1:
2: \documentclass[12pt]{article}
3: \usepackage{amsfonts}
4: \usepackage{amssymb}
5: \usepackage[usenames]{color}
6: \usepackage{epsfig}
7:
8: \begin{document}
9:
10: \input epsf.sty
11:
12:
13: \centerline{\bf S.P.Novikov\footnote{Sergey P. Novikov, IPST and
14: MATH Department, University of Maryland, College Park MD, USA and
15: Russian Academy of Sciences (Steklov Math Institute and Landau
16: Institute for Theoretical Physics, Moscow), e-mails
17: novikov@ipst.umd.edu, snovikov@mi.ras.ru; This work is partially
18: supported by the Russian Grant in the Nonlinear Dynamics. }}
19:
20: \vspace{0.3cm}
21:
22: \centerline{\Large New Discretization of Complex Analysis:}
23:
24: \centerline{\Large The Euclidean and Hyperbolic Planes}
25:
26: \vspace{0.5cm}
27:
28: {\it Abstract. Discretization of Complex Analysis on the Plane based
29: on the standard square lattice was started in 1940s. It was
30: developed by many people and also extended to the surfaces
31: subdivided by the squares. In our opinion, this standard
32: discretization does not preserve well-known remarkable features of
33: the Completely Integrable System. These features certainly
34: characterize the standard
35: Cauchy Continuous Complex Analysis. They played a key role in the
36: great success of
37: Complex Analysis in Mathematics and Applications. Few years ago we
38: developed jointly with I.Dynnikov a New Discretization of Complex
39: Analysis (DCA) based on the two-dimensional manifolds with colored
40: black/white triangulation (see \cite{ND}). Especially deep results
41: were obtained for the Euclidean plane with equilateral triangle
42: lattice in the works \cite{ND,GN}. Our approach preserves a lot of
43: features
44: of Completely Integrable Systems.
45: In the present work we develop a DCA theory for the analogs of
46: equilateral triangle lattice in the Hyperbolic plane. This case is
47: much more difficult than Euclidean. Many problems (easily solved for
48: the Euclidean Plane) are not solved here yet. Some specific very
49: interesting ''dynamical phenomena'' appear in this case: for
50: example, description of boundaries of the most fundamental geometric
51: objects (like the round ball) leads to dynamical problems. Mike
52: Boyle from the University of Maryland helped me to use here the
53: methods of symbolic dynamics. }
54:
55: \vspace{2cm}
56:
57: {\bf Introduction. History.} We do not discuss here ''geometric''
58: discretizations of conformal mappings started in early XX
59: century-see survey and history in \cite{Mer}. By the way, specific
60: topological properties of surfaces subdivided by the squares, was
61: started by the combinatorial geometry/topology group in the Steklov
62: Institute in 1980s by the suggestion of the present author (see
63: \cite{DSS}). This group called this object ''The Quadrillage''. I
64: formulated these problems under the influence of the Statistical
65: Physics of Lattice Models, after discussions with A.Polyakov. It is
66: mentioned also in my Encyclopedia volume number 13, Topology-I.--see
67: \cite{N2}. In the works of the group \cite{Mer} made in last decade
68: such surfaces subdivided into squares are called ''quad-graphs''.
69:
70: {\bf Our goal is to discretize Cauchy-Riemann operator
71: $\bar{\partial}$ as a Linear Difference First Order Operator on
72: Triangulated Surfaces.} Let us say at the very beginning of this
73: article that {\bf Neither in the Standard Approach nor in our New Discretization
74: The Discrete Analogs of Holomorphic Functions form a Commutative Associative Ring
75: (at least naturally).}
76:
77: The Discretization of Complex Analysis as a Linear Difference Operator
78: was done for the square lattice in $R^2$ by
79: Lelong-Ferrand. Her work was published in 1944 (see \cite{LF}). In
80: the work \cite{DUF} Duffin developed or rediscovered this approach.
81: He also extended it to the rombic lattices later.
82:
83: {\it Duffin claimed that
84: another mathematician already discovered this idea before and published his work
85: in 1941 in the notes of some small provincial South American University.
86: Our attempts to
87: find this obscure edition failed, so we continue to quote Lelong-Ferrand as a
88: first author
89: who invented this idea.}
90:
91: In the standard approach Discrete Analog of the Cauchy-Riemann Operator
92: $\bar{\partial}$ acts on the $C$-valued functions $\psi$ of vertices
93: in the square lattice on the plane:
94: $$Q_{square}\psi(m,n)=\psi(m,n)+i\psi(m+1,n)-\psi(m+1,n+1)-i\psi(m,n+1)$$
95: By definition, every solution to the equation $Q_{square}\psi=0$ is
96: a d-holomorphic function.
97:
98: A lot of people work here now developing this approach (see, for
99: example, in \cite{Mer}).
100:
101: {\it Let us present a hint of the idea: Why we are not satisfied by
102: the standard
103: discretization? Of course, every continuous model admits many difference
104: approximations which converge to it in the continuum limit.
105:
106: For the continuous systems with huge hidden algebraic symmetry
107: we wish to find optimal discretization preserving as much symmetry as possible.
108: As we already know, there exists no natural discretization preserving multiplication
109: of holomorphic functions. All attempts to invent multiplicative structure with
110: good properties failed. It became clear many years ago.}
111:
112:
113: Let us point out following:
114:
115: 1. In the standard approach discrete analog of the Cauchy-Riemann
116: operator is in fact a
117: second order difference operator: two lengths are
118: involved in the sum (the lengths of sides and diagonals).
119:
120: {\bf We wish to have
121: a first
122: order difference operator as a natural discrete analog of $\bar{\partial}$.
123: Is it possible? The answer is YES for the Equilateral Triangle Lattice. Corresponding theory
124: of the difference first order ''Triangle Operators'' on the simplicial complexes
125: is developed in the Part 1 in more general form. We define also nonstandard
126: discrete analogs of
127: $GL_n$-Connections for the triangulated n-manifolds}.
128:
129: In the Part 2 we develop a general theory of discrete analogs of holomorphic
130: functions
131: (the d-holomorphic functions) on the 2-manifolds with discrete
132: analog of conformal structure (DCS). {\bf We define DCS as a colored
133: black/white triangulation.}
134:
135: The d-analogs of Liouville Principle and Maximaum Principle will be clarified here
136: for the general 2-surfaces with DCS.
137:
138: Part 3 is dedicated to the theory of d-holomorphic functions on the Equilateral
139: Triangle Lattice.
140:
141: 2.Continuous 2D Laplace operator admits a natural factorization
142: $$\Delta=\bar{\partial}\partial$$ Nothing like that exists on the square lattice.
143:
144: {\it However, Equilateral Triangle Lattice is much better for this--see below.}
145:
146:
147: {\bf Problems:}
148:
149: 1.How to construct Discrete Analogs of Holomorphic Polynomials without multiplication?
150:
151: People did it with great efforts after many years in the standard approach. Their constructions
152: are non-
153: canonical and non-unique.
154:
155: {\bf It is very easy in our new approach.
156: Our construction is obvious and canonical.}
157:
158:
159:
160:
161: 2.How to construct rational functions without multiplication?
162:
163: In particular, in the standard continuous complex analysis rational functions
164: are especially important because the function $1/z$ (the Cauchy Kernel) is also a
165: unique fundamental solution to the equation $$\bar{\partial}(\psi(z))=2\pi i\delta(z)$$
166: decreasing at infinity?
167:
168: {\bf The answer to this question is also positive and canonical in our approach.
169: It was missed in our first
170: work with Dynnikov \cite{ND} and made by Grinevich and R.Novikov later in \cite{GN}.}
171:
172:
173:
174: Part 4 is dedicated to the Hyperbolic (Lobachevski) Plane $H$.
175:
176: Equilateral Triangle Lattice in $H$ is defined simply as a triangulation of plane such that
177: $p$ triangles hit every vertex and
178: $p\geq 7$. We need $p=2s$. For all $s\geq 4$ we have a triangulation of Lobachevski Plane $H$
179: such that it admits
180: a black/white coloring. Let us concentrate our attention on the case $s=4$.
181:
182: At the moment we can
183: construct neither d-analog of polynomials nor d-rational functions.
184:
185: Even boundary of
186: round ball $D_r$ of radius $r\in Z$ is a complicated object. How to
187: describe it? How many points does it contain? We succeeded in the
188: solution of this problem using methods of symbolic dynamics.
189: Mike Boyle
190: from the University of Maryland helped us. The boundary $\partial
191: D_r$ contains approximately $\lambda^r$ points where
192: $\lambda=2+\sqrt{3}$.
193:
194: It turns out that for the reconstruction of d-holomorphic function
195: in the ball $D_{r+1}$ we need to know values of the boundary
196: function in $[n(r+1)/2]+1$ points. Here $n(r)$ is a number of
197: vertices in the boundary. It exactly corresponds to the continuous
198: case where only set of coefficients at the nonnegative powers of $z$
199: (i.e. one half plus one Fourier coefficients of function on the
200: boundary circle) determines completely our function inside. The main
201: unsolved problem is:
202:
203: {\bf Problem. How to find a bounded basis of d-holomorphic functions
204: on Hyperbolic Plane?}
205:
206: \vspace{1cm}
207:
208: {\bf Part I. Definitions. Discrete $GL_n$-Connections. B/W
209: manifolds. The First Order Triangle Operators.}
210:
211: \vspace{0.5cm}
212:
213:
214: Take a simplicial complex $M$. We assume that this complex is given
215: with canonical metric where every simplex is a standard unit linear
216: subsimplex in euclidean space with natural euclidean metric. We fix
217: a family of $n$-simplices such that every vertex belongs at least to
218: one simplex of that family. Fix also collection of nonzero numbers
219: $b_{T:P}\neq 0$ for all $T\in X$ and $P\in T$.
220:
221: \newtheorem{defi}{Definition}
222: \begin{defi} We call following operator
223: the Triangle Operator $Q^X$ associated with
224: family $X$: $$Q^X\psi(T)=\sum_{P\in T}\psi (P)b_{T:P}$$ It maps
225: functions of vertices into functions of simplices $T\in X$. Our
226: coefficients normally belong to $R$ or $C$. The operators $Q^X$ are
227: the first order linear difference operators.
228: \end{defi}
229:
230: Now we define a Discrete Version of $GL_n$-Connections.
231:
232: \begin{defi}
233: Let $X$ be a family of all $n$-simplices in $n$-manifold $M$. We
234: call the equation $Q\psi=0$ Discrete Differentially-Geometrical
235: $GL_n$-Connection. The coefficients $b_{T:P}$ are defined in every
236: $n$-simplex up to nonzero factor. So every DG-Connection is defined
237: by the set of ratios $$\mu^T_{PP'}=b_{T:P}/b_{T:P'}$$
238: \end{defi}
239:
240: This
241: discretization is different from the standard Wilson Discretization
242: used by physicists studying Yang-Mills fields; There is no natural
243: way to select compact holonomy groups in our approach.
244:
245: The theory of
246: discrete $GL_n$ connections was constructed in the works
247: \cite{DN,N}, some first ideas appeared in \cite{N1,ND} in connection
248: with completely integrable systems. We are working only with
249: scalar-valued functions. At the same time we are going to define a
250: Nonabelian Curvature and Holonomy with values in the group $GL_n$
251: for the $n$-manifolds.
252:
253: As we demonstrated in the \cite{N}, two different Holonomy
254: Representations can be constructed here: Abelian (''Framed
255: Abelian'') and Nonabelian. We need here only Nonabelian Holonomy.
256:
257: What is a Nonabelian Discrete Curvature? A Nonabelian Holonomy
258: group is defined along the Thick Paths.
259:
260: \begin{defi}
261: We call by Thick Path any sequence of $n$-simplices $<T_1T_2...T_k>$
262: such that intersection $T_j\bigcap T_{j+1}$ is exactly $n-1$-dimensional
263: face $\Delta_j$, and $\Delta_j\neq\Delta_{j+1}$. We say that Thick Path is closed if
264: $T_k= T_1$. Its length is equal to $k-1$.
265: \end{defi}
266:
267: Therefore in every simplex $T_j$ of the thick path exactly two
268: different $n-1$-dimensional faces are fixed: $\Delta_j\subset
269: T_{j+1}$ is called an ''In-face'' in $T_{j+1}$, and $\Delta_J\subset
270: T_j$ is called an ''Out-face'' in $T_j$.
271:
272: Let us define a {\bf Parallel Transport} of the $n$-vector $\psi$
273: consisting of values in all vertices of the in-face $\Delta_1\subset
274: T_1$, along the thick path $<T_1...T_k$. Solving equation $Q=0$, we
275: define this function in all vertices of $T_1$. So we know values of
276: $\psi$ in all vertices of the out-face in $T_1$ equal to the in-face
277: in $T_2$. After that we are doing the same procedure for $T_2$ and
278: so on. Finally we define a ''Transported value'' of $n$-vector
279: $\psi$ in the out-face of the simplex $T_k$. This map is linear. For
280: the closed paths we are coming to the {\bf Nonabelian Holonomy
281: Representation of the Semigroup of the closed thick paths into the
282: Group $GL_n$.}
283:
284: $$\Omega_{thick}(M,\Delta):\rightarrow GL_n$$
285:
286:
287:
288:
289:
290:
291:
292:
293: \begin{defi} We call by the Nonabelian Discrete Curvature in the vertex $P$ a Holonomy
294: Linear Map along the Thick Closed Paths in the Simplicial Star
295: $St(P)$. It is enough to know holonomy for all thick closed paths
296: around all $n-2$=simplices--see Fig 1 for $n=2$.
297:
298: \end{defi}
299:
300:
301:
302:
303:
304:
305:
306:
307: \begin{center}
308: \color{black}Fig 1 \vspace{5mm}
309: \mbox{\epsfysize=4cm\epsffile{fig5.eps}}
310: \end{center}
311:
312:
313:
314:
315:
316:
317:
318:
319:
320:
321:
322:
323: However, the natural gauge group
324: $$\psi\rightarrow f\psi, Q\rightarrow g(T)Qf^{-1}(P)$$ is abelian,
325: $f\neq 0,g\neq 0$. There are very interesting specific features here
326: realizing non-abelian $GL_n$ Connections using only the spaces of
327: scalar functions of vertices. Some sort of mixing abelian and
328: non-abelian properties appears. Details of classification can be
329: found in the work \cite{N}. No theory of Characteristic Classes is
330: constructed yet.
331:
332:
333: \begin{defi}
334: We call special discrete $GL_n$-Connection with $b_{T:P}=1$, a
335: Canonical Connection
336: \end{defi}
337:
338:
339:
340: Consider now any $n$-manifold. Is it possible to color its $n$-simplices
341: by the
342: black and white colors
343: such that for every pair of $n$-simplices attached to each other
344: along $n-1$-face, the colors are opposite?
345:
346: \begin{defi} We call such colored manifolds a B/W
347: manifolds.
348: \end{defi}
349:
350: Such coloring exists if and only if every closed thick path consists
351: of even number of $n$-simplices. More general, fix any $n-1$-simplex
352: $\Delta$. We have a natural homomorphism
353: $$\Phi_2:\Omega_{thick}(M,\Delta)\rightarrow Z_2=Z/2Z$$
354: defined as a number of $n$-simplices in any thick path modulo 2.
355: This map also looks like simplest Connection whose Holonomy Group
356: is $Z_2$, but we do not realize it by the linear operator.
357:
358: \begin{defi} Let the map $\Phi_2$ is trivial for
359: all thick closed paths in the star of every vertex. We call such triangulation
360: ''Locally B/W''.
361: \end{defi}
362:
363: In this case following
364: simple lemmas are true:
365:
366:
367: \newtheorem{ex}{Example}
368:
369:
370:
371: \newtheorem{lem}{Lemma}
372:
373: \begin{lem}
374: Under this assumptions the map $\Phi_2$ defines a map $\phi_2:\pi_1(M)\rightarrow Z_2$.
375: \end{lem}
376:
377:
378:
379: \begin{lem}
380: Curvature of the Canonical Connection is trivial for every locally B/W
381: manifold.
382: \end{lem}
383:
384: Proof. Fix any $n-2$-simplex $S$ in the star $St(P)$. Let exactly
385: $s(S)$ $n$-simplices contain $S$. The Nonabelian Holonomy with
386: initial values $a_0,...,a_n$ in the vertices of $n$-simplex
387: $T_n=<S,P_{n-1},P_n>$ containing $S$ (and around it), is determined
388: by permutation $A^{s(P)}$. Here $A(a_0,...,a_{n-2},a_{n-1},a_n)=
389: (a_0,...,a_{n-2},a_n,a_{n-1})$. So $A^2=1$. We see that Nonabelian
390: Holonomy (Curvature) is equal to $1$ if and only if $s(S)$ is even.
391: Lemma is proved.
392:
393:
394: Every locally flat Connection (i.e. Nonabelian Curvature is trivial
395: for every vertex) defines a homomorphism of topological holonomy
396: $\pi_1(M)\rightarrow GL_n$.
397:
398: \begin{lem}
399: Every flat Canonical Connection defines a Topological
400: Holonomy Homomorphism. Its image belongs to the subgroup
401: $S_{n+1}\subset GL_n$.
402:
403: \end{lem}
404:
405: Proof. The imbedding is following: Take arbitrary distinct $n+1$
406: numbers $a_0,...,a_n$ such that $\sum a_j=0$. Put this set of
407: numbers in the vertices of initial $n$-simplex $T_1$. For every
408: thick closed path $T_1,....,T_k$ the values of Parallel Transport
409: along this path in every simplex $T_j$ consists of the very same
410: numbers $a_q$ because $\sum a_j=0$, and Connection is Canonical. In
411: the final simplex $T_k=T_1$ we obtain permutation belonging to
412: $S_{n+1}$ which does not depend on the choice of initial numbers
413: $a_j$. Lemma is proved.
414:
415: {\bf Now we are going to consider only Orientable Manifolds $M$.}
416:
417: \begin{lem} Let Curvature and Topological Holonomy are trivial for the
418: Canonical Connection (i.e. it is flat). Then B/W coloring exists
419: globally.
420: \end{lem}
421:
422:
423:
424:
425:
426:
427:
428:
429:
430:
431: Proof. Let a closed thick path $<T_1,...,T_k>$ defines a trivial
432: permutation corresponding to the Canonical Connection. We claim that
433: this path consists of the even number of $n$-simplices. Indeed, it
434: follows from the fact that our Nonabelian Holonomy is equal to the
435: product of elementary transpositions. Every one of them changes
436: orientation. The whole path preserves orientation. So our lemma
437: follows.
438:
439: \begin{lem}
440: For every manifold $M$ with flat Canonical Connection the equation
441: $Q\psi=0$ defines $n$-dimensional linear space of Covariant
442: Constants.
443: \end{lem}
444:
445: Proof obviously follows from the fact that every initial values in
446: the $n-1$-simplex $Delta$ defines a covariant constant correctly
447: because our connection is flat.
448:
449: So we fix a manifold $M$ with B/W structure. We always assume that
450: our Canonical Connection is flat. It is always true for simply
451: connected manifolds. It is locally flat because B/W structure is
452: given.For non-simply-connected case B/W structure implies that
453: Nonabelian Holonomy maps $\pi_1(M)$ into $S_{n+1}$. So Canonical
454: Connection became globally flat on some finite covering with number
455: of sheets no more than $n!$. We define d-analog of holomorphic
456: functions in every B/W manifolds (see below).
457:
458: {\bf How to construct Continuous Limit
459: for our Discretization of Complex Analysis?}
460:
461: For that we are doing the following procedure: Let $n=2$. Construct
462: a special Covariant Constant $f_0$ whose values in every triangle
463: are $1,\zeta,\zeta^2$ where $\zeta^3=1$. It is unique up to the
464: group $S_3$ permuting vertices of the initial triangle. Apply
465: following gauge transformation to the Canonical Connection:
466: $$Q\rightarrow f_0^{-1}Qf_0, \psi\rightarrow f^{-1}_0\psi$$
467: We still have two-dimensional space of covariant constants for this
468: connection but in this gauge one of covariant constants became an
469: ordinary constant. We work over the extended field containing
470: $\zeta$. For the case $R$ and extended field $C$ we used this
471: construction to get Continuum Limit of our theory to the ordinary
472: complex analysis. For $M=R^2$ exactly one half of our theory
473: converges to the ordinary complex analysis. The second half diverges
474: for small triangulations.
475:
476:
477:
478: {\bf Now we start to construct New Discretization of Complex
479: Analysis.}
480:
481: Let $M$ be any B/W-manifold. Following two families play fundamental
482: role in our theory: $X_1=b$ (all black simplices) and $X_2=w$ (all
483: white simplices)--see Fig 2.
484:
485:
486: \begin{center}
487: \color{black} Fig 2 \vspace{5mm} \mbox{\epsfysize=7cm
488: \epsffile{fig4.eps}}
489: \end{center}
490:
491: \begin{defi}The operators $Q^b,Q^w$ are called Black and White
492: Triangle Operators. The union of this families $b\bigcup w$ is
493: complete. Using the Canonical Connection, we define an operator
494: $Q=Q^b\bigoplus Q^w$.
495: \end{defi}
496:
497: Following lemma is very simple:
498:
499: \begin{lem} The operators $Q^b,Q^w,Q$ have following properties:
500: $$L= Q^*Q=2Q^{b*}Q^b=2Q^{w*}Q^w$$
501: Let $\Delta=dd^*$ be an ordinary scalar simplicial Laplace-Beltrami
502: Operator, $m_P$ is the number of edges entering vertex $P$. For
503: 2-manifolds we have $Q^*Q=-2\Delta+3m_P$.
504: \end{lem}
505:
506: Proof follows from thew elementary calculation.
507:
508: \begin{defi}Consider any function $\psi$ such that $Q^b\psi=0$. For
509: $n=2$ we call such functions {\bf $d$-holomorphic}. We call
510: solutions to the equation $Q^w\psi=0$ $d$-antiholomorphic.
511: \end{defi}
512:
513: Now we assume that the Canonical Connection is flat. For every black
514: triangle $T$ we define a unique covariant constant $E\psi(T)$ whose
515: values in $T$ coincide with $\psi$.
516:
517: \begin{defi} We call $E\psi$ an {\bf Evaluation of function $\psi$ in the
518: black triangle $T$}.
519: \end{defi}
520:
521: {\bf As a conclusion, we can view every d-holomorphic function
522: either as a real-valued function $\psi$ on the vertices or as a
523: $R^2$-valued function $E\psi(T)$ of the black triangle $T$.}
524:
525: \vspace{2cm}
526:
527:
528: {\bf Part II. D-holomorphic functions on 2-manifolds. Liouville
529: Principle and Maximum Principle}
530:
531:
532: \vspace{1cm}
533:
534: As above, we consider triangulated orientable B/W manifolds $M$ with
535: flat Canonical Connection. Our field here is $R$. Following general
536: properties of d-holomorphic functions were found in \cite{DN}:
537:
538: \newtheorem{thm}{Theorem}
539: \begin{thm}(The Liouville Principle).
540: For the compact closed 2-manifold every d-holomorphic function is a
541: covariant constant.
542:
543: \end{thm}
544:
545: Proof is based on the ''instanton phenomenon'' for the quadratic
546: functional $(L\psi,\psi)$ if $L$ is factorizable $L=Q^*Q$. Let the
547: global minima $\psi$ in the Hilbert space $L_2$ is a zero mode
548: $L\psi=0$. The order of Euler-Lagrange equation drops twice reducing
549: to the ''Self-Duality Equation'' which is in our case simply the
550: d-analog of the Cauchy-Riemann Equation. Let us remind here that
551: {\bf This Property is the most fundamental feature of the standard
552: continuous Complex Analysis unifying it with The Completely
553: Integrable Systems}. The proof of our theorem is following:
554: $Q^b\psi=0$ implies $(Q^b)^*Q^b\psi=0 $ implies $L\psi=Q^*Q\psi=0$
555: implies $(L\psi,\psi)=0$ implies $(Q\psi,Q\psi)=0$ implies
556: $Q\psi=0$, so our theorem is proved.
557:
558: In
559: principle, the famous Instanton Phenomena discovered by Polyakov,
560: Belavin, Schwarz and Tiupkin in 1974 in the Nonlinear Variational
561: Calculus (Yang-Mills Field Theory), is based on the similar
562: arguments.
563:
564: Let $D$ be a bounded domain in $M$ consisting of black triangles.
565: \begin{defi} We call $T\in D$ a boundary triangle if some of its
566: vertices belong to black triangle not belonging to the domain $D$.
567: \end{defi}
568:
569: \begin{thm} (The Maximum Principle)
570: For every d-holomorphic function $\psi$ in $D$ the set of covariant
571: constants $E\psi(T)\in R^2$ for all $T\in D$ is contained in the
572: convex hull of the image of boundary triangles.
573: \end{thm}
574:
575: Proof can be found in \cite{DN}.
576:
577: In the next chapters we consider specific results obtained for the
578: Euclidean (Part III) and Lobachevski or Hyperbolic Planes (Part
579: IV).
580:
581: \vspace{1cm}
582:
583: {\bf Part III. Equilateral Triangle Lattice in $R^2$.}
584:
585: \vspace{1cm}
586:
587: The standard regular equilateral lattice in $R^2$ (see Fig 3) admits
588: two basic (unitary) shift operators $t_1,t_2$ acting on vertices
589: $$t_1(m,n)=(m+1,n), t_2(m,n)=(m,n+1)$$.
590:
591: \begin{center}
592: \color{black} Fig 3 \vspace{5mm}
593: \mbox{\epsfysize=5cm\epsffile{fig2.eps}}
594: \end{center}
595:
596: All six elementary shifts
597: $t_i^{\pm 1}, (t_1t_2^{-1}))^{\pm 1}$ have equal length. The
598: operators $Q^b,Q^w$ are equal to
599: $$Q^b=1+t_1+t_2, Q^w=(Q^b)^*=1+t_1^{-1}+t_2^{-1}$$
600: They map the space of functions of vertices into itself and are
601: adjoint to each other. Our field here is $R$. In the works
602: \cite{N1,ND} we defined first time black and white triangle
603: operators and invented the idea of new type discrete $GL_2$
604: connection. We proved also that every difference second order
605: self-adjoint operator $L=a+bt_1+ct_2+dt_1t_2^{-1}+(adjoint)$
606: admits Laplace-type factorization $L=Q^*Q+W$ where $Q$ is some
607: triangle operator $Q=u+vt_1+wt_2$. and $W,a,b,c,d,u,v,w$ are some
608: real functions on this lattice. In particular, $-\Delta+9=Q^bQ^w$
609: for the standard Laplace-Beltrami operator. A theory of discrete
610: completely integrable systems based on the discretized second
611: order operators, was started.
612:
613: \begin{defi} We call d-holomorphic function $\psi$ polynomial (i.e. $\psi\in
614: Pol_k$)
615: of degree
616: $k$ if $Q^b\psi=0$ and $(Q^w)^{k+1}\psi=0$.
617: \end{defi}
618:
619: As it was established in \cite{DN}, these functions have a
620: $k$-polynomial growth at infinity in $R^2$. They are completely
621: determined by their values in any standard $k$-triangles $T_k$ ,
622: black from inside, with $2k+2$ points in each edge (see Fig 4).
623:
624:
625: \begin{center}
626: \color{black} Fig 4 \vspace{5mm}
627: \mbox{\epsfysize=5cm\epsffile{fig6.eps}}
628: \end{center}
629:
630: So
631: the dimension of the space $Pol_k$ of $k$-polynomials is $2k+2$. We
632: can choose the canonical $k$-polynomials
633: $\psi_{T_k,\alpha},\alpha=1,2,3$. They are equal to zero in $T_k$
634: everywhere except one boundary edge $\alpha$ where they have values
635: $\pm 1$ . Sum of them belongs to the space $Pol_{k-1}$:
636: $$\sum_{\alpha}\psi_{k,\alpha}\in Pol_{k-1}$$
637: Therefore for the choice of basis we have to select pair of edges in
638: $T_k$.
639:
640:
641:
642:
643: Following theorem (\cite{DN}) give natural analog of polynomial
644: approximation of holomorphic functions leading to the ''d-Taylor
645: Series'':
646:
647: \begin{thm} For every function $\psi$, d-holomorphic in $R^2$, and every
648: canonical triangle $T_k$, there exists a unique $k$-polynomial
649: $\phi$ such that $\psi-\phi$ is identically equal to zero in the
650: triangle $T_k$.
651:
652: \end{thm}
653:
654: For the Taylor decomposition of $\psi$ we have to choose an
655: increasing sequence of triangles $T_k,k=0,1,2,...$ and pair of
656: boundary edges in each of them. The choice of such sequence is
657: non=canonical.
658:
659:
660: {\bf How to get analog of the Cauchy formula?}
661:
662: We need to construct
663: a ''Cauchy Kernel'' (d-analog of $1/z$) satisfying to equation:
664: $$Q^bG(x,y)=\delta(x-y)$$ where $x,y$ are points in the lattice
665: $x=(m,n),y=(m',n')$, and difference operator $Q^b$ acts on the
666: variables $x=(m,n)$. Having any such function, we consider any
667: bounded domain $D$ and $d$-holomorphic function $\psi$ in $D$. Let
668: us extend this function to the function $\bar{\psi}$ such that
669: $$\bar{\psi}(x)=\psi(x), x\in D; \bar{\psi}(x)=0, x\in R^2 minus D$$
670:
671: We have $$\sum_y [Q^b\bar{\psi}(y)]G(x-y) =\psi(x)$$ for all $x\in
672: D$. Let us point out that the function $Q^b\bar{\psi}(x)$ is
673: concentrated along the ''boundary strip'', so it is really analog of
674: the Cauchy formula. In the work \cite{DN} we constructed following
675: ''hyperbolic-like'' Cauchy Kernel (see Fig 6). We call it Pascal
676: Triangle. It is equal to zero outside of infinite triangle and has
677: exponential growth inside of it.
678:
679:
680:
681: \begin{center}
682: \color{black} Fig 5 \vspace{5mm}
683: \mbox{\epsfysize=5cm\epsffile{fig7.eps}}
684: \end{center}
685:
686: As it was pointed out by P.Grinevich and R.Novikov in \cite{GN},
687: much better Green function can be obtained simply by the Fourier
688: Transform
689: $$G(x)=(const)\int_0^{2\pi}\int_0^{2\pi}dk_1dk_2e^{imk_1}e^{ink_2}
690: (1+\exp\{ik_1\}+\exp\{ik_2\})^{-1}$$
691: where $x=(m,n)$.
692:
693: They proved following theorem:
694:
695: \begin{thm}This integral converges. This Green Function decays at infinity as
696: $c/d(P)$ where $d(P)$ is equal to the distance of this point to
697: zero. In particular, every d-holomorphic function on the whole
698: equilateral triangle lattice whose grouth at infinity is no more
699: that polynomial, is a d-holomorphic polynomial.
700:
701: \end{thm}
702:
703: {\bf Conclusion. This result gives a unique function decaying like
704: $1/|x|$ at $|x|\rightarrow\infty$. All functions $(Q^w)^kG(x)$ has a
705: growth like $1/|x|^{k+1}$. Therefore we have also analogs of
706: rational function without any of multiplication (which does not
707: exist).}
708:
709: \vspace{0.5cm}
710:
711:
712: {\bf Part IV. D-holomorphic functions on The Hyperbolic
713: (Lobachevski) Plane.}
714:
715: \vspace{1cm}
716:
717: Every triangulated plane $H^2$ such that more than six triangles
718: (edges) hit every vertex, can be viewed as a negative curvature
719: plane. An Equilateral Triangle Lattice we get in the case then this
720: number is the same for all vertices (i.e. $m_P\geq 7$). We need also
721: $B/W$ structure so our Equilateral Triangle Lattices $H^2_m$ in
722: $H^2$ with d-conformal structure (i.e. with B/W coloring) correspond
723: to $m=8,10,12,...$. We consider here only the case $m=8$.
724:
725: Let us point out that the triangles of this lattice cannot be made
726: arbitrary small in the standard Lobachevski metric (their size is
727: fixed in $H^2$ by the number $m$) but our domain $D$ can be made
728: arbitrarily large.
729:
730: For example, for every finite set of vertices $K\subset H^2_m$ and
731: positive integer $r\in Z_+$ we define a domain $D_{K,r}$ consisting
732: of vertices $x\in D_{K,r}$ such that distance $d(x,K)$ is no more
733: than $r$. We measure distances between vertices (sets) counting
734: minimal number of edges needed for joining them. The simplest
735: important domains of that kind correspond to the cases: 1.$K$=vertex
736: $0$ (we call it ''standard ball'' $D_r$); 2.$K$=triangle $T$ i.e.
737: $K$=3 vertices of $T$. We denote it $D_{T,r}$. It is also like a
738: ball with center in the center of triangle. For $K=D_{K',l}$ we have
739: $D_{K,r}=D_{K',r+l}$ if $K$ is connected (i.e. no jumps of the
740: length more than one are needed to reach one point from another).
741:
742: There is very big automorphism group mapping this Lattice into
743: itself but this group is non-commutative. It does not contain big
744: enough commutative subgroups, so nothing like Fourier transform
745: exists here. We cannot construct good enough Green function. Our
746: operators $Q^b,Q^w$ map space of functions of vertices into the
747: space of functions of black and white triangles correspondingly.
748: Therefore we cannot iterate them. So we cannot construct analogs of
749: polynomials here.
750:
751: {\bf Problem: How to construct basis of d-holomorphic functions
752: $\psi_l(x)$ in $H^2_8$ which are globally bounded in all space?}
753:
754: In the continuous case our space is realized as a unit disc
755: $D^2:|z|<1$. We have basis of bounded holomorphic functions $z^k,
756: k=0,1,2,...$. Every rational function with poles outside of unit
757: disc is bounded in it.
758:
759: Easy to construct some d-holomorphic functions $z_{P,r}(x)$ equal to
760: zero inside of the $r$-ball $D_r$ (i.e. for $x\in D_l,l<r$) and
761: equal to zero along the boundary $\partial D_r$ except of the
762: specific place $P\subset \partial D_r$ looking like in Fig 6, a and
763: b:
764:
765: \begin{center}
766: \color{black} Fig 6 \vspace{5mm}
767: \mbox{\epsfysize= 3cm\epsffile{fig19.eps}}
768: \end{center}
769:
770: Its extension to the external domain is non-unique. {\bf Can we
771: construct a globally bounded extension?}.
772:
773: Let us introduce class of right-convex paths.
774: \begin{defi} We call oriented path consisting of edges Right Convex
775: if it bounds two or three triangles only from the right site in
776: every vertex
777: \end{defi}
778:
779: We are coding all right convex paths by the words in two symbols
780: $w,b$ assigning to every edge letter $w$ or $b$ depending on which
781: color has triangle from the right site of this path.-see Fig 7
782:
783: \begin{center}\color{black}Fig 7\vspace{5mm}
784: \mbox{\epsfysize= 5cm\epsffile{fig18.eps}}
785: \end{center}
786:
787:
788: \begin{thm}
789: Let $\psi$ be a d-holomorphic function and $R$ is a full set of
790: zeroes $\psi(x)=0$. Consider boundary of its complementary domain,
791: i.e. the set $D_{R,1} minus R$, which is a set of points-closest
792: neighbors of zeroes. Every connected component of this boundary set
793: is right convex choosing orientation such that corresponding
794: component of zeroes lies inside.
795: \end{thm}
796: Proof. In order to have zero value of $\psi$ in the point $x$ we
797: should have following values in the neighboring point outside of
798: zero set (see Fig 8, a and b).
799:
800: \begin{center}\color{black}Fig 8\vspace{5mm}
801: \mbox{\epsfysize= 5cm\epsffile{fig21.eps}}
802: \end{center}
803:
804: No other possibilities exist because of
805: the equation $Q^b\psi=0$. It means in this case that $c\neq 0$
806: implies either $d\neq 0$ or $d=0$ in the Fig 8. In the first case
807: our boundary contains only two triangles inside. In the second case
808: $(d=0)$ our boundary contains three triangles inside. Anyway, it is
809: convex. Our theorem is proved.
810:
811: \begin{center}
812: \color{black} Fig 9\vspace{5mm}
813: \mbox{\epsfysize=5cm\epsffile{fig20.eps}}
814: \end{center}
815:
816: A maximal right convex set has a boundary path coded by the infinite
817: sequence $...bbbbb...$. For every point $x$ and direction-edge $l$
818: started in $x$ and having black triangle from the right, we uniquely
819: construct such maximal path $...bbb...$ and denote it by
820: $\gamma_{x,l}$. Easy to construct d-holomorphic function
821: $\psi_{x,l}(y)$ such that $\psi=0$ in the domain to the right of the
822: path $\gamma_{x,l}$, and $\psi(x)=\pm 1$ along the path
823: $\gamma_{x,l}$--see Fig 9. Its continuation to the complementary
824: domain is non-unique. We can define it from the requirement that the
825: growth is minimal if distance $d(y,\gamma_{x,l})\rightarrow \infty$,
826: but this definition is noneffective. {\bf How to find this growth?
827: What is a ''minimal'' function $\psi_{x,l}$?}. This function and its
828: group shifts give basis in the space of all d-holomorphic functions.
829:
830: Every
831: right-convex path $\gamma$ can be shifted to the left side by the
832: distance equal to one. We get a new path $T(\gamma)$. Between these
833: two paths we have a strip containing all triangles $T$ having at
834: least one common vertex with path $\gamma$. New path $T(\gamma)$
835: consists of edges opposite to vertices belonging to $\gamma$. Only
836: triangles having exactly one common vertex with $\gamma$ participate
837: in this construction (see Fig 10).
838:
839: \begin{center}
840: \color{black} Fig 10 \vspace{5mm}
841: \mbox{\epsfysize=5cm\epsffile{fig22.eps}}
842: \end{center}
843:
844:
845: \begin{lem}
846: New path $T(\gamma)$ is also right convex. Its word can be obtained
847: from the word describing $\gamma$ by the following procedure: take
848: every pair of neighboring letters in $\gamma$ and put between them
849: words written below $$bw\rightarrow bwbw, wb\rightarrow wbwb$$
850: $$bb\rightarrow bwb,ww\rightarrow wbw$$
851: After that delete old letters.What remains is exactly a word
852: $T(\gamma)$.
853: \end{lem}
854:
855: Proof easily follows from the picture (see Fig 11).
856:
857: \begin{center}\color{black} Fig 11
858: \vspace{5mm}
859: \mbox{\epsfysize=10cm\epsffile{fig23.eps}}
860: \end{center}
861:
862: This map can be studied by the technic of symbolic dynamics. Mike
863: Boyle from the University of Maryland helped me a lot with this
864: business. We introduce new letters in order to describe this map.
865: Let $bw$ corresponds to the symbol $w_b$, $bb$ corresponds to $b_b$
866: $wb$ corresponds to $b_w$ and $ww$ corresponds to $w_w$. Written in
867: these symbols, our map $T$ has a form $$w_b\rightarrow w_bb_ww_bw_w,
868: w_w\rightarrow b_ww_bw_w$$ and $$b_w\rightarrow b_ww_bb_wb_b,
869: b_b\rightarrow w_bb_wb_b$$ After abelianization and replacing
870: product by sum, we are coming to the''Perron matrix'' $A$ whose
871: largest eigenvalue is $\lambda=2+\sqrt{3}$. So we proved following
872:
873: \begin{thm}The size of right convex path $\gamma$ shifted to the left side
874: by the distance one, increases asymptotically by the factor
875: $\lambda=2+\sqrt{3}$, so we have $|T(\gamma)|\sim
876: (2+\sqrt{3})|\gamma|$. In particular, this is true for the boundary
877: of $r$-ball $|\gamma|=|\partial D_r|\sim \lambda^{r-k}|\partial
878: D_k|, k\geq 1$
879: \end{thm}
880:
881: \begin{ex}We have $|\partial D_r|=8, 32, 120, 448, 1672, ...$ for
882: $r=1,2,3,4, 5,..$, so our asymptotic formula is practically exact
883: for $r\geq 3$.
884: \end{ex}
885:
886: {\bf How to calculate dimension of the space of d-holomorphic
887: functions in the ball $D_r$? How many data on the boundary $\partial
888: D_r$ are needed to recover d-holomorphic function in $D_r$?}
889:
890:
891: Consider strip between $D_r$ and $D_{r+1}$-see Fig 12, r=1.
892:
893:
894: \begin{center}
895: \color{black} Fig 12 \vspace{5mm}
896: \mbox{\epsfysize=5cm\epsffile{fig17.eps}}
897: \end{center}
898:
899: Every letter $b$ in $\partial D_{r+1}$ defines black triangle in the
900: strip touching $\partial D_r$ in one vertex. Every letter $w$ in
901: $\partial D_r$ defines black triangle in the same strip touching
902: $\partial D_{r+1}$ in one vertex. So total number of equations
903: $Q^b\psi=0$ in this strip is equal to $B_{r+1}+W_r$, the numbers of
904: black and white letters in the boundaries $\partial D_{r+1},\partial
905: D_r$. Unifying all strips with $k \leq r+1$, we get total number of
906: equations $Eq_{r+1}:Q^b\psi=0$:
907: $$Eq_{r+1}=(B_{r+1}+W_r)+...+(B_2+W_1)+B_1$$
908: For the number of points in $D_{r+1}$ we have
909: $$N_{r+1}=1+B_1+W_1+...+B_k+W_k+...+B_{r+1}+W_{r+1}$$
910: and $|\partial D_k|=B_k+W_k$. So we have $Eq_{r+1}=B_{r+1}+N_r-1$.
911: For the number of necessary data on the boundary $\partial D_{r+1}$
912: we obtain (taking into account $W_k=B_k$)
913: $$N_{r+1}-Eq_{r+1}=W_{r+1}+1=|\partial D_{r+1}|/2+1$$
914:
915: {\bf Conclusion: Our result that ''The Number of Necessary Data is
916: equal to $[|\partial D_{r+1}|/2]+1$'' is an exact analog of the
917: Continuous case: Only the set of Fourier coefficients on the
918: boundary circle corresponding to the exponents $\exp\{in\phi\}$ with
919: nonnegative $n\geq 0$, is needed for the reconstruction of
920: holomorphic function in the whole disc inside.}
921:
922:
923: \begin{thebibliography}{99}
924:
925: \bibitem{DN}{I.Dynnikov, S.Novikov. Geometry of Triangle Equation, Moscow Math Journal-MMJ
926: (2003) v 3, pp 410-438 }
927:
928:
929: \bibitem{LF} {J.(Lelong)-Ferrand. Fonctions preharmoniques et fonctions preholomorphes,
930: Bull Sci Math(1944) v 68 second series, pp 152-180}
931:
932: \bibitem{DUF} {R.Duffin. Basic properties of discrete analytic function, Duke Math Journal 1956 vol 23 pp 335-363}
933:
934: \bibitem{N}{S.Novikov. Discrete $GL_n$-Connections. Proceeding of Steklov Math Institute,
935: (2004) v 247, pp 186-201}
936:
937: \bibitem{Mer}{A.Bobenko, C.Mercat, Yu.Suris. Linear and nonlinear theories of discrete
938: analytic functions. Integrable structure and isomonodromic Green functions,
939: J.Reine Angew. Mathematics(2005) v 583, pp 117-161 }
940:
941: \bibitem{DSS}{N.Dolbilin, M.Shtan'ko, M.Shtogrin. a) Cubic subcomplexes in regular
942: lattices. Dokl. Akad. Nauk SSSR Dokl. 291, 277-279 (1986), English
943: translation: Sov. Math. Dokl. 34, 467-469 (1987); b) (jointly with
944: Sedrakyan, A.G.). A topology for the family of parametrizations of
945: two-dimensional cycles arising in the three-dimensional Ising model.
946: Dokl. Akad. Nauk SSSR 295, 12-23 (1987), English translation: Sov.
947: Math., Dokl. 36, No.1, 11-15 (1987); c) The problem of
948: parametrization of cycles modulo 2 in a three-dimensional
949: cubic lattice. Izv. Akad. Nauk SSSR, Ser. Mat., 52, No. 2, 355-377 (1988);
950: English translation:
951: Math. USSR, Izv. 52, No.2, 359-383 (1989); ñ) Quadrillages and
952: parametrizations of lattice cycles. Tr. Mat. Inst. Steklova 196,
953: 66-85 (1991); English translation: Proc. Steklov Inst. Math. 196,
954: 73-93 (1992); d) Cubic manifolds in lattices. Izv. Ross. Akad.
955: Nauk, Ser. Mat., 58, No. 2, 93-107 (1994), English translation:
956: Izv., Math. 44, No.2, 301-313 (1995). }
957:
958: \bibitem{N2}{S.Novikov. Topology I. Encyclopedia Math Sciences (Editor General-R.Gamkrelidze), vol 13
959: (S.Novikov, B.Botvinnik, R Burns--editors),
960: Springer Verlag (in english), page 41-42}
961:
962: \bibitem{N1}{S.Novikov. Algebraic properties of 2D difference operators, Russian Math Surveys
963: (1997) v 52, n 1, pp 225-226 }
964:
965: \bibitem{ND}{S.Novikov, I.Dynnikov. Discrete Spectral Symmetries of differential and difference
966: low dimensional operators, Russian Math Surveys (1997) v 52, n 5, pp
967: 175-234}
968:
969: \bibitem{GN}{ P.Grinevich, R.Novikov. The Cauchy kernel for Novikov-Dynnikov (DN) discrete
970: complex analysis in triangular lattioces, Russian Math Surveys
971: (2007) v 62, n 4, pp 799-801}
972:
973: \bibitem{GN1} {P.Grinevich, R.Novikov, to appear in 2010}
974:
975: \end{thebibliography}
976:
977: \end{document}
978: