the absolute; they are invariant under the
4366: isotropy group of $y$.
4367:
4368:
4369:
4370: Let us choose for any point $y$ such a volume form $\mu_y$.
4371: For two points $x,y$ the
4372: ratio $ \mu_x/\mu_y$ is a nonzero function on the absolute.
4373:
4374:
4375: Let $I(y_0,...,y_{n })$ be the geodesic simplex with vertices at
4376: $y_0,...,y_{n }$ where the points $y_i$ could be on the absolute.
4377: Denote by ${\rm vol}I(y_0,...,y_{n})$ the volume of this simplex with respect to the invariant volume form in ${\cal H}_n$ normalized by the following condition: if we realize the hyperbolic space as the interior of the unit ball $y_1^2 + ... + y_n^2 \leq 1$ then the volume form restricted to the tangent space at the origin $(0, ..., 0)$ is $dy_1 \wedge ... \wedge dy_n$.
4378:
4379: \begin{theorem} \label{2112}
4380: For any hyperbolic geodesic simplex
4381: $I(y_0,...,y_{n})$ one has
4382: \begin{equation} \label{1221}
4383: \frac{(n-1)^n{\rm vol}(S^{n-1})}{n} \cdot {\rm vol}I(y_0,...,y_{n}) =
4384: \int_{\partial {\cal H}_n}
4385: \log | \frac{\mu_{y_1}}{\mu_{y_0}}| d\log|\frac{\mu_{y_2}}{\mu_{y_0}} | \wedge ... \wedge d\log| \frac{\mu_{y_{n}}}{\mu_{y_0}} |
4386: \end{equation}
4387: \end{theorem}
4388:
4389:
4390:
4391: Let $\varphi(y_0, ..., y_{n})$ be the function defined by the right hand side of (\ref{1221}). Thanks to property 2) of Proposition \ref{1.2.} it does not depend on the choice of invariant volume forms $\mu_y$.
4392:
4393:
4394:
4395: \begin{proposition} \label{1.21.1}
4396: The function $\varphi(y_0,...,y_{n})$ has the following properties: It is
4397:
4398: 1) A smooth function on the vertices $y_i$.
4399:
4400: 2) Equal to zero if three of the vertices belong to the same geodesic.
4401:
4402: 3) Additive with respect to cutting of a simplex, i.e. if $y_0,...,y_{n}$ are points
4403: such that $y_0,y_1,y_2 $ are on the same geodesic, $y_1$ between $y_0$ and $y_2$, then
4404: $$
4405: \varphi(y_0,y_2,...,y_{n+1}) = \varphi(y_0,y_1,y_3, ...,y_{n+1}) + \varphi(y_1,y_2,...,y_{n+1})
4406: $$
4407:
4408: 4) Invariant under the action of the symmetry group $SO(n,1)$.
4409: \end{proposition}
4410:
4411: {\bf Proof}. 1) This is clear.
4412:
4413: 2) Let us realize the hyperbolic space as the interior part of
4414: the unit ball in $\R ^{n}$. Consider the geodesic $l$ passing through the center of the ball in the vertical direction. The subgroup $SO(n-1) \subset SO(n,1)$ preserves pointwise this geodesic. So for any point $y$ on the geodesic the invariant volume form $\mu_y$ is invariant under the action of the group $SO(n-1)$.
4415: The quotient of $\partial {\cal H}_n$ under the action of $SO(n-1)$ is given by the projection $p: \partial {\cal H}_n \longrightarrow \R $ onto the vertical axis.
4416: Take three points $y_1,y_2,y_3$ on the geodesic. Then $\frac{\mu_{y_2}}{\mu_{y_1}}$ and $\frac{\mu_{y_3}}{\mu_{y_1}}$ are lifted from the line $\R $. Therefore $d\log|\frac{\mu_{y_2}}{\mu_{y_1}}| \wedge d\log|\frac{\mu_{y_3}}{\mu_{y_1}}|$ = 0.
4417: So for a degenerate simplex $(y_0,...,y_n)$ one has
4418: $$
4419: \log|\frac{\mu_{y_0}}{\mu_{y_1}}| d\log|\frac{\mu_{y_2}}{\mu_{y_1}}|
4420: \wedge d\log|\frac{\mu_{y_3}}{\mu_{y_1}}|\wedge ... \wedge d\log|
4421: \frac{\mu_{y_n}}{\mu_{y_1}}| =0
4422: $$
4423: It remains to mention the skewsymmetry of the integral (\ref{1221}).
4424: The property 2) is proved.
4425:
4426: 3) Follows from 2) and the additivity property from Proposition
4427: \ref{1.2.}.
4428:
4429: 4) This is clear from 4) of Proposition
4430: \ref{1.2.}. The proposition is proved.
4431:
4432:
4433:
4434: The leading term of the Taylor expansion of the function $\varphi(y_0, y_1, ..., y_n)$
4435: when $y_0$ is fixed and $y_1, ..., y_n$ are near $y_0$ provides an exterior $n$-form in
4436: $T_{y_0}{\cal H}_n$ denoted $\varphi_{y_0}(Y_1, ..., Y_n)$, $Y_i \in T_{y_0}{\cal H}_n$.
4437: Let us compare it with the volume form
4438: ${\rm Vol}_{y_0}(Y_1, ..., Y_n)$ in $T_{y_0}{\cal H}_n$ normalized as
4439: before Theorem 7.1.
4440:
4441: \begin{lemma} \label{1.21.1}
4442: $\varphi_{y_0}(Y_1, ..., Y_n) = \frac{(n-1)^n}{n}{\rm vol}(S^{n-1}) \cdot
4443: {\rm Vol}_{y_0}(Y_1, ..., Y_n)$.
4444: \end{lemma}
4445:
4446: {\bf Proof}. Below we abuse notations by writing $y$ for $y'$.
4447: One has $\mu_{y_1}/\mu_{y_2} = \frac{(y_2,x)^{n-1}}{(y_1,x)^{n-1}}$. So
4448: $$
4449: \varphi(y_0, ..., y_n) = (n-1)^{n}\int_{\partial {\cal H}_n} \log\vert
4450: \frac{(y_1,x)}{(y_0,x)}\vert d \log\vert
4451: \frac{(y_2,x)}{(y_0,x)}\vert \wedge ... \wedge d \log\vert
4452: \frac{(y_n,x)}{(y_0,x)}\vert
4453: $$
4454: Thus
4455: $$
4456: \varphi_{y_0}(Y_1, ..., Y_n) = (n-1)^{n}\int_{\partial {\cal H}_n} (Y_1,x) d( Y_2, x) \wedge ... \wedge
4457: d(Y_n, x)
4458: $$
4459: To do the computation of this integral we may suppose that $y_0 = (0, ..., 0, 1)$, $Y_i =
4460: \frac{\partial}{\partial y_i}$, so $(Y_i, x) = x_i$. Then the last integral equals to
4461: $$
4462: (n-1)^{n}\int_{S^{n-1}} x_1 dx_2 \wedge ... \wedge dx_n = \frac{(n-1)^{n}}{n}{\rm vol}(S^{n-1})
4463: $$
4464: where $S^{n-1}$ is the sphere $x_1^2 + ... + x_n^2 =1$.
4465: The lemma follows.
4466:
4467:
4468: {\bf Proof of Theorem \ref{2112}}. Let us suppose first that the points $x_i$
4469: are inside of the hyperbolic space.
4470:
4471: The function $\varphi(x_0,...,x_{n})$ defines an
4472: $n$-density $\widetilde \varphi$ on
4473: $ {\cal H}_n$.
4474: Namely, to define the integral $\widetilde \varphi$ over a simplex $M$ one has to
4475: subdivide it into small simplices and take the sum of the functions
4476: $\varphi$ corresponding to their vertices.
4477: When the simplices are getting smaller the limit exists and is by definition
4478: $\int_M \widetilde \varphi$. Here we used the properties 1) - 3).
4479: More precisely 1) and 2) imply that $\varphi$ defines an additive volume form on
4480: ${\cal H}_n$, and 3) (together with 1)) guarantee that this volume form is $\sigma$-additive.
4481:
4482:
4483: The skewsymmetry property implies that $\widetilde \varphi$ is actually a differential $n$-form. It is invariant under the action of
4484: the group $SO(n,1)$. Therefore it is proportional to the standard volume form.
4485:
4486: Now suppose that the vertices $x_i$ can be on the absolute. Then
4487: it is easy to see that the corresponding integral
4488: (\ref{i2}) is still convergent. Moreover,
4489: if the vertices of the geodesic simplex are in general position
4490: then it is a continuous function of the vertices. This implies that the volume of an
4491: ideal geodesic simplex is finite (which is, of course, an elementary fact) and
4492: coincides with the corresponding integral (\ref{1221}).
4493:
4494: Completely similar results are valid for the complex $n$-dimensional hyperbolic
4495: space ${{\cal H}_n}^{\C}:= \{|z_1| + ... +|z_n|^2 < 1\}$, $z_i \in \C$, and the
4496: quaternionic hyperbolic space ${{\cal H}_n}^{\HH}:= \{|q_1| + ... +|q_n|^2 < 1\}$ ($q_i$ are quaternions). Indeed, a point $x$ in each of these spaces defines an invariant volume form $\mu_x$ on the boundary.
4497:
4498:
4499:
4500:
4501:
4502:
4503:
4504:
4505: {\bf 2. Calculation of the volume of a three dimensional ideal geodesic simplex}.
4506: If $n=3$ the absolute can be identified with $\C \PP^1$,
4507: and for the ideal simplex with vertices at
4508: the points $\infty, 0,1,a$ on the absolute we get
4509: $$
4510: {\rm vol} (I(\infty, 0,1,a)) = 3 c_3 \cdot
4511: \int_{\C \PP^1} (\log|z| d\log|1-z| - \log|1-z| d\log|z|) \wedge d\log|z-a| =
4512: $$
4513: $$
4514: 3 c_3 \cdot \int_{\C \PP^1} (\log|z| d\arg (1-z) - \log|1-z| d\arg (z))\wedge d\arg (z-a)
4515: $$
4516: because $d \log (z-1) \wedge d \log (z-a) = 0$.
4517: Here $\log f = \log |f| + i \arg(f)$.
4518: Using
4519: $$
4520: d {\cal L}_2(z) = \log|z| d\arg (1-z) - \log|1-z| d\arg (z)
4521: $$
4522: we rewrite the last integral as
4523: \begin{equation} \label{1331}
4524: 3 c_3 \cdot \int_{\C \PP^1} d {\cal L}_2(z) \wedge d\arg (z-a)
4525: \end{equation}
4526: Computing
4527: the differential in the sense of distributions we get
4528: $$
4529: d ({\cal L}_2(z) d\arg (z-a)) = 2\pi \cdot {\cal L}_2(z) \delta(z-a)dx d y +
4530: d {\cal L}_2(z)\wedge d\arg (z-a)
4531: $$
4532: So the integral of the right hand side over $\C \PP^1$ is zero, i.e. the integral (\ref{1331}) is equal to
4533: $- 6 \pi c_3 \cdot {\cal L}_2(a)$. ( $c_3 = -1/6\pi$).
4534:
4535:
4536:
4537:
4538:
4539: {\bf 3. Volumes of geodesic simplices in $SL_n(\C)/SU(n)$}. Recall
4540: the invariant differential $(2n-1)$--form $\omega_{D_n}$ in ${\Bbb H}_n$.
4541:
4542: {\bf Question}. Is it true that
4543: \begin{equation} \label{1221?q}
4544: {\rm vol}_{\omega_{D_n}}(I(x_0,...,x_{2n-1})) = \mbox{\rm constant}
4545: \times \psi_n(x_0,...,x_{2n-1})?
4546: \end{equation}
4547:
4548:
4549: One can show following the lines of s. 7.1 that the positive answer
4550: to this question is equivalent to the following statement: if $x_0, x_1, x_2$
4551: are on the same geodesic then $\psi_n (x_0,...,x_{2n-1}) =0$.
4552:
4553:
4554: {\bf 4. Another approach to Grassmannian polylogarithms}.
4555: The following construction was suggested to the author
4556: during the Fall of 1989, independently,
4557: by M. Kontsevich and by J. Nekovar.
4558: A hyperplane $h$ in an $n$--dimensional
4559: complex vector space $V$ determines an arrow in the space of degenerate
4560: non-negative definite
4561: hermitian forms in $V$ consisting of the forms with the kernel $h$.
4562: Let $h_1, ..., h_{2n}$
4563: be hyperplanes in $V$. Let $C(h_1, ..., h_{2n})$ be projectivization of
4564: the convex hull of the arrays corresponding to these
4565: hyperplanes. It is a simplex in $\overline {\Bbb H}_n$. The idea is to
4566: integrate
4567: the form $\omega_{D_n}$ over this simplex.
4568: If $n=2$ this construction provides an ideal geodesic simplex
4569: in the Cayley realization of the hyperbolic space, given by
4570: the interior part of a ball in $\R \PP^3$. However
4571: the convergence of this integral for $n>2$
4572: has not been established yet,
4573: although it does not seem to be a very difficult problem.
4574: If the integral is convergent, we get a function on configurations
4575: of $2n$ hyperplanes in $\CP^{n-1}$.
4576: It would be very interesting to investigate this construction further and
4577: compare it with our construction of the Grassmannian polylogarithms.
4578:
4579: {\bf 5. A $(2n-1)$-cocycle of $GL(\C)$}.
4580: Consider an infinite dimensional $\C$-vector space with a given basis
4581: $e_1,...,e_m,...$. The group $GL(\C)$ is the
4582: group of automorphisms of this space moving only finite number of basis
4583: vectors.
4584:
4585: Let us describe the restriction of the cocycle to the subgroup
4586: $GL_{n+m}(\C)$ acting on the subspace generated by first $n+m$
4587: vectors.
4588: Take $2n$ elements $g_1,..., g_{2n}$ of this group. Consider the corresponding
4589: $2n$ $(m+1)$-tuples of vectors:
4590: $$
4591: g_1(e_{n},...,e_{n+m}), \quad ... , \quad g_{2n}(e_{n},...,e_{n+m})
4592: $$
4593: The set of all $(m+1)$--tuples of vectors form a vector space.
4594: Let $\Delta_{2n-1}$ be the standard simplex $\sum^{2n}_{j=1} \lambda_j =
4595: 1$.
4596: Consider the set
4597: $
4598: C_n^{m} \subset \Delta_{2n-1}
4599: $
4600: consisting of $(\lambda_1,...,\lambda_{2n})$ such that
4601: $$
4602: \sum_{i=1}^{2n} \lambda_i \cdot g_i(e_{n},...,e_{n+m})
4603: $$
4604: is an $(m+1)$-tuple of vectors {\it not} in generic position. It is a cycle of
4605: codimension $n$. The Chow polylogarithm function evaluated
4606: on it provides the desired (measurable) cocycle.
4607:
4608: The cocycle property follows from the functional equation for the Chow
4609: polylogarithm and the following general fact.
4610: The set of those $(\lambda_1,...,\lambda_{2n+1})$ such that
4611: $
4612: \sum_i \lambda_i \cdot g_i(e_{n},...,e_{n+m})
4613: $
4614: is an $(m+1)$-tuple of vectors {\it not} in generic position is also of
4615: codimension $n$.
4616:
4617: The construction is consistent with the restriction to $GL_n$ just by definition.
4618:
4619: {\bf Problem}. Show that the cohomology class of this cocycle is nontrivial and proportional to the Borel class.
4620:
4621:
4622:
4623:
4624: \begin{thebibliography}{BGSV}
4625:
4626:
4627:
4628: \bibitem[A]{A} Arnold V.I.: Mathematical methods of classical mechanics.
4629: Springer Verlag, 1978.
4630:
4631: \bibitem[B1]{B1} Beilinson A.A: {\it Higher regulators and values of
4632: $L$-functions}, VINITI, 24 (1984), 181--238.
4633:
4634: \bibitem[B2]{B2} Beilinson A.A: {\it Height pairing between
4635: algebraic cycles}, Springer Lect. Notes in Math. 1289.
4636:
4637: \bibitem[B3]{B3} Beilinson A.A: {\it Notes on absolute Hodge cohomology}.
4638: Applications of algebraic $K$-theory to algebraic geometry and number theory,
4639: Part I, II (Boulder, Colo., 1983),
4640: 35--68, Contemp. Math., 55, Amer. Math. Soc., Providence, RI, 1986.
4641:
4642: \bibitem[BD]{BD} Beilinson A.A., Deligne P: {\it Interpr\'etation motivique de la conjecture de Zagier} in Symp. in Pure Math., v. 55, part 2, 1994, 23-41.
4643:
4644:
4645: \bibitem[BMS]{BMS} Beilinson A.A., MacPherson R., Schechtman,
4646: V.V: {\it Notes on motivic cohomology}, Duke Math.\ J.\ 54 (1987),
4647: 679--710.
4648:
4649:
4650:
4651: \bibitem[Bl]{Bl} Bloch S: {\it Algebraic cycles and higher K-theory}, Adv. in
4652: Math., (1986), v. 61, 267--304.
4653:
4654: \bibitem[Bl2]{Bl2} Bloch S: {\it The moving lemma for higher Chow groups},
4655: J. Alg. geometry. 3 (1994).
4656:
4657: \bibitem[Bl3]{Bl3} Bloch S: {\it Algebraic cycles and the
4658: Beilinson conjectures}, Cont. Math. 58 (1986) no 1, 65-79.
4659:
4660: \bibitem[BK]{BK} Bloch S. Kriz I: {\it Mixed Tate motives},
4661: Ann. of Math., 140 (1994) vol. 3, 557-605.
4662: 1992.
4663:
4664: \bibitem[Bo1]{Bo1} Borel A: {\it Stable real cohomology of arithmetic groups}, Ann. Sci. Ec. Norm. Super., (4) 7 (1974), 235--272
4665: \
4666: \bibitem[Bo2]{Bo2} Borel A: {\it Cohomologie de $SL_{n}$ et valeurs de
4667: fonctions z\^{e}ta aux points entiers}, Ann. Sc. Norm.
4668: Sup. Pisa 4 (1977), 613--636.
4669:
4670: \bibitem[Bott]{Bott} Bott R.: {\it On the characteristic classes of groups
4671: of diffeomorphisms}, L'ensiegnement Mathematique, T. XXIII, fsc. 3-4
4672: (1977) p. 209-220.
4673:
4674: \bibitem[Bu]{Bu} Burgos, J. I.: {\it
4675: Arithmetic Chow rings and Deligne-Beilinson cohomology},
4676: J. Algebraic Geom. 6 (1997), no. 2, 335--377.
4677:
4678: \bibitem[Con]{Con} Consani C.: {\it Double complexes and Euler's L-factors},
4679: Compositio Math. 111 (1998) 323-358.
4680:
4681: \bibitem[D1]{D1} Dupont J.: {\it Simplicial De Rham cohomolgy and
4682: characteristic classes of flat bundles}, Topology 15, 1976, p. 233-245.
4683:
4684: \bibitem[Del]{Del} Deligne, P. {\it Le d\'eterminant de la cohomologie}.
4685: Current trends in arithmetical algebraic geometry (Arcata, Calif., 1985),
4686: 93--177, Contemp. Math., 67, Amer. Math. Soc., Providence, RI, 1987.
4687:
4688: \bibitem[F]{F} Fulton W.:. {\it Intersection theorey}. Second edition. Springer 1998.
4689:
4690: \bibitem[Dyn]{Dyn} Dynkin E.B.: {\it Topological characteristics of compact Lie groups homomorphisms} Math. Sb. NS 35 (1954), 129-173
4691:
4692:
4693:
4694: \bibitem[GGL]{GGL} Gabrielov A., Gelfand I.M., Losik M.:
4695: {\it Combinatorial computation of characteristic classes} I, II,
4696: Funct. Anal. i ego pril. 9 (1975) 12-28, ibid 9 (1975) 5-26.
4697:
4698: %\bibitem[Gel]{Gel}Gelfand I.M.: {\it General theory of hypergeometric
4699: %functions}. Doklady Acad. Nauk USSR, 1986, 288, N 1, p.14-18.
4700:
4701: \bibitem[GM]{GM} Gelfand I.M., MacPherson R: {\it Geometry in
4702: Grassmannians and a generalisation of the dilogarithm}, Adv.
4703: in Math., 44 (1982) 279--312.
4704:
4705: \bibitem[GS]{GS} Gillet, H.; SoulŽé, Ch.:
4706: {\it Arithmetic intersection theory},
4707: Inst. Hautes ŽÉtudes Sci. Publ. Math. No. 72, (1990), 93--174 (1991).
4708:
4709: \bibitem[G1]{G1} Goncharov A.B: {\it Geometry of configurations,
4710: polylogarithms and motivic cohomology},
4711: Adv. in Math., 114, (1995), 197--318.
4712:
4713: \bibitem[G2]{G2} Goncharov A.B.: {\it Polylogarithms and motivic Galois
4714: groups} in Symp. in Pure Math., v. 55, part 2, 1994, 43 - 97.
4715:
4716: \bibitem[G3]{G3} Goncharov A.B.: {\it Explicit construction of
4717: characteristic classes}, I.M. Gelfand seminar
4718: Adv. Soviet Math., 1993,
4719: v. 16, p. 169-210.
4720:
4721: \bibitem[G4]{G4}
4722: Goncharov, A.B.: {\it Deninger's conjecture on special values of
4723: $L$-functions of elliptic curves at $s=3$}, J. Math. Sci. 81 (1996), N3, 2631-2656,
4724: alg-geom/9512016.
4725:
4726:
4727: \bibitem[G5]{G5} Goncharov A.B.: {\it Chow polylogarithms and
4728: regulators}. Math. Res. Letters, 2, (1995), 99-114.
4729:
4730:
4731:
4732:
4733: \bibitem[G6]{G6}
4734: Goncharov, A.B.: {\it Geometry of trilogarithm and the motivic Lie
4735: algebra of a field}, Regulators in analysis, geometry and number theory,
4736: 127--165, Progr. Math., 171,
4737: Birkhauser Boston, Boston, MA, 2000. math.AG/0011168.
4738:
4739: \bibitem[G7]{G7}
4740: Goncharov, A.B.: {\it Explicit regulator maps on the polylogarithmic
4741: motivic complexes}, Motives, polylogarithms and Hodge theory, Part I (Irvine, CA, 1998), 245--276, Int.
4742: Press Lect. Ser., 3, I, Int. Press, Somerville, MA, 2002.
4743: math.AG/0003086.
4744:
4745: \bibitem[GZ]{GZ} Goncharov A. B., Zhao J.: {\it The Grassmannian trilogarithms}.
4746: Compositio Math. 127 (2001), no. 1, 83--108. math.AG/0011165.
4747:
4748:
4749: \bibitem[HM]{HM} Hain R, MacPherson R: {\it Higher Logarithms},
4750: Ill. J.\ of Math,, vol. 34, (1990) N2, 392--475.
4751:
4752: \bibitem[H]{H} Hain R: {\it The existence of higher logarithms}. Compositio Math. 100 (1996) N3, 247- 276
4753:
4754:
4755: \bibitem[HY]{HY} Hain R, Yang R.: {\it Real Grassmann polylogarithms and Chern classes}. math. Annalen, 304, (1996),
4756: N1, 157-201.
4757:
4758: \bibitem[HM1]{HM1} Hanamura M, MacPherson R: {\it Geometric construction
4759: of polylogarithms}, Duke Math. J. 70 ( 1993)481-516.
4760:
4761:
4762:
4763: \bibitem[HM2]{HM2} Hanamura M, MacPherson R.:
4764: {\it Geometric construction of polylogarithms, II},
4765: Progress in Math. vol. 132 (1996) 215-282.
4766:
4767: \bibitem[Lei]{Lei} Gerhardt C.I. (ed). G.W.Leibniz Mathematische
4768: Schriften III/1 pp. 336-339 Georg Olms Verlag, Hildesheim. New York
4769: 1971.
4770:
4771: \bibitem[Le]{Le} Levin A.M.: {\it Notes on $\R$-Hodge-Tate sheaves}.
4772: Preprint MPI 2001.
4773: \bibitem[Lev]{Lev} Levine, M.: {\it Bloch's higher Chow groups revisited. $K$-theory}. Astérisque No. 226 (1994), 10, 235--320.
4774:
4775: \bibitem[Li]{Li} Lichtenbaum S: {\it Values of zeta functions at
4776: non-negative integers}, Lect. Notes in Math., 1068, Springer Verlag,
4777: 1984, 127-138.
4778:
4779: \bibitem[M]{M} MacPherson R.: {\it Gabrielov, Gelfand, Losic
4780: combinatorial formula for the first Pontryagin class} Seminar Bourbaki 1976
4781:
4782: \bibitem[NS]{NS} Nesterenko Yu.P., Suslin A.A.: {\it
4783: Homology of the general linear group over a local ring, and
4784: Milnor's $K$-theory}.
4785: Math. USSR-Izv. 34 (1990), no. 1, 121--142.
4786:
4787: \bibitem[S]{S} Soul\'e C., D. Abramovich, J. F. Burnol, J. K. Kramer:
4788: Lectures on Arakelov geometry. Cambridge University Press,
4789: 1992.
4790:
4791: \bibitem[Su]{Su} Suslin A.A: {\it Homology of $GL_{n}$,
4792: characteristic classes and Milnor's $K$-theory}. Lect. Notes in Math.
4793: 1046 (1989), 357--375.
4794:
4795: \bibitem[Z1]{Z1} Zagier D: {\it Polylogarithms, Dedekind zeta
4796: functions and the algebraic $K$-theory of fields}, Arithmetic
4797: Algebraic Geometry (G.v.d.Geer, F.Oort, J.Steenbrink, eds.),
4798: Prog. Math., Vol 89, Birkhauser, Boston, 1991, pp. 391--430.
4799:
4800: \end{thebibliography}
4801:
4802:
4803:
4804:
4805:
4806:
4807:
4808:
4809:
4810:
4811: \bigskip
4812:
4813:
4814:
4815:
4816: Department of Mathematics, Brown University, Providence, RI 02912, USA.
4817:
4818: e-mail sasha@math.brown.edu
4819:
4820:
4821:
4822:
4823:
4824:
4825:
4826:
4827:
4828:
4829: \end{document}
4830:
4831:
4832:
4833:
4834:
4835:
4836: {\bf 8. Cocycles for continuous cohomology classes of $GL_N(\C)$}.
4837: {\it A $(2n-1)$-cocycle of $GL_n(\C)$}.
4838: Choose a hyperplane $h$ in $\C \PP^{n-1}$. Then the
4839: function
4840: \begin{equation} \label{c10}
4841: c^{n}_{2n-1}(g_1,...,g_{2n}):= {\cal P}_n(g_1 h, ... ,g_{2n} h)
4842: \end{equation}
4843: is a (2n-1)-cocycle of $GL_n(\C)$. The cocycle
4844: condition is just the property c) in Theorem (\ref{0.3a}). Notice that
4845: the cocycle $c^{n}_{2n-1}(g_1,...,g_{2n})$
4846: is defined everywhere on
4847: $GL_n(\C)^{2n}$.
4848: It is, however, discontinuous near the identity and only a measurable
4849: function on $GL_n(\C)^{2n}$.
4850:
4851: This cocycle essentially coincides with the cocycle
4852: described in the s.5 of the Introduction, see Remark in s.2 below.
4853: In particular it represents (a multiple of) the Borel class.
4854:
4855:
4856:
4857: {\it A $(2n-1)$-cocycle for the Borel class of $GL_{n+m}(\C)$}.
4858: The property b) in Theorem (\ref{0.3a}) guarantee that this cocycle can
4859: be extended to the group $GL(\C)$. Namely, to do this for the group
4860: $GL_{n+m}(\C)$ we should proceed as follows.
4861:
4862: The projective duality provides a bijection
4863: $$
4864: \left \{ \mbox{ Configurations of 2n hyperplanes in } \C \PP^{n-1} \right \} \quad <--> \quad $$
4865: $$
4866: \left \{ \mbox{ Configurations of 2n points in } \C \PP^{n-1} \right \}
4867: $$
4868: So we can consider the dual function $\widehat {\cal P}^G_n$
4869: on configurations of $2n$ points in $\C \PP^{n-1}$. By definition its
4870: value value at a
4871: configuration of points is equal to the value of ${\cal P}_n$
4872: on the projectively dual configuration of hyperplanes in $\C \PP^{n-1}$
4873:
4874: {\bf Remark}. The configuration of $2n$
4875: vectors corresponding to an $(n-1)$- plane $h$ according to Lemma 1.1
4876: is different from (dual to) the one constructed above. Nevertheless
4877: the function $\widetilde {\cal P}_n$ coincides with the function
4878: $\widehat {\cal P}_n$ constructed in s.5 of the Introduction.
4879:
4880:
4881: Let us call by $m$-flag in $\C \PP^{k}$ a sequence of subspaces
4882: $L_{\bullet}:= L_0 \subset L_1 \subset ... \subset L_{m-1}$ where $dim
4883: L_i = i$.
4884:
4885: Let $H_1\ast H_2$ be the joining of planes $H_1, H_2$. In general
4886: $dim(H_1 \ast H_2) = dimH_1 + dimH_2 -1$.
4887:
4888: For $2n$ generic $(m+1)$-flags $L_{\bullet}^{(1)}, ... , L_{\bullet}^{(2n)}$ in $\C \PP^{n+m-1}$ set
4889: $$
4890: \widetilde {\cal P}_n^{(m)}(L_{\bullet}^{(1)}, ... , L_{\bullet}^{(2n)}):=
4891: \sum_{j_1+...+j_{2n} = m} \widetilde {\cal P}_n\Bigl((L_{j_1-1}^{(1)}\ast ... \ast L_{j_{2n}-1}^{(2n)}\vert L_{j_1}^{(1)}, ... , L_{j_{2n}}^{(2n)})\Bigr)
4892: $$
4893: Here $(L_{j_1-1}^{(1)}\ast ... \ast L_{j_{2n}-1}^{(2n)}\vert L_{j_1}^{(1)}, ... , L_{j_{2n}}^{(2n)})$
4894: is the configuration of $2n$ points in $\C \PP^{n-1}$ obtained by the
4895: projection of $L_{j_k}^{(k)}$ with the center at $L_{j_1-1}^{(1)}\ast
4896: ... \ast L_{j_{2n}-1}^{(2n)}$ (see fig 2). More precisely, the set of
4897: all planes of
4898: dimension $j_1+...+j_{2n}$ containing $L_{j_1-1}^{(1)}\ast
4899: ... \ast L_{j_{2n}-1}^{(2n)}$ is a projective space of dimension $n-1$.
4900: Each $L_{j_k}^{(k)}$ provides a point in this space.
4901:
4902: Choose an $(m+1)$-flag $L_{\bullet}$ in $\C \PP^{n+m-1}$. Set
4903: \begin{equation} \label{8as}
4904: c_{2n-1}^{n+m}(g_1,...,g_{2n}):= \widetilde {\cal P}_n^{(m)}(g_1
4905: L_{\bullet}^{(1)}, ... , g_{2n} L_{\bullet}^{(2n)})
4906: \end{equation}
4907: This function is defined only on a Zariski open subset of $GL_{n+m}(\C)^{2n}$ for which the flags in
4908: the right hand side of (\ref{8as}) are in generic position.
4909:
4910:
4911: \begin {Theorem}
4912: \label {5.5}
4913: a) $c_{2n-1}^{n+m}(g_1,...,g_{2n})$ is a measurable cocycle of the Lie
4914: group $GL_{n+m}(\C)$.
4915:
4916: b) The cohomology class of $(2 \pi i)^n c^{n+m}_{2n-1}$ in
4917: is a non
4918: zero rational multiple of the Borel class.
4919: \end{Theorem}
4920:
4921: (The measurable and continuous cohomology of a Lie group $G$ are
4922: isomorphic).
4923:
4924:
4925: {\bf Proof}. a) follows from the Key Lemma in s.2.1 in[G3] and b) from
4926: Theorem 5.12 in [G3]. The existence of the bi-Grassmannian $n$-logarithm
4927: is the main ingredient of the proof.
4928:
4929:
4930:
4931: {\bf 9. A sketch of a direct proof of Theorem (\ref{2112})}.
4932: Let $Q $ be a nondegenerate quadric in $\PP^n$ and $L$
4933: be a nondegenerate simplex in $\PP^n$.
4934: Assume that $L$ and $Q$ are in generic position.
4935:
4936:
4937: Let $\widetilde Q(x)$ be a quadratic
4938: equation of the quadric $Q$ in homogeneous coordinates. Set
4939: $$
4940: \omega_Q := \frac{{\sqrt det (\widetilde Q)}}{(2\pi i)^n}\frac{\sigma_n(x,dx)}{\widetilde Q(x)^n}
4941: $$
4942: It does not depend on the choice of $\widetilde Q(x)$. Let $\omega_L$ be the canonical differential form with logarithmic singularities at
4943: $L$ with integer periods.
4944:
4945: Let $\Delta_{L}$ be an $n$-chain (or an $n$-membrane) in ${\Bbb C} \PP^n$ such that $[\Delta_{L}]$ is a generator of $H_n({\Bbb C} \PP^n,L;\Z)$.
4946:
4947: Similarly let $\Delta_{Q}$ be an $n$-chain in ${\Bbb C} \PP^n$ such that $[\Delta_{Q}]$ is a generator of $H_n({\Bbb C} \PP^n,Q;\Z)$ corresponding to the primitive class under the isomorphism
4948: $$
4949: H_{n}({\Bbb C} \PP^{n},Q;\Z) = Ker \Bigl(H_{n-1}(Q;\Z)
4950: \longrightarrow H_{n-1}({\Bbb C} \PP^{n};\Z)\Bigr)
4951: $$
4952: The choice of a generator corresponds to the choice of the sign in the definition of the form $\omega_Q$.
4953:
4954:
4955:
4956: Let $\widehat Q$ be the quadric in
4957: $\widehat \PP^n$ dual to $Q$ and $\widehat L$ be the dual simplex in $\widehat \PP^n$. The vertices of $\widehat L$ correspond to the codimension one faces of $L$.
4958:
4959:
4960: \begin{theorem}
4961: a)
4962: $
4963: \int_{\Delta_{ \widehat Q}}\omega_D = \omega_Q, \qquad \int_{\Delta_{ \widehat L}}\omega_D = \omega_L
4964: $
4965:
4966: b)
4967: $
4968: \int_{\Delta_{L \times Q}}\omega_D = \int_{\Delta_{\widehat L}}\omega_Q = \int_{\Delta_{ \widehat Q}}\omega_L
4969: $
4970: \end{theorem}
4971:
4972:
4973: {\bf Proof}
4974: b) follows from a).
4975:
4976: a)Now consider a sphere $S^{n-1}$ in $\R \PP ^n$. It cuts the projective space on two domains, $
4977: {\cal B}_+ \cup {\cal B}_-$ one of which is convex. Consider a simplex
4978: $L$ inside of the convex ball. Let us think about the dual projective space as of the set of all hyperplanes. Then the hyperplanes which do not intersect the convex ball form a convex domain in $\widehat \R \PP ^n$, which we will denote $\widehat
4979: {\cal B}_+$.
4980:
4981: The simplex $L$ defines a dual simplex $\widehat L$ in $\widehat \R \PP^n$. Namely, each vertex $l_i$ of $L$ produces a face of $\widehat L_i$ of the dual simplex, namely
4982: $\widehat L_i$ consists of all hyperplanes passing through the vertex $L_i$ which do not intersect the interior part of $L$. Therefore the dual simplex $\widehat L$ lies outside of the dual ball $\widehat {\cal B}_+$
4983:
4984: The cycle
4985: $$
4986: L \times \widehat {\cal B}_+ \subset \R \PP ^n \times \widehat {\R \PP^n}
4987: $$
4988: does not intersect the incidence variety $\Delta \subset \R \PP^n \times \widehat {\R
4989: \PP^n}$. In particular the integral of our form
4990: over this cycle converges.
4991: $\int_{L} \omega_D = \omega_{\widehat L}$. Since the dual simplex $\widehat L$ does not intersect the interior part of the dual ball, the restriction of the form $\omega_{\widehat L}$ to the dual ball is exact. So we can use the Stokes theorem to reduce the calculation of the integral of $\omega_{\widehat L}$ over the dual ball to the integral over the sphere.
4992:
4993:
4994: {\bf Second proof}. We will suppose that $S$ is rational and connected.
4995: The starting point is the Lemma \ref{2.14}
4996:
4997: Let $s$ and $t$ be two (generic) points
4998: of $S$. Choose a path $\gamma$ from $s$ to $t$. Then
4999: $$
5000: {\cal P}_2(Y^s;f^s_1,f^s_2,f^s_3) - {\cal P}_2(Y^t;f^t_1,f^t_2,f^t_3) = \int_{\gamma}
5001: {\rm Alt}_3 \Bigl(v_x(f_1) \log|f_2(x)| d_s \arg f_3(x) \Bigr)
5002: $$
5003: Let us prove that the right hand side is expressed by ${\cal L}_2$. The integrand on the right is supported on a divisor $D \subset Y$, see proof of Lemma \ref{2.14}. Let us apply
5004: the transfer map
5005: $K_2(\C(D)) \longrightarrow K_2(\C(S))$ to the element $Res(f_1 \wedge f_2 \wedge f_3 )$. Let $D_1$ be an irreducible component of $D$. Denote by $\gamma_1$ the preimage of the path $\gamma$ in $D_1$.
5006: The same argumentation as in the first proof shows that we can assume that $ \C(D_1)$ is a Galois extension of $\C(S) $ and reduce the problem to the calculation of two integrals: $\int_{\gamma'} r_1(1-f,f)$ and $\int_{\gamma} r_1(f,g)$. Noting that
5007: $r_1(1-f,f) = d{\cal L}_2(f)$,
5008: we compute the first integral. Computing the second we may suppose that $\gamma$ lies in $\C {\Bbb P}^1 $. Then the problem reduces to the calculation of the integral $\int_0^z r_1(z-a,z-b)$. Writing
5009: $$
5010: (z-a) \wedge (z-b) = \left(1-\frac{(z-b)}{(a-b)}\right) \wedge \frac{(z-a)}{(a-b)}\quad \mbox{modulo} \quad \C^* \wedge \C(z)^*
5011: $$ and using that $\int_0^z r_1(c,f(z))=0$ for a constant $c$ we reduce the problem to calculation of the first integral. Theorem is proved.
5012:
5013:
5014:
5015: c) The special values of $L$-functions
5016: (Conjectures of Birch-Swinnerton-Dyer, Birch-Tate and Lichtenbaum and Deligne, the theorems of Borel, the regulators of Bloch, Beilinson and philosophy of mixed motives.
5017:
5018:
5019: