1: \documentclass[12pt]{amsart}
2: \usepackage{amsmath}
3: \usepackage{amsfonts}
4: \usepackage{amssymb}
5: \usepackage{graphics}
6: \usepackage{epstopdf}
7: \usepackage{epsf}
8: \usepackage{epsfig}
9: \usepackage{url}
10:
11: \textwidth 14cm
12: \textheight 21cm
13: \oddsidemargin=1.2cm
14: \evensidemargin=1.2cm
15:
16: \newtheorem{theorem}{Theorem}
17: \newtheorem{lemma}[theorem]{Lemma}
18: \newtheorem{claim}[theorem]{Claim}
19: \newtheorem{problem}[theorem]{Problem}
20: \newtheorem{question}[theorem]{Question}
21: \newtheorem{corollary}[theorem]{Corollary}
22: \newtheorem{proposition}[theorem]{Proposition}
23: \newtheorem{assumption}[theorem]{Assumption}
24: \newtheorem{conjecture}[theorem]{Conjecture}
25: \newtheorem{algorithm}[theorem]{Algorithm}
26: \newtheorem{property}[theorem]{Property}
27: \newtheorem{exercise}[theorem]{Exercise}
28: \theoremstyle{definition}
29: \newtheorem{definition}[theorem]{Definition}
30: \newtheorem{example}[theorem]{Example}
31: \newtheorem{remark}[theorem]{Remark}
32:
33: \newcommand{\CC}{\mathbb C}
34: \newcommand{\RR}{\mathbb R}
35: \newcommand{\QQ}{\mathbb Q}
36: \newcommand{\ZZ}{\mathbb Z}
37: \newcommand{\NN}{\mathbb N}
38: \newcommand{\PP}{\mathbb P}
39: \newcommand{\N}{\mathcal N}
40:
41: \begin{document}
42:
43: \title[The Hyperdeterminant and Triangulations of the 4-Cube]{The
44: Hyperdeterminant and \\ Triangulations of the 4-Cube}
45:
46: \author[Peter Huggins, Bernd Sturmfels, Josephine Yu,
47: and Debbie Yuster]{Peter Huggins,
48: Bernd Sturmfels, \\ Josephine Yu
49: and Debbie S. Yuster}
50:
51: \begin{abstract}
52: The hyperdeterminant of format $2\times 2 \times 2 \times 2$
53: is a polynomial of degree $24$ in $16$ unknowns which has
54: $ 2894276$ terms. We compute the Newton polytope of this
55: polynomial and the secondary polytope of the $4$-cube.
56: The $87959448 $ regular triangulations
57: of the $4$-cube are classified
58: into $25448$ $D$-equivalence classes,
59: one for each vertex of the Newton polytope.
60: The $4$-cube has $80876$ coarsest regular subdivisions,
61: one for each facet of the secondary polytope, but
62: only $268$ of them come from the hyperdeterminant.
63:
64:
65: \end{abstract}
66: \maketitle
67: \section{Introduction}
68:
69: This article makes three contributions to the mathematics of computation:
70: \begin{itemize}
71: \item solution to a computational challenge problem posed by I.M.~Gel'fand,
72: \item new theorems in algebra (hyperdeterminant) and
73: geometry ($4$-cube),
74: \item new computational methodology
75: (algorithms and software) for large polynomials
76: and large convex polytopes in the presence of symmetry.
77: \end{itemize}
78:
79: \smallskip
80:
81: We begin our discussion with a small familiar polynomial in four unknowns:
82: $$ D_{22} \quad = \quad c_{00} c_{11} - c_{01} c_{10}. $$
83: This irreducible quadratic polynomial is the determinant of a $2 \times 2$-matrix $(c_{ij})$.
84: Next, the {\em hyperdeterminant} of a $2 \times 2 \times 2$-tensor is the following irreducible
85: polynomial in eight unknowns which is the sum of
86: twelve terms of degree four:
87: \begin{eqnarray*}
88: D_{222} \quad =
89: & \quad 4 c_{000}c_{011}c_{101}c_{110}+ 4c_{001}c_{010}c_{100}c_{111} \\
90: &+ c_{000}^2c_{111}^2 + c_{001}^2c_{110}^2 + c_{010}^2c_{101}^2 +
91: c_{011}^2c_{100}^2
92: \\
93: & - 2 c_{000}c_{001}c_{110}c_{111}
94: -2 c_{000}c_{010}c_{101}c_{111}
95: -2 c_{000}c_{011}c_{100}c_{111} \\
96: & -2 c_{001}c_{010}c_{101}c_{110}
97: -2 c_{001}c_{011}c_{110}c_{100}
98: -2 c_{010}c_{011}c_{101}c_{100}.
99: \end{eqnarray*}
100: This polynomial is known as the {\em tangle} in the
101: physics literature.
102: Hyperdeterminants of arbitrary format were
103: studied by Gel'fand, Kapranov, and Zelevinsky
104: \cite{GKZ}. They have numerous applications ranging
105: from quantum information theory \cite{Lev, LT} to
106: computational biology \cite{AR, SJ} and numerical analysis \cite{DSL, tBe}.
107: Basic definitions concerning hyperdeterminants will be reviewed in Section 3.
108:
109: The starting point of this paper is a problem stated by
110: I.~M.~Gel'fand in his Fall 2005 research seminar at Rutgers University:
111: find the monomial expansion of the
112: hyperdeterminant $D_{2222}$ of a $2 \times 2 \times 2 \times 2$-tensor.
113: Back in 1852, Schl\"afli
114: had given a nested formula (see \cite[\S 14.4]{GKZ}) for $D_{2222}$, which
115: is a polynomial of degree $24$ in $16$ unknowns
116: $\,c_{0000}, c_{0001}, \ldots, c_{1111}$, and, 150 years later,
117: Luque and Thibon \cite{LT} expressed
118: $D_{2222}$ in terms of the fundamental tensor invariants.
119: Gel'fand's challenge to the computer algebra community was to
120: expand these formulas into monomials.
121: We solved Gel'fand's problem:
122:
123:
124: \begin{theorem}
125: The hyperdeterminant $D_{2222}$ is the sum of
126: $ 2894276$ monomials in $9617$ orbits.
127: The Newton polytope of $D_{2222}$ is an $11$-dimensional polytope with
128: $268$ facets in $8$ orbits, and
129: $25448$ vertices in $111$ orbits. It contains
130: the exponent vectors of $20992$ monomials in $69$ orbits
131: which do not appear in $D_{2222}$.
132: \end{theorem}
133:
134: In this theorem, the term {\em orbits} refers to the full
135: symmetry group $B_4$ of the $4$-dimensional cube, which
136: has order $384$, and the {\em Newton polytope}
137: is the convex hull in $\RR^{16}$ of the exponent vectors
138: of all monomials appearing in $D_{2222}$.
139:
140: Our computational proof of Theorem~1 is explained in Sections
141: 3 and 4. We list representatives for each orbit
142: of facets, vertices, and ``missing monomials,'' and
143: we discuss key properties found in these data.
144: Complete data are available at our
145: supplementary materials website \
146: \url{bio.math.berkeley.edu/4cube/}.
147:
148:
149:
150: According to the general theory of Gel'fand, Kapranov and Zelevinsky
151: \cite[\S 11]{GKZ}, there is a beautiful and deep relationship between the
152: monomials in the $2 \times 2 \times \cdots \times 2$-hyperdeterminant
153: and the triangulations of the $n$-dimensional cube. In the language of
154: polyhedral geometry, this relationship can be stated succinctly as follows:
155: the {\em secondary polytope} of the $n$-cube equals the Minkowski sum
156: of the Newton polytopes of the hyperdeterminants corresponding to all the faces of the $n$-cube, up to dimension $n$.
157: Readers unfamiliar with these geometric concepts will find an elementary
158: and self-contained introduction in Section 2, where we
159: explain the correspondence between the twelve monomials in
160: $D_{222}$ and the $74$ triangulations of the $3$-cube.
161:
162: Theorem 1 opened up the possibility of determining the
163: same correspondence for the $4$-cube. However, the
164: challenges were formidable, in light of the size of the secondary polytope of the 4-cube:
165:
166: \begin{theorem}
167: The $4$-cube has $87959448 $ regular triangulations
168: in $235277$ orbits. The secondary polytope of the
169: $4$-cube has $80876$ facets in $334$ orbits.
170: \end{theorem}
171:
172: The first statement in Theorem 2 is a known result
173: due to Pfeifle and Rambau \cite{PR}, who had
174: computed all regular triangulations of the 4-cube
175: using Rambau's software {\tt TOPCOM} \cite{Ram}.
176: We independently verified their enumeration.
177: The second statement in Theorem 2 is one of the
178: contributions of this paper.
179:
180: Equipped with Theorems 1 and 2, it took us two more months
181: of serious computational work to obtain what we consider to be
182: the main mathematical result of this paper: the extension of the material
183: in Section 2 from dimension three to dimension four.
184: Each regular triangulation of the $4$-cube maps
185: to a monomial in $D_{2222}$, and the fibers of this
186: map are the {\em $D$-equivalence classes} of \cite{GKZ}.
187: We found that, up to symmetry, there are $111$ $D$-equivalence
188: classes. They are discussed in
189: Section~6. Section~7 studies the $80876$ facets in Theorem 2,
190: and it classifies the coarsest regular polyhedral subdivisions of the $4$-cube.
191: Most of the algorithms and softwares we developed and used here can easily be generalized and applied to other polynomials and polytopes with symmetry.
192:
193: Section~5 places our computations into a larger mathematical
194: context. The hyperdeterminant $D_{2222}$ is
195: the {\em $A$-discriminant} for the matrix of the $4$-cube
196: $$
197: A \quad = \quad
198: \left[
199: \begin{array}{cccccccccccccccc}
200: 1&1&1&1&1&1&1&1&1&1&1&1&1&1&1&1\\
201: 0&0&0&0&0&0&0&0&1&1&1&1&1&1&1&1\\
202: 0&0&0&0&1&1&1&1&0&0&0&0&1&1&1&1\\
203: 0&0&1&1&0&0&1&1&0&0&1&1&0&0&1&1\\
204: 0&1&0&1&0&1&0&1&0&1&0&1&0&1&0&1
205: \end{array}
206: \right].
207: $$
208: Dickenstein, Feichtner and Sturmfels \cite{DFS}
209: recently gave a recipe for computing the
210: Newton polytope of the $A$-discriminant for any
211: point configuration $A$. Our computations
212: are complementary to their tropical approach.
213: We represent subdivisions of the cube $A$
214: by their dual polyhedral complexes, here called
215: {\em tight spans} as in \cite{Hir}.
216: The related cases
217: when $A$ is a second hypersimplex
218: or a product of two simplices correspond to
219: finite metrics \cite{SY} and to tropical convexity \cite{DS}
220: respectively. We also study the irreducible factorizations of all
221: leading forms of $D_{2222}$ that are supported on
222: facets of the Newton polytope.
223:
224:
225: \section{The secondary polytope of the $3$-cube}
226:
227: As a warm-up for our study of the
228: $4$-cube, we first discuss our primary
229: objects of
230: interest for the $3$-cube. Recall that the {\em Newton polytope} $\mathcal{N}(G)$ of a
231: polynomial $G$ is the convex hull
232: of the exponent vectors of the monomials which appear in the
233: expansion of $G$. The Newton polytope $\mathcal{N}(D_{222})$
234: of the hyperdeterminant $D_{222}$ is the convex hull
235: in $\RR^8$ of the six rows of the matrix
236: $$ \bordermatrix{
237: & x_{000} & x_{001} & x_{010} & x_{011} & x_{100} & x_{101} & x_{110} &
238: x_{111} \cr
239: c_{000} c_{011} c_{101} c_{110} & 1 & 0 & 0 & 1 & 0 & 1 & 1 & 0 \cr
240: c_{001} c_{010} c_{100} c_{111} & 0 & 1 & 1 & 0 & 1 & 0 & 0 & 1 \cr
241: c_{000}^2c_{111}^2 & 2 & 0 & 0 & 0 & 0 & 0 & 0 & 2 \cr
242: c_{001}^2c_{110}^2 & 0 & 2 & 0 & 0 & 0 & 0 & 2 & 0 \cr
243: c_{010}^2c_{101}^2 & 0 & 0 & 2 & 0 & 0 & 2 & 0 & 0 \cr
244: c_{011}^2c_{100}^2 & 0 & 0 & 0 & 2 & 2 & 0 & 0 & 0 \cr
245: }.
246: $$
247: These six monomials labeling the rows are the {\em extreme monomials} of $D_{222}$, which means that their exponent vectors are vertices of
248: the Newton polytope $\mathcal{N}(D_{222})$.
249: The other six monomials in $D_{222}$ are not extreme monomials,
250: since their exponent vectors lie in the relative interiors
251: of edges of the four-dimensional polytope
252: $\mathcal{N}(D_{222})$.
253: Combinatorially, $\mathcal{N}(D_{222})$
254: is a bipyramid over a tetrahedron. Its {\em $f$-vector}
255: records the number of faces of dimensions $0$, $1$, $2$, and $3$
256: respectively:
257: \begin{equation}
258: \label{smallfvector}
259: f\bigl(\mathcal{N}(D_{222})) \quad = \quad (6,14,16,8).
260: \end{equation}
261: The Newton polytope of the $2 \times 2 \times 2$-hyperdeterminant has
262: the following irredundant
263: presentation by linear equations and inequalities:
264: \begin{eqnarray*}
265: & \mathcal{N}(D_{222}) \,\, = \,\,
266: \{(x_{000},x_{001}, \ldots, x_{111} )\in\mathbb{R}^8 \,:\, \qquad
267: \qquad \qquad \qquad \qquad \\
268: & x_{000}, x_{001}, x_{010}, x_{011}, x_{100}, x_{101}, x_{110}, x_{111}
269: \geq 0 , \\
270: & \! x_{000} + x_{001} + x_{010} + x_{011} \,=\,
271: x_{000} + x_{001} + x_{100} + x_{101} \,=\, 2 ,\\
272: & \, \, x_{000} + x_{010} + x_{100} + x_{110} \,=\,
273: x_{001} + x_{011} + x_{101} + x_{111} \,=\, 2 \,\,
274: \}.
275: \end{eqnarray*}
276: In Section 4, we determine the analogous presentation for $\mathcal{N}(D_{2222})$.
277:
278: We consider the {\em principal determinant}
279: of the $3$-cube.
280: By~\cite[Thm.~10.B.1.2]{GKZ}, this is the
281: following product of determinants associated to
282: faces and vertices:
283: \begin{eqnarray*}
284: E_{222} \,\,\, = & D_{222} \cdot
285: (c_{000} c_{011} - c_{001} c_{010}) \cdot
286: (c_{000} c_{101} - c_{001} c_{100}) \cdot \\
287: & (c_{000} c_{110} - c_{010} c_{100}) \cdot
288: (c_{001} c_{111} - c_{011} c_{101}) \cdot \\
289: & (c_{010} c_{111} - c_{011} c_{110}) \cdot
290: (c_{100} c_{111} - c_{101} c_{110}) \cdot \\
291: & c_{000} \cdot c_{001} \cdot c_{010} \cdot c_{011} \cdot
292: c_{100} \cdot c_{101} \cdot c_{110} \cdot c_{111}.
293: \end{eqnarray*}
294: The expansion of this polynomial of degree $24$ has
295: $231$ monomials, of which $74$ are extreme monomials. The symmetry group
296: of the $3$-cube, which is the Weyl group $B_3$ of order 48, acts on these $231$ monomials. The
297: $74$ extreme monomials come in six orbits:
298: $$ \begin{matrix}
299: {\rm Type} & {\rm Monomial} & \hbox{GKZ Vector} &
300: \hbox{Orbit Size} \\
301: 1 & -4 c_{000} c_{001}^5 c_{010}^5 c_{011} c_{100}^5 c_{101} c_{110}
302: c_{111}^5 &
303: (1\, 5\, 5\, 1\, 5\, 1\, 1\, 5) & 2 \\
304: 2 &
305: -c_{000} c_{001}^4 c_{010}^4 c_{011}^3 c_{100}^6 c_{101} c_{110} c_{111}^4 &
306: (1 \, 4\, 4\, 3\, 6\, 1\, 1\, 4 ) & 8 \\
307: 3 & c_{000} c_{001}^3 c_{010}^4 c_{011}^4 c_{100}^6 c_{101}^2 c_{110}
308: c_{111}^3 &
309: (1 \, 3 \, 4 \, 4 \, 6 \, 2 \, 1 \, 3) & 24 \\
310: 4 & c_{000} c_{001}^3 c_{010}^3 c_{011}^5 c_{100}^5 c_{101}^3 c_{110}^3
311: c_{111} &
312: (1 \, 3 \, 3 \, 5 \, 5 \, 3 \, 3 \, 1) & 12 \\
313: 5 & -c_{000} c_{001}^3 c_{010}^3 c_{011}^5 c_{100}^6 c_{101}^2 c_{110}^2
314: c_{111}^2 &
315: (1 \, 3 \, 3 \, 5 \, 6 \, 2 \, 2 \, 2) & 24 \\
316: 6 & c_{000}^2 c_{001}^2 c_{010}^2 c_{011}^6 c_{100}^6 c_{101}^2 c_{110}^2
317: c_{111}^2 &
318: (2 \, 2 \, 2 \, 6 \, 6\, 2 \, 2 \, 2) & 4
319: \end{matrix}
320: $$
321: The Newton polytope $\mathcal{N}(E_{222})$ of the principal
322: determinant $E_{222}$ is the
323: {\em secondary polytope} of the $3$-cube.
324: It is $4$-dimensional and has the $f$-vector $(74,152,$ $100, 22)$.
325: A Schlegel diagram of this polytope
326: was posted by Julian Pfeifle at
327: \begin{verbatim}
328: www.eg-models.de/models/Discrete_Mathematics/
329: Polytopes/Secondary_Polytopes/2000.09.031/.
330: \end{verbatim}
331:
332: Using {\tt POLYMAKE} \cite{GJ}, we found that
333: the secondary polytope of the $3$-cube has the following irredundant
334: presentation by linear equations and inequalities:
335: \begin{equation}
336: \begin{array}{cc}
337: & \mathcal{N}(E_{222}) \,\, = \,\,
338: \bigl\{ (x_{000},x_{001}, \ldots, x_{111} )\in\mathbb{R}^8 \,:\, \qquad
339: \qquad \qquad \qquad \qquad \\
340: & x_{000} + x_{001} + x_{010} + x_{011} \,=\,
341: x_{000} + x_{001} + x_{100} + x_{101} \,=\, 12 ,\\
342: & x_{000} + x_{010} + x_{100} + x_{110} \,=\,
343: x_{001} + x_{011} + x_{101} + x_{111} \,=\, 12 ,\\
344: & 1 \, \leq \, x_{000}, \,x_{001}, \,x_{010}, \,x_{011}, \,x_{100},
345: \,x_{101},
346: \,x_{110}, \,x_{111}
347: \,\leq \,6, \\ &
348: x_{000} + x_{001} \geq 4,\,
349: x_{000} + x_{010} \geq 4,\,
350: x_{000} + x_{100} \geq 4 ,\,\\ &
351: x_{001} + x_{011} \geq 4 ,\,
352: x_{010} + x_{011} \geq 4,\,
353: x_{001} + x_{101} \geq 4 \,\bigr\}.
354: \end{array}
355: \label{NE222ineq}
356: \end{equation}
357:
358: According to the theory of Gel'fand, Kapranov, and Zelevinsky \cite{GKZ},
359: the $74$ vertices of the secondary polytope $\mathcal{N}(E_{222})$
360: are in bjiection with the {\em regular triangulations} of the $3$-cube.
361: For the $3$-cube all triangulations are
362: regular, so the $3$-cube has $74$ triangulations which come in $6$ orbits.
363: The vertex of $\mathcal{N}(E_{222})$ corresponding to
364: a given triangulation $\Pi$ is called the {\em GKZ vector} of $\Pi$.
365: The $i$th coordinate of the GKZ vector of $\Pi$ is the sum of the
366: normalized volumes of all tetrahedra in $\Pi$
367: which contain the $i$th vertex of the cube.
368: For a general introduction to triangulations, their
369: GKZ vectors, and secondary polytopes,
370: see the forthcoming book by De Loera, Rambau and Santos \cite{DRS}.
371: The six types of triangulations of the $3$-cube are
372: depicted in \cite[Figure 1.38]{DRS} and \cite[Figure 3]{DeL}. Their tight spans, which are polyhedral complexes dual to the triangulations, are shown in Figure 1 below.
373: Here is a detailed description:
374:
375:
376: \smallskip
377:
378: \noindent {\bf Type 1:} These two triangulations divide the cube into five
379: tetrahedra. There is
380: one central tetrahedron of normalized volume two, and four
381: unimodular tetrahedra which border the central tetrahedron.
382:
383: \smallskip
384:
385: The remaining five types of triangulations divide the 3-cube into
386: six unimodular tetrahedra. Each of these triangulations
387: uses a diagonal of the cube.
388:
389: \noindent {\bf Type 2:}
390: We get these $8$ triangulations by slicing off the
391: three vertices adjacent to a fixed vertex.
392: The remaining bipyramid is cut into three tetrahedra.
393:
394: \noindent {\bf Type 3:}
395: These $24$ triangulations also use a diagonal. Of the
396: other six vertices we pick two that are diagonal on a facet, and we
397: slice them off.
398:
399: \noindent {\bf Type 4:}
400: These twelve triangulations are indexed by ordered pairs of diagonals.
401: The end points of the first diagonal are sliced off, and the
402: remaining octahedron is triangulated using the second diagonal.
403:
404: \noindent {\bf Type 5:}
405: These $24$ triangulations are indexed by a diagonal and one other vertex.
406: That vertex is sliced off, and the remaining polytope is divided into
407: a pentagonal ring of tetrahedra around the diagonal.
408:
409: \noindent {\bf Type 6:}
410: These four triangulations are indexed by the diagonals.
411: The cube is divided into a hexagonal ring of tetrahedra around the diagonal.
412:
413: \smallskip
414:
415: We depict each triangulation of the $3$-cube
416: by its corresponding tight span, which is
417: the planar graph dual to the triangulation.
418: Each vertex of the tight span corresponds to
419: a tetrahedron in the triangulation, and two
420: vertices are connected by an edge if the
421: corresponding tetrahedra share a triangle.
422: Regions of the tight span
423: correspond to interior edges of the triangulation.
424: The six types of tight spans are shown in Figure~1. Compare it to Figure 2 in \cite{San}.
425:
426: \begin{figure}
427: \includegraphics[scale=0.65]{all3cubeTspans.eps}
428: %\vskip 5cm
429: \caption{The tight spans dual to the six types of triangulations of the
430: $3$-cube.
431: Each tight span represents an orbit of extreme monomials of the principal
432: determinant $\,E_{222}$.}
433: \label{cubetriangs}
434: \end{figure}
435:
436: The $22$ facets of $\mathcal{N}(E_{222})$ correspond to
437: proper subdivisions of the $3$-cube which are as coarse as possible.
438: The eight facet inequalities
439: like $x_{000} \geq 1$ correspond to
440: slicing off one vertex. The eight inequalities
441: like $x_{000} \leq 6$ correspond to subdividing
442: the $3$-cube into three pyramids whose bases
443: are the square facets disjoint from one vertex.
444: The six inequalities like $x_{000} + x_{001} \geq 4$
445: correspond to subdividing the $3$-cube into two triangular prisms.
446:
447: Since the hyperdeterminant $D_{222}$ is a factor of the principal
448: determinant $E_{222}$, its Newton polytope
449: $\mathcal{N}(D_{222})$ is a {\em Minkowski summand}
450: of the secondary polytope $\mathcal{N}(E_{222})$.
451: This implies the existence of a natural map from the
452: $74$ vertices of $\mathcal{N}(E_{222})$
453: onto the six vertices of $\mathcal{N}(D_{222})$.
454: Equivalently, we have a map from the
455: regular triangulations of the $3$-cube to
456: the extreme monomials of $D_{222}$.
457: A formula for this map can be derived
458: from Theorem 3.2 in \cite[\S 11.A]{GKZ}.
459:
460: The extreme monomials of $D_{222}$ come in two $B_3$-orbits. We now
461: describe the corresponding
462: two orbits of {\em $D$-equivalence classes}. Each class consists of all
463: triangulations which are mapped to a fixed
464: extreme monomial of $D_{222}$.
465: The $D$-equivalence class of the monomial $c_{001} c_{010} c_{100} c_{111} $
466: consists of only one triangulation, namely the triangulation
467: of type 1 with GKZ vector $(1,5,5,1,5,1,1,5)$.
468: The $D$-equivalence class of $c_{000}^2 c_{111}^2$ consists
469: of all $18$ triangulations which use the diagonal
470: from $(0,0,0)$ to $(1,1,1)$. Thus the number of triangulations
471: of the $3$-cube decomposes as follows into the sizes of the
472: $D$-equivalence classes:
473: \begin{equation}
474: \label{seventyfour} 74 \quad = \quad 2 \cdot 1 + 4 \cdot 18 \quad = \quad
475: 1 + 1 + 18 + 18 + 18 + 18.
476: \end{equation}
477:
478: We emphasize that all of these results on the 3-cube are
479: easy and well-known. This section served a purely expository purpose,
480: namely, to set the stage for the new results on the $4$-cube
481: to be presented in the later sections. For instance,
482: Table \ref{decompose} generalizes the identity (\ref{seventyfour})
483: from the $3$-cube to the $4$-cube.
484:
485:
486: \section{Schl\"{a}fli's formula and its expansion}
487:
488: We now consider a multilinear polynomial in four variables $x,y,z,$ and $w$,
489: \begin{equation*}
490: \begin{split}
491: F \,\,= & \,\, c_{0000} + c_{0001} w + c_{0010} z + c_{0011} z w +
492: c_{0100} y +
493: c_{0101} y w \\
494: &+ c_{0110} y z + c_{0111} y z w + c_{1000} x + c_{1001} x w + c_{1010} x
495: z \\
496: &+ c_{1011} x z w + c_{1100} x y + c_{1101} x y w + c_{1110} x y z +
497: c_{1111} x y z w,
498: \end{split}
499: \end{equation*}
500: where the $16$ coefficients $c_{ijkl}$ are
501: regarded as unknowns.
502:
503: \begin{definition}
504: The {\em hyperdeterminant}
505: $D_{2222}$ is the unique irreducible polynomial (up to sign) of content
506: one in the $16$ unknowns $\,c_{ijkl}\,$ that vanishes whenever the system
507: of equations
508: \begin{equation}
509: \label{fiveeqns}
510: F \,\,\, = \,\,\, \frac{\partial F}{\partial x}
511: \,\,\, = \,\,\, \frac{\partial F}{\partial y}
512: \,\,\, = \,\,\, \frac{\partial F}{\partial z}
513: \,\,\, = \,\,\, \frac{\partial F}{\partial w} \quad = \quad 0
514: \end{equation}
515: has a solution $(x_0,y_0,z_0,w_0)$ in $\CC^4$.
516: \end{definition}
517:
518: In theory, we can compute $D_{2222}$ by eliminating
519: the four variables $x,y,z,w$ from the five equations (\ref{fiveeqns}),
520: say by using Gr\"obner bases or resultants
521: \cite{CLO}, but in practice this is infeasible.
522: The analogous computation for a multilinear polynomial
523: in three variables, however, is easy to do, and it yields
524: the expression for $D_{222}$ stated in the Introduction.
525:
526: Schl\"afli's formula for $D_{2222}$ is obtained as follows.
527: We replace each of the eight unknowns $\,c_{ijk}\,$
528: in the $2 \times 2 \times 2$-hyperdeterminant $D_{222}$ by
529: $\,c_{ijk0} + c_{ijk1} w$. The resulting expression
530: is a polynomial of degree four in the variable $w$. Its
531: coefficients are polynomials of degree four in the $c_{ijkl}$. The
532: discriminant of this polynomial with respect to the variable $w$ is an expression of degree $24$
533: in the $c_{ijkl}$. This expression is
534: Schl\"afli's formula for
535: the hyperdeterminant $D_{2222}$.
536:
537: \begin{proposition}
538: The hyperdeterminant $D_{2222}$
539: coincides with the discriminant of
540: $\,D_{222}( c_{ijk0} + c_{ijk1} w)$,
541: considered as a polynomial in $w$,
542: divided by $256$.
543: \end{proposition}
544:
545: This proposition is a special case of \cite[Theorem 14.4.1]{GKZ}.
546: Its proof is based on the fact that both polynomials have
547: degree $24$, which we know for $D_{2222}$ by \cite[Corollary 14.2.10]{GKZ}.
548: For tensors of larger format (e.g., $2 \! \times \! 2 \! \times\! 2
549: \!\times \! 2 \!\times \! 2)$
550: the same recursive method does not work so well.
551: In general, Schl\"afli's formula yields the desired
552: hyperdeterminant times a large extraneous factor.
553:
554: The first assertion in Theorem 1 was proved by expanding
555: Schl\"afli's formula, as follows. Using ${\tt MAPLE}$ we expressed
556: $\,D_{222}( c_{ijk0} + c_{ijk1} w)$ as
557: $$ D_{222}( c_{ijk0} + c_{ijk1} w) \quad = \quad
558: b_4 w^4 + b_3 w^3 + b_2 w^2 + b_1 w + b_0 $$
559: where each $b_i$ is a degree 4 polynomial in the 16 unknowns $c_{ijkl}$.
560: These expressions for the $b_i$ in terms of the $c_{ijkl}$
561: were then substituted into the discriminant
562: \begin{equation}
563: \label{quarticDisc}
564: \begin{split}
565: &
566: 256 b_0^3 b_4^3
567: -192 b_0^2 b_1 b_3 b_4^2
568: -128 b_0^2 b_2^2 b_4^2
569: +144 b_0^2 b_2 b_3^2 b_4
570: -27 b_0^2 b_3^4 \\ &
571: +144 b_0 b_1^2 b_2 b_4^2
572: -6 b_0 b_1^2 b_3^2 b_4
573: -80 b_0 b_1 b_2^2 b_3 b_4
574: +18 b_0 b_1 b_2 b_3^3
575: +16 b_0 b_2^4 b_4 \\ &
576: -4 b_0 b_2^3 b_3^2
577: -27 b_1^4 b_4^2
578: +18 b_1^3 b_2 b_3 b_4
579: -4 b_1^3 b_3^3
580: -4 b_1^2 b_2^3 b_4
581: +b_1^2 b_2^2 b_3^2.
582: \end{split}
583: \end{equation}
584:
585: After substituting, ${\tt MAPLE}$ was unable to expand and combine the
586: discriminant's $16$
587: products due to memory constraints. Instead, we used ${\tt MAPLE}$ to
588: substitute and expand each
589: of the $16$ products (such as $\,b_1^2 b_2^2 b_3^2 \,)$ separately.
590: We then wrote a
591: C++ program to merge the $16$ expansions and divide all coefficients by
592: $256$.
593: The result of this merge was a lexicographically sorted list of
594: $2894276$ monomials of degree $24$ in the $16$ unknowns $c_{ijkl}$,
595: each with its integer coefficient. The largest absolute
596: value of any coefficient is $ 112464$.
597:
598: The symmetry group of the $4$-cube is the
599: Weyl group $B_4$ of order $384$.
600: It acts on the set of monomials in $D_{2222}$ giving
601: $9617$ orbits.
602: At our website, the $9617$ lexicographically smallest
603: monomials in each orbit are listed in lexicographic order.
604: Each monomial is listed on a separate row, in the format $\,{\tt
605: [[ExponentVector], \,Coefficient, \,FaceDimension, \,OrbitSize]}$.
606: Here ${\tt FaceDimension}$ is the dimension of the smallest face of
607: $\mathcal{N}(D_{2222})$ containing the exponent vector.
608: For example, the $192$ monomials in the orbit of
609: $$ -2 \cdot c_{0011}^2 c_{0101}^2 c_{0110}^3 c_{0111}^5 c_{1000}^7
610: c_{1001} c_{1010}
611: c_{1011} c_{1100} c_{1101}
612: $$
613: are represented by the row
614: $\, {\tt [ [0, 0, 0, 2, 0, 2, 3, 5, 7, 1, 1, 1, 1, 1, 0, 0], -2, 3, 192] }$.
615: The distribution of the $9617$ orbits according to orbit size is as follows:
616: $$
617: \begin{array}{cccccccccccc}
618: 2 & 8 & 12 & 16 & 24 & 32 & 48 & 64 & 96 & 192 & 384 \\
619: \underline{2} & 6 & 6 & 16 & 27 & 24 & 142 & 90 & 577 & 2743 & 5984 \\
620: \end{array}
621: $$
622: The two orbits of size two are represented by the monomial
623: $$
624: 2008 \cdot c_{0001}^3 c_{0010}^3 c_{0100}^3 c_{0111}^3
625: c_{1000}^3 c_{1011}^3 c_{1101}^3 c_{1110}^3,
626: $$
627: whose exponent vector lies on a $3$-dimensional face, and
628: the monomial
629: \begin{eqnarray*} & 112464 \cdot
630: c_{0000} c_{0001}^2 c_{0010}^2 c_{0011} c_{0100}^2 c_{0101} c_{0110}
631: c_{0111}^2 \\
632: & \qquad \quad \, \cdot \,\, c_{1000}^2 c_{1001} c_{1010}
633: c_{1011}^2 c_{1100} c_{1101}^2 c_{1110}^2 c_{1111},
634: \end{eqnarray*}
635: which has the largest coefficient and whose exponent vector lies in the relative interior of $\mathcal{N}(D_{2222})$. The largest odd coefficient appears in the monomial
636: $$
637: -5811 \cdot c_{0001} c_{0010} c_{0011}^4 c_{0100}^2 c_{0101} c_{0110}^2
638: c_{0111}
639: c_{1000}^2
640: c_{1001}^2 c_{1010}^2 c_{1100}^2 c_{1101}^2 c_{1110} c_{1111}
641: $$
642: with orbit size $384$ whose exponent vector lies in the relative interior of a $9$-face.
643:
644: The distribution of the $9617$ orbits according to face dimension is as follows:
645: $$
646: \begin{array}{cccccccccccc}
647: 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 \\
648: 111 & 230 & 269 & 540 & 1145 & 1862 & 2138 & 1845 & 976 & 405 & 70 & 26
649: \end{array}
650: $$
651:
652: The computation of the above face dimensions and the computation in the rest of this section were all performed after we computed the facet representation of the Newton polytope $\N(D_{2222})$. In the next section, we present all eight classes of facet inequalities and explain how we obtained them.
653:
654: In the remainder of this section we discuss the
655: $20992$ monomials whose exponent vectors lie in the Newton polytope
656: $\mathcal{N}(D_{2222})$ but which have coefficient
657: zero in the expansion of
658: hyperdeterminant $D_{2222}$. They come in $69$ $B_4$-orbits.
659: In the following table we list one representative from each orbit:
660: \smaller
661: \begin{equation*}
662: \begin{array}{cccc}
663: 0 0 0 2 1 1 4 4 4 5 1 0 1 0 1 0 &
664: 0 0 0 2 1 1 4 4 5 4 0 1 1 0 1 0 &
665: 0 0 0 2 1 1 4 4 5 4 1 0 0 1 1 0 &
666: 0 0 0 2 1 1 4 4 6 3 0 1 0 1 1 0 \\
667: 0 0 0 2 1 1 5 3 4 5 1 0 1 0 0 1 &
668: 0 0 0 2 1 1 5 3 5 4 0 1 1 0 0 1 &
669: 0 0 0 2 1 1 5 3 5 4 1 0 0 1 0 1 &
670: 0 0 0 2 1 1 5 3 6 3 0 1 0 1 0 1 \\
671: 0 0 0 2 1 2 2 5 7 1 1 1 1 0 0 1 &
672: 0 0 0 2 1 2 6 1 2 6 2 0 0 1 1 0 &
673: 0 0 0 2 1 3 6 0 2 5 3 0 0 1 0 1 &
674: 0 0 0 2 1 4 4 1 1 4 5 0 1 1 0 0 \\
675: 0 0 0 2 2 1 6 1 2 6 1 1 0 1 1 0 &
676: 0 0 0 2 2 2 3 3 5 3 2 0 0 0 0 2 &
677: 0 0 0 2 2 2 6 0 2 6 2 0 0 0 0 2 &
678: 0 0 0 2 3 1 4 2 3 5 2 0 0 0 0 2 \\
679: 0 0 0 2 4 1 1 4 5 0 1 4 1 1 0 0 &
680: 0 0 0 3 1 1 3 4 4 4 1 0 2 0 1 0 &
681: 0 0 0 3 1 1 4 3 4 4 0 1 2 0 1 0 &
682: 0 0 0 3 1 1 4 3 5 3 0 1 2 0 0 1 \\
683: 0 0 0 3 1 1 4 3 5 3 1 0 0 2 1 0 &
684: 0 0 0 3 1 1 4 3 6 1 1 1 0 3 0 0 &
685: 0 0 0 3 1 1 5 2 6 2 0 1 0 2 0 1 &
686: 0 0 0 3 1 3 4 1 1 4 4 0 2 1 0 0 \\
687: 0 0 0 3 1 4 4 0 5 0 2 2 0 2 0 1 &
688: 0 0 0 3 2 2 2 3 5 0 2 2 1 2 0 0 &
689: 0 0 0 3 2 2 3 2 2 5 2 0 1 0 2 0 &
690: 0 0 0 3 2 2 3 2 4 2 2 1 1 1 0 1 \\
691: 0 0 0 3 2 2 5 0 3 3 1 2 1 1 0 1 &
692: 0 0 0 3 3 2 4 0 4 1 1 3 0 2 0 1 &
693: 0 0 0 3 4 0 1 4 4 1 1 3 2 1 0 0 &
694: 0 0 0 4 1 1 3 3 7 1 0 0 0 2 1 1 \\
695: 0 0 0 4 1 1 4 2 5 2 1 0 0 3 1 0 &
696: 0 0 0 4 1 1 5 1 5 3 0 0 0 2 1 1 &
697: 0 0 0 4 1 1 5 1 6 1 0 1 0 3 0 1 &
698: 0 0 0 4 1 2 5 0 4 3 0 1 1 1 1 1 \\
699: 0 0 0 4 2 1 5 0 4 3 1 0 0 2 0 2 &
700: 0 0 0 5 1 1 4 1 5 1 1 0 0 4 1 0 &
701: 0 0 0 5 1 1 5 0 5 0 1 1 0 5 0 0 &
702: 0 0 0 5 1 1 5 0 6 0 0 1 0 4 0 1 \\
703: 0 0 0 6 1 1 3 1 4 1 1 0 2 3 1 0 &
704: 0 0 0 6 1 1 4 0 6 0 0 0 0 4 1 1 &
705: 0 0 0 6 2 1 2 1 2 2 1 1 4 1 1 0 &
706: \underline{0 0 1 1 1 1 1 7 7 1 1 1 1 1 0 0} \\
707: 0 0 1 1 1 1 3 5 5 3 1 1 1 1 0 0 &
708: 0 0 1 1 1 2 1 6 7 2 1 0 0 0 1 1 &
709: 0 0 1 1 1 2 1 6 8 1 0 1 0 0 1 1 &
710: 0 0 1 1 1 4 4 1 5 1 1 3 0 1 0 1 \\
711: 0 0 1 1 2 2 2 4 6 2 0 2 0 0 1 1 &
712: 0 0 1 2 1 2 5 1 3 6 0 0 0 0 2 1 &
713: 0 0 1 2 1 4 2 2 4 1 3 1 1 1 0 1 &
714: 0 0 1 2 2 2 0 5 5 1 2 1 2 0 0 1 \\
715: 0 0 1 2 3 1 0 5 4 2 2 1 2 0 0 1 &
716: 0 0 1 2 4 1 0 4 4 1 0 4 2 0 1 0 &
717: 0 0 1 3 1 4 3 0 5 2 1 0 0 0 1 3 &
718: 0 0 1 3 2 2 1 3 3 3 2 0 1 1 2 0 \\
719: 0 0 1 3 2 2 2 2 2 3 2 1 2 1 1 0 &
720: 0 0 1 3 3 3 0 2 5 1 1 1 0 0 2 2 &
721: 0 0 1 4 2 2 0 3 5 1 0 1 1 1 3 0 &
722: 0 0 1 4 2 3 1 1 5 1 0 1 0 1 3 1 \\
723: 0 0 1 4 3 2 0 2 5 2 0 0 0 0 3 2 &
724: 0 0 1 4 3 3 0 1 5 1 1 0 0 0 2 3 &
725: 0 0 1 4 3 3 1 0 5 1 0 1 0 0 2 3 &
726: 0 0 2 3 2 3 1 1 5 2 0 0 0 0 2 3 \\
727: 0 1 1 1 2 2 4 1 3 2 1 3 0 2 1 0 &
728: 0 1 1 2 2 1 4 1 3 2 0 3 1 2 1 0 &
729: 0 1 1 3 2 1 3 1 3 2 2 0 1 2 0 2 &
730: 0 1 1 4 2 1 3 0 3 1 1 1 2 2 0 2 \\
731: 0 1 2 1 2 1 2 3 2 4 2 0 2 0 0 2
732: \end{array}
733: \end{equation*}
734: \normalsize
735: For instance, the underlined vector
736: represents the monomial
737: $$
738: c_{0010} c_{0011} c_{0100} c_{0101} c_{0110} c_{0111}^7
739: c_{1000}^7 c_{1001} c_{1010} c_{1011} c_{1100} c_{1101}. $$
740: The orbit of this monomial has size $96$.
741: Each of these $69$ monomials listed above does appear during
742: the expansion of Schl\"afli's formula, namely, when we expand
743: each of the $16$ terms of the discriminant (\ref{quarticDisc}). But
744: the coefficients of these monomials sum to zero when we merge to form $D_{2222}$.
745:
746: The distribution of the $69$ orbits of ``missing monomials''
747: according to the face dimension in $\mathcal{N}(D_{2222})$ is as follows:
748: $$
749: \begin{array}{cccccccccccc}
750: 0 & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 \\
751: 0 & 0 &15 & 3 & 20 & 13 & 7 & 5 & 6 & 0 & 0 & 0 \\
752: \end{array}
753: $$
754:
755: The following method was used to generate all lattice points in
756: $\mathcal{N}(D_{2222})$ and hence find
757: the $20992$ ``missing monomials'' in $69$ orbits. We used the five equations $A \cdot x = (24,12,12,12,12)^T$ and 268 facet defining inequalities of $\N(D_{2222})$ which will be presented in the next section. By symmetry,
758: it suffices to generate at least one lattice point in each $B_4$-orbit. We first listed all
759: ways to assign eight nonnegative integers summing to $12$ on the first facet, i.e., the first eight coordinates. There were $50388$ ways in all. Per the symmetry remark above, we kept only the $1349$ ways which were lex-min under $B_3$. We then extended each of these $1349$ facet assignments to the entire $4$-cube as follows.
760: We fixed two disjoint non-parallel edges on the opposite facet, assigned all possible values for those four entries, and solved for the remaining four entries via the five linear relations above. Of the resulting $4$-cube assignments, $87435$ have
761: non-negative entries. We then tested these against the facet inequalities and found that $80788$ of them lied in the polytope. We now had at least one representative from each orbit of lattice points in $\N(D_{2222})$. From these we found $9686$ orbits of lattice points. Removing the $9617$ orbits of terms which do appear in $D_{2222}$,
762: we found precisely the
763: $ 69$ additional orbits listed above.
764:
765:
766: \section{Computing the Newton polytope}
767:
768: In this section we present our census of the vertices, facets, and other faces
769: of the Newton polytope of the
770: $2 \times 2 \times 2 \times 2$-hyperdeterminant.
771: The following result completes the proof of Theorem 1.
772: Being the analogue of equation (\ref{smallfvector}), it shows the increase
773: in complexity when passing from the $3$-cube to the $4$-cube.
774:
775: \begin{proposition}
776: \label{fD2222}
777: The $f$-vector of the Newton polytope of $D_{2222}$ equals
778: \begin{eqnarray*}
779: f\bigl(\mathcal{N}(D_{2222})) \quad = &
780: (\, 25448,\,
781: 178780,\,
782: 555280,\,
783: 1005946,\,
784: 1176976, \\ & \quad
785: 927244,\,
786: 495936,\,
787: 176604, \,39680,\,
788: 5012,\, 268 \, ).
789: \end{eqnarray*}
790: \end{proposition}
791:
792: We computed the vertices of the Newton polytope $\N(D_{2222})$
793: starting from the exponent vectors of the
794: 2894276 monomials in 9617 orbits described in Section 3. We took advantage of the symmetry group using the following heuristics:
795: First, we sorted the set of 9617 lex-min elements, one from each orbit, lexicographically and then chose a small lexicographically contiguous subset.
796: Within this subset, we removed points that were redundant
797: (i.e.,~not vertices) using {\tt POLYMAKE} \cite{GJ}.
798: Whenever a point was found to be redundant, we removed
799: its entire $B_4$-orbit from our original set of points.
800: After a few iterations we obtained the convex hull of the 9617 lex-min elements, which has 1794 vertices. The union of the orbits of those 1794 points contains 484804 points.
801: We repeated the same process on these 484804 points, finding redundant points in a small subset and removing their whole orbits. We eventually reached
802: a subset of $ 25448$ points in $111$ orbits which appeared to be irredundant.
803: We verified that these points are vertices
804: by solving $111$ linear programming feasibility problems.
805:
806: The {\em tangent cone} at a vertex $v$
807: is the set $\{v+w \in \RR^{16} : v + \epsilon \cdot w
808: \in \N(D_{2222}) \text{ for some } \epsilon > 0\}$.
809: The faces of the tangent cone are in
810: natural bijection with the faces of
811: $\N(D_{2222})$ containing $v$.
812: The {\em vertex figure} of a vertex $v$
813: is the $10$-dimensional polytope obtained by cutting off
814: $v$ from its tangent cone.
815:
816: \begin{proof}[Proof of Proposition \ref{fD2222}:]
817: The facets and $f$-vector of $\mathcal{N}(D_{2222})$ were computed as follows:
818: For each of the $111$ vertex classes, we chose
819: a representative $v$ and
820: subtracted it from each of the other $25447$ vertices.
821: The cone generated by these $25447$ difference vectors is
822: (a translate of) the tangent
823: cone at $v$. Using {\tt POLYMAKE} we computed the faces of all dimensions of the $111$ tangent cones. These computations were fairly fast since $\N(D_{2222})$ is close to being simple. We now had at least one representative from the $B_4$-orbit of every face of $\mathcal{N}(D_{2222})$. Each face is represented by the set of facets
824: containing that face. With careful relabeling, we merged the 111 lists of faces and computed their
825: orbits, thus obtaining all faces of $\N(D_{2222})$.
826: \end{proof}
827:
828: Here is a complete list of representatives for
829: the $111$ orbits of vertices. The upper and lower indices attached to each vector
830: are the coefficient and the orbit size respectively.\label{Dvertices}
831: \begin{eqnarray*}
832: 0 0 0 2 0 2 2 6 6 2 2 0 2 0 0 0^{1}_{32} &
833: 0 0 0 2 0 2 2 6 7 1 1 1 2 0 0 0^{-1}_{192} &
834: 0 0 0 2 0 2 2 6 8 0 1 1 1 1 0 0^{1}_{192} \\
835: 0 0 0 2 0 2 2 6 9 0 0 1 0 1 1 0^{-1}_{64} &
836: 0 0 0 2 0 2 3 5 7 1 0 2 2 0 0 0^{1}_{384} &
837: 0 0 0 2 0 2 3 5 7 2 1 0 1 0 0 1^{1}_{192}\\
838: 0 0 0 2 0 2 3 5 8 0 0 2 1 1 0 0^{-1}_{384} &
839: 0 0 0 2 0 2 3 5 8 1 0 1 1 0 0 1^{-1}_{384} &
840: 0 0 0 2 0 2 3 5 9 0 0 1 0 1 0 1^{1}_{192} \\
841: 0 0 0 2 0 2 4 4 8 0 0 2 0 2 0 0^{1}_{192} &
842: 0 0 0 2 0 2 6 2 2 6 2 0 2 0 0 0^{1}_{96} &
843: 0 0 0 2 0 2 7 1 2 6 1 1 2 0 0 0^{-1}_{384} \\
844: 0 0 0 2 0 2 7 1 2 7 1 0 1 0 1 0^{1}_{192} &
845: 0 0 0 2 0 2 7 1 3 5 0 2 2 0 0 0^{1}_{384} &
846: 0 0 0 2 0 2 7 1 3 6 0 1 1 0 1 0^{-1}_{384} \\
847: 0 0 0 2 0 2 7 1 4 5 0 1 0 1 1 0^{1}_{192} &
848: 0 0 0 2 0 2 8 0 2 6 1 1 1 1 0 0^{1}_{192} &
849: 0 0 0 2 0 2 8 0 2 7 1 0 1 0 0 1^{-1}_{192} \\
850: 0 0 0 2 0 2 8 0 3 5 0 2 1 1 0 0^{-1}_{384} &
851: 0 0 0 2 0 2 8 0 3 6 0 1 1 0 0 1^{1}_{384} &
852: 0 0 0 2 0 2 8 0 4 5 0 1 0 1 0 1^{-1}_{192} \\
853: 0 0 0 2 0 3 3 4 7 0 0 3 2 0 0 0^{-1}_{192} &
854: 0 0 0 2 0 3 3 4 8 1 1 0 0 0 0 2^{1}_{96} &
855: 0 0 0 2 0 3 3 4 9 0 0 1 0 0 0 2^{-1}_{96} \\
856: 0 0 0 2 0 3 6 1 3 6 1 0 0 0 2 0^{1}_{384} &
857: 0 0 0 2 0 3 6 1 4 5 0 1 0 0 2 0^{-1}_{384} &
858: 0 0 0 2 0 3 7 0 2 5 1 2 2 0 0 0^{1}_{192} \\
859: 0 0 0 2 0 3 7 0 2 6 1 1 1 0 1 0^{-1}_{384} &
860: 0 0 0 2 0 3 7 0 3 4 0 3 2 0 0 0^{-1}_{192} &
861: 0 0 0 2 0 3 7 0 3 5 0 2 1 0 1 0^{1}_{384} \\
862: 0 0 0 2 0 3 7 0 3 6 1 0 0 0 1 1^{-1}_{384} &
863: 0 0 0 2 0 3 7 0 4 5 0 1 0 0 1 1^{1}_{384} &
864: 0 0 0 2 0 3 7 0 4 5 1 0 0 0 0 2^{1}_{192} \\
865: 0 0 0 2 0 3 7 0 5 4 0 1 0 0 0 2^{-1}_{192} &
866: 0 0 0 2 0 4 6 0 3 5 1 1 0 0 2 0^{-1}_{384} &
867: 0 0 0 2 0 4 6 0 4 4 0 2 0 0 2 0^{1}_{384} \\
868: \underline{ 0 0 0 2 0 4 6 0 6 0 0 4 0 2 0 0}^{16}_{192} &
869: \underline{ 0 0 0 2 0 5 5 0 5 0 0 5 2 0 0 0}^{-16}_{96} &
870: \underline{ 0 0 0 2 0 5 5 0 7 0 0 3 0 0 0 2}^{-16}_{96} \\
871: 0 0 0 2 1 1 1 7 7 1 1 1 2 0 0 0^{1}_{96} &
872: 0 0 0 2 1 1 1 7 7 1 2 0 1 1 0 0^{1}_{192} &
873: 0 0 0 2 1 1 1 7 8 0 1 1 1 1 0 0^{-1}_{384} \\
874: 0 0 0 2 1 1 1 7 8 1 1 0 0 1 1 0^{-1}_{192} &
875: 0 0 0 2 1 1 1 7 9 0 0 1 0 1 1 0^{1}_{192} &
876: 0 0 0 2 1 1 2 6 7 1 0 2 2 0 0 0^{-1}_{384} \\
877: 0 0 0 2 1 1 2 6 7 1 2 0 0 2 0 0^{-1}_{384} &
878: 0 0 0 2 1 1 2 6 7 2 1 0 1 0 0 1^{-1}_{384} &
879: 0 0 0 2 1 1 2 6 8 0 0 2 1 1 0 0^{1}_{384} \\
880: 0 0 0 2 1 1 2 6 8 0 1 1 0 2 0 0^{1}_{384} &
881: 0 0 0 2 1 1 2 6 8 1 0 1 1 0 0 1^{1}_{192} &
882: 0 0 0 2 1 1 2 6 8 1 1 0 0 1 0 1^{1}_{384} \\
883: 0 0 0 2 1 1 2 6 9 0 0 1 0 1 0 1^{-1}_{192} &
884: 0 0 0 2 1 1 7 1 1 7 1 1 2 0 0 0^{1}_{96} &
885: 0 0 0 2 1 1 7 1 1 7 2 0 1 1 0 0^{1}_{192} \\
886: 0 0 0 2 1 1 7 1 1 8 1 0 1 0 1 0^{-1}_{384} &
887: 0 0 0 2 1 1 7 1 2 6 0 2 2 0 0 0^{-1}_{192} &
888: 0 0 0 2 1 1 7 1 2 6 2 0 0 2 0 0^{-1}_{384} \\
889: 0 0 0 2 1 1 7 1 2 7 0 1 1 0 1 0^{1}_{384} &
890: 0 0 0 2 1 1 7 1 2 7 1 0 0 1 1 0^{1}_{384} &
891: 0 0 0 2 1 1 7 1 3 6 0 1 0 1 1 0^{-1}_{384} \\
892: 0 0 0 2 1 1 8 0 1 8 1 0 1 0 0 1^{1}_{192} &
893: 0 0 0 2 1 1 8 0 2 7 0 1 1 0 0 1^{-1}_{384} &
894: 0 0 0 2 1 1 8 0 2 7 1 0 0 1 0 1^{-1}_{384} \\
895: 0 0 0 2 1 1 8 0 3 6 0 1 0 1 0 1^{1}_{384} &
896: 0 0 0 2 1 2 2 5 7 0 0 3 2 0 0 0^{1}_{192} &
897: 0 0 0 2 1 2 2 5 8 1 1 0 0 0 0 2^{-1}_{96}\\
898: 0 0 0 2 1 2 2 5 9 0 0 1 0 0 0 2^{1}_{48} &
899: 0 0 0 2 2 0 1 7 6 3 1 0 1 0 1 0^{-1}_{384} &
900: 0 0 0 2 2 0 1 7 7 2 0 1 1 0 1 0^{1}_{192} \\
901: 0 0 0 2 2 0 1 7 7 2 1 0 0 1 1 0^{1}_{384} &
902: 0 0 0 2 2 0 2 6 6 2 0 2 2 0 0 0^{1}_{96} &
903: 0 0 0 2 2 0 2 6 6 2 2 0 0 2 0 0^{1}_{192} \\
904: \end{eqnarray*} \begin{eqnarray*}
905: 0 0 0 2 2 0 2 6 6 3 1 0 1 0 0 1^{1}_{384} &
906: 0 0 0 2 2 0 2 6 7 2 1 0 0 1 0 1^{-1}_{384} &
907: 0 0 0 2 2 1 7 0 2 7 1 0 0 0 0 2^{1}_{96} \\
908: 0 0 0 2 2 1 7 0 3 6 0 1 0 0 0 2^{-1}_{192} &
909: 0 0 0 2 3 0 0 7 5 2 3 0 1 1 0 0^{1}_{384} &
910: 0 0 0 2 3 0 0 7 6 1 2 1 1 1 0 0^{-1}_{384} \\
911: 0 0 0 2 3 0 0 7 6 2 2 0 0 1 1 0^{-1}_{192} &
912: 0 0 0 2 3 0 0 7 7 1 1 1 0 1 1 0^{1}_{192} &
913: 0 0 0 2 3 0 1 6 6 3 1 0 0 0 1 1^{1}_{384} \\
914: 0 0 0 2 3 0 2 5 6 3 1 0 0 0 0 2^{-1}_{192} &
915: 0 0 0 2 3 0 6 1 3 6 0 1 0 0 0 2^{1}_{96} &
916: 0 0 0 2 4 0 0 6 5 2 3 0 0 1 0 1^{-1}_{384} \\
917: 0 0 0 2 4 0 0 6 6 1 2 1 0 1 0 1^{1}_{192} &
918: 0 0 0 2 4 3 3 0 5 0 0 5 0 0 0 2^{-1}_{96} &
919: 0 0 0 2 5 0 0 5 5 2 2 1 0 0 0 2^{1}_{48} \\
920: 0 0 0 3 0 3 3 3 9 0 0 0 0 0 0 3^{1}_{16} &
921: 0 0 0 3 0 3 5 1 4 5 0 0 0 0 3 0^{1}_{192} &
922: 0 0 0 3 0 3 6 0 2 6 1 0 1 0 2 0^{1}_{96} \\
923: 0 0 0 3 0 3 6 0 4 5 0 0 0 0 2 1^{-1}_{192} &
924: 0 0 0 3 0 3 6 0 6 3 0 0 0 0 0 3^{-1}_{64} &
925: 0 0 0 3 0 4 5 0 3 5 1 0 0 0 3 0^{1}_{192} \\
926: 0 0 0 3 1 1 1 6 7 1 1 0 2 0 0 1^{-1}_{192} &
927: 0 0 0 3 1 1 1 6 8 0 1 0 1 1 0 1^{1}_{192} &
928: 0 0 0 3 1 1 1 6 9 0 0 0 0 1 1 1^{-1}_{64} \\
929: 0 0 0 3 1 1 7 0 1 7 1 0 1 1 1 0^{1}_{96} &
930: 0 0 0 3 1 1 7 0 2 6 1 0 0 2 1 0^{-1}_{192} &
931: 0 0 0 3 1 1 7 0 3 6 0 0 0 1 1 1^{-1}_{192} \\
932: 0 0 0 4 0 4 4 0 4 0 0 4 4 0 0 0^{1}_{96} &
933: \underline{0 0 0 4 0 4 4 0 8 0 0 0 0 0 0 4}^{-27}_{16} &
934: 0 0 0 8 1 1 1 1 1 1 1 1 8 0 0 0^{1}_{48} \\
935: 0 0 0 9 1 1 1 0 1 1 1 0 7 1 1 0^{1}_{48} &
936: 0 0 0 9 1 1 1 0 1 1 1 0 8 0 0 1^{-1}_{96} &
937: 0 0 1 1 1 1 0 8 8 1 0 1 1 0 1 0^{1}_{92} \\
938: 0 0 1 1 1 1 1 7 7 2 1 0 1 0 0 1^{1}_{192} &
939: 0 0 1 1 1 1 1 7 8 1 0 1 1 0 0 1^{-1}_{192} &
940: 0 0 1 1 1 2 0 7 7 1 2 0 1 0 0 1^{1}_{192} \\
941: 0 0 1 1 1 2 0 7 8 1 1 0 0 0 1 1^{-1}_{192} &
942: \underline{0 1 1 0 1 0 0 9 9 0 0 1 0 1 1 0}^{1}_{8} &
943: 0 1 1 0 1 0 6 3 3 6 0 1 0 1 1 0^{1}_{32}
944: \end{eqnarray*}
945:
946: The distribution of the $111$ orbit sizes is
947: $$
948: \begin{array}{cccccccc}
949: 8 & 16 & 32 & 48 & 64 & 96 & 192 & 384 \\
950: 1 & 2 & 2 & 4 & 3 & 16 & 44 & 39 \\
951: \end{array}
952: $$
953: The distribution of the $111$ corresponding extreme monomial
954: coefficients is
955: $$
956: \begin{array}{ccccc}
957: 1 & -1 & 16 & -16 & -27 \\
958: 60 & 47 & 1 & 2 & 1 \\
959: \end{array}
960: $$
961: All but four of the monomials have coefficient $\pm 1$.
962: The largest absolute value of a
963: coefficient appears in the monomial
964: $-27 \cdot c_{0011}^4 c_{0101}^4 c_{0110}^4 c_{1000}^8 c_{1111}^4$
965: which has orbit size 16. Its $D$-equivalence class will
966: be discussed in Example \ref{fatclass}. The monomial
967: $16 \cdot c_{0011}^2 c_{0101}^4 c_{0110}^6 c_{1000}^6 c_{1011}^4 c_{1101}^2$
968: has orbit size $192$, and there are two monomials with
969: coefficient $-16$, each having orbit size $96$. Their $D$-equivalence classes will be discussed in Example \ref{coeff16}.
970:
971: Among the $111$ types of vertices, $35$ are simple.
972: The last underlined vertex, representing the monomial $8 \cdot c_{0001} c_{0010} c_{0100} c_{0111}^9 c_{1000}^9 c_{1011} c_{1101} c_{1110}$, has the largest vertex figure.
973: That vertex figure has the $f$-vector
974: $$(67,873,4405,11451,17440,16452,9699,3446,667,56).$$
975: Thus this distinguished vertex is adjacent to
976: $67$ other vertices, and it lies in $56$ facets.
977: In Example \ref{largestclass} we shall see that this vertex
978: corresponds to the largest $D$-equivalence class
979: of triangulations.
980: Table 1 shows the distribution of the $111$ types of vertices
981: according to the number of incident edges and facets.
982:
983:
984: \begin{table}
985: \footnotesize
986: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|}
987: \hline
988: &11&12&13&14&15&16&17&18&19&20&21&22&26&28&29&39&56&\!total\!\\ \hline
989: 11&35& & & & & & & & & & & &
990: & &
991: & & & 35\\ \hline
992: 12 & &14& & & & & & & & & & &
993: & &
994: & & & 14\\ \hline
995: 13 & &13& 5 & 1 & & & & & & & & & &
996: & &
997: & & 19\\ \hline
998: 14 & & & 1 & 2 & & & & & & & & & &
999: &
1000: & & & 3\\ \hline
1001: 15 & &1 & 3 & 1 & 1 & & 1 & & & & & & &
1002: & &
1003: & & 7 \\ \hline
1004: 16 & & & 1 & 1 & 2 & & & & & & & & &
1005: &
1006: & & & 4 \\ \hline
1007: 17 & & & 1 & 1 & 3 & 1 & 1 & & & & & & &
1008: & &
1009: & & 7\\ \hline
1010: 18 & & & & & 1 & 3 & & 1 & 1 & & & & &
1011: & &
1012: & & 6 \\ \hline
1013: 19 & & & & 1 & & &1 & & & & & &
1014: & &
1015: & & & 2 \\ \hline
1016: 20 & & & & & & & 1 & & & & & &
1017: & &
1018: & & & 1 \\ \hline
1019: 21 & & & & & & & & & 1 &1 &1 & &
1020: & &
1021: & & &3 \\ \hline
1022: 22 & & & & & & & & & & & & 1 &
1023: & &
1024: & & & 1 \\ \hline
1025: 23 & & & & & 1 & & & & & & & &
1026: & &
1027: & & & 1 \\ \hline
1028: 25 & & & & & & & & & & & 1 & & 1
1029: & &
1030: & & &2 \\ \hline
1031: 26 & & & & & & & & & & & & &
1032: & 1&
1033: & & & 1 \\ \hline
1034: 27 & & & & & & & & & & & & 1 &
1035: & &
1036: & & & 1 \\ \hline
1037: 30 & & & & & & & & & & & & &
1038: & 1 &
1039: & & & 1 \\ \hline
1040: 32 & & & & & & & & & & & & &
1041: & &
1042: 1 & & & 1 \\ \hline
1043: 42 & & & & & & & & & & & & &
1044: & &
1045: & 1 & & 1 \\ \hline
1046: 67 & & & & & & & & & & & & &
1047: &
1048: & & & 1 & 1 \\ \hline
1049: \!total\!&35&28&11&7&8&4&4&1&2&1&2&2&1&2&1&1&1 & 111 \\ \hline
1050: \end{tabular}
1051: \vskip .2cm
1052: \caption{Distribution of the 111 vertices
1053: according to the number of facets (columns) and
1054: edges (rows) containing the vertex.}
1055: \label{tab:vertices}
1056: \end{table}
1057:
1058: The $268$ facets of $\N(D_{2222})$
1059: come in eight orbits. In Table 2
1060: we present the $f$-vector
1061: and the orbit size for each facet.
1062: We discuss the geometry of
1063: each of the eight facet types.
1064: Since $\N(D_{2222})$ is a Minkowski summand of the
1065: secondary polytope $\N(E_{2222})$, each
1066: facet normal of $\N(D_{2222})$ is a facet normal
1067: of $\N(E_{2222})$ and hence corresponds to
1068: a {\em coarsest regular subdivision} of the $4$-cube.
1069: We depict each subdivision by its {\em tight span}, which
1070: is the complex of bounded faces of the polyhedron
1071: $\, \{ u \in \RR^5 : u \cdot A \leq w \}$,
1072: where $A$ is the $5 \times 16$-matrix
1073: in Section 1 and $w \in \RR^{16}$ is the facet normal in question.
1074:
1075:
1076: \begin{table}
1077: \begin{tabular}{|c|c|c|}
1078: \hline
1079: & \text{$f$-vector} & \!\text{orbit}\! \\ \hline
1080: 1& \!(11625,72614,197704,308238,303068,194347,80874,20906,3021,187)\!\!&
1081: 16 \\ \hline
1082: 2& (4112, 25746, 71456,115356,119228,81590,36802,10488, 1704, 122)& 12 \\ \hline
1083: 3& (363, 2289, 6538, 10996, 11921, 8581, 4080, 1239, 225, 22) & 8 \\ \hline
1084: 4& (938, 5226, 13182, 19716, 19263, 12765, 5758, 1721, 318, 31) & 32 \\ \hline
1085: 5& (336, 1937, 5126, 8121, 8468, 6022, 2928, 950, 194, 22)& 48 \\ \hline
1086: 6& (289, 1624, 4228, 6636, 6894, 4914, 2413, 798, 168, 20) & 96 \\ \hline
1087: 7& (450, 2526, 6522, 10103, 10315, 7195, 3440, 1099, 220, 24) & 48\\ \hline
1088: 8& (681, 3906, 10323, 16407, 17194, 12264, 5933, 1877, 357, 34) & 8 \\ \hline
1089: \end{tabular}
1090: \vskip .2cm
1091: \label{tab:facets}
1092: \caption{The eight types of facets of the Newton polytope $\N(D_{2222})$}
1093: \end{table}
1094:
1095:
1096: \medskip
1097:
1098: \noindent {\bf Facet 1:}
1099: These $16$ facets are defined by inequalities like
1100: $\,x_{0000} \geq 0 $. The tight span of the corresponding
1101: subdivision is a line segment.
1102: The two maximal cells are the simplex obtained by slicing off one vertex
1103: and the convex hull of the other 15 vertices, which has $f$-vector $(15,34,28,9)$.
1104:
1105: \smallskip
1106: \noindent {\bf Facet 2:}
1107: These $12$ facets are given by inequalities like
1108: $$
1109: x_{0000}+x_{0001}+x_{0010}+x_{0011} \geq 2.
1110: $$
1111: The tight span is a line segment. The two maximal cells
1112: in the corresponding subdivision
1113: have $f$-vector $(12,24,19,7)$. Here we
1114: cut the 4-cube with the hyperplane containing the eight vertices
1115: in a pair of opposite 2-faces.
1116:
1117: \smallskip
1118: \noindent {\bf Facet 3:}
1119: These $8$ facets are given by inequalities like
1120: $$
1121: 2 \cdot x_{0000}+x_{0001}+x_{0010}+x_{0100}+x_{1000} \geq 2.
1122: $$
1123: The tight span is a line segment, and the $f$-vector of both
1124: maximal cells is $(11,28,26,9)$.
1125: Here we cut the 4-cube with the hyperplane spanned
1126: by the six vertices which have coordinate sum two, whose convex hull is an octahedron.
1127:
1128: \smallskip
1129: \noindent {\bf Facet 4:}
1130: These $32$ facets are given by inequalities like
1131: $\, x_{0000}+x_{0001} \leq 9 $.
1132: The tight span is a triangle. Each of the three maximal cells is a prism
1133: over a square pyramid and has $f$-vector $(10,21,18,7)$. They are
1134: formed by fixing an edge of the $4$-cube and joining that
1135: edge to the three facets which are disjoint from that edge.
1136: Any two cells intersect in a triangular prism, and all
1137: three cells intersect in the rectangle formed by the fixed
1138: edge and its opposite edge.
1139:
1140: \begin{figure}
1141: \includegraphics[scale=0.5]{allFacetTspans.eps}
1142: %\vskip 6cm
1143: \caption{The tight spans dual to the subdivisions
1144: of the 4-cube corresponding to the
1145: facets of $\N(D_{2222})$. Each vertex of a tight span
1146: is labeled by the normalized volume and the number of vertices of the
1147: corresponding
1148: maximal cell in the subdivision.}
1149: \end{figure}
1150:
1151:
1152: \smallskip
1153: \noindent {\bf Facet 5:}
1154: These 48 facets are given by inequalities like
1155: $$
1156: x_{0000}+x_{0001}+x_{0010}+x_{0100}+x_{0110}+x_{1000}+x_{1001} \geq 3 .
1157: $$
1158: The tight span is a triangle. Two of the three maximal cells have $f$-vector
1159: $(10,23,21,8)$ and intersect in a square pyramid; the other cell has
1160: $f$-vector
1161: $(9,18,15,6)$ and intersects each of the first two in a triangular prism.
1162: The intersection of all three cells is a square.
1163:
1164: \smallskip
1165: \noindent {\bf Facet 6:}
1166: These 96 facets are given by inequalities like
1167: $$
1168: x_{0000}+x_{0001}+x_{0010}+x_{0011}+x_{0100}+x_{0110}+x_{1000}+x_{1001}
1169: \geq 4.
1170: $$
1171: The tight span is a tetrahedron. Two of the maximal cells have
1172: $f$-vector $(8,18,17,7)$ and the other two have $f$-vector
1173: $(9,21,20,8)$. The intersection of all four cells is a segment which is a
1174: diagonal of the $4$-cube.
1175:
1176: \smallskip
1177: \noindent {\bf Facet 7:}
1178: These 48 facets are given by inequalities like
1179: $$
1180: x_{0000}+x_{0001}+x_{0010}+x_{0011}+x_{0100}+x_{1000}+x_{1100} \leq 19.
1181: $$
1182: The tight span is a square pyramid.
1183: The maximal cell corresponding to the tip of the pyramid is the convex hull of two squares in complementary dimensions that share a vertex. It has $f$-vector $(7,17,18,8)$ and meets each of the other four cells in a triangle.
1184: The other four maximal cells have $f$-vector $(8,18,17,7)$, and their intersection is a triangle. All the cells meet along a diagonal segment.
1185:
1186: \smallskip
1187: \noindent {\bf Facet 8:}
1188: These 8 facets are given by inequalities like
1189: $$
1190: 2 \cdot x_{0000}+x_{0001}+x_{0010}+x_{0100}+ x_{1000} \leq 18.
1191: $$
1192: These facets are indexed by diagonals. We can identify
1193: the $4$-cube with a boolean lattice when we fix a diagonal. The maximal
1194: antichain in this lattice has the structure of an octahedron. Each point in the
1195: antichain uniquely determines a maximal cell in the subdivision. Hence the tight span is also an octahedron. Each maximal cell is the convex hull of two squares in complementary dimensions joined
1196: at a vertex, and each has $f$-vector $(7,17,18,8)$.
1197:
1198: \smallskip
1199:
1200: This concludes our discussion of the facets of
1201: the Newton polytope $\mathcal{N}(D_{2222})$.
1202: The $B_4$-orbits of the eight
1203: stated inequalities yield $268$ facet inequalities. These together with the identity
1204: $\, A \cdot x \, = \, (24,12,12,12,12)^T \,$
1205: gives an irredundant presentation
1206: of $\mathcal{N}(D_{2222})$
1207: by linear equations and inequalities.
1208:
1209: \medskip
1210:
1211: \section{Tight spans and $A$-discriminants of other
1212: $0/1$-polytopes}
1213:
1214: The concepts and computations presented in this paper
1215: make sense for any $d \times n$-integer matrix $A$ whose row span
1216: contains the all-ones vector $(1,1,\ldots,1) $. Following
1217: \cite{GKZ}, the $A$-discriminant $D_A$ is an irreducible factor of
1218: the principal $A$-determinant $E_A$. The
1219: vertices of the secondary polytope $\mathcal{N}(E_A)$
1220: correspond to regular triangulations of $A$, and these
1221: map onto the vertices of the Newton polytope
1222: $\mathcal{N}(D_A)$. The fibers of
1223: that map (i.e., the $D$-equivalence classes) and the
1224: normal fan of $\mathcal{N}(D_A)$ have
1225: recently been characterized in \cite{DFS}.
1226:
1227: For fixed $A$ and an arbitrary row vector $w \in \RR^n$,
1228: we consider the polyhedron
1229: $$ P_w \quad = \quad \{\,u \in \RR^d \,: \, u \cdot A \leq w \,\}. $$
1230: This polyhedron is dual to the regular triangulation of $A$
1231: defined by $w$. The {\em tight span} of $(A,w)$ is the
1232: complex of bounded faces of the polyhedron $P_w$. For
1233: experts in tropical geometry, we note that the
1234: $(d-2)$-skeleton of $P_w$ is the {\em tropical hypersurface} of $(A,w)$.
1235: The polyhedron $P_w$ also occurs naturally when
1236: relating regular triangulations to linear programming
1237: duality as in \cite[\S 1.2]{DRS}.
1238:
1239: We propose that, for many matrices $A$ of
1240: interest in combinatorics and discrete convexity \cite{Hir},
1241: the tight span is an excellent geometric
1242: representation of the data $(A,w)$.
1243: In Section 2 we studied the case when $A$ is the
1244: $4 \times 8$-matrix representing
1245: the $3$-cube, and in Sections 4, 6 and 7
1246: we are concerned with the case when $A$ is the
1247: $5 \times 16$-matrix representing the $4$-cube.
1248: In what follows we present two
1249: related situations which have appeared
1250: in the recent literature.
1251:
1252: \begin{example}[\bf Tropical Polytopes]
1253: \label{ex1} \rm
1254: Let $A$ be the $(r+s) \times (r \cdot s)$-matrix
1255: which represents the direct product of an
1256: $(r-1)$-simplex $\Delta_{r-1}$
1257: with an $(s-1)$-simplex $\Delta_{s-1}$.
1258: The principal $A$-determinant $E_A$
1259: is the product of all subdeterminants
1260: (of all sizes) of an $r \times s$-matrix
1261: of unknowns $(c_{ij})$; see \cite[page 303]{GKZ}.
1262:
1263:
1264: For instance, if $r=s=3$ then $A$ represents the product of two triangles,
1265: \begin{equation}
1266: \label{triangletimestriangle}
1267: A \quad = \quad
1268: \left[
1269: \begin{array}{ccccccccc}
1270: 1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\
1271: 0 & 0 & 0 & 1 & 1 & 1 & 0 & 0 & 0 \\
1272: 0 & 0 & 0 & 0 & 0 & 0 & 1 & 1 & 1 \\
1273: 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 \\
1274: 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 \\
1275: 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1
1276: \end{array}
1277: \right],
1278: \end{equation}
1279: and the $A$-discriminant $D_A$ is the determinant of the matrix
1280: $$ \begin{pmatrix} c_{11} & c_{12} & c_{13}\\
1281: c_{21} & c_{22} & c_{23}\\
1282: c_{31} & c_{32} & c_{33}
1283: \end{pmatrix}.$$
1284: The degree $30$ polynomial $E_A$ is the product
1285: of all $19$ subdeterminants of this matrix.
1286: The extreme monomials of $E_A$ appear in
1287: \cite[Figure 39, page 250]{GKZ}.
1288:
1289:
1290: The Newton polytope $\mathcal{N}(E_A)$ of the product of all
1291: subdeterminants is the secondary polytope whose faces
1292: correspond to regular polyhedral subdivisions of
1293: $\Delta_{r-1} \times \Delta_{s-1}$. The corresponding
1294: tight spans are precisely the {\em tropical polytopes},
1295: which are obtained as tropical convex hulls \cite{DS}
1296: of $r$ points in $(s-1)$-space. The combinatorial classification of
1297: such polytopes is the analogue of what will be
1298: accomplished for the $4$ cube in
1299: the next section. We refer to recent work of Santos \cite{San}
1300: for a discussion. His
1301: Section 5 deals with tropical polytopes,
1302: and his Figure 2 is essentially the same as our Figure 1.
1303: Our number $235277$ in Theorem 2 is the $4$-cube analogue
1304: of the numbers in the table at the end of
1305: Section 4 in \cite{DS} including the number $35$ in
1306: \cite[Fig.~1]{DeL} and in \cite[Fig.~6]{DS}.
1307: \qed
1308: \end{example}
1309:
1310: \begin{example} [\bf Injective hulls of finite metric spaces]
1311: \label{ex2}
1312: The term ``tight span'' is derived from the special case when
1313: $A$ represents the {\em second hypersimplex}.
1314: As shown in \cite{Hir, SY}, here the
1315: tight spans are precisely the {\em injective hulls
1316: of finite metric spaces}, which play an important
1317: role in phylogenetic combinatorics.
1318: The case studied in \cite{SY}
1319: concerns metrics on six points, where
1320: $$ A \quad = \quad
1321: \left[
1322: \begin{array}{ccccccccccccccc}
1323: 1 & 1 & 1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\
1324: 1 & 0 & 0 & 0 & 0 & 1 & 1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\
1325: 0 & 1 & 0 & 0 & 0 & 1 & 0 & 0 & 0 & 1 & 1 & 1 & 0 & 0 & 0 \\
1326: 0 & 0 & 1 & 0 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & 1 & 0 \\
1327: 0 & 0 & 0 & 1 & 0 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 1 & 0 & 1 \\
1328: 0 & 0 & 0 & 0 & 1 & 0 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 1 & 1 \\
1329: \end{array}
1330: \right] .
1331: $$
1332: Here the $A$-discriminant $D_A$ is the determinant of the
1333: symmetric matrix
1334: $$ \begin{pmatrix}
1335: 0 & c_{12} & c_{13} & c_{14} & c_{15} & c_{16} \\
1336: c_{12} & 0 & c_{23} & c_{24} & c_{25} & c_{26} \\
1337: c_{13} & c_{23} & 0 & c_{34} & c_{35} & c_{36} \\
1338: c_{14} & c_{24} & c_{34} & 0 & c_{45} & c_{46} \\
1339: c_{15} & c_{25} & c_{35} & c_{45} & 0 & c_{56} \\
1340: c_{16} & c_{26} & c_{36} & c_{46} & c_{56} & 0 \\
1341: \end{pmatrix}.
1342: $$
1343: Up to a constant, the principal $A$-determinant $E_A$ is
1344: the product of all principal
1345: minors of size $\geq 3$. This is a polynomial
1346: of degree $156$ having $194160$ extreme terms in
1347: $339$ symmetry classes \cite[Theorem 1]{SY}.
1348: Thus our number $235277$ is the $4$-cube analogue
1349: to the number $339$ of generic six-point metrics.
1350: \qed
1351: \end{example}
1352:
1353: The prominent role of the second hypersimplex
1354: in phylogenetic combinatorics
1355: raises the question of what happens for other
1356: hypersimplices. We propose the following
1357: problem for further
1358: mathematical and computational research.
1359:
1360:
1361: \begin{problem} [\bf The hypersimplex $\Delta(6,3)$]
1362: \rm
1363: \label{ex3} \hfill \break
1364: Let $A$ be the $6 \times 20$-matrix whose
1365: columns are the vertices
1366: $e_i+e_j+e_k$ ($1 \leq i < j < k \leq 6$) of
1367: the third hypersimplex $\Delta(6,3)$.
1368: What is the degree of the $A$-discriminant $D_A$~?
1369: Can the $14$-dimensional polytope $\N(D_A)$ be computed~?
1370: Can the monomial expansion of the $A$-discriminant $D_A$
1371: be computed~?
1372: \end{problem}
1373:
1374:
1375:
1376: All matrices $A$ in the examples above have their entries
1377: in $\{0,1\}$, so they represent subpolytopes
1378: of cubes of appropriate dimensions.
1379: In what follows we examine the $A$-discriminants of
1380: various subpolytopes of the $4$-cube.
1381: These arise naturally as irreducible factors in the initial forms
1382: of the hyperdeterminant.
1383:
1384: For any vector $w$ in $\RR^{16}$ we define the
1385: {\em initial form} $\,{\rm in}_w(D_{2222})\,$
1386: as the sum of all terms in $D_{2222}$ having minimal $w$-weight.
1387: If $w$ is generic then $\,{\rm in}_w(D_{2222})\,$
1388: is a monomial, and we have classified all of these
1389: monomials in Section 4. We now consider the other
1390: extreme case when $w$ is as non-generic as possible.
1391: More precisely, we pick $w$ among the normal vectors
1392: to the facets
1393: of $\mathcal{N}(D_{2222})$.
1394:
1395:
1396: \begin{proposition} [\bf All maximal initial forms of the hyperdeterminant]
1397: \label{all max}
1398: The following list is the classification of the initial forms
1399: of hyperdeterminant $D_{2222}$ corresponding to all facets of its
1400: Newton polytope. The eight symmetry classes of facets
1401: of $\N(D_{2222})$ are listed
1402: in the same order as in Section~4. \rm
1403:
1404: \smallskip
1405:
1406: \noindent {\bf Facet 1:}
1407: % $$e_{0000} \geq 0.$$
1408: If $w = e_{0000} $ then
1409: ${\rm in}_w(D_{2222})$ is obtained from
1410: $D_{2222}$ by setting $c_{0000}=0$.
1411: This initial form is irreducible: It is the
1412: $A$-discriminant where $A$ is the
1413: $15$-point configuration
1414: obtained from the $4$-cube by removing
1415: one vertex.
1416:
1417: \smallskip
1418: \noindent {\bf Facet 2:}
1419: % $$ e_{0000}+e_{0001}+e_{0010}+e_{0011} \geq 2. $$
1420: If $w = e_{0000}+e_{0001}+e_{0010}+e_{0011}$ then
1421: ${\rm in}_w(D_{2222})$ has $67230$ terms and
1422: factors as
1423: $c_{1010} c_{1001} - c_{1000} c_{1011}$ times
1424: $c_{0101} c_{0110} - c_{0100} c_{0111}$ times
1425: the square of the $2 \times 2 \times 2$-hyperdeterminant of
1426: $[c_{1000}, c_{1001}, c_{1010}, c_{1011}, c_{0100} ,c_{0101} , c_{0110} ,c_{0111}]
1427: $
1428: times the product of two larger factors,
1429: each having $66$ terms of degree six,
1430: which are the $A$-discriminants of
1431: the two maximal cells in the subdivision.
1432:
1433:
1434: \smallskip
1435: \noindent {\bf Facet 3:}
1436: % $$ 2 \cdot e_{0000}+e_{0001}+e_{0010}+e_{0100}+e_{1000} \geq 2.$$
1437: If $ w = 2 \cdot e_{0000}+e_{0001}+e_{0010}+e_{0100}+e_{1000}$ then
1438: ${\rm in}_w(D_{2222})$ equals
1439: \begin{smaller}
1440: \begin{eqnarray*}
1441: & (1/4) \cdot
1442: c_{0011}
1443: c_{0101}
1444: c_{0110}
1445: c_{1001}
1446: c_{1010}
1447: c_{1100} \cdot
1448: \biggl\{
1449: {\rm det} \left[ \begin{matrix}
1450: 0 & c_{0011} & c_{0101} & c_{1001} \\
1451: c_{0011} & 0 & c_{0110} & c_{1010} \\
1452: c_{0101} & c_{0110} & 0 & c_{1100} \\
1453: c_{1001} & c_{1010} & c_{1100} & 0
1454: \end{matrix} \right] \biggr\}^{\mathbf 2} \times \\ &
1455: \!\!\!
1456: {\rm det} \! \left[ \begin{matrix}
1457: 2 c_{0000} & c_{0001} & c_{0010} & c_{0100} & c_{1000} \\
1458: c_{0001} & 0 & c_{0011} & c_{0101} & c_{1001} \\
1459: c_{0010} & c_{0011} & 0 & c_{0110} & c_{1010} \\
1460: c_{0100} & c_{0101} & c_{0110} & 0 & c_{1100} \\
1461: c_{1000} & c_{1001} & c_{1010} & c_{1100} & 0
1462: \end{matrix} \right] \!\!
1463: {\rm det} \! \left[ \begin{matrix}
1464: 2 c_{1111} & c_{1110} & c_{1101} & c_{1011} & c_{0111} \\
1465: c_{1110} & 0 & c_{1100} & c_{1010} & c_{0110} \\
1466: c_{1101} & c_{1100} & 0 & c_{1001} & c_{0101} \\
1467: c_{1011} & c_{1010} & c_{1001} & 0 & c_{0011} \\
1468: c_{0111} & c_{0110} & c_{0101} & c_{0011} & 0
1469: \end{matrix}\right].
1470: \end{eqnarray*}
1471: \end{smaller}
1472:
1473: \noindent The squared factor is the $A$-discriminant of the octahedron,
1474: and the $5 \times 5$-determinants are $A$-discriminants
1475: of the two maximal cells in this subdivision.
1476:
1477: \medskip
1478: \noindent {\bf Facet 4:}
1479: If $ w = - e_{0000} - e_{0001} $ then
1480: ${\rm in}_w(D_{2222})$ equals
1481: \begin{smaller}
1482: \begin{eqnarray*} &
1483: (c_{0000} c_{1111}-c_{0001} c_{1110})^3 \\ &
1484: (c_{1100} c_{1111}-c_{1101} c_{1110})
1485: (c_{1010} c_{1111}-c_{1011} c_{1110})
1486: (c_{0110} c_{1111}-c_{0111} c_{1110}) \\ & \!\!\!
1487: (c_{0000}^2 c_{1001} c_{1111} - c_{0000}^2 c_{1011} c_{1101}
1488: - c_{0000} c_{0001} c_{1000} c_{1111}
1489: - c_{0000} c_{0001} c_{1001} c_{1110} \\ &
1490: + c_{0000} c_{0001} c_{1010} c_{1101}
1491: + c_{0000} c_{0001} c_{1011} c_{1100}
1492: + c_{0001}^2 c_{1000} c_{1110} - c_{0001}^2 c_{1010} c_{1100}) \\ & \!\!\!
1493: (c_{0000}^2 c_{0101} c_{1111}-c_{0000}^2 c_{0111} c_{1101}
1494: -c_{0000} c_{0001} c_{0100} c_{1111}
1495: -c_{0000} c_{0001} c_{0101} c_{1110} \\ &
1496: +c_{0000} c_{0001} c_{0110} c_{1101}
1497: +c_{0000} c_{0001} c_{0111} c_{1100}
1498: +c_{0001}^2 c_{0100} c_{1110} - c_{0001}^2 c_{0110} c_{1100}) \\ & \!\!\!
1499: (c_{0000}^2 c_{0011} c_{1111} - c_{0000}^2 c_{0111} c_{1011}
1500: -c_{0000} c_{0001} c_{0010} c_{1111}
1501: -c_{0000} c_{0001} c_{0011} c_{1110} \\ &
1502: +c_{0000} c_{0001} c_{0110} c_{1011}
1503: +c_{0000} c_{0001} c_{0111} c_{1010}
1504: +c_{0001}^2 c_{0010} c_{1110}
1505: -c_{0001}^2 c_{0110} c_{1010}).
1506: \end{eqnarray*}
1507: \end{smaller}
1508: The quartic factors are the $A$-discriminants of the
1509: three maximal cells.
1510:
1511: \medskip
1512: \noindent {\bf Facet 5:}
1513: % $$ e_{0000}+e_{0001}+e_{0010}+e_{0100}+e_{0110}+e_{1000}+e_{1001} \geq 3 .$$
1514: If $w = e_{0000}+e_{0001}+e_{0010}+e_{0100}+e_{0110}+e_{1000}+e_{1001} $ then
1515: ${\rm in}_w(D_{2222})$
1516: equals
1517: $\, c_{0000} c_{0011} c_{0101} c_{1010} c_{1100} \,$
1518: times $\,(c_{0011} c_{1100} - c_{0101} c_{1010})^3 \,$
1519: times
1520: \begin{eqnarray*}
1521: & c_{0011} c_{1100} c_{1111}-c_{0011} c_{1101} c_{1110}
1522: -c_{0101} c_{1010} c_{1111} \\ & +c_{0101} c_{1011} c_{1110}
1523: +c_{0111}c_{1010} c_{1101} - c_{0111} c_{1011} c_{1100}
1524: \end{eqnarray*}
1525: times the product of two larger factors,
1526: each having $15$ terms of degree five
1527: which are the $A$-discriminants of
1528: the two cells with $f$-vector $(10,23,21,8)$.
1529:
1530: \medskip
1531:
1532: In the remaining cases the tight spans
1533: are three-dimensional (Figure 2), and
1534: the initial form ${\rm in}_w(D_{2222})$ factors as a
1535: monomial times the product of the $A$-discriminants
1536: of the maximal cells in that subdivision of the 4-cube.
1537:
1538: \smallskip
1539:
1540: \noindent {\bf Facet 6:}
1541: %e_{0000}+e_{0001}+e_{0010}+e_{0011}+e_{0100}+e_{0110}+e_{1000}+e_{1001}\geq 4.
1542: If $ w = e_{0000}+e_{0001}+e_{0010}+e_{0011}
1543: +e_{0100}+e_{0110}+e_{1000}+e_{1001}$ then
1544: ${\rm in}_w(D_{2222})$ has $404$ terms and factors as a monomial
1545: times the product of the $A$-discriminants of the four maximal cells:
1546: \begin{eqnarray*}
1547: & \,\,\,\,\,{\rm in}_w(D_{2222}) \quad = \quad c_{0000} \cdot
1548: c_{1100} \cdot
1549: c_{1010}^4 \cdot
1550: c_{0101}^4 \cdot
1551: \qquad \qquad \qquad \qquad \qquad
1552: \\ &
1553: (c_{0101} c_{1010} c_{1111} - c_{0101} c_{1011} c_{1110}
1554: -c_{0111} c_{1010} c_{1101} + c_{0111} c_{1011} c_{1100})
1555: \\ &
1556: (c_{0000} c_{0111} c_{1011} - c_{0001} c_{0111} c_{1010}
1557: - c_{0010} c_{0101} c_{1011} + c_{0011} c_{0101} c_{1010})
1558: \\ & \!\!\!\!
1559: (c_{0000} c_{0101} c_{0111} c_{1110}
1560: -c_{0000} c_{0111}^2 c_{1100}-c_{0010} c_{0101}^2 c_{1110} \\ & \qquad \qquad
1561: +c_{0010} c_{0101} c_{0111} c_{1100}
1562: -c_{0100} c_{0101} c_{0111} c_{1010}
1563: +c_{0101}^2 c_{0110} c_{1010} )
1564: \\ & \!\!\!\!
1565: (c_{0000} c_{1010} c_{1011} c_{1101}
1566: -c_{0000} c_{1011}^2 c_{1100}
1567: -c_{0001} c_{1010}^2 c_{1101} \\ & \qquad \qquad
1568: +c_{0001} c_{1010}c_{1011} c_{1100}
1569: -c_{0101} c_{1000} c_{1010} c_{1011}
1570: +c_{0101} c_{1001} c_{1010}^2).
1571: \end{eqnarray*}
1572:
1573: \smallskip
1574: \noindent {\bf Facet 7:}
1575: If $w = - e_{0000}-e_{0001}-e_{0010}-e_{0011}-e_{0100}-e_{1000}-e_{1100}$ then
1576: \begin{eqnarray*}
1577: & {\rm in}_w(D_{2222}) \quad = \quad
1578: c_{1111} \cdot
1579: c_{0011}^4 \cdot
1580: c_{1100}^4 \cdot
1581: \qquad \qquad \qquad \qquad \qquad \qquad \qquad
1582: \\ &
1583: (c_{0000} c_{0011} c_{1100}
1584: -c_{0001} c_{0010} c_{1100}
1585: -c_{0011} c_{0100} c_{1000})
1586: \\ &
1587: (c_{0010} c_{1000} c_{1111}
1588: -c_{0010} c_{1011} c_{1100}
1589: -c_{0011} c_{1000} c_{1110}
1590: +c_{0011} c_{1010} c_{1100})
1591: \\ &
1592: (c_{0010} c_{0100} c_{1111}
1593: -c_{0010} c_{0111} c_{1100}
1594: -c_{0011} c_{0100} c_{1110}
1595: +c_{0011} c_{0110} c_{1100})
1596: \\ &
1597: (c_{0001} c_{1000} c_{1111}
1598: -c_{0001} c_{1011} c_{1100}
1599: -c_{0011} c_{1000} c_{1101}
1600: +c_{0011} c_{1001} c_{1100})
1601: \\ &
1602: (c_{0001} c_{0100} c_{1111}
1603: -c_{0001} c_{0111} c_{1100}
1604: -c_{0011} c_{0100} c_{1101}
1605: +c_{0011} c_{0101} c_{1100}).
1606: \end{eqnarray*}
1607:
1608:
1609: \smallskip
1610: \noindent {\bf Facet 8:}
1611: % $$2 \cdot e_{0000}+e_{0001}+e_{0010}+e_{0100}+ e_{1000} \leq 18. $$
1612: If $w = - 2 \cdot e_{0000}-e_{0001}-e_{0010}-e_{0100}- e_{1000} $ then
1613: ${\rm in}_w(D_{2222})$ equals
1614: \begin{eqnarray*}
1615: c_{0000}^3 \cdot
1616: c_{1111}^3 & \!\!\!\! \! \cdot \,
1617: (c_{0000} c_{1100} c_{1111}
1618: -c_{0000} c_{1101} c_{1110}-c_{0100} c_{1000} c_{1111}) \\ &
1619: (c_{0000} c_{1010} c_{1111}
1620: -c_{0000} c_{1011} c_{1110}-c_{0010} c_{1000} c_{1111}) \\ &
1621: (c_{0000} c_{0110} c_{1111}
1622: -c_{0000} c_{0111} c_{1110}-c_{0010} c_{0100} c_{1111}) \\ &
1623: (c_{0000} c_{1001} c_{1111}
1624: -c_{0000} c_{1011} c_{1101}-c_{0001} c_{1000} c_{1111}) \\ &
1625: (c_{0000} c_{0101} c_{1111}
1626: -c_{0000} c_{0111} c_{1101}-c_{0001} c_{0100} c_{1111}) \\&
1627: (c_{0000} c_{0011} c_{1111}
1628: -c_{0000} c_{0111} c_{1011}-c_{0001} c_{0010} c_{1111}) .
1629: \end{eqnarray*}
1630: \end{proposition}
1631:
1632: These formulas show that $A$-discriminants for
1633: subconfigurations of the cube appear
1634: naturally as irreducible factors of
1635: leading forms of the hyperdeterminant.
1636: This suggests the general problem of
1637: studying $A$-discriminants and the related
1638: tight spans for various families of $0/1$-polytopes.
1639: Other natural classes of configurations
1640: for which a study of $A$-discriminants would be
1641: interesting include generalized permutohedra,
1642: Birkhoff polytopes, and reflexive polytopes.
1643:
1644:
1645: \section{Vertices of the secondary polytope}
1646:
1647: The $4$-cube is given by the $5 \times 16$-matrix
1648: $A$ in the Introduction. For any generic vector
1649: $w \in \RR^{16}$ we consider the tight span of $(A,w)$
1650: whose cells are the bounded faces of the simple polyhedron
1651: $\,P_w= \{ u \in \RR^5 : u \cdot A \leq w \}$.
1652: Each vertex of $P_w$ is indexed by five columns of $A$.
1653: This collection of $5$-tuples,
1654: each regarded as a $4$-simplex, is
1655: a {\em regular triangulation} $\Pi_w$ of the $4$-cube
1656: (cf.~\cite[\S 5]{DRS}).
1657:
1658: The regular triangulations of the 4-cube are in bijection
1659: with the vertices of the secondary polytope.
1660: As in Section 2, we define the secondary polytope as
1661: the Newton polytope $\N(E_{2222})$ of the {\em principal determinant} $E_{2222}$,
1662: which is the following product of determinants associated to all
1663: faces of the $4$-cube:
1664: \begin{itemize}
1665: \item $16$ linear factors $c_{ijkl}$ corresponding to the vertices,
1666: \item $24$ quadratic factors $D_{22}$ corresponding to the $2$-faces,
1667: \item $8$ quartic factors $D_{222}$ corresponding to the facets,
1668: \item the hyperdeterminant $D_{2222}$ corresponding to the solid $4$-cube.
1669: \end{itemize}
1670: Thus the principal determinant $E_{2222}$ is a polynomial
1671: in the $16$ unknowns $c_{ijkl}$. of degree
1672: $\,120 = 16 \cdot 1 + 24 \cdot 2 + 8 \cdot 4 + 1 \cdot 24 $.
1673: The vertices of its Newton polytope $\N(E_{2222})$
1674: are the GKZ vectors of the regular triangulations
1675: of the 4-cube.
1676:
1677:
1678: Recall from the first part of Theorem 2 that the 4-cube has
1679: $87959448$ regular triangulations, or, equivalently,
1680: $\N(E_{2222})$ has $87959448$ vertices. Up to
1681: the action of the Weyl group $B_4$, there are
1682: $235277$ types of regular triangulations. Thus,
1683: this is the number of combinatorially distinct
1684: simple polyhedra $P_w$, and also
1685: the number of generic tight spans.
1686:
1687:
1688: \begin{figure}
1689: \vskip-0.5in
1690: %\vskip2in
1691: \begin{tabular}{ccc}
1692: \begin{tabular}{c}
1693: \hskip-0.7in \includegraphics[scale=0.35]{staircase.eps}
1694: \end{tabular}
1695: &
1696: \begin{tabular}{c}
1697: \hskip-1.6in \includegraphics[scale=0.45]{minimal.eps}
1698: \end{tabular}
1699: &
1700: \begin{tabular}{c}
1701: \hskip-1.8in \includegraphics[scale=0.4]{trun3cube.eps}
1702: \end{tabular}
1703: \end{tabular}
1704: \vskip-0.7in
1705: \caption{Tight spans of the three triangulations
1706: in Example \ref{mostsymmetric}}
1707: \label{triangpic1}
1708: \end{figure}
1709:
1710:
1711: The main objective of this section is to provide
1712: a detailed study of the regular triangulations of the
1713: $4$-cube, with emphasis on their tight spans.
1714: We present a range of results
1715: which were computed from our data.
1716:
1717: \begin{corollary}
1718: The distribution of the $235277$ symmetry classes of
1719: vertices of the secondary polytope $\N(E_{2222})$
1720: according to orbit size is as follows:
1721: $$
1722: \begin{array}{cccccccccccc}
1723: 8 & 16 & 24 & 32 & 48 & 64 & 96 & 192 & 384 \\
1724: \underline{3} & 7 & 2 & 13 & 48 & 102 & 516 & 11357 & 223229 \\
1725: \end{array}
1726: $$
1727: \end{corollary}
1728:
1729:
1730:
1731: \begin{table}
1732: \begin{tabular}{|c|ccc|c|}
1733: \hline
1734: \# maximal
1735: & \multicolumn{3}{c|}{\# simplices of}
1736: & {\begin{tabular}{c}\# $B_4$-orbits of\end{tabular}} \\
1737: simplices & volume 1 & volume 2 & volume 3 & triangulations\\
1738: \hline
1739: 16&8&8&0&1\\ \hline
1740: 17&10&7&0&1\\ \hline
1741: 18&12&6&0&6\\
1742: 18&13&4&1&11\\ \hline
1743: 19&14&5&0&25\\
1744: 19&15&3&1&48\\ \hline
1745: 20&16&4&0&628\\
1746: 20&17&2&1&344\\ \hline
1747: 21&18&3&0&5847\\
1748: 21&19&1&1&1263\\ \hline
1749: 22&20&2&0&24499\\
1750: 22&21&0&1&1967\\ \hline
1751: 23&22&1&0&48648\\ \hline
1752: 24&24&0&0&151989\\
1753: \hline
1754: \end{tabular}
1755: \medskip
1756: \caption{Classification of regular triangulations
1757: of the 4-cube according to the number and volumes of maximal simplices}
1758: \label{tab:Nsimplices}
1759: \end{table}
1760:
1761:
1762: \begin{example}[\bf The most symmetric triangulations]
1763: \label{mostsymmetric}
1764: One of the triangulations with orbit size $8$
1765: is the {\em staircase triangulation}. Its
1766: tight span is a solid permutohedron, and
1767: its GKZ vector is $(4,6,6,24,6,4,4,$ $6,6,4,4,6,24,6,6,4)$.
1768: Another is the unique triangulation that
1769: uses 16 maximal simplices.
1770: Eight simplices have volume one and
1771: eight have volume two. Its GKZ vector is
1772: $(1,12,12,1,12,1,1,20,20,1,1,$ $12,1,12,12,1)$, and
1773: its tight span is a $3$-cube with tentacles attached
1774: to its vertices. The third triangulation with orbit size
1775: $8$ has a solid truncated $3$-cube as its tight span.
1776: Its GKZ vector is $(3,8,8,3,8,3,3,24,24,3,3,8,3,8,8,3)$.
1777: The tight spans of these symmetric triangulations
1778: are shown in
1779: Figure \ref{triangpic1}. \qed
1780: \end{example}
1781:
1782: Table 3 classifies the regular
1783: triangulations according to the number of maximal simplices used.
1784: It is well-known that the minimal number is sixteen (cf.~\cite{Cot}).
1785: We see that any triangulation can only use at most one simplex of volume
1786: three, since all such simplices contain the centroid of the cube
1787: as an interior point. There are $16$ simplices of volume three,
1788: all of which are $B_4$-equivalent; a representative is $\,\{(1000),(1111),(0011),(0101), (0110)\}$.
1789:
1790: \begin{example}[\bf Triangulating the 4-cube with 17 simplices]
1791: \label{sweet17}
1792: Table 3 shows that there is a unique regular triangulation with
1793: $17$ maximal simplices, ten of volume one and seven of volume two.
1794: Its tight span, shown in
1795: Figure \ref{triangpic2},
1796: is a cube with a truncated vertex and tentacles
1797: attached to the seven original vertices.
1798: Its GKZ vector is $(1,11,12,1,12,1,1,21,20,3,1,11,1,11,12,1)$. \qed
1799: \end{example}
1800:
1801: \begin{table}
1802: \begin{tabular}{|c|c|c|c|c|c|c|c|}
1803: \hline
1804: & \{3\} & \{3,2\} & \{3,1\} &\{3,2,1\} &\{2\} &\{2,1\} & total\\
1805: \hline
1806: (16, 20, 6, 1) & & & 1 & & & & 1 \\
1807: \hline
1808: (17, 22, 7, 1) & & & 1 & & & &1 \\
1809: \hline
1810: (18, 24, 8, 1) & & & 4 &2 & & & 6 \\
1811: (18, 23, 6, 0) & & & & & & 11 & 11 \\
1812: \hline
1813: (19, 26, 9, 1) & & & 6 & 19 & & & 25 \\
1814: (19, 25, 7, 0) & & & & & & 48 & 48 \\
1815: \hline
1816: (20, 28, 10, 1) & & 1 & 23 & 209 & & & 233 \\
1817: (20, 27, 8, 0) & & & & & 24 & 715 & 739 \\
1818: \hline
1819: (21, 30, 11, 1) & & 5 & 34 & 1372 & & & 1411 \\
1820: (21, 29, 9, 0) & & & & & 392 & 5307 & 5699\\
1821: \hline
1822: (22, 32, 12, 1) & & 112 & 84 & 9342 & & & 9538\\
1823: (22, 31, 10, 0) & & & & & 2156 & 14772 & 16928\\
1824: \hline
1825: (23, 34, 13, 1) & & 2116 & 100 & 46432 & & & 48648\\
1826: \hline
1827: (24, 36, 14, 1) & 125 & 27054 & & 124810 & & & 151989\\
1828: \hline
1829: total & 125 & 29288 & 253 & 182186 & 2572 & 20853 & 235277\\
1830: \hline
1831: \end{tabular}
1832: \medskip
1833: \caption{Distribution of the 235277 types of triangulations. The rows correspond to the $f$-vectors and the columns correspond to the dimensions of maximal faces of the tight span.}
1834: \label{tab:TSfvector}
1835: \end{table}
1836:
1837: Table \ref{tab:TSfvector} classifies the tight spans
1838: according to their $f$-vectors and their {\em signature},
1839: by which we mean the set of dimensions of the maximal cells.
1840: Note that every triangulation of the 4-cube has at most one interior
1841: edge (namely, a diagonal), so each tight span has
1842: at most one $3$-dimensional cell.
1843: The $f$-vector of a tight span is determined
1844: by the number of vertices and 3-cells, i.e.,
1845: the number of maximal simplices and diagonals used in the triangulation.
1846:
1847: \begin{example}[\bf The smallest tight spans of
1848: signature $\{3,2\}$ and $\{3,2,1\}$]
1849: \label{smallesttight}
1850: The unique tight span of signature $\{3,2\}$ with
1851: $20$ vertices contains a solid tetrahedron with six hexagons attached, one to each of the tetrahedron's edges. See Figure \ref{triangpic2}. Its triangulation has
1852: GKZ vector $\,(5,2,2,15,2,15,15,4,24,5,5,2,5,2,$ $2,15)\,$
1853: and lies in the 88th $D$-equivalence class.
1854:
1855: There are two tight spans of signature
1856: $\{3,2,1\}$ with $18$ vertices. Their GKZ vectors are
1857: $( 1,9,13,2,12,1,1,21,19,6,1,9,1,11,12,1)$ and
1858: $(1,12,12,1,$ $11,1,1,21,19,1,1,13,5,10,10,1)$.
1859: They are both in the largest $D$-equivalence class.
1860: One of them is shown in Figure \ref{triangpic2}. The other has a triangle and a tentacle in place of the rectangular maximal 2-face.
1861: \qed
1862: \end{example}
1863:
1864:
1865: \begin{figure}
1866: \vskip-0.5in
1867: \begin{tabular}{ccc}
1868: \hskip-0.9in \includegraphics[scale=0.4]{seventeen.eps}&
1869: \hskip-1.3in \includegraphics[scale=0.35]{20V32.eps}&
1870: \hskip-1in \includegraphics[scale=0.35]{18V321a.eps}
1871: \end{tabular}
1872: \vskip-0.5in
1873: \caption{The triangulations
1874: in Examples \ref{sweet17} and \ref{smallesttight}}
1875: \label{triangpic2}
1876: \end{figure}
1877:
1878:
1879: The first column in Table \ref{tab:TSfvector} is
1880: especially interesting as it ties in
1881: with an active area of research in
1882: combinatorial commutative algebra. Using the
1883: language introduced in \cite{CHT}, the $d$-cube is a
1884: {\em Gorenstein polytope},
1885: and the tight spans of signature $\{d-1\}$ are precisely its
1886: {\em Gorenstein triangulations}.
1887: The 3-cube has only one Gorenstein
1888: triangulation, corresponding to the last type in
1889: Figure~\ref{cubetriangs}.
1890:
1891: \begin{example}[\bf The Gorenstein triangulations of the 4-cube]
1892: \label{gorenstein}
1893: The 4-cube has precisely $125$ types of Gorenstein
1894: triangulations. Two of them are pictured
1895: in Figure \ref{triangpic1}.
1896: The Stanley-Reisner ideal of a Gorenstein
1897: triangulation corresponds to an initial ideal of the Segre variety
1898: $\PP^1 \times\PP^1 \times \PP^1 \times \PP^1 \subset \PP^{15}$
1899: that is squarefree and Gorenstein. The minimal free
1900: resolution of the Alexander dual of this monomial ideal
1901: is cellular and is supported on the tight span itself.
1902: For tropical polytopes (Example \ref{ex1}) this was
1903: shown in \cite{BY}, but the same holds
1904: for unimodular regular triangulations of arbitrary
1905: polytopes, including the 4-cube.
1906: All $125$ Gorenstein tight spans
1907: are {\em deformed permutohedra}, and it would
1908: be interesting to study their combinatorics
1909: in arbitrary dimensions $d \geq 4$.
1910: \qed
1911: \end{example}
1912:
1913:
1914: \begin{table}
1915: \begin{equation*}
1916: \begin{array}{rrrrrrrrrrrrrrrr}
1917: \multicolumn{4}{l}{87959448} &&&&&&&&&&&&\\
1918: =& 32&\!\cdot\!\!& 2116 &+& 192&\!\cdot\!\!& 644 &+& 192&\!\cdot\!\!& 2752
1919: &+&
1920: 64&\!\cdot\!\!& 30390\\
1921: +& 384 &\!\cdot\!\!& 1550 &+& 192 &\!\cdot\!\!& 368 &+& 384 &\!\cdot\!\!&
1922: 3742 &+&
1923: 384 &\!\cdot\!\!& 716 \\
1924: +& 192 &\!\cdot\!\!& 10384 &+& 192 &\!\cdot\!\!& 20552 &+& 96
1925: &\!\cdot\!\!& 1444 &+&
1926: 384 &\!\cdot\!\!& 380 \\
1927: +& 192 &\!\cdot\!\!& 4408 &+& 384 &\!\cdot\!\!& 1318 &+& 384 &\!\cdot\!\!&
1928: 940 &+&
1929: 192 &\!\cdot\!\!& 468 \\
1930: +& 192 &\!\cdot\!\!& 2584 &+& 192 &\!\cdot\!\!& 18050 &+& 384
1931: &\!\cdot\!\!& 4122 &+&
1932: 384 &\!\cdot\!\!& 11614 \\
1933: +& 192 &\!\cdot\!\!& 9090 &+& 192 &\!\cdot\!\!& 5276 &+& 96 &\!\cdot\!\!&
1934: 2440 &+&
1935: 96 &\!\cdot\!\!& 22788 \\
1936: +& 384 &\!\cdot\!\!& 3182 &+& 384 &\!\cdot\!\!& 1358 &+& 192 &\!\cdot\!\!&
1937: 1444 &+&
1938: 384 &\!\cdot\!\!& 760 \\
1939: +& 192 &\!\cdot\!\!& 5140 &+& 384 &\!\cdot\!\!& 940 &+& 384 &\!\cdot\!\!&
1940: 4068 &+&
1941: 384 &\!\cdot\!\!& 2810 \\
1942: +& 192 &\!\cdot\!\!& 7956 &+& 192 &\!\cdot\!\!& 15260 &+& 384
1943: &\!\cdot\!\!& 1318 &+&
1944: 384 &\!\cdot\!\!& 3278\\
1945: +& 192 &\!\cdot\!\!& \underline{18964} &+& 96 &\!\cdot\!\!& \underline{18340} &+& 96
1946: &\!\cdot\!\!& \underline{21192}
1947: &+& 96 &\!\cdot\!\!& 196 \\
1948: +& 192 &\!\cdot\!\!& 392 &+& 384 &\!\cdot\!\!& 1616 &+& 192 &\!\cdot\!\!&
1949: 1428 &+&
1950: 192 &\!\cdot\!\!& 18042 \\
1951: +& 384 &\!\cdot\!\!& 454 &+& 384 &\!\cdot\!\!& 336 &+& 384 &\!\cdot\!\!&
1952: 112 &+&
1953: 384 &\!\cdot\!\!& 2112 \\
1954: +& 384 &\!\cdot\!\!& 780 &+& 192 &\!\cdot\!\!& 212 &+& 384 &\!\cdot\!\!&
1955: 336 &+&
1956: 192 &\!\cdot\!\!& 4624 \\
1957: +& 96 &\!\cdot\!\!& 100 &+& 192 &\!\cdot\!\!& 200 &+& 384 &\!\cdot\!\!&
1958: 1360 &+&
1959: 192 &\!\cdot\!\!& 338\\
1960: +& 384 &\!\cdot\!\!& 240 &+& 384 &\!\cdot\!\!& 288 &+& 384 &\!\cdot\!\!&
1961: 200 &+&
1962: 384 &\!\cdot\!\!& 96\\
1963: +& 192 &\!\cdot\!\!& 8770 &+& 384 &\!\cdot\!\!& 5852 &+& 384
1964: &\!\cdot\!\!& 2180
1965: &+& 384 &\!\cdot\!\!& 1692 \\
1966: +& 192 &\!\cdot\!\!& 1490 &+& 96 &\!\cdot\!\!& 392 &+& 48 &\!\cdot\!\!&
1967: 3344 &+&
1968: 384 &\!\cdot\!\!& 908 \\
1969: +& 192 &\!\cdot\!\!& 4056 &+& 384 &\!\cdot\!\!& 1392 &+& 96
1970: &\!\cdot\!\!& 1152 &+&
1971: 192 &\!\cdot\!\!& 576 \\
1972: +& 384 &\!\cdot\!\!& 232 &+& 384 &\!\cdot\!\!& 480 &+& 96 &\!\cdot\!\!&
1973: 900 &+&
1974: 192 &\!\cdot\!\!& 1690 \\
1975: +& 384 &\!\cdot\!\!& 1096 &+& 384 &\!\cdot\!\!& 956 &+& 192
1976: &\!\cdot\!\!& 1508 &+&
1977: 192 &\!\cdot\!\!& 1856 \\
1978: +& 384 &\!\cdot\!\!& 280 &+& 192 &\!\cdot\!\!& 676 &+& 96 &\!\cdot\!\!&
1979: 2304 &+&
1980: 384 &\!\cdot\!\!& 676 \\
1981: +& 192 &\!\cdot\!\!& 676 &+& 96 &\!\cdot\!\!& 2304 &+& 48 &\!\cdot\!\!&
1982: 1152 &+&
1983: 16 &\!\cdot\!\!& 141888 \\
1984: +& 192 &\!\cdot\!\!& 4600 &+& 96 &\!\cdot\!\!& 2888 &+& 192
1985: &\!\cdot\!\!& 2700 &+&
1986: 64 &\!\cdot\!\!& 23584\\
1987: +& 192 &\!\cdot\!\!& 5140 &+& 192 &\!\cdot\!\!& 112 &+& 192
1988: &\!\cdot\!\!& 424 &+&
1989: 64 &\!\cdot\!\!& 6570 \\
1990: +& 96 &\!\cdot\!\!& 400 &+& 192 &\!\cdot\!\!& 200 &+& 192 &\!\cdot\!\!&
1991: 580 &+& 96
1992: &\!\cdot\!\!& 18340 \\
1993: +& 16 &\!\cdot\!\!& \underline{82832} &+& 48 &\!\cdot\!\!& 11764 &+& 48
1994: &\!\cdot\!\!& 6200 &+& 96 &\!\cdot\!\!& 58150 \\
1995: +& 192 &\!\cdot\!\!& 4860 &+& 192 &\!\cdot\!\!& 96 &+& 192 &\!\cdot\!\!&
1996: 288 &+&
1997: 192 &\!\cdot\!\!& 352 \\
1998: +& 192 &\!\cdot\!\!& 1240 &+& 8 &\!\cdot\!\!& \underline{349555}&+& 32
1999: &\!\cdot\!\!& \underline{64}&\\
2000: \end{array}
2001: \end{equation*}
2002: \caption{Decomposition into $D$-equivalence classes}
2003: \label{decompose}
2004: \end{table}
2005:
2006: We now return to our discussion of the hyperdeterminant $D_{2222}$.
2007: Since $D_{2222}$ is a factor
2008: of the principal determinant $E_{2222}$,
2009: there is a natural many-to-one map from the vertices
2010: of $\N(E_{2222})$ to the vertices of $\N(D_{2222})$.
2011: Two regular triangulations are $D$-equivalent if they are
2012: mapped to the same vertex of $\N(D_{2222})$.
2013: For the 3-cube the $D$-equivalence classes
2014: were described in Section~2.
2015: Table \ref{decompose} shows the decomposition of
2016: the number $87959448$ of regular triangulations of the
2017: 4-cube into the cardinalities of the orbits of $D$-equivalence classes.
2018: Thus the identity expressed in Table \ref{decompose} is the 4-cube version
2019: of the identity (\ref{seventyfour}) for the 3-cube. The $111$ orbits of $D$-equivalence classes are listed in the order of the corresponding orbits of vertices of $\N(D_{2222})$ in Section 4.
2020:
2021: In the triangulations of the 4-cube, the centroid $(1,\frac{1}{2},\frac{1}{2},\frac{1}{2},\frac{1}{2})$ is in the relative interior of one of the following three kinds of simplices: a 4-simplex of normalized volume 3, a tetrahedron of normalized volume 2, or a diagonal.
2022:
2023: \begin{example}[\bf The ``coefficient $-27$ class'']
2024: \label{fatclass}
2025: The triangulations which use a fixed simplex of normalized volume
2026: three form a single $D$-equivalence class
2027: of cardinality $82832$.
2028: This class corresponds to the vertex
2029: $ 0 0 0 4 0 4 4 0 8 0 0 0 0 0 0 4^{-27}_{16} $
2030: of $\N(D_{2222})$. The corresponding
2031: monomial in $D_{2222}$ has the largest absolute value of any coefficient,
2032: namely $-27$, among all the extreme monomials. \qed
2033: \end{example}
2034:
2035: \begin{example}[\bf The ``coefficient $\pm 16$ classes'']
2036: \label{coeff16}
2037: The triangulations in which the centroid is in a fixed tetrahedron of volume two form a $D$-equivalence class corresponding to one of the extreme monomials of $D_{2222}$ with coefficient $\pm 16$, which come in three symmetry classes underlined in Table \ref{decompose} and on page \pageref{Dvertices}. There are 24 such tetrahedra, all lying in a single $B_4$ orbit. The hyperplane spanned by the special tetrahedron contains eight vertices of the 4-cube, and there are four vertices on each side. In each triangulation, one vertex from each side is joined to the special tetrahedron. The Hamming distance between these two vertices can be two, three, or four, and they correspond precisely to the three $D$-equivalence classes with coefficient $\pm 16$. The first underlined class corresponds to Hamming distance three where there are two choices on one side after the vertex on the other side is chosen. Hence it has orbit size $24 \times 4 \times 2 = 192$. The second and third underlined classes correspond to Hamming distances four and two respectively and have orbit size $24 \times 4 = 96$. \qed
2038: \end{example}
2039:
2040: The other $D$-equivalence classes contain only triangulations using a diagonal of the 4-cube, and all those classes contain some unimodular triangulations.
2041:
2042: \begin{example}[\bf The smallest $D$-equivalence class]
2043: \label{smallestclass}
2044: This class has only $64$ triangulations and corresponds to the last summand
2045: in Table \ref{decompose}.
2046: The tight span of each of these 64 triangulations contains
2047: a hexagonal prism and six lower-dimensional pieces attached
2048: to alternating edges of the prism, as shown on
2049: the left in Figure \ref{triangpic3}.
2050: Each of these six lower-dimensional pieces
2051: is either a rectangle or a triangle with a tentacle,
2052: whence the number $\,2^6 = 64$. \qed
2053: \end{example}
2054:
2055: \begin{example}[\bf The largest $D$-equivalence class]
2056: \label{largestclass}
2057: This class has $349555$ triangulations and corresponds to
2058: the vertex of $\N(D_{2222})$ which has the largest vertex figure and the smallest orbit size.
2059: It contains all the triangulations in Examples
2060: \ref{mostsymmetric}, \ref{sweet17} and \ref{gorenstein}.
2061: \qed
2062: \end{example}
2063:
2064:
2065: \begin{figure}
2066: \vskip-0.3in
2067: \begin{tabular}{ccc}
2068: \hskip-0.7in \includegraphics[scale=0.35]{111D10.eps}&
2069: \hskip-1.1in \includegraphics[scale=0.35]{101D3479.eps}&
2070: \hskip-1.1in \includegraphics[scale=0.35]{mostDL1.eps}
2071: \end{tabular}
2072: \vskip-0.4in
2073: \caption{Tight spans of triangulations
2074: in Examples \ref{smallestclass}, \ref{fatclass} and \ref{delightful}}
2075: \label{triangpic3}
2076: \end{figure}
2077:
2078: Using our data, it is easy to study other questions concerning regular
2079: triangulations of the 4-cube. Answering a question which
2080: was left open in \cite{SS}, we found that the
2081: 4-cube does not possess any {\em delightful triangulations}.
2082: Such a triangulation would correspond to a square-free
2083: initial ideal of the first secant variety of
2084: $\PP^1 \times \PP^1 \times \PP^1 \times \PP^1 \subset \PP^{15}$,
2085: which has dimension $9$ and degree $64$.
2086: For every regular triangulation $\Pi$ we computed the
2087: number of $10$-element subsets of the vertices
2088: of the 4-cube which are the union of vertices of two simplices
2089: in $\Pi$. That number is bounded above by $64$,
2090: and $\Pi$ would be delightful if equality holds.
2091:
2092: \begin{example} [\bf The most delightful triangulation]
2093: \label{delightful}
2094: We found that the maximum number of
2095: $10$-sets of vertices which are the unions of vertices of two simplices
2096: in a regular triangulation is $56$.
2097: This number is attained by two triangulations which are both in
2098: $D$-equivalence classes of type 11.
2099: Their GKZ vectors are $(1,4,4,11,6,13,9,12,12,9,13,6,11,4,4,1)$ and
2100: $(1,4,4,11,6,13, 9,$ $12, 12, 11,13, 4,9,4,6,1).$ Both of the corresponding tight spans have $f$-vector $(24,36,14,1)$ and consist of two maximal edges, eight maximal 2-faces, and one 3-cube. The tight span of the first triangulation is shown in Figure \ref{triangpic3}.
2101: \qed
2102: \end{example}
2103:
2104: \smallskip
2105:
2106: \section{Facets of the secondary polytope}
2107:
2108: The facets of the secondary polytope
2109: correspond to the coarsest regular polyhedral
2110: subdivisions $\Pi$.
2111: Here {\em coarsest} means that $\Pi$ refines no other proper subdivisions.
2112: We computed all such subdivisions for the $4$-cube.
2113: This result gives an irredundant
2114: inequality representation of the secondary polytope $\N(E_{2222})$ of the 4-cube, analogous to that of the 3-cube in expression (\ref{NE222ineq}).
2115:
2116: Since the Newton polytope $\N(D_{2222})$ is a
2117: summand of the secondary polytope $\N(E_{2222})$,
2118: all facet normals of $\N(D_{2222})$ are also
2119: facet normals of $\N(E_{2222})$. Thus we already
2120: know $268$ facets of the secondary polytope from
2121: Section 4. Furthermore the linear equations
2122: $Ax = (120,60,60,60,60)^T $ hold on $\N(E_{2222})$.
2123:
2124: In order to compute the facets of the secondary polytope of the 4-cube, we first wrote a program to compute the GKZ vector of each regular triangulation of the 4-cube. The information about the triangulations themselves was obtained using TOPCOM. We then used the following method to compute the facets of the secondary polytope, which can be applied to any point configuration with symmetry.
2125:
2126: Since $\N(E_{2222})$ is an $11$-dimensional polytope with $87959448$
2127: vertices, computing its facets is currently out of scope of any
2128: general polyhedral software. Thus, to compute the facets, we took advantage of the relationship between the secondary polytope and subdivisions of the 4-cube. Namely, we
2129: know that each vertex of $\N(E_{2222})$ corresponds to a regular triangulation of the 4-cube,
2130: and, by \cite[\S 7.2.C]{GKZ}, each edge $(v,w)$ of $\N(E_{2222})$
2131: corresponds to the circuit connecting the two triangulations represented by the vertices $v$ and $w$.
2132:
2133: Note that, given any triangulation of the 4-cube, there are relatively few
2134: circuits which contain a face of the triangulation. As a result,
2135: each tangent cone of $\N(E_{2222})$ is generated by relatively few vectors.
2136: Hence we computed the facets of $\N(E_{2222})$ by equivalently computing the
2137: facets of its tangent cones. Although there are $87959448$ tangent cones, they come in $235227$ orbits under the $B_4$-action. Thus, we only
2138: computed the facets of $235227$ tangent cones; all other facets of
2139: $\N(E_{2222})$ were obtained by applying the $B_4$-action.
2140:
2141: We found that $\N(E_{2222})$ has $80876$ facets in $334$ orbits.
2142: Thus the 4-cube admits exactly $80876$ coarsest regular subdivisions.
2143: The distribution of the types of coarsest regular subdivisions according to orbit size is
2144: $$
2145: \begin{array}{cccccccccc}
2146: 8 & 12 & 16 & 24 & 32 & 48 & 64 & 96 & 192 & 384 \\
2147: 2 & 1 & 4 & 2 & 4 & 14 & 16 & 26 & 132 & 133 \\
2148: \end{array}
2149: $$
2150:
2151:
2152: \begin{example}[\bf The most symmetric coarsest regular subdivisions]
2153: The three most symmetric coarsest regular subdivisions come from
2154: facets of $\N(D_{2222})$.
2155: The two coarsest subdivisions with orbit size 8
2156: correspond to Facet 3 and Facet 8
2157: in Sections 4 and 5. Their tight spans are a
2158: segment and an octahedron respectively. The unique subdivision with orbit size 12 corresponds to Facet 2 in Sections 4 and 5.
2159: Its tight span is also a line segment.
2160: \qed
2161: \end{example}
2162:
2163: The distribution of the orbits according to the number of maximal cells in the subdivision and the dimensions of maximal faces of the tight span are shown in Table \ref{NEfacets}.
2164: A coarsest subdivision of the 4-cube can have up to 13 maximal cells. Here it is no longer the case that the number of vertices and 3-faces determine the $f$-vector of the tight span. There are 33 types of $f$-vectors.
2165:
2166: \begin{table}
2167: \begin{tabular}{|c|c|c|c|c|c|}
2168: \hline
2169: & \{3\} & \{3,2\} & \{2\} & \{1\} & total \\
2170: \hline
2171: 2& & & &4& 4\\
2172: 3& & &5& & 5\\
2173: 4&3& &7& & 10\\
2174: 5&4&3&17& & 24\\
2175: 6&5&14&12& & 31\\
2176: 7&13&21&39& & 73\\
2177: 8&7&31&51& & 89\\
2178: 9&7&26&24& & 57\\
2179: 10&2&14&21& & 37\\
2180: 11&1&1& & & 2\\
2181: 12&1& & & & 1 \\
2182: 13&1& & & & 1 \\
2183: \hline
2184: total & 44 & 176 & 110 & 4 & 334\\
2185: \hline
2186: \end{tabular}
2187: \vskip .2cm
2188: \caption{Distribution of the 334 types of coarsest regular subdivisions.
2189: The rows correspond to the number of maximal cells
2190: and the columns correspond to the
2191: signatures of the tight span.}
2192: \label{NEfacets}
2193: \end{table}
2194:
2195:
2196: \begin{figure}
2197: \vskip-0.5in
2198: \begin{tabular}{ccc}
2199: \begin{tabular}{c}
2200: \includegraphics[scale=0.55]{4thsplit.eps}
2201: \end{tabular}
2202: &
2203: \hskip-0.2in
2204: \begin{tabular}{c}
2205: \includegraphics[scale=0.3]{largeCoarse.eps}
2206: \end{tabular}
2207: &
2208: \hskip-0.8in
2209: \begin{tabular}{c}
2210: \includegraphics[scale=0.35]{11V32.eps} \\
2211: \end{tabular}
2212: \end{tabular}
2213: \vskip-0.3in
2214: \caption{The tight spans for Examples \ref
2215: {split4}, \ref{largestcoarse}, and \ref{nonpurecoarse}}
2216: \label{coarsefig}
2217: \end{figure}
2218:
2219:
2220: \begin{example}[\bf The missing split]
2221: \label{split4}
2222: The 4-cube admits 4 splits, i.e., subdivisions obtained by slicing
2223: the 4-cube with a hyperplane, whose tight spans are line segments. Three of them correspond to Facets 1, 2, and 3 of $\N(D_{2222})$. The remaining split subdivides the 4-cube into 2 cells with 8 and 14 vertices each, having normalized volumes 4 and 20 respectively. The corresponding facets of $\N(E_{2222})$ are given by inequalities of the form
2224: $\,
2225: x_{0000} + x_{0001} \geq 5 $.
2226: \qed
2227: \end{example}
2228:
2229: \begin{example}[\bf The largest tight span]
2230: \label{largestcoarse}
2231: There is a unique coarsest regular subdivision with 13 maximal cells. Its tight span is a 3-dimensional polytope with $f$-vector $(13,24,13,1)$, as shown in Figure \ref{coarsefig}. The corresponding facets are given by inequalities
2232: like $\, x_{0000}+ x_{1111} - x_{0011} \leq 44 $.
2233: \qed
2234: \end{example}
2235:
2236: \begin{example}[\bf The largest tight span with non-pure dimension]
2237: \label{nonpurecoarse}
2238: Almost half of the coarsest regular subdivisions have a pure dimensional tight span. The largest non-pure dimensional tight span has $f$-vector $(11,20,11,1)$. It contains one 3-dimensional face and two maximal 2-faces. A figure is depicted in Figure \ref{coarsefig}.
2239: The corresponding facets are given by inequalities like
2240: $$
2241: x_{0000}+x_{1010}+x_{1111} - x_{1000} - x_{1001} - x_{1011} \leq 47 .
2242: $$
2243: \vskip -.5cm
2244: \qed
2245: \end{example}
2246:
2247: All the 334 facet inequalities and all the data we have discussed in this paper are available on our website
2248: \url{bio.math.berkeley.edu/4cube/}.
2249:
2250:
2251: \bigskip
2252: \bigskip
2253:
2254: \noindent {\bf Acknowledgements:}
2255: This work began during Debbie Yuster's
2256: Fall 2005 visit to the UC Berkeley Mathematics Department.
2257: Peter Huggins was supported by an ARCS Foundation Fellowship.
2258: Bernd Sturmfels was partially supported by the National Science
2259: Foundation (DMS-0456960), and
2260: Josephine Yu was supported by an NSF
2261: Graduate Research Fellowship. We also thank Francisco Santos for helpful discussions and comments.
2262:
2263: \begin{thebibliography}{10}
2264:
2265:
2266: \bibitem{AR} E.~Allman and J.~Rhodes:
2267: Phylogenetic invariants for stationary base composition,
2268: {\em Journal Symbolic Computation} {\bf 41} (2006) 138--150.
2269:
2270: \bibitem{BY} F.~Block and J.~Yu:
2271: Tropical convexity via cellular resolutions,
2272: {\em Journal of Algebraic Combinatorics} {\bf 24} (2006) 103--114.
2273:
2274: \bibitem{Cot} R.~W.~Cottle:
2275: Minimal triangulation of the 4-cube, {\em Discrete Math.}
2276: {\bf 40} (1982) 25--29.
2277:
2278: \bibitem{CHT} A.~Conca, S.~Hosten and R.~Thomas:
2279: Nice initial complexes of some classical ideals,
2280: {\tt ArXiv:math.AC/0512283}.
2281:
2282: \bibitem{CLO} D.~Cox, J.~Little and D.~O'Shea:
2283: {\em Using Algebraic Geometry},
2284: Graduate Texts in Mathematics {\bf 185}, second edition,
2285: Springer Verlag, New York, 2005.
2286:
2287: \bibitem{DeL} J.~De Loera:
2288: Nonregular triangulations of products of simplices,
2289: {\em Discrete Comput. Geom.} {\bf 15} (1996) 253--264
2290:
2291: \bibitem{DSL} V.~de Silva and L.-H.~Lim:
2292: Tensor rank and the ill-posedness of the best low-rank approximation problem,
2293: {\tt ArXiv:math.NA/0607647}.
2294:
2295: \bibitem{DFS} A.~Dickenstein, E.M.~Feichtner and
2296: B.~Sturmfels: Tropical discriminants, \hfill \break
2297: {\tt arXiV:math.AG/0510126}.
2298:
2299: \bibitem{DS} M.~Develin and B.~Sturmfels: Tropical convexity,
2300: {\em Documenta Math.} {\bf 9} (2004) 1-27.
2301:
2302: \bibitem{DRS} J.~De Loera, J.~Rambau and F.~Santos:
2303: {\em Triangulations: Applications, Structures and Algorithms},
2304: Algorithms and Computation in Mathematics, Springer Verlag, Heidelberg,
2305: to appear.
2306:
2307: \bibitem{GJ} E.~Gawrilow and M.~Joswig: Polymake: a framework for
2308: analyzing convex
2309: polytopes,
2310: in {\em Polytopes --- Combinatorics and Computation},
2311: eds.~G.~Kalai and G.M.~Ziegler,
2312: Birkh\"auser, 2000, pp.~43--74.
2313:
2314:
2315: \bibitem{GKZ} I.M.~Gelfand, M.M.~Kapranov, A.V.~Zelevinsky:
2316: {\em Discriminants, Resultants, and Multidimensional Determinants\/};
2317: Birkh\"auser, Boston, MA, 1994.
2318:
2319: \bibitem{Hir} H.~Hirai: A geometric study of the split decomposition,
2320: Discrete and Computational Geometry, 2006, in press.
2321:
2322: \bibitem{Lev} P. L\'evay:
2323: On the geometry of a class of N-qubit entanglement monotones,
2324: {\em Journal of Physics A} {\bf 38} (2005) 9075--9085.
2325:
2326: \bibitem{LT}
2327: J.-G. Luque and J.-Y. Thibon:
2328: The polynomial invariants of four qubits,
2329: {\em Phys. Rev.} A 67, 042303 (2003).
2330:
2331: \bibitem{PR} J.~Pfeifle and J.~Rambau:
2332: Computing triangulations using oriented matroids,
2333: in {\em Algebra, Geometry and Software Systems} (M.~Joswig and N.~Takayama, eds.)
2334: Springer Verlag, 2003, pp.~49--76.
2335:
2336: \bibitem{Ram} J.~Rambau:
2337: TOPCOM: triangulations of point configurations and oriented matroids.
2338: {\em Mathematical Software} (Beijing, 2002),
2339: World Sci. Publishing, River Edge, NJ, 2002, 330--340.
2340:
2341: \bibitem{San}
2342: F.~Santos:
2343: The Cayley trick and triangulations of products of simplices,
2344: In {\em Integer Points in Polyhedra - Geometry, Number Theory, Algebra, Optimization},
2345: (eds. A. Barvinok, M. Beck, C. Haase, B. Reznick, and V. Welker),
2346: Contemporary Mathematics {\em 374}, Amer. Math. Soc., Providence, 2005
2347:
2348: \bibitem{SS}
2349: B.~Sturmfels and S.~Sullivant:
2350: Combinatorial secant varieties, to appear in
2351: {\em Quarterly Journal of Pure and Applied Mathematics},
2352: {\bf 2} (2006) 285-309, (Special issue: In Honor of Robert MacPherson).
2353:
2354: \bibitem{SY}
2355: B.~Sturmfels and J.~Yu:
2356: Classification of six-point metrics,
2357: {\em Electronic Journal of Combinatorics} {\bf 11} (2004/05) R44.
2358:
2359:
2360: \bibitem{SJ}
2361: J.G.~Sumner and P.D.~Jarvis:
2362: Using the tangle: a consistent construction
2363: of phylogenetic distance matrices for quartets,
2364: {\tt arXiV:q-bio.PE/0510035}.
2365:
2366: \bibitem{tBe}
2367: J.M.F.~ten Berge: Kruskal's polynomial for $2 \times 2 \times 2$-arrays and a generalization
2368: to $2 \times n \times n$-arrays, {\em Psychometrika} {\bf 58} (1991) 631--636.
2369:
2370: \end{thebibliography}
2371:
2372:
2373:
2374: \bigskip
2375:
2376: \noindent {\bf Authors' addresses:}
2377:
2378: \bigskip
2379:
2380: \noindent Peter Huggins, Bernd Sturmfels and Josephine Yu,
2381:
2382: Department of Mathematics, University of California,
2383:
2384: Berkeley, CA 94720, USA,
2385:
2386: {\tt [phuggins,bernd,jyu]@math.berkeley.edu}
2387:
2388: \medskip
2389:
2390: \noindent Debbie S. Yuster,
2391:
2392: Department of Mathematics, Columbia University,
2393:
2394: New York, NY 10027, USA,
2395:
2396: {\tt debbie@math.columbia.edu}
2397:
2398: \end{document}
2399: