1:
2: \documentclass{amsart}
3: \usepackage{fullpage,amssymb,epic,eepic,epsfig}
4: %,amscd,pb-diagram,lamsarrow,pb-lams}
5: \usepackage{ipe}
6: %\usepackage{graphicx}
7: \usepackage{latexsym}
8: \newcommand{\lipefig}[4]
9: { % #1 picture, #2 caption, #3 placement, #4scale
10: \begin{figure*}[#3]
11: \begin{center}
12: \IpeNoScaleFont
13: \IpeScale{#4}
14: \Ipe{#1.ipe}
15: \end{center}
16: \caption{#2}
17: \label{f:#1}
18: \end{figure*}
19: } % labelled Ipe figure
20:
21: \newcommand{\lipefignocap}[3]
22: { % #1 picture, #2 placement, #3 scale
23: \begin{figure*}[#2]
24: \refstepcounter{figure}
25: \begin{center}
26: \IpeNoScaleFont
27: \IpeScale{#3}
28: \Ipe{#1.ipe}
29: \\
30: Figure \thefigure
31: \label{f:#1}
32: \end{center}
33: \end{figure*}
34: }
35:
36: %%%%%%%%%%% macros
37:
38: \theoremstyle{plain}
39: \newtheorem{axiom}{Axiom}
40: \newtheorem{theorem}{Theorem}
41: \newtheorem{proposition}{Proposition}[section]
42: \newtheorem{lemma}[proposition]{Lemma}
43: \newtheorem{corollary}[proposition]{Corollary}
44: \newtheorem{claim}[proposition]{Claim}
45: \newtheorem{conjecture}{Conjecture}
46: \newtheorem{observation}{Observation}
47: \newtheorem{add}{Addendum}
48: \theoremstyle{definition}
49: \newtheorem{definition}[proposition]{Definition}
50: \newtheorem{problem}[proposition]{Problem}
51: \newtheorem{question}{Question}
52: \newtheorem{prize}{Prize}
53:
54: \theoremstyle{remark}
55: \newtheorem{example}[proposition]{Example}
56: \newtheorem{exercise}[proposition]{Exercise}
57: \newtheorem{hint}[proposition]{Hint}
58: \newtheorem{remark}[proposition]{Remark}
59:
60: \def\printname#1{
61: \if\draft y
62: \smash{\makebox[0pt]{\hspace{-0.5in}
63: \raisebox{8pt}{\tt\tiny #1}}}
64: \fi
65: }
66:
67: \newcommand{\mathmode}[1]{$#1$}
68: \newcommand{\psdraw}[2]
69: {\begin{array}{c} \hspace{-1.3mm}
70: \raisebox{-4pt}{\epsfig{figure=draws/#1.eps,width=#2}}
71: \hspace{-1.9mm}\end{array}}
72: \def\ppsdraw#1#2{\epsfig{figure=draws/#1.eps,height=#2in,silent=}}
73:
74: \newlength{\standardunitlength}
75: \setlength{\standardunitlength}{0.0125in}
76: \newcommand{\eepic}[2]
77: {\printname{#1}
78: \setlength{\unitlength}{#2\standardunitlength}
79: \begin{array}{c} \hspace{-1.7mm}
80: \raisebox{-8pt}{\input draws/#1.eepic }
81: \hspace{-1.9mm}
82: \end{array}
83: }
84: \newcommand{\silenteepic}[2]
85: {\setlength{\unitlength}{#2\standardunitlength}
86: \begin{array}{c} \hspace{-1.7mm}
87: \raisebox{-3pt}{\input draws/#1.tex }
88: \hspace{-1.9mm}
89: \end{array}
90: }
91:
92: \catcode`\@=11
93: \long\def\@makecaption#1#2{%
94: \vskip 10pt
95:
96: \setbox\@tempboxa\hbox{%\ifvoid\tinybox\else\box\tinybox\fi
97: \small\sf{\bfcaptionfont #1. }\ignorespaces #2}%
98: \ifdim \wd\@tempboxa >\captionwidth {%
99: \rightskip=\@captionmargin\leftskip=\@captionmargin
100: \unhbox\@tempboxa\par}%
101: \else
102: \hbox to\hsize{\hfil\box\@tempboxa\hfil}%
103: \fi}
104: \font\bfcaptionfont=cmssbx10 scaled \magstephalf
105: \newdimen\@captionmargin\@captionmargin=2\parindent
106: \newdimen\captionwidth\captionwidth=\hsize
107: \catcode`\@=12
108:
109: \newcommand{\End}{\operatorname{End}}
110: \newcommand{\Span}{\operatorname{span}}
111: \newcommand{\tr}{\operatorname{tr}}
112: \newcommand{\im}{\operatorname{Im}}
113: \newcommand{\noi}{\noindent}
114:
115: \def\lbl#1{\label{#1}\printname{#1}}
116: \def\biblbl#1{\bibitem{#1}\printname{#1}}
117:
118: \def\eqdef{\overset{\text{def}}{=}}
119:
120: \def\BTclear{\def\tensorname{}
121: \def\basea{}\def\collectora{}\def\emittera{}
122: \def\baseb{}\def\collectorb{}\def\emitterb{}}
123:
124:
125: %%%%%%%%%%%%% Stavros's defs
126:
127: \def\BN{\mathbb N}
128: \def\BZ{\mathbb Z}
129: \def\BQ{\mathbb Q}
130: \def\BR{\mathbb R}
131: \def\BC{\mathbb C}
132: \def\BE{\mathbb E}
133:
134: \def\AC{\mathcal AC}
135: \def\A{\mathcal A}
136: \def\B{\mathcal B}
137: \def\C{\mathcal C}
138: %\def\D{\mathcal D}
139: \def\E{\mathcal E}
140: \def\D{\Delta}
141: \def\K{\mathcal K}
142: \def\cL{\mathcal L}
143: \def\G{\mathcal G}
144: \def\O{\mathcal O}
145: \def\T{\mathcal T}
146: \def\M{\mathcal M}
147: \def\N{\mathcal N}
148: \def\F{\mathcal F}
149: \def\V{\mathcal V}
150: \def\W{\mathcal W}
151: \def\P{\mathcal P}
152: \def\R{\mathcal R}
153: \def\T{\mathcal T}
154:
155: \def\I{\mathcal I}
156: \def\CI{\mathcal C}
157: \def\Err{\mathcal Err}
158:
159: \def\a{\alpha}
160: \def\bb{\beta}
161: \def\La{\Lambda}
162: \def\l{\lambda}
163: \def\Ga{\Gamma}
164: \def\S{\Sigma}
165: \def\Sp{\Sigma^p}
166: \def\s{\sigma}
167: \def\varsi{\varsigma}
168: \def\ga{\gamma}
169:
170: \def\as{algebraically split}
171: \def\ihs{integral homology 3-sphere}
172: \def\qhs{rational homology 3-sphere}
173: \def\hs{homology 3-sphere}
174: \def\fti{finite type invariant}
175: \def\uf{unit-framed}
176: \def\rf{rationally framed}
177: \def\Heg{Heegaard}
178: \def\cok{\mathrm{cok}}
179: \def\torsion{\mathrm{torsion}}
180:
181: \def\la{\langle}
182: \def\ra{\rangle}
183:
184:
185: \def\Fb#1{\mathcal F^{b}_{#1}\mathcal M} %%%% filtrations on \mathcal M.
186: \def\FY#1{\mathcal F^Y_{#1}\mathcal M}
187:
188: \def\Gb#1{\mathcal G^{b}_{#1}\mathcal M} %%%% filtrations on \mathcal M.
189: \def\GY#1{\mathcal G^Y_{#1}\mathcal M}
190:
191: \def\w{\omega}
192: \def\ov#1{\overline{#1}}
193:
194: \def\we{\wedge}
195: \def\gl{\mathfrak{gl}}
196:
197: \def\g{\gamma}
198: \def\t{\tau}
199: \def\e{\epsilon}
200: \def\Ga{\Gamma}
201: \def\d{\delta}
202: \def\b{\beta}
203: \def\th{\theta}
204: \def\Th{\Theta}
205: \def\s{\sigma}
206: \def\p{\prime}
207: \def\sub{\subseteq}
208:
209: \def\lk{{\text{lk}}}
210: \def\sgn{\operatorname{sgn}}
211:
212: \newcommand{\Ker}{\operatorname{Ker}}
213: \newcommand{\con}{\equiv}
214: \newcommand{\done}{\hfill$\Box$\par}
215: \def\Sei{\mathrm{Sei}}
216:
217: \def\iso{\cong}
218: \def\pt{\mathbb Z[\![t_1 ,\dots,t_m ]\!]}
219: \def\sms{\smallsmile} %%\smallsmile}
220: \def\smf{\smallfrown} %%\smallfrown}
221:
222: \def\sminus{\smallsetminus}
223:
224: \def\ti{\widetilde}
225:
226: \def\circle{\operatorname{\psfig{figure=draws/circle.eps,height=0.1in}}}
227: \def\line{\operatorname{\psfig{figure=draws/line.eps,height=0.1in}}}
228: \def\wheel{\operatorname{\psfig{figure=draws/w2.eps,height=0.1in}}}
229: \def\ostar{\circledast}
230: \def\tcircle{\circlearrowleft}
231: \def\tline{\uparrow}
232:
233: %%%%%%%% the .eps definition of \strutb
234: % \def\strutb#1#2#3{\overset{#1}{\underset{#2}{
235: % \begin{array}{c} \hspace{-1.3mm}
236: % \raisebox{-4pt}{\psfig{figure=draws/strutb.eps,height=0.2in} }
237: % \hspace{-1.9mm}\end{array} }}#3}
238:
239: %%%%%%%% the .tex definition of \strutb
240: \def\strutb#1#2#3{\overset{#1}{\underset{#2}{
241: \begin{array}{c} \vspace{0.0cm}
242: \uparrow
243: \vspace{-0.25cm} \\ %\vspace{-0.25cm}
244: | \vspace{-0.45cm} \\ %\vspace{-0.45cm}
245: \bullet \vspace{0.00cm} %\vspace{0.00cm}
246: \end{array} }}\! #3}
247:
248: \def\st#1#2#3{\overset{#1}{\underset{#2}{
249: \begin{array}{c} \hspace{-1.3mm}
250: \raisebox{-4pt}{\psfig{figure=draws/strut.eps,height=0.2in} }
251: \hspace{-1.9mm}\end{array} }}#3}
252:
253: \def\AS{\mathrm{AS}}
254: \def\IHX{\mathrm{IHX}}
255: \def\GA{\A^{\mathrm{gp}}}
256: \def\GAz{\A^{\mathrm{gp},0}}
257:
258: \def\mat#1#2#3#4{\left(
259: \begin{matrix}
260: #1 & #2 \\
261: #3 & #4
262: \end{matrix}
263: \right)}
264:
265: \def\matt#1#2#3#4#5#6#7#8#9{\left[
266: \begin{matrix}
267: #1 & #2 & #3 \\
268: #4 & #5 & #6 \\
269: #7 & #8 & #9
270: \end{matrix}
271: \right]}
272:
273: \def\Sym{\mathrm{Sym}}
274: \def\pt{\partial}
275:
276: \def\Per{\mathrm{Per}}
277: \def\fsl{\mathfrak s\mathfrak l}
278: \def\LIG{\mathrm{LIG}}
279: \def\LID{\mathrm{LID}}
280: \def\TID{\mathrm{TID}}
281: \def\IM{\mathrm{IM}}
282: \def\IMJ{\mathrm{IM}_J}
283: \def\ot{\otimes}
284: \newcommand{\cl}{$\clubsuit$}
285: \def\bt{\bar{t}}
286: \def\rot{\mathrm{rot}}
287: \def\longto{\longrightarrow}
288: \def\Sev{{\mathcal S}}
289:
290: \def\Ges{\G^{ev,s}}
291: \def\sign{\mathrm{sign}}
292: \def\exc{\mathrm{exc}}
293: \def\Cab{\mathrm{Cab}}
294: \def\dgn{\mathrm{dgn}}
295: \def\udot{\mathaccent\cdot\cup}
296: \def\mult{\mathrm{mult}}
297: \def\as{\mathrm{as}}
298: \def\smooth{\mathrm{smooth}}
299: \def\bq{\bar{q}}
300: \def\err{\mathrm{err}}
301: \def\dec{\mathrm{dec}}
302: \def\vert{\mathrm{vert}}
303: \def\circ{\mathrm{circ}}
304: \def\longto{\longrightarrow}
305: \def\unknot{\mathrm{unknot}}
306: \def\defect{\mathrm{def}}
307: \def\Per{\mathrm{Per}}
308: \def\vphi{\varphi}
309: \def\Edges{\mathrm{Edges}}
310:
311: \begin{document}
312:
313: %%%%%%%%%%%%%%%%%%%%%%{page1}
314:
315: \title[The 3D Ising Problem Revisited]
316: {The 3D Dimer and Ising Problems Revisited}
317:
318: \author{Martin Loebl}
319: \address{Dept.~of Applied Mathematics and\\
320: Institute of Theoretical Computer Science (ITI)\\
321: Charles University \\
322: Malostransk\' e n. 25 \\
323: 118 00 Praha 1 \\
324: Czech Republic.}
325: \email{loebl@kam.mff.cuni.cz}
326: \author{Lenka Zdeborov\' a}
327: \address{Institute of Physics \\
328: Academy of Science of the Czech Republic\\
329: Na Slovance 2\\
330: 182 21 Praha 8\\
331: Czech Republic.}
332: \email{zdeborl@fzu.cz}
333:
334: \thanks{The second author gratefully acknowledges the support of project AVOZ10100520
335: of the Academy of Sciences of the Czech Republic.
336: % This and related preprints can be obtained at
337: %{\tt http: ...}
338: %\newline
339: 2000 {\em Mathematics Classification.} Primary 82B20. Secondary 82B41.
340: \newline
341: {\em Key words and phrases: 3D Ising model, 3D Dimer, 2D surfaces, Kasteleyn's orientations}
342: }
343:
344: \date{
345: This edition: \today \hspace{0.5cm} First edition: October 2004.}
346:
347: %\dedicatory{Preliminary notes. Please do not distribute under
348: %any circumstances!}
349:
350: \begin{abstract}
351: We express the finite 3D Dimer partition function as a linear combination of determinants
352: of oriented adjacency matrices, and the finite 3D Ising partition sum as a linear
353: combination of products over aperiodic closed walks. The methodology we use is embedding
354: of cubic lattice on 2D surfaces of large genus.
355: \end{abstract}
356:
357: \maketitle
358:
359: %\tableofcontents
360:
361:
362:
363: %%%%%%%%%%%%%%%%% the text file
364:
365: \section{Introduction}
366: \label{sec.int}
367:
368: The motivation for this work has been a renewed interest of theoretical physicists
369: in basic statistical physics models, the Ising and the Dimer models.
370: Let us mention two such recent promising works.
371: The first one is a new relation (duality) of topological strings and the Dimer model, see e.g.
372: \cite{ORV}.
373: The second one is a relation of the theory of discrete Riemann surfaces \cite{M}
374: and Theta functions with criticality of the Ising problem \cite{AMV, CM1, CM2}.
375:
376:
377:
378: In this paper we present new formulas for the finite 3D Dimer and 3D Ising problems.
379: The formulas are obtained in a combinatorial way. We hope that a profitable relation
380: with the work mentioned above may be found.
381:
382: A model we consider here is the
383: {\it Ising version of the Edwards-Anderson model}.
384: It can be described as follows.
385: A {\it coupling constant} $J_{ij}$ is assigned to each bond $\{i, j\}$
386: of a given lattice graph G;
387: the coupling constant characterizes the interaction between the particles
388: represented by sites $i$ and $j$.
389: A physical state of the system is an assignment of spin
390: $\sigma _i \in \{+1,-1\}$ to each site $i$.
391: The {\it Hamiltonian} (or energy function) is defined as
392: $H(\sigma) = -\sum_{\{i,j\} \in E} {J_{ij} \sigma_i \sigma_j}$.
393: The distribution of physical states over all possible energy levels
394: is encapsulated in the {\it partition function}
395: $Z(\beta) = \sum_{\sigma} {e^{-\beta H(\sigma)}}$ from which
396: all fundamental physical quantities may be derived.
397:
398: We may reformulate the Ising problem
399: in graph theoretic terms as follows.
400: A graph is a pair $G=(V,E)$ where $V$ is a set of {\it vertices} and $E$ is
401: now the set of edges (not the energy).
402: A graph with some regularity properties may be called a {\it lattice graph}.
403: We associate with each edge $e$ of $G$
404: a weight $w(e)$ and for a subset of edges $A \subset E$,
405: $w(A)$ will denote the sum
406: of the weights $w(e)$ associated with the edges in $A$.
407:
408: An {\it even subgraph} of a graph $G=(V,E)$ is a set of edges $U \subset E$
409: such that each vertex of $V$ is incident with an even number of edges from $U$.
410: The {\it generating function of even subgraphs} ${\mathcal H}(G,x)$
411: equals the sum of $\alpha^{w(U)}$ over all even subgraphs $U$ of $G$. A classic
412: relation between the Ising partition function and the generating function of
413: even subgraphs of the same graph states that
414: $$
415: Z(\beta) = {2^n} \left( {\prod_{\{i,j\} \in E} \cosh({\beta} J_{ij})}
416: \right) {\mathcal H} (G, \tanh(\beta J_{ij})).
417: $$
418:
419: A subset of edges $P\subset E$ is called
420: a {\it perfect matching} or {\it dimer arrangement} if each vertex belongs
421: to exactly one element of $P$.
422: The {\it dimer partition function} on graph $G$ may be viewed as a polynomial
423: ${\mathcal P}(G,\alpha)$ which equals the sum of $\alpha^{w(P)}$ over all perfect
424: matchings $P$ of $G$. This polynomial is also called the {\it generating function
425: of perfect matchings}.
426: \medskip
427:
428:
429: The generating functions of even subgraphs and perfect matchings may
430: be defined in
431: a more general way as follows: associate a variable $x_e$ with each edge
432: $e$ of graph $G$, let $x(A)=\prod_{e\in A} x_e$ and let e.g. the generating
433: function of perfect matchings be the sum of $x(P)$, $P$
434: perfect matching of $G$.
435: All results introduced in this paper also hold in this more general setting;
436: however the presentation using weights rather than variables is perhaps
437: more natural.
438: \medskip\noindent
439:
440:
441: \subsection*{Convention for the cubic lattice}
442: \label{sub.conv}
443:
444: This paper studies properties of finite cubic lattices. Let us now fix
445: some notation for them.
446:
447: \begin{figure}[ht]
448: \begin{minipage}{0.45\linewidth}
449: \includegraphics[width=6.5cm]{unper2.eps}
450: \caption{\label{fig-orde} Illustration of the linear order of vertices
451: on a lattice $Q(3,3,4)$.}
452: \end{minipage}
453: \begin{minipage}{0.50\linewidth}
454:
455: Let $m_1, m_2, k$ be positive integers, where $k$ is even and $m_1$
456: and $m_2$ odd.
457: The cubic lattice $Q= Q(m_1m_2k)$ is the following graph: \\
458:
459:
460:
461: \noindent The vertices are: \\
462: \noindent $V(xyz)$, $x= 1,\dots, m_1$, $y= 1,\dots, m_2$,
463: $z=1,\dots, k$. \\
464:
465:
466:
467: \noindent The edges are:
468: \begin{itemize}
469: \item{the {\it vertical} edges $v_{xyz}=\{V_{xyz},V_{xy(z+1)}\}$,\\
470: $z=1,\dots, k-1$,}
471: \item{the {\it width} edges $w_{xyz}=\{V_{xyz},V_{x(y+1)z}\}$,\\ $y=1,\dots, m_2-1$, }
472: \item{the {\it horizontal} edges $h_{xyz}=\{V_{xyz},V_{(x+1)yz}\}$,\\
473: $x=1,\dots, m_1-1$.}
474: \end{itemize}
475: \end{minipage}
476: \end{figure}
477:
478:
479:
480: \begin{definition}
481: \label{lin_order}
482: We now fix as in Fig. \ref{fig-orde} a {\it linear order} '$<$' on
483: the set of the vertices of $Q$ as
484: $$
485: V(11), {\bar V(12)},\dots, { V(1m_2)}, {\bar V(2m_2)}, {V(2(m_2-1))},
486: \dots, { V(m_1 m_2)}
487: $$
488: where $V(xy)= V(xy1),\dots, V(xyk)$ and ${\bar V_{xy}}$
489: denotes the reversal of $V_{xy}$.
490: \end{definition}
491:
492:
493:
494: \begin{figure}[ht]
495: \begin{minipage}{0.38\linewidth}
496: \includegraphics[width=6cm]{sign1.eps}
497: \caption{\label{fig-sign}For fixed $e$ ($h(L,e)$ is even) in a given
498: layer $L$ are marked
499: all possible $e'$ ($h(L,e')$ is odd) to be in pair with $e$.}
500: \end{minipage}
501: \begin{minipage}{0.61\linewidth}
502:
503: \begin{definition}
504: \label{R(Q)}
505: Let us now define a {\it set of pairs of edges} $R(Q)$.
506: We let $H$ be the set of all horizontal edges. For each $x\leq m_1$
507: we let $W_x$ consist of all width edges with their x-coordinate equal to $x$.
508: We call sets $H$ and $W_x$ {\it layers} of $Q$. It is convenient to depict
509: the edges of a layer as the horizontal edges of a square grid, where lower
510: left corner is the smallest in the fixed ordering of vertices. The
511: first column of the layer $H$ is ordered according to the fixed
512: ordering of vertices. Size of layer $H$ is $m_1 \times km_2$, sizes of
513: all $W_x$ are $m_2 \times k$.
514:
515: If $e$ is an edge
516: of such a layer $L$ then let $h(L,e)$ ($v(L,e)$ respectively) denote
517: the horizontal (vertical respectively) coordinate of the first vertex
518: of $e$ in $L$. For a layer $L$ we let $R(L)$ be the set of all pairs
519: $(e,e')$ of edges of $L$ such that $h(L,e')< h(L,e), v(L,e)\leq
520: v(L,e') \leq v(L,e)+1$, and $h(L,e)$ is even while $h(L,e')$ is
521: odd, example on Fig. \ref{fig-sign}. Finally we let $R(Q)$ to be the
522: union of $R(L)$ over all layers of $Q$.
523: \end{definition}
524:
525: \end{minipage}
526: \end{figure}
527:
528:
529: \section{Statement of the main results: The Dimer Problem}
530: \label{sec.statD}
531: Our result here is a simplification
532: of the main result of \cite{L1}. The reader
533: is heartily encouraged to check the original statement for a comparison.
534: Before we write the main theorems let us introduce some necessary notation.
535:
536:
537: \subsection*{Determinants}
538: \label{sub.det}
539: $Q$ is a bipartite graph, which means that its vertices may
540: be partitioned into two sets $V_1,V_2$ such that if $e$ is an edge
541: of $Q$ then $|e\cap V_1|=|e\cap V_2|=1$.
542: We have $|V_1|=|V_2|=m_1m_2k/2$. Let ${\mathcal Z}$ be the square
543: $(|V_1| \times |V_2|)$ matrix defined by ${\mathcal Z}_{ij}=0$ if
544: $\{i,j\}$ is not any edge of $Q$. If $e=\{ij\}$ is an edge of $Q$
545: ($i \in V_1$, $j \in V_2$) then ${\mathcal Z}_{ij}=\alpha^{w(ij)}$ if $i<j$ (in
546: fixed linear order) and ${\mathcal Z}_{ij}= - \alpha^{w(ij)}$ if $i>j$.
547: %${\mathcal Z}_{ij}=\alpha^{w(ij)}$ if $e=\{ij\}$
548: %is an edge of $Q$ with weight $w(e)=w(ij)$, and ${\mathcal Z}_{ij}=0$
549: %otherwise.
550:
551: We will consider matrix ${\mathcal Z}$ with its rows and columns ordered
552: in agreement with the fixed ordering
553: and we will assume that $V_{111}\in V_1$.
554: A {\it signing} of a matrix is obtained by multiplying some of
555: the entries of the matrix by $-1$.
556:
557: An orientation of a graph $G=(V,E)$ is a {\it digraph} $D=(V,A)$
558: obtained from $G$ by assigning an orientation to each edge of $G$,
559: i.e. by ordering the elements of each edge of $G$. The elements of $A$ are called {\it arcs}.
560: We say that signing $Z$ of ${\mathcal Z}$ corresponds to orientation $D$
561: of $Q$ if $Z_{ij}= -{\mathcal Z}_{ij}$ for $(ij)\in D$ and $i>j$.
562: For arc $(ij)$ of orientation $D$ of $Q$ we let
563: $\sign(D,(ij))= 1$ if $i<j$, and $\sign(D,(ij))=-1$ otherwise.
564:
565:
566: \begin{theorem}
567: \label{thm.uno}
568: The dimer partition function ${\mathcal P}(Q, \alpha)$ is
569: $$
570: {\mathcal P}(Q, \alpha)=2^{-C}\sum_D(-1)^{|\{(e,e')\in R(Q); \hspace{2mm} \sign(D,e)= \sign(D,e')=-1\}|}
571: {\rm det}(Z(D)),
572: $$
573: where the sum is over all orientations of $Q$ with all vertical edges positive,
574: and $C=km_1(m_2-1)/2 + km_2(m_1-1)/2$.
575: \end{theorem}
576:
577: \noindent Knowing the proof Theorem \ref{thm.uno} can be easily
578: rewritten as
579:
580: \begin{theorem}
581: \label{thm.dos}
582: $$
583: {\mathcal P}(Q, \alpha)=2^{C}\alpha^{w(M)}- (2^{C}+1)\E,
584: $$
585: where $M$ is the unique perfect matching of $Q$ consisting of vertical edges
586: only and $\E$ equals the average of ${\rm det}(Z(D))$ over all orientations $D$
587: of $Q$ satisfying: $|\{(e,e')\in R(Q); \hspace{2mm} \sign(D,e)=
588: \sign(D,e')=-1\}|$
589: is even (and again all vertical edges are positive).
590: \end{theorem}
591:
592: There are $2^{2C}$ orientations to be sum
593: over in Theorem \ref{thm.uno} and $2^{C-1} (2^C+1)$ terms in Theorem
594: \ref{thm.dos} to be averaged over. Therefore the exact numerical
595: analysis is probably not possible. We just tested
596: Theorem~\ref{thm.uno} for small planar lattices. However it could be
597: interesting to study statistical
598: properties of determinants ${\rm det}(Z(D))$. Let us remark here also that
599: Theorem \ref{thm.uno} for a planar graph (2D Ising model, $m_1=1$) is
600: still nontrivial, a question remains
601: how to reduce it to the Kasteleyn's one determinant. And would it be
602: possible to apply similar reduction even for the non-planar graph?
603:
604:
605:
606: \section{Statement of the main results: The Ising Problem}
607: \label{sec.stmI}
608:
609: The theory described in section \ref{sec.draw} combined with the results of \cite{S, L2}
610: yields an expression of the 3D Ising partition function as a linear
611: combination of products over aperiodic closed walks on $Q$. We get a
612: remarkably simple formula, analogous to the statement for the
613: dimer partition function.
614:
615: \subsection*{Products over aperiodic closed walks}
616: \label{sub.prod}
617: Let $G=(V,E)$ be a planar graph embedded in the plane and for each edge $e$ let $x_e$ be an
618: associate variable. Let
619: $A=(V,A(G))$ be an arbitrary orientation of $G$.
620: If $e\in E$ then $a_e$ will denote the orientation of $e$ in $A(G)$
621: and $a^{-1}$ will be the reversed directed edge to $a$. We let $x_{a_e}= x_{a_e^{-1}}=x_e$.
622: A circular sequence $p=v_1,a_1,v_2,a_2,...,a_n,(v_{n+1}=v_1)$ is called
623: {\it non-periodic closed walk} if the following
624: conditions are satisfied:
625: $a_i \in \{a_e, a_e^{-1}: e\in E\}$, $a_i\neq a_{i+1}^{-1}$ and $(a_1,...,a_n)\neq Z^m$
626: for some sequence $Z$ and $m>1$.
627: We let $X(p)=\prod_{i=1}^n x_{a_i}$.
628: We further let $\sign(p)=(-1)^{1+n(p)}$,
629: where $n(p)$ is a {\it rotation number} of $p$,
630: i.e. the number of integral revolutions of the tangent vector.
631: Finally let $W(p)=\sign(p)X(p)$.
632:
633: Now assume $p$ is a closed aperiodic walk in 3D cubic lattice $Q$. Can we define its rotation number?
634: The following construction provides a solution.
635:
636: Assume some vertices of planar graph $G$ embedded in the plane of degree $4$ are marked.
637: Then we call a walk $p$ {\it correct} if it satisfies the 'crossover condition'
638: at each marked vertex, i.e. it never enters and exits a marked vertex
639: in a pair of neighbouring edges along the vertex.
640:
641: We naturally get a planar graph from $Q$ if we consider $Q$
642: drawn as bold $Q(3,3,2)$ in Figure~\ref{cube} and put a new vertex to each edge-crossing. These new
643: vertices will form the set of the marked vertices. Clearly there is
644: a natural bijection which associates to each walk $p$ of $Q$ the corresponding correct
645: walk $p'$ in the new graph and we define rotation of $p$ to be equal to rotation of $p'$.
646: This defines $W(p)$ for each closed aperiodic walk $p$ in $Q$.
647:
648: There is a natural equivalence on non-periodic closed walks:
649: $p$ is equivalent with reversed $p$.
650: Each equivalence class has two elements and will be denoted by $[p]$.
651: We let $W([p])=W(p)$ and note that this definition is correct since
652: equivalent walks have the same sign.
653:
654: We denote by $\prod(1+W([p])$ the formal infinite product of $(1+W([p])$ over
655: all equivalence classes of non-periodic closed walks of $G$.
656:
657: Let $D$ be an orientation of $Q$. We further let
658: $$
659: \prod_D(1+W([p])= \prod(1+W([p])|_{x_e= \sign(D,e)\alpha^{w_e}}.
660: $$
661:
662: \begin{theorem}
663: \label{thm.tres}
664: Generating function of even subgraphs is
665: $$
666: {\mathcal H}(Q,\alpha)= 2^{-C}\sum_D (-1)^{|\{(e,e')\in R(Q); \hspace{2mm} \sign(D,e)= \sign(D,e')=-1\}|}
667: \prod_D(1+W([p]),
668: $$
669: where the sum is over the orientations of $Q$ with all the vertical edges positive.
670: \end{theorem}
671:
672: \noindent {\bf Remark.} Theorem \ref{thm.tres} expresses the 3D Ising partition function as a
673: linear combination of infinite products over aperiodic closed walks. For planar lattice this reduces
674: to just one infinite product (conjectured by Feynman, proved in \cite{S}) which provides
675: an equivalence of planar Ising model with the quantum field theory of free fermion. There
676: have been several attempts to generalize this equivalence for 3D. Earlier papers
677: (e.g. \cite{D}) attempt to replace closed aperiodic walks by 2D surfaces. Recent
678: promising development mentioned in the introduction deals with discrete Riemann surfaces
679: and the theta function:
680:
681: It is known that the planar Ising partition function in the thermodynamical limit
682: behaves like a 'common term' times a Riemann theta function.
683: For torus, \cite{AMV} asserts that in the thermodynamic
684: limit and near to criticality,
685: the Ising partition function behaves like a 'common term' times a linear
686: combination of four Riemann theta functions corresponding to torus.
687: In \cite{M} and \cite{CM1, CM2}, there is evidence that
688: for the critical Ising model, the dependence of the determinants of
689: adjacency matrices on the Kasteleyn orientations is exactly the same as
690: the dependence of the determinants of the Dirac operator, of the corresponding conformal
691: field theory, on the spin structures of the 2D Riemann surface; this is given
692: in terms of theta functions of half-integer characteristics.
693:
694: We believe that theorem \ref{thm.tres} may provide a new insight into these efforts.
695:
696:
697: \section{Drawing of cubic lattices}
698: \label{sec.draw}
699:
700: In this section we describe how we draw cubic lattices on 2-dimensional surfaces.
701:
702: \subsection{Theory of generalised g-graphs.}
703: \label{sub.ge}
704:
705: \begin{definition}
706: \label{def.surface}
707: A {\it surface polygonal representation} $S_g$ of an orientable 2D surface of genus $g$
708: consists of a {\it base} $B_0$ and $2g$
709: {\it bridges} $B^i_j$, $i=1,...,g$ and $j=1,2$, where
710:
711: \begin{itemize}
712: \item [i)] $B_0$ is a convex $4g$-gon with vertices
713: $a_1, ...,a_{4g}$ numbered clockwise;
714:
715: \item [ii)] $B^i_1$, $i=1,\dots,g$, is a $4$-gon
716: with vertices $x^i_1, x^i_2, x^i_3, x^i_4$ numbered clockwise. It is glued
717: with $B_0$ so that the edge $[x^i_1,x^i_2]$ of $B^i_1$ is identified with
718: the edge $[a_{4(i-1)+1},a_{4(i-1)+2}]$ of $B_0$ and the edge $[x^i_3,x^i_4]$ of
719: $B^i_1$ is identified with the edge $[a_{4(i-1)+3},a_{4(i-1)+4}]$ of $B_0$;
720:
721: \item [iii)] $B^i_2$, $i=1,\dots,g$, is a $4$-gon with vertices
722: $y^i_1, y^i_2, y^i_3, y^i_4$ numbered clockwise. It is glued
723: with $B_0$ so that the edge $[y^i_1,y^i_2]$ of $B^i_2$ is identified with
724: the edge $[a_{4(i-1)+2},a_{4(i-1)+3}]$ of $B_0$ and the edge
725: $[y^i_3,y^i_4]$ of $B^i_2$ is identified with
726: the edge $[a_{4(i-1)+4},a_{4(i-1)+5}]$ of $B_0$. Indexing is $ {\rm
727: mod}\, 4g $.
728: \end{itemize}
729: \end{definition}
730:
731: \begin{definition}
732: \label{def.embed}
733: A graph $G$ is called a {\it $g$-graph} if it is embedded on $S_g$
734: so that all the vertices belong to the base $B_0$, and
735: the embedding of each edge uses at most one bridge.
736: We denote the set of the edges embedded entirely on the base by
737: $E_0$ and the set of the edges embedded on each bridge $B^i_j$ by $E^i_j$,
738: $i=1,\dots,g,$ $j=1,2$.
739: %We also let $G_0=(V,E_0)$ and $G^i_j=(V,E_0\cup E^i_j)$.
740: \noindent
741: Moreover the following conditions need to be satisfied.
742: \begin{enumerate}
743: \item[1.]
744: the outer face of $G_0=(V,E_0)$ is a cycle,
745: and it is embedded on the boundary of $B_0$,
746: \item[2.]
747: if $e\in E^i_1$ then $e$ is embedded entirely on $B^i_1$ with
748: one end-vertex belonging to $[x^i_1,x^i_2]$ and the other one
749: to $[x^i_3,x^i_4]$. Analogously for $e\in E^i_2$.
750: \end{enumerate}
751: \end{definition}
752:
753: We need a generalization of the notion of a $g$-graph.
754:
755: \begin{definition}
756: \label{def.geng}
757: Any graph $G$ obtained by the following construction will be called
758: {\it generalized g-graph}.
759: \begin{enumerate}
760: \item[1.] Let $g=g_1+...+g_n$ be a partition of $g$ into positive integers.
761: \item[2.] Let $S_{g_i}$ be a polygonal representation of a surface of genus
762: $g_i$, $i=1,...,n$. Let us denote the basis and the bridges of $S_{g_i}$ by $B_0^i$ and
763: $B^i_{j,k}$, $i=1,...,n$, $j=1,...,g_i$ and $k=1,2$.
764: \item[3.] For $i=1,...,n$ let $H_i$ be a $g_i$-graph with the property
765: that the subgraph of $H_i$ embedded on $B^i_0$ is a cycle, embedded on the boundary
766: of $B^i_0$. Let us denote it by $C^i$.
767: \item[4.] Let $G_0$ be a $2$-connected graph properly embedded on the plane and let $F_1,...,F_n$
768: be a subset of the faces of $G_0$. Let $K^i$ be the
769: cycle bounding $F_i$, $i=1,...,n$. Let each $K^i$ be isomorphic
770: to $C^i$.
771: \item[5.] Then $G$ is obtained by gluing the $H_i$'s into $G_0$ so
772: that each $K^i$ is identified with $C_i$.
773: \end{enumerate}
774: \end{definition}
775:
776: \noindent {\bf Orientations.}
777: Let $G$ be a $g$-graph and let $G^i_j=(V,E_0\cup E^i_j)$.
778: An orientation $D_0$ of $G_0$ such that each inner face of
779: each $2$-connected component of $G_0$ is clockwise odd in $D_0$
780: is called a {\it basic orientation} of $G_0$.
781: Note that a basic orientation always exists for a planar graph.
782: Further we define the orientation $D^i_j$ of each $G^i_j$ as follows:
783: We consider $G^i_j$ embedded on the plane by the planar projection of $E^i_j$
784: outside $B_0$, and complete the basic orientation $D_0$ of $G_0$ to an orientation
785: of $G^i_j$ so that each inner face of each $2$-connected
786: component of $G^i_j$ is clockwise odd.
787: The orientation $-D^i_j$ is defined by complete reversing of the orientation $D^i_j$ of
788: $G^i_j$.
789:
790: Observe that after fixing a basic orientation $D_0$,
791: the orientation $D^i_j$ is uniquely determined for each $i,j$.
792:
793:
794: \begin{definition}
795: \label{def.r}
796: Let $G$ be a $g$-graph, $g \geq 1$. An orientation $D$ of $G$
797: which equals the {\it basic orientation} $D_0$ on $G_0$ and which equals
798: $D^i_j$ or $-D^i_j$ on $E^i_j$
799: is called {\it relevant}. We define its {\it type}
800: $r(D) \in \{+1,-1\}^{2g}$ as follows:
801: For $i =0,\dots,g-1$ and $j =1,2$, $r(D)_{2i + j}$ equals $+1$ or $-1$
802: according to the sign of $D^{i+1}_j$ in $D$.
803:
804: Moreover we let $c(r(D))$ equal the product
805: of $c_i$, $i=0,...,g-1$, where $c_i=c(r_{2i+1},r_{2i+2})$ and
806: $c(1,1)=c(1,-1)=c(-1,1)=1/2$ and $c(-1,-1)=-1/2$.
807: \end{definition}
808:
809:
810: Observe that $c(r(D))=(-1)^n 2^{-g}$, where
811: $n=|\{i;r_{2i+1}=r_{2i+2}=-1\}|$.
812:
813: For each generalized g-graph $G$ we can define $4^g$ relevant orientations
814: $D_1,...,D_{4^g}$ with respect to a fixed basic orientation of $G_0$,
815: and coefficients $c(r(D_i))$, $i=1,...,n$
816: in the same way as for a g-graph. Now we write theorem proved in \cite{GL}
817: which will be essential in proof of Theorem~\ref{thm.uno}.
818:
819: \begin{theorem}
820: \label{thm_GL}
821: Let G be a generalized g-graph with a perfect matching $M_0$ of $G_0$. Let
822: $D_0$ be a basic orientation of $G_0$. If we order the vertexes of $G$ so
823: that $s(D_0,M_0)$ is positive then
824: $$
825: {\mathcal P}(G, \alpha) = \sum_{i=1}^{4^g} c(r(D_i))
826: Pf_{G}(D_i,\alpha),
827: $$
828: where $D_i$ are all the relevant orientations of $G$.
829: \end{theorem}
830:
831: Definition of $s(D_0,M_0)$ follows from definition of {\it Pfaffian}.
832:
833: Let $G=(V,E)$ be a graph with $2n$ vertices and $D$ an orientation of $G$.
834: Denote by $A(D)$
835: the skew-symmetric matrix with the rows and the columns indexed by
836: $V$, where $a_{uv}=\alpha^{w({u,v})}$ in case $(u,v)$ is an arc of $D$,
837: $a_{u,v}= -\alpha^{w({u,v})}$ in case $(v,u)$ is an arc of $D$, and $a_{u,v}=0$
838: otherwise.
839:
840: \begin{definition}
841: \label{def_pf}
842: The {\it Pfaffian} is defined as
843: $$
844: Pf_G(D,\alpha)=\sum_{P} s^*(P)a_{i_1j_1} \cdots a_{i_nj_n},
845: $$
846: where $P=\{\{i_1j_1\},\cdots,\{i_nj_n\}\}$ is a partition of the
847: set $\{1,\dots,2n\}$ into pairs, $i_k<j_k$ for $k=1,\dots,n$, and
848: $s^*(P)$ equals the sign
849: of the permutation $i_1j_1\dots i_nj_n$ of $12\dots (2n)$.
850:
851: Each nonzero term of the expansion of the Pfaffian
852: equals $\alpha^{w(P)}$ or $-\alpha^{w(P)}$ where $P$ is a perfect
853: matching of $G$. If $s(D,P)$ denote the sign of the term
854: $\alpha^{w(P)}$ in the expansion, we may write
855: $$
856: Pf_G(D,\alpha)=\sum_{P} s(D,P)\alpha^{w(P)}.
857: $$
858: \end{definition}
859:
860:
861:
862: \subsection{Cubic lattices as generalized g-graphs.}
863:
864: In this subsection we will describe how to realize $Q(m_1,m_2,k)$ as
865: a subgraph of a generalized g-graph. Let us denote by $Q^L$ the larger
866: lattice $Q(m_1, m_2, 2k+1)$.
867: \\
868:
869: \noindent {\bf How to draw $Q^L$ on the plane.}
870: First draw the paths $V_{xy}$ along a cycle in the linear order as in
871: Def. \ref{lin_order}.
872: Next, draw the horizontal edges inside this cycle, and the width edges
873: outside of this cycle as depicted in Fig.~\ref{cube} below where $Q^L=Q(3,3,5)$
874: is properly drawn.
The figure also shows how $Q = Q(3,3,2)$ is embeded in
875: $Q^L=(3,3,5)$.
876:
877: We keep the following rule: the interiors of the curves
878: representing $h_{xyz}$ and $h_{(x+1)yz}$ ($w_{xyz}$ and $w_{x(y+1)z}$
879: respectively)
880: intersect if and only if $z$ is even. We denote by $C(e)$ the curve
881: representing edge $e$.
882:
883: \begin{figure}[ht]
884: \includegraphics[width=9cm]{cube4.eps}
885: \caption{\label{cube} Drawing of a cube $Q^L=Q(3,3,3)$ in a plane.}
886: \end{figure}
887:
888: % \lipefignocap{cube}{h}{50}
889:
890: \noindent {\bf Now we modify $Q^L$ into a generalized g-graph $Q'$.}
891: We introduce new vertices to some edge-crossings of $Q^L$ and delete
892: and subdivide some edges.
893: In this way we obtain
894: a generalised g-graph $Q'$ with an even subdivision of $Q(m_1, m_2, k)$ as its subgraph.
895: The construction is analogous
896: and independent for each plane, and so we describe it only for a plane $W_x$.
897:
898: The construction is described by Fig.~\ref{cons_Q} for the edges between $V_{x(y-1)}, V_{xy}$ and
899: between $V_{xy}, V_{x(y+1)}$, for $x$ odd and $y<m_2-1$ even.
900:
901: \begin{figure}[ht]
902: \includegraphics[width=11cm]{width.ipe}
903: \caption{\label{cons_Q} Construction of $Q'$ for part of the plane $W_x$
904: ($x$ odd), $V_{x(y-1)}, V_{xy}, V_{x(y+1)}$ ($y$ even).}
905: \end{figure}
906:
907: %\lipefignocap{width}{h}{50}
908:
909: \begin{enumerate}
910: \item For each $y$ even let $Aux_1=\{w_{xyz}; z$ odd $\}$.
911: For each edge $e$ of $Aux_1$ introduce a new vertex to each
912: intersection of $C(e)$ with the curves representing the edges
913: of $W_{x(y-1)} \cup W$, where $W=W_{x(y+1)}$ in case $y < m-1$ and
914: $W=\emptyset$ otherwise. By this operation, each $e \in Aux_1$
915: is replaced by a path. Call each edge of this path {\it auxiliary}.
916:
917: \item For each $y$ even let $Aux_2=\{w_{x(y-1)1}, w_{x(y-1)n}\}\cup A$, where
918: $A=\{w_{x,(y+1)1}, w_{x(y+1)n}\}$ in case $y < m-1$ and $A=\emptyset$ otherwise.
919: For each edge $e$ of $Aux_2$ introduce a new vertex to each
920: intersection of $C(e)$ with the curves representing the edges
921: of $W_{xy}$. Hence each $e \in Aux_2$
922: is replaced by a path. Call each edge of this path {\it auxiliary}.
923: For each $y$ even the edges $v_{xy1}, v_{xy(n-1)}$
924: and also $v_{x(y+1)1}, v_{x(y+1)(n-1)}$ will also be called {\it auxiliary}.
925: In Figure~\ref{cons_Q}, the auxiliary edges are represented by dashed lines.
926:
927: \item The edges $w_{xyz}$, $y$ even and $z$ even will be called
928: {\it relevant} for $Q$. If $y < m-1$ then the relevant edges are subdivided
929: by two vertices (added in 2.) into three edges of $Q'$. The middle one will
930: be called {\it special} and the other two {\it long}.
931: If $y=m-1$ then the relevant edge $w_{xyz}$ is subdivided by one vertex
932: into two edges of $Q'$. The one incident to $V_{xm}$
933: will be called {\it special} and the other one {\it long}.
934: If $e$ is a relevant edge of $Q$, then we choose a
935: initial long edge $f$ and we let $w(e)=w(f)$. We let the weight of the
936: special edge and of the remaining long edge be equal to 0.
937:
938: \item The edges of $W_{x(y-1)}\cup W$ also got subdivided by new
939: vertices introduced in step~1 and step~2.
940:
941: \item We delete all edges of the paths
942: obtained from $w_{x(y-1)z}$ and $w_{x(y+1)z}$,
943: $1 < z < n$ odd, as well as we delete all vertexes in intersections of those
944: edges with auxiliary edges.
945: In Figure~\ref{cons_Q}, the deleted edges are represented by dotted lines.
946:
947: \item Each edge $e \in \{w_{x(y-1)z}, w_{x(y+1)z}; z$ even $\}$,
948: is subdivided by new vertices introduced in step 1 into a path.
949: We let the weights assigned to the edges of the path equal 0 except of one
950: initial edge whose weight is let equal $w(e)$.
951: The edge $e$ of this path such that the interior of $C(e)$ does not intersect interior of
952: any curve representing a long edge will also be {\it special}.
953: The others are called {\it short}.
954:
955: \item All vertical edges which are not auxiliary (see Fig.\ref{cons_Q})
956: will be called {\it special}.
957: In Figure~\ref{cons_Q}, the special edges are represented by normal lines.
958:
959: \item
960: Let $Aux$ denote the set of all auxiliary edges. Then deletion of $Aux$ results
961: in a subdivision of $Q_{m_1m_2k}$. We subdivide some special edges (vertical and border
962: width ones) so that the graph
963: $\ti Q = Q'-Aux$ is an even subdivision of $Q_{m_1m_2k}$. All these new edges will be
964: special, and we set their weights equal~0.
965: \end{enumerate}
966:
967: This finishes the construction. In Figure~\ref{cons_Q}, the long and short edges
968: are represented by fat lines.
969:
970: {\bf $Q'$ is a generalised g-graph.} Its planar part $Q^p$ is set of all the
971: auxiliary and special edges. Other edges (i.e. the short and long) are drawn on
972: a face of $Q^p$ and they may be drawn onto a pair of bridges above this face.
973: One bridge contains one long edge, and the other bridge contains all the short
974: edges.
975:
976: Now let us set appropriate weights to all the auxiliary edges. In order to do
977: that first subdivide and add some auxiliary edges and vertices to $Q^p$
978: in such a way that there exists a matching $M_{Aux}$ of
979: $V(Q')-V(\ti Q)$ consisting of auxiliary edges, and a perfect matching $M_0$ of $Q^p$.
980: We set weights of edges from
981: $M_{Aux}$ as $w(e)=0$. Weights of other auxiliary edges $w(e)= -\infty$.
982: With such properties of $Q'$ we have an important result: ${\mathcal
983: P}(Q',\alpha)= {\mathcal P}(Q(m_1,m_2,k),\alpha)$.
984:
985: Now we introduce a {\it basic orientation} $D^p$ of $Q^p$ in order to prepare
986: our stage to use the Theorem~\ref{thm_GL} for the generalised g-graph $Q'$.
987: Orientation $D^p$ has the following properties:
988: \begin{enumerate}
989: \item[1.]
990: $D^p$ on special edges is in agreement with the natural ordering
991: (Def. \ref{lin_order}).
992: \item[2.] All the signs $s(D^p,P)$ in Pfaffian $Pf_{Q^p}(D^p,\alpha)$
993: (Def.~\ref{def_pf}) are positive.
994: \item[3.]
995: The orientation of edges on a bridge has positive sign if and only if
996: it is in agreement with the natural ordering (Def.~\ref{lin_order}).
997: \end{enumerate}
998: The construction of such a $D^p$ is possible according to Kasteleyn
999: \cite{K}.
1000:
1001: \section{The proofs}
1002:
1003: %In this chapter we prove Theorems \ref{thm.uno}-\ref{thm.tres}.
1004: %\\
1005:
1006: \subsection{\bf Proof of Theorem \ref{thm.uno}:}
1007:
1008: \noindent We use
Theorem \ref{thm_GL} for the graph $Q'$.
1009: Then we may write
1010: \begin{eqnarray}
1011: {\mathcal P}(Q{(m_1m_2k)},\alpha) &=& {\mathcal P}(Q',\alpha) =
1012: \sum_{i=1}^{4^g} c(r(D_i)) Pf_{Q'}(D_i,\alpha) = \nonumber \\
1013: &=& \sum_{i=1}^{4^g}
1014: c(r(D_i)) Pf_{\ti Q \cup M_{Aux}}(D_i,\alpha) = \sum_{i=1}^{4^g}
1015: c(r(D_i)) Pf_{Q}(D^*_i,\alpha), \nonumber
1016: \end{eqnarray}
1017: where the first equality is the result of previous section. The second
1018: equality holds because of the Theorem \ref{thm_GL}. The third equality
1019: holds because of the setting of the weights of edges in $Q'$ in the
1020: previous section. And finally, orientation $D^*_i$ being induced on
1021: $Q$ by $D_i$ on $\ti Q$, the fourth equality follows from the
1022: three facts: setting of the weights of edges, $\ti Q$ is even
1023: subdivision of $Q$, it is possible to permute vertexes covered by
1024: $M_{Aux}$ so that the sign is correct.
1025:
1026: In Section \ref{sub.det} we have defined for a bipartite graph $G$ a matrix
1027: $Z(D)$ depending on a orientation $D$. It is easy to check that for
1028: $Q$ holds $Pf_{Q}(D^*_i,\alpha)={\rm det} (Z(D^*_i))$.
1029:
1030: To proceed in proof of Theorem \ref{thm.uno} we have to establish the
1031: dependence of $c(r(D_i))$ on the orientation $D^*_i$ of $Q$ and
1032: specify the orientations we sum over in the language of $Q$ not $Q'$.
1033:
1034: Each relevant orientation $D_i$ of $Q'$ is determined by the fixed basic
1035: orientation $D^p$ of $Q^p$, and by a pair of signs for each pair of bridges.
1036: Each pair of bridges is associated with one long edge and set of short
1037: edges of $Q'$. Hence these signs
1038: may be given by specifying $(d_{D_i}^1(e),d_{D_i}^2(e)) \in \{+-\}^2$,
1039: for each long edge $e$,
1040: where $d_{D_i}^1(e)$ denotes the sign of the bridge containing $e$, and
1041: $d_{D_i}^2(e)$ denotes the sign of the other bridge containing the set
1042: of short edges.
1043:
1044: The relevant edges $w_{x(m_2-1)z}(Q{(m_1m_2k)})$
1045: and $h_{(m_1-1)yz}(Q{(m_1m_2k)})$ are associated with only
1046: one long edge of $Q'$. If $e$ is such relevant edge of $Q$,
1047: we will call it {\it border edge}, and we denote by $e_1$ the corresponding
1048: long edge. We let $d_{D_i}(e)=(d_{D_i}^1(e_1),d_{D_i}^2(e_1),+,+)$.
1049: Each relevant non-border edge $e$ of $Q$ has two long edges
1050: $e_1,e_2$ associated with it. We let
1051: $d_{D_i}(e)=(d_{D_i}^1(e_1),d_{D_i}^2(e_1),d_{D_i}^1(e_2),d_{D_i}^2(e_2)))$.
1052: A {\it relevant vector} is any element $r$ of $[\{+,-\}^4]^{\mathcal R}$
1053: such that $r(e)_3=r(e)_4= +$ for each relevant border edge $e$ of $Q$.
1054:
1055: From the definition of $c(r(D_i))$ and from properties of the
1056: generalized g-graph $Q'$ of genus $g$ follows that $c(r(D_i))=2^{-g} (-1)^n$,
1057: where $n$ is number of faces of $Q^p$ above them edges on both bridges
1058: are oriented in negative sense, i.e. against fixed ordering.
1059: There is a natural bijection between relevant orientations of $Q'$
1060: and relevant vectors.
1061: If $r$ is a relevant vector, then let $D_i(r)$ denote the corresponding
1062: relevant orientation of $Q'$ and let $\sgn(r)$ of a relevant vector $r$
1063: be calculated as follows:
1064: $\sgn(r)=2^g c(r(D_i)) =(-1)^{|\{(e,i); \quad i=0,1; \quad
1065: r(e)_{2i+1}=r(e)_{2i+2}=-1\}|}$.
1066:
1067: There is $C_r = km_1(m_2-1)/2 +km_2 (m_1 -1)/2$ relevant edges and
1068: $C_b=k(m_1+m_2)$ border edges. Genus of $Q'$ is therefore
1069: $g=2C_r-C_b$. There are $4^{2C_r-C_b}$ relevant vectors.
1070:
1071: Let us consider a set $\mathcal D$ of relevant orientations $D_i$ such that
1072: there is at least one relevant non-border edge $e$ with $r(e)_2 \neq r(e)_4$,
1073: i.e. orientation of bridges with short edges incident to $e$ are
1074: different.
1075:
1076: \begin{proposition}
1077: \label{sum_zero} Sum of contribution of all orientations from $\mathcal D$
1078: to ${\mathcal P}(Q,\alpha)$ is zero.
1079: \end{proposition}
1080: \noindent {\bf Proof:}
1081: For given orientation $D \in \mathcal D$ define
1082: a set $A_D = \{ e \quad {\rm relevant}, r(e)_2 \neq r(e)_4 \}$. Now
1083: let fix a set o relevant edges $A_f$ and consider all orientations
1084: $D_f$ such that $A_{D_f} = A_f$. Fix also one relevant edge $e_f\in
1085: A_f$.
1086:
1087: Remark: $c(r(D_1)) +c(r(D_2)) =0$ if $r(D_1)$ and $r(D_2)$ differs
1088: only in one relevant edge $e$ in such a way that $r_{D_1}(e)_2
1089: = r_{D_2}(e)_2 \neq r_{D_1}(e)_4= r_{D_2}(e)_4$ and $r_{D_1}(e)_1= -
1090: r_{D_2}(e)_1$, $r_{D_1}(e)_3 = - r_{D_2}(e)_3$. It is also that
1091: induced orientations $D^*_1 = D^*_2$.
1092:
1093: Using this remark for $e=e_f$ we have for every $A_f$:
1094: $$
1095: \sum_{D_i \in D_f} c(r(D_i)) Pf_{Q}(D^*_i,\alpha) = 0.
1096: $$
1097: The claim in Proposition \ref{sum_zero} follows directly.
1098: \done
1099: \medskip
1100:
1101: Let us call all $D_i \not\in \mathcal D$ {\it useful orientations} and
1102: from now we sum only over such orientations. There is $ 2^{3C_r-2C_b}$
1103: of useful orientations. Similarly we define useful relevant vectors.
1104:
1105: \begin{proposition}
1106: \label{all_D} Each orientation $D$ of $Q$, which agrees
1107: with the fixed order on vertical edges, corresponds to $2^{C_r-C_b}$ of
1108: useful orientations $D_i$ of $Q'$ and all those $D_i$ have the same
1109: sign $c(r(D_i))$.
1110: \end{proposition}
1111: \noindent {\bf Proof:} Let $r$ be a useful vector. Then $D(r)$
1112: determines uniquely $r(e)_2$ and $r(e)_4$ for each relevant edge $e$
1113: and also $r(f)_1$ for each relevant border edge $f$. Hence $D(r)$
1114: determines uniquely $r(f)$ for each relevant border edge
1115: $f$. Moreover $D(r)$ determines uniquely the product $r(e)_1\times
1116: r(e)_3$ for each relevant non-border edge $e$. Since there are
1117: $C_r-C_b$ relevant non-border edges, there is $2^{C_r-C_b}$
1118: orientations $D_i$ corresponding to it.
1119:
1120: Let $r,s$ be useful and both lead to the same $D$ of $Q$. Then
1121: $r(e)_2=s(e)_2=s(e)_4=r(e)_4$ for each relevant non-border edge $e$
1122: and $r(e)_2=s(e)_2$ and $s(e)_1=r(e)_1$ for each relevant border
1123: edge. This implies that $\sgn(r)=\sgn(s)$.
1124: \done
1125: \medskip
1126:
1127: Concluding from what we did till now we can write the dimer partition
1128: function as
1129: $$
1130: {\mathcal P}(Q,\alpha) = 2^{-C_r} \sum_{D} \sgn(D) {\rm det}(Z(D)),
1131: $$
1132: where the sum is over all orientations of $Q$ with all vertical edges
1133: positive. At this point the only thing missing in proof of Theorem
1134: \ref{thm.uno} is to concretize $\sgn(D)$.
1135: \begin{proposition}
1136: \label{sgn_D} $\sgn(D)=(-1)^{|\{(e,e')\in R(Q); \hspace{2mm}
1137: \sign(D,e)= \sign(D,e')=-1\}|}$, where the set of edges $R(Q)$ is
1138: defined in \ref{R(Q)}.
1139: \end{proposition}
1140: \noindent {\bf Proof:} Let us remind that $\sgn(D)=(-1)^n$ where $n$
1141: is number of faces with both bridges above them negative. Each
1142: non-border relevant edge $e$ corresponds to two faces with
1143: bridges. The contribution of the two corresponding faces to sign
1144: could be negative only if orientation of $e$ is negative
1145: ($r(e)_1\neq r(e)_3 $). And moreover if orientation of
1146: corresponding shot edges is negative ($r(e)_2= r(e)_4 = -1 $).
1147: It follows from the construction that this happen only if product
1148: of orientation of all $e'$ (as defined in \ref{R(Q)} ) is
1149: negative.
1150: Considering all the relevant edges we get the claim of Proposition
1151: \ref{sgn_D}
1152: \done \medskip
1153:
1154:
1155:
1156: \subsection{\bf Proof of Theorem \ref{thm.dos}:}
1157:
1158: \begin{proposition}
1159: \label{av}
1160: The average of ${\rm det}(Z(D))$ over all orientations of $Q$ with all
1161: vertical edges fixed positive equals $\alpha^{w(M)}$, where $M$ is
1162: perfect matching of $Q$ consisting only from vertical edges.
1163: \end{proposition}
1164: \noindent {\bf Proof:}
1165: By the linearity of expectation % and the definition of orientations $D$,
1166: the contribution of other than vertical edges cancel out when we calculate
1167: the average of ${\rm det}(Z(D))$. Since $Q{(m_1m_2k)}$ has exactly one
1168: perfect matching consisting of vertical edges only, Proposition
1169: \ref{av} follows.
1170: \done \medskip
1171:
1172: There is $2^{2C}$ orientation D in \ref{thm.uno} to be summed over.
1173: And $2^{C-1}(2^C+1)$ of them have positive sign. Proof: Looking at
1174: Figure \ref{fig-sign} we can see that as soon as there is at least one
1175: relevant (shaded) edge negative there is so many orientation with
1176: positive as with negative sign. Only if all the relevant edges are
1177: positive the sign is also positive. It means there is $2^C$ more
1178: positive orientation then negative.
1179:
1180: By Theorem \ref{thm.uno} and Proposition \ref{av} we have then
1181: $$
1182: {\mathcal P}(Q,\alpha)=2^{-C}[-2^{2C} \alpha^{w(M)} +2\mathcal
1183: E (2^{C-1}(2^{C}+1))]= -2^{C} \alpha^{w(M)} + \mathcal
1184: E (2^{C}+1),
1185: $$
1186: where $\mathcal E$ equals the average of ${\rm det}(Z(D))$ over all $D$
1187: of $Q$ with positive sign. And Theorem \ref{thm.dos} is proved.
1188: \done \medskip
1189:
1190:
1191:
1192: \subsection{\bf Proof of Theorem \ref{thm.tres}:}
1193:
1194: We use a straightforward reformulation of a theorem of \cite{L2}:
1195:
1196: \begin{theorem}
1197: \label{thm.prd}
1198: Let G be a generalized g-graph. Then
1199: $$
1200: {\mathcal H}(G, \alpha) = \sum_{i=1}^{4^g} c(r(D_i))
1201: \prod_{D_i}(1+W([p]) ,
1202: $$
1203: where $D_i$ are all the relevant orientations of $G$, and rotation of $p$
1204: and $\prod_{D_i}(1+W([p])$ are defined in the same way as described for $Q$
1205: in section \ref{sub.prod}.
1206: \end{theorem}
1207:
1208: \noindent Proof of this theorem could be found in \cite{L2} and for planar graphs in \cite{S}.
1209:
1210: Let $D_i, i=1,\dots, 4^g$ be the relevant orientations of generalised g-graph $Q'$.
1211: If we apply theorem \ref{thm.prd} to $Q'$, from the choice of weights we get
1212: $$
1213: {\mathcal H}(Q, \alpha) = \sum_{i=1}^{4^g} c(r(D_i))
1214: \prod_{D^*_i}(1+W([p]).
1215: $$
1216: The same analysis as in the previous section finishes the proof of Theorem \ref{thm.tres}.
1217:
1218: \section{Conclusions}
1219: \label{sec.con}
1220:
1221: We express the finite 3D Dimer partition function as a linear combination of determinants
1222: of oriented adjacency matrices in Theorem \ref{thm.uno}, and the finite 3D Ising partition
1223: function as a linear combination of products over aperiodic closed walks in Theorem \ref{thm.tres}.
1224:
1225: The methodology we use is embedding of cubic lattice on 2D surfaces of large genus. By this embedding
1226: cubic lattice is transformed to a generalized g-graph. Then we use theorems from \cite{GL, L2}, which
1227: generalize approach of \cite{K}.
1228:
1229: We believe this approach may relate to recent developments in conformal field theory
1230: \cite{M,CM1,CM2} and topological string theory \cite{ORV}, and ultimately teach us something
1231: about 3D basic statistical physics problems.
1232:
1233:
1234: \medskip
1235: \ifx\undefined\bysame
1236: \newcommand{\bysame}{\leavevmode\hbox
1237: to3em{\hrulefill}\,}
1238: \fi
1239: \begin{thebibliography}{[EMSS]}
1240:
1241: \bibitem[ORV]{ORV} A. Okounkov, N. Reshetikhin, C. Vafa,
1242: {\em Quantum Calabi-Yau and Classical Crystals},
1243: hep-th/0309208
1244:
1245: \bibitem[AMV]{AMV} Alvarez-Gaume L., Moore G., Vafa C.,
1246: {\em Theta-functions, Modular invariance, and Strings},
1247: Commun. Math. Phys. 106 (1): 1-40 1986.
1248:
1249: \bibitem[CM1]{CM1} Costa-Santos R., McCoy M.B.,
1250: {\em Dimers and the Critical Ising Model on lattices of genus
1251: $> 1$},
1252: Nuclear Physics B 623 (3): 439-473 FEB 25 2002.
1253:
1254: \bibitem[CM2]{CM2} R. Costa-Santos, B.M. McCoy,
1255: {\em Finite size corrections for the Ising model on higher
1256: genus triangular lattices},
1257: Journal of Statistical Physics 112 (5-6): 889-920 SEP 2003.
1258:
1259: \bibitem[M]{M} Mercat C.,
1260: {\em Discrete Riemann surfaces and the Ising model},
1261: Commun. Math. Phys. 218 (1): 177-216 APR 2001.
1262:
1263:
1264: \bibitem[S]{S} S. Sherman,
1265: {\em Combinatorial Aspects of the Ising Model of Ferromagnetism I},
1266: J. Math. Phys. 1, 1960.
1267:
1268: \bibitem[D]{D} Distler J.,
1269: {\em A note on the 3D Ising model as a string theory},
1270: Nucl. Phys.B 388 (3): 648-670 DEC 28 1992.
1271:
1272: \bibitem[L1]{L1} Loebl M.,
1273: {\em On the Dimer Problem and the Ising Problem in Finite
1274: 3-dimensional Lattices},
1275: Electr. J. Comb. R30 9 (1): 2002.
1276:
1277: \bibitem[L2]{L2} Loebl M.,
1278: {\em A Discrete Non-Pfaffian Approach to the Ising Problem},
1279: DIMACS, Series in Discrete Math and Theor. Comp. Science, 63 2004.
1280:
1281: \bibitem[GL]{GL} A. Galluccio, M. Loebl,
1282: {\em A Theory of Pfaffian Orientations I,II},
1283: Electr. J. Comb. 6 (1): 1999.
1284:
1285:
1286: \bibitem[K]{K} P. W. Kasteleyn,
1287: {\em The statistics of dimers on a lattice},
1288: Physics 27: 1209-1225, 1961.
1289:
1290: \end{thebibliography}
1291: \end{document}
1292: \endinput
1293: