1: %\documentclass[pre,superscriptaddress,,showpacs,twocolumn]{revtex4}
2: \documentclass[12pt,pre,aps,,showpacs,preprint]{revtex4}
3:
4: \usepackage{psfig}
5: \usepackage{amsfonts}
6: \begin{document}
7:
8: \title{Exactly Solvable Disordered Sphere-Packing Model
9: in Arbitrary-Dimension Euclidean Spaces}
10: %An Exactly Solvable Disordered Sphere Packing Model in Arbitrary
11: %Euclidean Dimension
12:
13:
14: \author{S. Torquato}
15:
16: \email{torquato@electron.princeton.edu}
17:
18:
19: \affiliation{\emph{Department of Chemistry}, \emph{Princeton University}, Princeton
20: NJ 08544}
21:
22: \affiliation{\emph{Program in Applied and Computational Mathematics}, \emph{Princeton
23: University}, Princeton NJ 08544}
24:
25: \affiliation{\emph{PRISM, Princeton University}, Princeton NJ 08544}
26:
27: \author{F. H. Stillinger}
28:
29: \affiliation{\emph{Department of Chemistry}, \emph{Princeton University}, Princeton
30: NJ 08544}
31:
32: \begin{abstract}
33:
34: We introduce a generalization of the well-known random sequential addition (RSA) process
35: for hard spheres in $d$-dimensional Euclidean space $\mathbb{R}^d$. We show that
36: all of the $n$-particle correlation functions
37: of this nonequilibrium model, in a certain limit called the ``ghost" RSA packing, can be obtained analytically for
38: all allowable densities and in any dimension. This represents the first exactly
39: solvable disordered sphere-packing model in arbitrary dimension.
40: The fact that the maximal density $\phi(\infty)=1/2^d$ of the ghost RSA packing implies
41: that there may be disordered sphere packings in sufficiently high $d$ whose density exceeds
42: Minkowski's lower bound for Bravais lattices, the dominant asymptotic
43: term of which is $1/2^d$. Indeed, we report on
44: a conjectural lower bound on the density whose asymptotic behavior
45: is controlled by $2^{-(0.77865\ldots) d}$, thus providing the putative
46: exponential improvement on Minkowski's 100-year-old bound.
47: Our results suggest that the densest packings in sufficiently
48: high dimensions may be disordered rather than periodic, implying
49: the existence of disordered classical ground states for some continuous potentials.
50:
51:
52: \end{abstract}
53: \pacs{05.20.-y, 61.20.-p}
54:
55:
56: \maketitle
57:
58: \section{Introduction}
59:
60: A collection of congruent spheres in $d$-dimensional Euclidean space $\mathbb{R}^d$
61: is called a sphere packing if no two spheres overlap.
62: The {\it packing density} or simply density $\phi$ of a sphere packing is the fraction of
63: space $\mathbb{R}^d$ covered by the spheres.
64: Hard-sphere packings have been used to model a variety of systems, including liquids \cite{Ha86},
65: amorphous and granular media \cite{To02a}, and crystals \cite{Chaik95}. Nonetheless,
66: there is great interest in understanding sphere packings
67: in high dimensions in various fields. For example, it is known
68: that the optimal way of sending digital signals over noisy channels correspond
69: to the densest sphere packing in a high dimensional space \cite{Co93}.
70: These ``error-correcting" codes underlie a variety of systems in digital
71: communications and storage, including compact disks, cell phones and the Internet.
72: Physicists have studied hard-sphere packings in high dimensions to gain insight
73: into ground and glassy states of matter as well as
74: phase behavior in lower dimensions \cite{Fr99,Pa00}. The determination of the densest packings
75: in arbitrary dimension is a problem of long-standing interest in discrete geometry \cite{Co93}.
76:
77:
78:
79: It is instructive to note that upper and
80: lower bounds on the {\it maximal density}
81: \begin{equation}
82: \phi_{\mbox{\scriptsize max}}= \sup_{P \subset \Re^d} \phi(P)
83: \end{equation}
84: exist in all dimensions \cite{Co93}, where the supremum is taken over
85: all packings $P$ in $\mathbb{R}^d$.
86: For example, Minkowski \cite{Mi05} proved that the maximal density
87: $\phi^L_{\mbox{\scriptsize max}}$ among all Bravais lattice packings
88: for $d \ge 2$ satisfies the lower bound
89: \begin{equation}
90: \phi^L_{\mbox{\scriptsize max}} \ge \frac{\zeta(d)}{2^{d-1}},
91: \label{mink}
92: \end{equation}
93: where $\zeta(d)=\sum_{k=1}^\infty k^{-d}$ is the Riemann zeta function.
94: One observes that for large values of $d$,
95: the asymptotic behavior of the {\it nonconstructive} Minkowski lower bound is controlled by $2^{-d}$.
96: Interestingly,
97: the density of a {\it saturated} packing of congruent spheres
98: in $\mathbb{R}^d$ for all $d$ satisfies
99: \begin{equation}
100: \phi \ge \frac{1}{2^d}.
101: \label{sat}
102: \end{equation}
103: A saturated packing of congruent spheres
104: of unit diameter and density $\phi$ in $\Re^d$ has the property that each point in space lies
105: within a unit distance from the center of some sphere. Thus, a covering
106: of the space is achieved if each center is encompassed by a sphere
107: of unit radius and the density of this covering is $2^d \phi \ge 1$, which proves the
108: so-called {\it greedy} lower bound (\ref{sat}). Note that it has the same
109: dominant exponential term as (\ref{mink}).
110:
111:
112: A statistically homogeneous (i.e., translationally invariant) packing is completely configurationally characterized
113: by specifying all of the $n$-particle correlation functions.
114: For such packings in $\mathbb{R}^d$, these correlation functions
115: are defined so that $\rho^n g_n({\bf r}_1,{\bf r}_2,\dots,{\bf r}_n)$ is proportional to
116: the probability density for simultaneously finding $n$ particles at
117: locations ${\bf r}_1,{\bf r}_2,\dots,{\bf r}_n$ within the system,
118: where $\rho$ is the number density. Thus, each $g_n$ approaches
119: unity when all particle positions become widely separated within $\mathbb{R}^d$,
120: indicating no spatial correlations.
121: To date, an exact determination of all of the $n$-particle correlation functions for a
122: packing has only been possible for $d=1$ in the special case of an equilibrium ensemble
123: of such particles \cite{Sa53}. Observe that in the limit $d \rightarrow \infty$,
124: it is known that the pressure of an equilibrium hard-sphere fluid
125: is exactly given by the low-density expansion up to the second-virial level for
126: a positive range of densities \cite{Fr99},
127: which implies a simplified form for all of the correlation functions \cite{To05b}.
128:
129: We present in Section II a generalization of the well-known random sequential
130: addition (RSA) process of hard particles \cite{Re63,To02a}. In a particular limit
131: of this nonequilibrium model that we call the ``ghost" RSA process, we are able to obtain the $g_n$ for
132: all allowable densities exactly for any $n$ and dimension $d$.
133: The key geometric quantity that determines $g_n$ is the union volume of $n$ overlapping
134: exclusion spheres of radius equal to the sphere diameter.
135: We show that this construction of
136: a disordered but {\it unsaturated} packing realizes the greedy
137: lower bound (\ref{sat}). This implies
138: that there may be disordered sphere packings in sufficiently high $d$ whose density exceeds
139: Minkowski's lower bound (\ref{mink}). Indeed, in Section III, we report on
140: a conjectural lower bound on the density whose asymptotic behavior
141: is controlled by $2^{-(0.77865\ldots) d}$, thus providing the putative
142: exponential improvement on Minkowski's 100-year-old bound.
143: Our results lead to the counterintuitive possibility that
144: optimal packings in sufficiently high dimensions may be disordered
145: and thus have implications
146: for our fundamental understanding of classical ground states of matter.
147:
148:
149:
150:
151:
152:
153: \section{Generalized Random Sequential Addition Model}
154:
155: We introduce a disordered sphere-packing model in $\mathbb{R}^d$ that is a subset of
156: the Poisson point process and is a generalization of the standard random RSA process.
157: The centers of ``test" spheres of unit diameter arrive continually
158: throughout $\mathbb{R}^d$ during time $t\ge 0$ according to a translationally invariant Poisson process
159: of density $\eta$ per unit time, i.e., $\eta$ is the number of
160: points per unit volume and time. Therefore, the expected number of
161: centers in a region of volume $\Omega$ during time $t$ is $\eta \Omega t$
162: and the probability that this region is empty of centers is $\exp(-\eta \Omega t)$.
163: However, this Poisson distribution of test spheres is not a packing because
164: the spheres can overlap. To create a packing from this point process,
165: one must remove test spheres such that no sphere center can lie within a spherical
166: region of unit radius from any sphere center. Without loss of generality,
167: we will set $\eta=1$.
168:
169: There is a variety of ways of achieving this ``thinning" process such that the subset of points
170: correspond to a sphere packing. One obvious rule is to retain a test
171: sphere at time $t$ only if it does not overlap a sphere that was successfully
172: added to the packing at an earlier time. This criterion defines the standard RSA process in $\mathbb{R}^d$
173: \cite{To02a,Re63}, which generates a homogeneous and isotropic
174: sphere packing in $\mathbb{R}^d$ with a time-dependent density
175: $\phi(t)$. In the limit $t \rightarrow \infty$, the RSA process corresponds to a saturated
176: packing with a maximal or {\it saturation} density $\phi_s(\infty) \equiv \lim_{t\rightarrow
177: \infty} \phi(t)$. In one dimension, the RSA process is commonly known as the ``car parking problem",
178: which Re{\' n}yi showed has a saturation density $\phi_s(\infty)= 0.7476\ldots$ \cite{Re63}.
179: For $2 \le d < \infty$, an exact determination of $\phi_s(\infty)$ is not
180: possible, but estimates for it have been obtained via computer experiments
181: for low dimensions \cite{To02a}.
182:
183: Another thinning criterion retains a test sphere centered at position $\bf r$ at time $t$
184: if no other test sphere is within a unit radial distance from $\bf r$ for
185: the time interval $\kappa t$ prior to $t$, where $\kappa$ is a positive constant
186: in the interval $[0,1]$. This packing is a subset of the RSA packing, and
187: hence we refer to it as the generalized RSA process. Note
188: that when $\kappa=0$, the standard RSA process is recovered, and when $\kappa=1$,
189: a model due to Mat{\' e}rn \cite{Ma86} is recovered \cite{footnote1}. The latter is amenable
190: to exact analysis and is the main focus of this paper.
191: For any $0 < \kappa \le 1$, the generalized
192: RSA process is always an {\it unsaturated} packing.
193: Figure \ref{processes} illustrates the differences
194: between the generalized RSA process at the
195: two extremes of $\kappa=0$ and $\kappa=1$.
196: In remainder of this section, we will focus on the case $\kappa=1$.
197:
198: \begin{figure}
199: \centerline{\psfig{file=rsa-process.eps,height=1.5in}
200: \hspace{0.2in}\psfig{file=grsa-process.eps,height=1.5in} }
201: \caption{The addition of four successfully added particles (in the numerical
202: order indicated) in the generalized RSA process at the
203: two extremes of $\kappa=0$ (left panel) and $\kappa=1$ (right panel).
204: In both cases, the rejected particles have dashed boundaries. For the
205: case $\kappa=1$, a test sphere cannot overlap a ghost sphere.
206: Here $3^\prime$ represents the second attempt to add a third sphere.}
207: \label{processes}
208: \end{figure}
209:
210: The time-dependent density $\phi(t)$ in the case of the generalized RSA process
211: with $\kappa=1$ is easily obtained. In this packing,
212: a test sphere at time $t$ is retained only if does not overlap
213: an existing sphere in the packing as well as any previously rejected
214: test sphere, which we will call ``ghost" spheres. The model itself
215: will be referred to as the ghost RSA process. An overlap cannot
216: occur if a test sphere is outside a unit radius of any successfully
217: added sphere or ghost sphere. Because of the underlying Poisson process,
218: the probability that a trial sphere is retained
219: at time $t$ is given by $\exp(-v_1(1) t)$, where $v_1(1)=\pi^{d/2}/\Gamma(1+d/2)$ is the volume
220: of a sphere of unit radius. Therefore, the expected time-dependent number density $\rho(t)$ and packing density
221: $\phi(t)=\rho(t)v_1({1/2})$ at any time $t$ are given by
222: \begin{equation}
223: \rho(t)=\int_0^t \exp(-v_1(1) t^\prime) dt^\prime=\frac{1-\exp(-v_1(1) t)}{v_1(1)},
224: \quad \phi(t)=\frac{1-\exp(-v_1(1) t)}{2^d}.
225: \label{rho(t)}
226: \end{equation}
227: In the limit $t \rightarrow \infty$, we therefore have that
228: \begin{equation}
229: \rho(\infty) \equiv \lim_{t \rightarrow \infty} \rho(t)=\frac{1}{v_1(1)}, \quad \phi(\infty) \equiv \lim_{t \rightarrow \infty} \phi(t)=\frac{1}{2^d}.
230: \label{rho}
231: \end{equation}
232: Observe that the greedy lower bound (\ref{sat}) on the density is achieved in
233: the infinite-time limit for this sequential but {\it unsaturated} packing,
234: which was pointed out only recently \cite{To05b}.
235: Although the limiting packing density $\phi(\infty)=1/2^d$ is far
236: from optimal in low dimensions, it is relatively large in high dimensions,
237: as discussed in our concluding remarks. Obviously, for any $0 \le \kappa < 1$, the maximum (infinite-time)
238: density of the generalized RSA packing is bounded from below by $1/2^d$
239: (i.e., the maximum density for $\kappa=1$). Henceforth, we write $v_1\equiv v_1(1)$.
240:
241: \begin{figure}[bthp]
242: \centerline{\psfig{file=2-int.eps,height=0.9in}
243: \hspace{0.75in}\psfig{file=3-int.eps,height=0.9in}}
244: \caption{ Left panel: Relevant subvolumes for two overlapping spheres of unit radius
245: associated with the arrivals of two test spheres. The labels refer to
246: distinct, nonoverlapping regions.
247: Right panel: Relevant subvolumes for three overlapping spheres of unit radius
248: associated with the arrivals of three test spheres. }
249: \label{2-int}
250: \end{figure}
251:
252: The derivation of the expression of $g_2(r;t)$ is actually a simple extension of the
253: aforementioned one for $\rho(t)$. Two test spheres that arrive at times $t_1$ and $t_2$
254: and whose centers are separated by a distance $r$ can only be retained if no other test
255: spheres arrived before $t_1$ and $t_2$, respectively (see Fig. \ref{2-int}).
256: Thus, the key geometrical object is
257: the union volume $v_2(r)$ of two spheres of unit radius whose centers are separated by a
258: distance $r$, which can be expressed in terms of the intersection volume $v_2^{int}(r)$
259: \cite{footnote2} between two such spheres via the relation
260: \begin{displaymath}
261: v_2(r)=2 v_1 -v_2^{int}(r).
262: \end{displaymath}
263: For $r \ge 2$, there is no volume common to two such spheres ($v_2^{int}(r)=0$) and therefore
264: $g_2(r;t)=1$, i.e., pair correlations vanish.
265: However, if $1 \le r \le 2$, the two spheres have a common volume and
266: \begin{eqnarray}
267: \rho^2(t)g_2(r;t)&=&\int_0^t \int_0^t \exp\Big[-t_1[v_1-v_2^{int}(r)]
268: -t_2[v_1-v_2^{int}(r)] -\max{(t_1,t_2)} v_2^{int}(r) \Big]dt_1dt_2 \nonumber \\
269: &=& 2\int_0^t dt_2\exp[-t_2 v_1] \int_0^{t_2} dt_1 \exp\Big[-t_1[v_1-v_2^{int}(r)]\Big] \nonumber \\
270: &=& \frac{2}{v_2(r)-v_1}\Big[\frac{1-e^{-tv_1}}{v_1}-\frac{1-e^{-tv_2(r)}}{v_2(r)}\Big].
271: \label{g2-1}
272: \end{eqnarray}
273: In relation (\ref{g2-1}), the terms within the first three brackets are the
274: distinct volumes of the regions
275: labeled 1, 2 and 12 in the left panel of Fig. \ref{2-int}.
276: Therefore, the time-dependent
277: pair correlation function for all $r$ and $t$ is given by
278: \begin{equation}
279: \rho^2(t)g_2(r;t)= \frac{2\Theta(r-1)}{v_2(r)-v_1}
280: \Big[ \frac{1-e^{-tv_1}}{v_1}-\frac{1-e^{-tv_2(r)}}{v_2(r)}\Big],
281: \label{g2-2}
282: \end{equation}
283: where $\Theta(x)$ is the unit step function,
284: equal to zero for $x<0$ and unity for $x \ge1$.
285: It is useful to note that at small times or, equivalently, low densities,
286: formula (\ref{rho(t)}) yields the asymptotic expansion
287: $\phi(t)=t- 2^{d-1} t^2+{\cal O}(t^3)$,
288: which when inverted yields $t=\phi+ 2^{d-1} \phi^2+{\cal O}(\phi^3)$.
289: Substitution of this last result into (\ref{g2-2}) gives
290: \begin{equation}
291: g_2(r;\phi)=\Theta(r-1)+ {\cal O}(\phi^3),
292: \label{g2-t-grsa}
293: \end{equation}
294: which implies that $g_2(r;\phi)$ tends to the unit step function
295: $\Theta(r-1)$ as $\phi \rightarrow 0$ for any $d$.
296:
297: In the limit $t \rightarrow \infty$, we have from (\ref{g2-2}) that
298: $\rho^2(\infty)g_2(r;\infty)=2\Theta(r-1)/[v_1v_2(r)]$
299: or, using (3),
300: \begin{equation}
301: g_2(r;\infty)=\frac{2\Theta(r-1)}{\beta_2(r)},
302: \label{g2-3}
303: \end{equation}
304: where $\beta_2(r)=v_2(r)/v_1$.
305: The radial distribution function $g_2(r;\infty)$ is plotted in Fig. \ref{grsa}
306: for the first five space dimensions.
307: Because $\beta_2(r)$ is equal to 2 for $r \ge 2$,
308: $g_2(r;\infty)=1$ for $r \ge 2$, i.e., spatial correlations
309: vanish identically for all pair distances except
310: those in the small interval $[0,2)$. Even the positive correlations
311: exhibited for $1 < r <2$ are rather weak and decrease
312: exponentially fast with increasing dimension \cite{To05b}, i.e., $g_2(r;\infty)$ tends to the unit step
313: function as $d \rightarrow \infty$, i.e., beyond the hard core (a constrained
314: correlation), spatial correlations
315: vanish.
316:
317: Mat{\' e}rn originally gave an expression for the time-dependent density $\phi(t)$ and
318: and a formal expression (as opposed to explicit expression for any $d$) for
319: the time-dependent radial distribution function $g_2(r;t)$ when $\kappa=1$
320: using a completely different approach. However, he did not consider obtaining any
321: of the higher-order correlation functions.
322:
323:
324: \begin{figure}[bthp]
325: \centerline{\psfig{file=g2-grsa.eps,height=2.0in}}
326: \caption{ Radial distribution function for the first five space dimensions
327: at the maximum density $\phi=1/2^d$ for the generalized RSA model with $\kappa=1$, i.e.,
328: the ``ghost RSA process."}
329: \label{grsa}
330: \end{figure}
331:
332:
333:
334: Let us now derive the time-dependent triplet correlation function
335: $g_3({\bf r}_{12},{\bf r}_{13};\infty)$.
336: Here the relevant geometrical object
337: is the union volume $v_3(r_{12},r_{13},r_{23})$ of three spheres of unit radius whose centers are
338: separated by the distances $r_{12}$, $r_{13}$ and $r_{23}$, which can be expressed in terms
339: of the intersection volume $v_3^{int}(r_{12},r_{13},r_{23})$ between three such spheres via
340: the relation
341: \begin{equation}
342: v_3(r_{12},r_{13},r_{23})=3 v_1 -v_2^{int}(r_{12}) -v_2^{int}(r_{13})
343: -v_2^{int}(r_{23}) +v_3^{int}(r_{12},r_{13},r_{23}).
344: \end{equation}
345: Whenever there is no overlap between the three spheres, $g_3=1$, i.e., triplet correlations vanish. On the
346: other hand, whenever the spheres overlap such that each pair distance is
347: greater than or equal to unity, there are triplet correlations.
348: In such situations, it is convenient to introduce
349: the time-dependent triplet function
350: \begin{eqnarray}
351: &&F(r_{12},r_{13},r_{23};t_1,t_2,t_3)= -t_1[v_1-v_2^{int}(r_{12})-v_2^{int}(r_{13})
352: + v_3^{int}(r_{12},r_{13},r_{23})] \nonumber \\
353: && -t_2[v_1-v_2^{int}(r_{12})-v_2^{int}(r_{23})+v_3^{int}(r_{12},r_{13},r_{23})]
354: -t_3[v_1-v_2^{int}(r_{13})-v_2^{int}(r_{23})+v_3^{int}(r_{12},r_{13},r_{23})] \nonumber \\
355: &&-\max{(t_1,t_2)}[v_2^{int}(r_{12})-v_3^{int}(r_{12},r_{13},r_{23})]
356: -\max{(t_1,t_3)}[v_2^{int}(r_{13})-v_3^{int}(r_{12},r_{13},r_{23})] \nonumber \\
357: &&-\max{(t_2,t_3)}[v_2^{int}(r_{23})-v_3^{int}(r_{12},r_{13},r_{23})]
358: -\max{(t_1,t_2,t_3)}v_3^{int}(r_{12},r_{13},r_{23}).
359: \end{eqnarray}
360: The terms within the first three brackets are the volumes of the regions
361: labeled 1, 2 and 3 in the right panel of Fig. \ref{2-int}. The terms within the fourth through
362: sixth brackets are the volumes labeled 12, 13 and 23 in the right panel of Fig. \ref{2-int}.
363: Of course, the region labeled 123 denotes the intersection volume of three spheres.
364: The triplet correlation function at time $t$ is
365: given by
366: \begin{displaymath}
367: \rho^3(t)g_3(r_{12},r_{13},r_{23};t)=\int_0^t \int_0^t \int_0^t
368: \exp\Big[-F(r_{12},r_{13},r_{23};t_1,t_2,t_3) \Big]dt_1dt_2 dt_3
369: \end{displaymath}
370: and therefore at infinitely large times we have, using (\ref{rho}, (\ref{g2-3}) and (\ref{g3-1}), that
371: \begin{eqnarray}
372: &&\rho^3(\infty)g_3(r_{12},r_{13},r_{23};\infty)= 2\int_0^\infty dt_3\exp[-t_3 v_1] \int_0^{t_3} dt_2
373: \exp\Big[-t_2[v_1-v_2^{int}(r_{23})]\Big] \nonumber \\ &&\times
374: \int_0^{t_2} dt_1 \exp\Big[-t_1[v_1-v_2^{int}(r_{12})-v_2^{int}(r_{13})+v_3^{int}(r_{12},r_{13},r_{23})] \Big] \nonumber \\
375: &+& 2\int_0^\infty dt_1\exp[-t_1 v_1] \int_0^{t_1} dt_3 \exp\Big[-t_3[v_1-v_2^{int}(r_{13})]\Big] \nonumber \\
376: &&\times
377: \int_0^{t_3} dt_2 \exp\Big[-t_2[v_1-v_2^{int}(r_{12})-v_2^{int}(r_{23})+v_3^{int}(r_{12},r_{13},r_{23})] \Big] \nonumber \\
378: &+& 2\int_0^\infty dt_2\exp[-t_2 v_1] \int_0^{t_2} dt_1 \exp\Big[-t_1[v_1-v_2^{int}(r_{12})]\Big] \nonumber \\
379: &&\times
380: \int_0^{t_1} dt_3
381: \exp\Big[-t_3[v_1-v_2^{int}(r_{13})-v_2^{int}(r_{23})+v_3^{int}(r_{12},r_{13},r_{23})] \Big]
382: \nonumber\\
383: &=& \frac{2}{v_1v_3(r_{12},r_{13},r_{23})}\Bigg[\frac{1}{v_2(r_{12})}+
384: \frac{1}{v_2(r_{13})}+\frac{1}{v_2(r_{23})}\Bigg].
385: \end{eqnarray}
386: Combination of (\ref{rho}, (\ref{g2-3}) and (\ref{g3-1}) yields
387: the following expression for the triplet correlation function for arbitrary
388: positions at infinitely large times:
389: \begin{eqnarray}
390: g_3({\bf r}_{12},{\bf r}_{13};\infty)&=&
391: \frac{\Theta(r_{12}-1)\Theta(r_{13}-1)\Theta(r_{23}-1)}{\beta_3(r_{12},r_{13},r_{23})}
392: \Big[g_2(r_{12};\infty)+g_2(r_{13};\infty)+g_2(r_{23};\infty)\Big],
393: \label{g3-1}
394: \end{eqnarray}
395: where $\beta_3(r_{12},r_{13},r_{23})=v_3(r_{12},r_{13},r_{23})/v_1$ and
396: $g_3({\bf r}_{12},{\bf r}_{13};\infty) \equiv g_3(r_{12},r_{13},r_{23};\infty)$.
397:
398: A similar analysis reveals that the four-particle correlation function
399: in the limit $t \rightarrow \infty$ is given by
400: \begin{equation}
401: g_4({\bf r}_{12},{\bf r}_{13},{\bf r}_{14};\infty)=\frac{\prod_{i<j}^4
402: \Theta(r_{ij}-1)}{\beta_4({\bf r}_{12},{\bf r}_{13},{\bf r}_{14};\infty)}\Big[
403: g_3({\bf r}_{12},{\bf r}_{13};\infty)+g_3({\bf r}_{12},{\bf r}_{14};\infty)+
404: g_3({\bf r}_{13},{\bf r}_{14};\infty)+g_3({\bf r}_{23},{\bf r}_{24};\infty)\Big]
405: \end{equation}
406: By induction, the $n$-particle correlation function for arbitrary
407: positions at infinitely large times is given by
408: \begin{equation}
409: g_n({\bf r}_{1}, \ldots, {\bf r}_{n};\infty)=
410: \frac{\prod_{i<j}^n\Theta(r_{ij}-1)}{
411: \beta_n({\bf r}_{1},\ldots,{\bf r}_{n})}
412: \Big[\sum_{i=1}^{n}g_{n-1}(Q_i;\infty)\Big],
413: \label{gn-grsa}
414: \end{equation}
415: where the sum is over all the $n$ distinguishable ways of choosing
416: $n-1$ positions from $n$ positions ${\bf r}_1,\ldots {\bf r}_n$ and the arguments of $g_{n-1}$
417: are the associated $n-1$ positions, which we denote by $Q_i$.
418: Moreover, $\beta_n({\bf r}_{12},{\bf r}_{13},\ldots,{\bf r}_{1n})=
419: v_n({\bf r}_{12},{\bf r}_{13},\ldots,{\bf r}_{1n})/v_1$,
420: where $v_n({\bf r}_{12},{\bf r}_{13},\ldots,{\bf r}_{1n})$ is the union volume
421: of $n$ congruent spheres of unit radius whose centers are located at
422: ${\bf r}_1,\ldots, {\bf r_n}$, ${\bf r}_{ij}={\bf r}_j -{\bf r}_i$
423: for all $1 \le i <j \le n$
424: and $r_{ij} =|{\bf r}_{ij}|$.
425:
426:
427: It can be shown \cite{To05b}
428: that in the limit $d \rightarrow \infty$ and for $\phi=1/2^d$
429: \begin{equation}
430: g_n({\bf r}_{12}, \ldots, {\bf r}_{1n};\infty) \sim \prod_{i<j}^n g_2(r_{ij};\infty),
431: \end{equation}
432: where $g_2(r;\infty) \sim \Theta(r-1)$.
433: We see that unconstrained spatial correlations vanish asymptotically.
434: Specifically, (i) the high-dimensional asymptotic behavior of $g_2$ is the same as the asymptotic
435: behavior in the low-density limit for any $d$ [cf. (\ref{g2-t-grsa})],
436: i.e., {\it unconstrained} spatial
437: correlations, which exist for positive densities at fixed $d$, vanish
438: asymptotically for pair distances beyond the hard-core diameter in the high-dimensional limit;
439: and (ii) $g_n$ for $n \ge 3$ asymptotically can be inferred from
440: a knowledge of only the pair correlation function $g_2$ and number density $\rho$.
441: These two asymptotic properties,
442: which we have called the {\it decorrelation principle} \cite{To05b},
443: apply more generally to any disordered packing, as discussed in Ref. \cite{To05b}.
444: Asymptotically, unconstrained correlations vanish (i.e., statistical
445: independence is established) because we know from
446: the Kabatiansky and Levenshtein asymptotic upper bound on the maximal density
447: $\phi_{\mbox{\scriptsize max}}$ of any sphere packing that the density must go to zero at least
448: as fast as $2^{-0.5990d}$ for large $d$ \cite{Ka78}.
449:
450:
451:
452: \section{Discussion}
453:
454:
455:
456: The fact that the maximal density $\phi(\infty)=1/2^d$ of the ghost RSA packing
457: coincides with the greedy lower bound (\ref{sat}) strongly suggests that there
458: are saturated disordered packings that have larger densities, i.e., the greedy
459: lower bound is a weak bound for saturated packings \cite{footnote3}. This implies
460: that there may be disordered sphere packings in sufficiently high $d$ whose density exceeds
461: Minkowski's lower bound (\ref{mink}) for Bravais lattices, the dominant asymptotic
462: term of which is $1/2^d$. Our results
463: already give insight into this fascinating possibility. For example, consider the so-called
464: checkerboard lattice $D_d$ in $d$ dimensions \cite{Co93}, which is a $d$-dimensional
465: generalization of the optimal (densest) face-centered cubic lattice in three dimensions,
466: and thought to be the optimal packing for $d=4$ and $d=5$. Its packing density
467: $\phi=\pi^{d/2}/[\Gamma(1+d/2)2^{(d+2)/2}]$ exponentially decreases with
468: increasing $d$ (because it quickly becomes
469: unsaturated) and falls below the ghost-RSA-process value of $1/2^d$ for the first time
470: at $d=28$ \cite{footnote4}. The ratio of densities of the ghost RSA process to the
471: checkerboard at $d=100$ is given by $\phi_{ghost}/\phi_{checker} \approx 7.5\times
472: 10^{25}$. Although both packings are unsaturated in such high dimensions,
473: the fact that $g_2(r)$ for the ghost RSA process is effectively uniform (unity)
474: for all $r>1$ but for the checkerboard lattice involves Dirac delta functions
475: of {\it weak} strength at {\it widely spaced} discrete distances explains why the former
476: is enormously denser than the latter.
477:
478: Over the last century, many extensions and generalizations of Minkowski's
479: lower bound (\ref{mink}) have been obtained \cite{Co93}, but none of these
480: investigations have been able to improve upon the dominant exponential term $2^{-d}$. In another
481: work \cite{To05b}, we will present comprehensive rigorous evidence that this exponential improvement
482: may be provided by considering specific disordered sphere packings.
483: Here we simply sketch the procedure leading to this putative improvement over
484: Minkowski's lower bound. The basic ideas underlying our new approach to the derivation of lower bounds on
485: $\phi_{\mbox{\scriptsize max}}$ were actually described in our earlier work \cite{To02c}
486: in which we studied so-called $g_2$-invariant processes.
487: A {\it $g_2$-invariant process} is one in which a given nonnegative pair correlation
488: $g_2({\bf r})$ function remains invariant for all ${\bf r}$ over the range of
489: densities
490: \begin{equation}
491: 0 \le \phi \le \phi_*.
492: \end{equation}
493: The terminal density $\phi_*$ is the maximum achievable density
494: for the $g_2$-invariant process subject to satisfaction of
495: certain necessary conditions on the pair correlation.
496: In particular, we considered those ``test" $g_2(r)$'s that are distributions on $\mathbb{R}^d$ depending
497: only on the radial distance $r$.
498: For any test $g_2(r)$, we want to maximize
499: the corresponding density $\phi$ satisfying the following three conditions:
500:
501:
502: \noindent (i) \hspace{0.25in} $g_2(r) \ge 0 \qquad \mbox{for all}\quad r,$
503:
504: \noindent (i) \hspace{0.2in} $g_2(r)= 0 \qquad \mbox{for}\quad r<1,$
505:
506:
507: \noindent (iii)
508: \begin{displaymath}
509: \hspace{-0.5in}S(k)= 1+ \rho\left(2\pi\right)^{\frac{d}{2}}\int_{0}^{\infty}r^{d-1}h(r)
510: \frac{J_{\left(d/2\right)-1}\!\left(kr\right)}{\left(kr\right)^{\left(d/2\right
511: )-1}}dr \ge 0 \qquad \mbox{for all}\quad k,
512: \end{displaymath}
513: where $h(r)=g_2(r)-1$ is the total correlation function.
514: Condition (i) is a trivial consequence of the fact that $g_2$ is a probability density
515: function. Condition (ii) is just the {\it hard-core} constraint for
516: spheres of unit diameter. Condition (iii)
517: states that the structure factor $S(k)$ in $d$ dimensions
518: must be nonnegative for all $k$.
519: When there exist sphere packings with $g_2$ satisfying conditions
520: (i)-(iii) for $\phi$ in the interval $[0,\phi_*]$, then we have the lower
521: bound on the maximal density given by
522: \begin{equation}
523: \phi_{\mbox{\scriptsize max}} \ge \phi_*.
524: \label{true-bound-phi}
525: \end{equation}
526:
527: It is rather remarkable that the optimization problem defined above is identical
528: to one formulated by Cohn \cite{Co02}. Specifically, it is the {\it dual}
529: of the {\it primal} infinite-dimensional linear program that Cohn employed with Elkies \cite{Co03}
530: to obtain upper bounds on the maximal packing density. Thus, even if there does
531: not exist a sphere packing with $g_2$ satisfying conditions
532: (i)-(iii), the terminal density $\phi_*$ can never exceed the
533: Cohn-Elkies upper bound and, more generally, our formulation has implications for upper bounds
534: on $\phi_{\mbox{\scriptsize max}}$.
535:
536:
537: In addition, to the
538: structure factor condition, there are generally many other conditions that a pair correlation
539: function corresponding to a point process must obey \cite{Cos04}.
540: One such additional necessary condition, obtained by Yamada \cite{Ya61},
541: is concerned with the variance $\sigma^2(\Omega) \equiv
542: \langle (N(\Omega)^2- \langle N(\Omega) \rangle)^2\rangle$,
543: in the number $N(\Omega)$ of particle centers contained within a region or ``window"
544: $\Omega \subset \mathbb{R}^d$:
545: \begin{equation}
546: \sigma^2(\Omega)=\rho |\Omega| \Big[ 1+\rho \int_{\Omega} h({\bf r}) d{\mathbf r}\Big] \ge \theta(1-\theta),
547: \label{yamada}
548: \end{equation}
549: where $\theta$ is the fractional part of the expected number of points
550: $\rho |\Omega|$ contained in
551: the window. This is a consequence of the fact that the number of particles
552: in any window must be an integer.
553:
554: In Ref. \cite{To02c}, a five-parameter test family of $g_2$'s had been considered,
555: which incorporated the known features of core exclusion, contact pairs, and damped oscillatory short-range
556: order beyond contact that are features intended to
557: describe {\it disordered} jammed sphere packings for $d=3$. However, because
558: of the functional complexity of this test $g_2$, the terminal density could only be determined numerically.
559: The general optimization procedure outlined above was employed in Ref. \cite{To05b} to obtain
560: analytical estimates of the terminal density in high dimensions that together with
561: the following conjecture provide the putative exponential improvement
562: on Minkowski's lower bound on $\phi_{\mbox{\scriptsize max}}$:
563:
564: \noindent{\bf Conjecture 1}: A hard-core nonnegative tempered distribution $g_2({\bf r})$ is a pair correlation function
565: of a translationally invariant disordered sphere packing in $\mathbb{R}^d$ at number density $\rho$
566: for sufficiently large $d$ if and only if
567: $S({\bf k})\ge 0$. The maximum achievable density
568: is the terminal density $\phi_*$.
569:
570: In other words, $g_2(r)$ that meets the conditions (i) - (iii),
571: at or above a critical dimension $d_c$, packings exist with such a $g_2$.
572: A {\it disordered packing} in $\mathbb{R}^d$ is defined in Ref. 9 to be
573: one in which the pair correlation function $g_2({\bf r})$ decays to its
574: long-range value of unity faster than $|{\bf r}|^{-d-\epsilon}$ for some
575: $\epsilon >0$."
576: Employing the aforementioned optimization procedure with a certain
577: test function $g_2$ and Conjecture 1, we obtain in what follows
578: conjectural lower bounds that yield the long-sought
579: asymptotic exponential improvement on Minkowski's bound.
580: An important feature of any dense packing is that the particles
581: form contacts with one another.
582: Experience with disordered jammed packings in low dimensions reveals that the contact or kissing
583: number as well as the density can be substantially increased if there
584: is there is a low probability of finding noncontacting particles from a typical
585: particle at radial distances just larger than the nearest-neighbor distance.
586: It is desired to idealize this small-distance negative correlation (relative to the
587: uncorrelated value of unity)
588: in such a way that it is amenable to exact asymptotic analysis.
589: Accordingly, a test radial distribution function was considered in Ref. 8
590: in which there is a gap between the location of a unit step function and the
591: delta function at finite $d$, i.e.,
592: \begin{equation}
593: g_2(r)=\Theta(r-\sigma)+\frac{Z}{s_1(1)\rho}\delta(r-1),
594: \label{step3}
595: \end{equation}
596: where $s(r)$ is the surface area of a $d$-dimensional sphere of radius $r$
597: and $Z$ is a parameter, which is the average contact or kissing number,
598: and unity is the sphere diameter.
599: The expression contains two adjustable parameters, $\sigma \ge 1$ and $Z$,
600: which must obviously be constrained to
601: be nonnegative.
602:
603:
604: Before reporting the main results of this optimization, it is instructive
605: to examine the test function (\ref{step3}) for two special cases: (1)
606: one in which $\sigma=1$ and $Z=0$ and (2) the other in which $\sigma=1$
607: and $Z>0$ (which were first considered in Ref. 17). In the first special instance,
608: there are no parameters to be optimized here, and the terminal density $\phi_*$ is given by
609: $\phi_*=\frac{1}{2^d}$. It is simple to show that the Yamada condition ((\ref{yamada}) is satisfied in any dimension
610: for $0 \le \phi \le 2^{-d}$.
611: We already established in the previous section that there exist sphere packings that
612: asymptotically have radial distribution functions given by the simple
613: unit step function for $\phi \le 2^{-d}$. Nonetheless,
614: invoking Conjecture 1 and the obtained terminal density,
615: implies the asymptotic lower bound on the maximal density is given by
616: \begin{equation}
617: \phi_{\mbox{\scriptsize max}} \ge \frac{1}{2^d},
618: \end{equation}
619: which provides an alternate derivation of the elementary bound (\ref{sat}).
620: Using numerical simulations with a finite but large
621: number of spheres on the torus, we have been able to construct particle configurations
622: in which the radial distribution function
623: is given by the test function (\ref{step3}) with $\sigma=1$ and $Z=0$ in one, two and three dimensions for densities
624: up to the terminal density \cite{Cr03,Ou05}. The existence of such
625: a discrete approximation to this test $g_2$ is suggestive that the standard nonnegativity
626: conditions may be sufficient to establish existence in this case
627: for densities up to $\phi_*$. In the second special case ($\sigma=1$
628: and $Z>0$) and under the constraint that the minimum of $S(k)$ occurs at $k=0$, then we have the exact results
629: $\phi_*=\frac{d+2}{2^{d+1}}$ and $Z_*=\frac{d}{2}$, where $Z_*$ is the optimized average kissing number.
630: The Yamada condition ((\ref{yamada}) is violated here only for $d=1$
631: and becomes less restrictive as the dimension increases from $d=2$.
632: Interestingly, we have also shown via numerical simulations
633: that there exist sphere packings possessing radial distribution functions
634: given by this test function in two and three dimensions for densities
635: up to the terminal density \cite{Ou05}. This is suggestive that the Conjecture 1
636: for this test function may in fact be stronger than is required.
637: In the high-dimensional limit, invoking Conjecture 1 and the
638: obtained terminal density, yields the conjectural lower bound
639: \begin{equation}
640: \phi_{\mbox{\scriptsize max}} \ge \frac{d+2}{2^{d+1}}.
641: \label{lower-2}
642: \end{equation}
643: This lower bound provides the same type of linear improvement over Minkowski's lower
644: bound as does Ball's rigorous lower bound \cite{Ball92} obtained using
645: a completely different approach.
646:
647: Now let us consider the problem when both $\sigma$ and $Z$ in (\ref{step3})
648: must be optimized. The presence of a gap between the unit step function and delta function will indeed lead
649: asymptotically to substantially higher terminal densities.
650: For sufficiently small $d$ ($d \le 200$), the optimization procedure is carried out numerically \cite{To05b}.
651: The Yamada condition (\ref{yamada}) is
652: violated only for $d=1$ for the test function
653: (\ref{step3}) for the terminal density $\phi_*$ and associated optimized
654: parameters $\sigma_*$ and $Z_*=(2\sigma_* \phi_*)^d -1$. One can
655: again verify directly that the Yamada condition becomes less restrictive as the
656: dimension increases from $d=2$. However, although the test function
657: (\ref{step3}) for $d=2$ with optimized parameters $\phi_*=0.74803$,
658: $\sigma_*=1.2946$ and $Z_*=4.0148$ satisfies the Yamada condition,
659: it cannot correspond to a sphere packing because it violates
660: local geometric constraints specified by $\sigma_*$ and $Z_*$ \cite{To05b}.
661: To our knowledge, this is the first example of a test radial
662: distribution function that satisfies the two standard non-negativity
663: conditions (i) and (iii) and the Yamada condition (\ref{yamada}), but cannot
664: correspond to a point process. Thus, there is at least one previously
665: unarticulated necessary condition that has been violated in the low
666: dimension $d=2$. As is the case with the Yamada condition (\ref{yamada}),
667: this additional necessary condition appears to lose
668: relevance in low dimensions because
669: we have shown that there is no analogous local geometric constraint
670: violation for $d \ge 3$. For $d \le 56$, the terminal density lies below the density of the densest
671: known packing (a Bravais lattice) \cite{Co93}. However,
672: for $d >56$, $\phi_*$ can be larger than the density of the densest
673: known arrangements, which are ordered.
674: Our numerical results for $d$ between 3 and 200, reveal exponential improvement of the terminal density
675: $\phi_*$ over the one for the gapless case, where $\phi_*=(d+2)/2^{d+1}$.
676:
677:
678: For large $d$, an exact (but nontrivial) asymptotic analysis can be performed \cite{To05b},
679: yielding the optimal terminal density. This result in conjunction
680: with Conjecture 1 yields the conjectural asymptotic lower bound
681: \begin{equation}
682: \phi_* \sim
683: \frac{3.276100896d^{1/6}}{2^{[3-\log_2(e)]d/2}}=\frac{3.276100896
684: ~d^{1/6}}{2^{0.7786524795\ldots \,d}},
685: \label{den}
686: \end{equation}
687: This putatively provides the long-sought exponential improvement on Minkowski's
688: lower bound. We call this a conjectural lower bound because it relies on Conjecture 1 being true,
689: which a number of results support.
690: First, the decorrelation principle states that unconstrained
691: correlations in disordered sphere packings vanish asymptotically in high dimensions
692: and that the $g_n$ for any $n \ge 3$ can be inferred entirely from a knowledge
693: of $\rho$ and $g_2$. Second, the necessary
694: Yamada condition appears to only have relevance in very low dimensions.
695: Third, we have demonstrated that other new necessary conditions
696: also seem to be germane only in very low dimensions.
697: Fourth, we recover the form of known rigorous bounds [cf. (21) and (22)]
698: in special cases of the test radial distribution function (20) when we
699: invoke Conjecture 1. Finally, in these two instances,
700: configurations of disordered sphere packings on the torus
701: have been numerically constructed with such $g_2$ in low dimensions for densities up to the terminal density \cite{Cr03,Ou05}.
702:
703: A byproduct of the bound (\ref{den}) is the
704: conjectural asymptotic lower bound on the maximal kissing number \cite{To05b}
705: \begin{equation}
706: Z_{\mbox{\scriptsize max}} \ge Z_* \sim
707: 40.24850787 \,d^{1/6}\, 2^{[\log_2(e)-1]d/2}=
708: 40.24850787 \,d^{1/6} \, 2^{0.2213475205\ldots\, d},
709: \end{equation}
710: This result is superior to the best known
711: asymptotic lower bound on the maximal kissing number of $2^{0.2075\ldots d}$
712: \cite{Wy65}.
713:
714:
715: The work described above suggests that the densest packings in sufficiently
716: high dimensions may be disordered rather than periodic, implying
717: the existence of disordered classical ground states for some continuous potentials.
718: In fact, there is no fundamental reason why disordered
719: ground states are prohibited in low dimensions \cite{Ru82}. A case in point
720: are the ``pinwheel" tilings of the plane, which possess both statistical translational
721: and rotational invariance \cite{Ra94}.
722:
723: This work was supported by the National Science Foundation
724: under Grant No. DMS-0312067.
725:
726: \begin{thebibliography}{20}
727:
728: \bibitem{Ha86}
729: J.~P. Hansen and I.~R. McDonald,
730: {\em Theory of Simple Liquids} (Academic Press, New York, 1986).
731:
732: \bibitem{To02a}
733: S.~Torquato, {\em Random Heterogeneous Materials: Microstructure and Macroscopic
734: Properties} (Springer-Verlag, New York, 2002).
735:
736: \bibitem{Chaik95}
737: P.~M. Chaikin and T.~C. Lubensky, {\em Principles of Condensed Matter Physics}.
738: (Cambridge University Press, New York, 1995).
739:
740: \bibitem{Co93}
741: J.~H. Conway and N.~J.~A. Sloane, {\em Sphere Packings, Lattices and Groups}
742: (Springer-Verlag, New York, 1993).
743:
744: \bibitem{Fr99}
745: H.~L. Frisch and J.~K. Percus,
746: {\em Phys. Rev. E}, {\bf 60}, 2942 (1999).
747:
748: \bibitem{Pa00}
749: G. Parisi and F. Slanina, {\em Phys. Rev. E}, {\bf 62}, 5644 (2000).
750:
751: \bibitem{Mi05} H.~Minkowski,
752: {\em J. {r}eine {a}ngew. {M}ath.}, {\bf 129}, 220 (1905).
753:
754: \bibitem{Sa53}
755: Z.~W. Salsburg, R.~W. Zwanzig, and J.~G. Kirkwood,
756: {\em J. Chem. Phys.}, {21}, 1098 (1953).
757:
758: \bibitem{To05b} S.~Torquato and F.~H. Stillinger, {\em Experimental Math.}, in press.
759: See also arXiv:math.MG/0508381 for the preprint version of the paper.
760:
761: \bibitem{Re63} A.~Re{\'n}yi, {\em Sel. Trans. Math. Stat. Prob.}, {\bf 4}, 203 (1963);
762: B.~Widom, {\em J. Chem. Phys.}, {\bf 44}, 3888 (1966); J. Feder, {\em J. Theor. Biol.}, {\bf 87} 237 (1980);
763: R. M. Ziff and R. D. Vigil, J. Phys. A: Math. Gen. {\bf 23}, 5103 (1990);
764: P.~Viot, G.~Tarjus, and J.~Talbot, {\em Phys. Rev. E}, {\bf 48}, 480 (1993).
765:
766:
767: \bibitem{Ma86} B.~Mat{\'e}rn, Spatial variation.
768: in {\em Lecture Notes in Statistics}, {\bf 36} (Springer-Verlag, New York, 1986).
769:
770: \bibitem{footnote1} Interestingly, this model can be mapped exactly to a
771: irreversible multilayer adsorption problem in which spheres are deposited from the
772: ($d+1$)-th dimension onto a $d$-dimensional substrate.
773: See P.~R. Van~Tasell and P.~Viot, {\em Europhys. Lett.}, {40}, 293 (1997).
774:
775: \bibitem{footnote2} S. Torquato and F. H. Stillinger, {\em Phys. Rev. E},
776: {\bf 68}, 041113 (2003); ibid. 069901. This paper gives an explicit formula for the
777: intersection volume $v_2^{int}(r)$ in any $d$. See also Ref. 9 for another exact representation of this intersection volume for
778: any $d$.
779:
780:
781:
782: \bibitem{Ka78} G.~A. Kabatiansky and V.~I. Levenshtein, {\em Problems of Information
783: Transmission}, {\bf 14}, 1 (1978).
784:
785: \bibitem{footnote3}Indeed, we will show in a future
786: work that the standard RSA packing ($\kappa=0$)
787: has a substantially higher density than $1/2^d$ in high dimensions.
788:
789: \bibitem{footnote4} In such dimensions, denser Bravais lattice packings
790: exist, but it is conjectured that in sufficiently high dimensions
791: all Bravais lattices become unsaturated (see Ref. \cite{Co93})
792: and therefore inefficient packings.
793:
794:
795: \bibitem{To02c}
796: S.~Torquato and F.~H. Stillinger.
797: \newblock {\em J. Phys. Chem. B} {\bf 106}, 8354 (2002); ibid.
798: \newblock {\em J. Phys. Chem. B} {\bf 106}, 11406 (2002)
799:
800: \bibitem{Co02}
801: H.~Cohn.
802: \newblock {\em Geom. Topol.} {\bf 6}, 329 (2002).
803:
804: \bibitem{Co03}
805: H.~Cohn and N.~Elkies.
806: \newblock {\em Annals Math.} {\bf 157}, 689 (2003).
807:
808: \bibitem{Cos04}
809: O.~Costin and J.~Lebowitz.
810: \newblock {\em J. Phys. Chem. B.} {\bf 108}, 19614 (2004).
811:
812: \bibitem{Ya61}
813: M.~Yamada
814: \newblock {\em Prog. Theor. Phys.} {\bf 25}, 579 (1961).
815:
816: \bibitem{Cr03}
817: J.~R. Crawford, S.~Torquato, and F.~H. Stillinger.
818: \newblock {\em J. Chem. Phys.} {\bf 119}, 7065 (2003).
819:
820: \bibitem{Ou05}
821: O.~U. Ouche, F.~H. Stillinger, and S.~Torquato,
822: Physica A {\bf 360}, 21 (2006).
823:
824:
825:
826: \bibitem{Ball92}
827: K.~Ball
828: \newblock {\em Duke J. Math.} {\bf 68}, 217 (1992).
829:
830: %\bibitem{footnote2} A {\it disordered packing} in $\Re^d$
831: %is one in which the pair correlation function $g_2({\bf r})$ decays
832: %to its long-range value of unity faster than $|{\bf r}|^{-d+\epsilon}$ for
833: %some $\epsilon >0$.
834:
835: %\bibitem{footnote3} The expected cumulative coordination number
836: %$Z(r)$ is related to the radial distribution function $g_2(r)$ via
837: %$Z(r)=\rho \Omega(d)\int_0^r x^{d-1} g_2(x) dx$, where $\Omega(d)$
838: %is the total solid angle contained in a $d$-dimensional sphere.
839:
840: \bibitem{Wy65}
841: A.~D. Wyner
842: \newblock {\em Bell System Tech. J.}, 44:1061--1122, 1965.
843:
844:
845:
846: \bibitem{Ru82} D. Ruelle, {\em Physica A}, {\bf 113}, 619 (1982);
847: J. Miekisz and C. Radin, {\em Phys. Rev. B}, {\bf 39}, 1950 (1989).
848:
849:
850: \bibitem{Ra94} C. Radin, {\em Annals Math.} {\bf 139}, 661 (1994).
851:
852:
853: \end{thebibliography}
854:
855: \end{document}
856: