1: % this version has black and white figs.
2: \documentclass[groupedaddress,nofootinbib, showpacs]{revtex4}
3: \usepackage{graphics}
4: \renewcommand{\wr}{\omega_r}
5: \newcommand{\wb}{\omega_b}
6: \begin{document}
7:
8: \voffset=0.2in
9: \title{Geometric structure of coexisting phases found in the\\
10: Biham-Middleton-Levine traffic model}
11:
12: \author{Raissa M. D'Souza} \email{raissa@microsoft.com}
13: \affiliation{Microsoft Research, Redmond WA 98052} \date{September 2003}
14:
15: \begin{abstract}
16: The Biham-Middleton-Levine traffic model is perhaps the simplest
17: system exhibiting phase transitions and self-organization. Moreover,
18: it is an underpinning to extensive modern studies of traffic flow.
19: The general belief is that the system exhibits a sharp phase
20: transition from freely flowing to fully jammed, as a function of
21: initial density of cars. However we discover intermediate stable
22: phases, where jams and freely flowing traffic coexist. The geometric
23: structure of such phases is highly regular, with bands of free flowing
24: traffic intersecting at jammed wavefronts that propagate smoothly
25: through the space. Instead of a phase transition as a function of
26: density, we see bifurcation points, where intermediate phases begin
27: coexisting with the more conventionally known phases. We show that
28: the regular geometric structure is in part a consequence of the finite
29: size and aspect ratio of the underlying lattice, and that for certain
30: aspect ratios the asymptotic intermediate phase is on a short periodic
31: limit cycle (the exact microscopic configuration recurs each $\tau$
32: timesteps, where $\tau$ is small compared to the system size). Aside
33: from describing these intermediate states, which previously were
34: overlooked, we derive simple equations to describe the geometric
35: constraints, and predict their asymptotic velocities.
36: \end{abstract}
37:
38: \pacs{64.60.My, 64.60.Cn, 89.40.Bb, 05.20.Dd}
39: \maketitle
40:
41: %\input{intro}
42: \section{Introduction}\label{sec:intro}
43: Modeling vehicular and Internet traffic, thereby gaining an
44: understanding of congestion patterns and jamming phenomena, is an
45: extremely relevant problem, with obvious practical ramifications. One
46: popular approach is the use of simple discrete cellular automata (CA)
47: models, which capture aspects of the dynamics of discrete vehicles or
48: packets. One of the most cited examples of such a CA model is the
49: Biham, Middleton, and Levine model (BML) of two-dimensional traffic
50: flow\cite{BML92}. At the time of this writing, Ref.~\cite{BML92} has
51: received over two-hundred citations in the scientific literature, and
52: it serves as a theoretical underpinning for the physicists' approach
53: to modeling traffic. Note that using techniques from physics to model
54: traffic has been a fruitful research area for more than a decade, and
55: continues to be. For recent reviews see
56: Refs.~\cite{NagaRev2002,SchadRev2002,ChowSanSchad-Rev2000,NagelRev96}.
57:
58: The BML model describes two species of ``cars'' moving on a
59: two-dimensional square lattice, with periodic boundary conditions.
60: The model is extremely simple, yet the behaviors it displays,
61: extraordinarily complex. The system shows what appears to be a phase
62: transition from having all cars freely moving at all time steps, to
63: complete gridlock, where no car can ever move again. In addition, in
64: all the phases, the system becomes fully correlated, forming a range
65: of interesting stable self-organized patterns. It is perhaps the
66: simplest model where one can study both phase transitions and self
67: organization. This model has recently become increasingly of interest
68: to the combinatorial mathematics community, as it continues to elude
69: rigorous theoretical analysis\cite{combinatorics-guys}.
70:
71: We implement the BML model and experimentally study its behaviors. We
72: discover stable intermediate states that have never been reported
73: before for the BML model, with highly structured geometric patterns of
74: wave fronts of jams moving through otherwise freely flowing
75: traffic. We show the geometry of these patterns arise due to the
76: finite size and periodic boundary conditions of the underlying
77: lattice. We also show that the aspect ratio of the lattice imposes
78: geometric constraints which restrict the patterns, and derive simple
79: equations describing these geometric constraints which allow us to
80: calculate the asymptotic velocities. For certain aspect ratios we can
81: prove that the intermediate configurations end up on a short periodic
82: limit cycle---the exact microscopic configurations recur each $\tau$
83: timesteps (where $\tau$ is small compared to the system size)---hence
84: these states are stable for all time. For the other aspect ratios, we
85: show the intermediate states are at least metastable, lasting as long
86: as we could simulate them. By establishing the existence of these
87: intermediate states, we show that the conventional beliefs about this
88: model need to be reexamined. Contrary to the evidence published
89: elsewhere, only on smaller spaces do we see evidence for a sharp
90: transition from freely flowing to fully jammed configurations as a
91: function of the initial density of cars, $\rho$. Instead we observe
92: bifurcations as a function of $\rho$, where different phases can begin
93: to coexist. The bifurcation points, the range of the windows for
94: phase coexistence, and the number of coexisting phases, depend on the
95: size and the aspect ratio of the underlying lattice. Considering the
96: amount of ongoing work on this model, and its use in large scale,
97: complex simulations of traffic, we believe it is important to
98: understand these new observations.
99:
100: This manuscript is organized as follows. In Sec.~\ref{sec:review} we
101: review the BML model and relevant past work. In
102: Sec.~\ref{sec:results} we describe our simulations and empirical
103: results. Section~\ref{sec:geom} contains a discussion of the kinetic
104: pathways, geometric constraints, and derivation of the velocities for
105: the intermediate states. Finally, in Sec.~\ref{sec:disc}, we
106: summarize and discuss open questions and areas for further inquiry.
107:
108: %\input{review}
109: \section{The BML model}\label{sec:review}
110:
111: Consider two species of particles ({\em i.e.}, ``cars''), east-bound
112: and north-bound (which we also interchangeably call ``red'' and
113: ``blue'' respectively), which populate a two-dimensional square
114: lattice with periodic boundary conditions. Each lattice site can be
115: in one of three states: empty, occupied by an east-bound car, or
116: occupied by a north-bound car. The cars are initially distributed
117: uniformly at random along the lattice sites, with spatial density,
118: $\rho$ (usually taken to be the same for both north and east-bound
119: cars). The discrete time dynamics has two phases. On even steps, all
120: east-bound cars synchronously attempt to advance one lattice site
121: towards the east. If the site eastward of a car is currently empty, it
122: advances. Otherwise, it remains stationary (even if the eastward site
123: is to become empty during the current time step). On odd time steps,
124: the north-bound cars follow the analogous dynamics, only attempting to
125: advance to the north-bound site. The dynamics is fully deterministic.
126: The only randomness is in the initial condition. Furthermore, the
127: dynamics conserves cars, and does not allow for an east-bound car to
128: change its row, nor for a north-bound car to change its column. So on
129: a $L\times L$ lattice, there are $2L$ conservation laws.
130:
131: If initialized with a low enough density of cars, the system
132: eventually self-organizes into a configuration where all cars can move
133: at each time step (each car has asymptotic velocity equal to unity).
134: A typical such configuration is shown in Fig.~\ref{fig:final}(a). If
135: initialized at slightly higher density, the cars are blocked by other
136: cars, until eventually all cars end up participating in one large
137: global jam, where no car can move (asymptotic velocity equal to zero).
138: A typical global jam is shown in Fig.~\ref{fig:final}(b). The
139: transition between the two behaviors has been thought of as sharp,
140: showing characteristics of a first order phase transition.
141: Initialized at much higher densities, small jams begin simultaneously
142: throughout the lattice and merge almost immediately with other small
143: jams, leaving all cars blocked (with all velocities equal to zero).
144: In this high density phase, the system has no time to self organize,
145: and instead of one global jam, we observe a collection of small random
146: jams. This latter type of jam has been compared to traffic in a large
147: city during ``rush hour'': a car might escape one jam, only to quickly
148: join the tail of the next. We expect that for an infinite size
149: system, the fully jammed state resembles this random type of jam.
150:
151: \begin{figure}[tb]
152: {\hfill\resizebox{2in}{!}{\includegraphics{final-free-256.ps}}\hfill\hfill
153: \resizebox{2in}{!}{\includegraphics{final-blocked2-256.ps}}\hfill\hfill
154: \resizebox{2in}{!}{\includegraphics{randJam256.ps}}\hfill}
155: %\vspace{-0.25in}
156: \caption{Typical configurations observed for the BML model on an
157: $L\times L$ system of size $L=256$. (a) The free flowing stage, where
158: all particles advance during each update ($v=1$). Note the ordered
159: stripes of alternating east and north-bound cars. The width of the
160: stripes increase, on average, with density $\rho$ while in the low
161: density phase. (b) A fully jammed configuration, consisting of one
162: global jam. Note the jam length, $\sqrt{2} L$, is larger than the
163: system size. (c) A high density, random jam configuration.}
164: \label{fig:final}
165: \end{figure}
166:
167: Most of the understanding of the BML model has come from numerical
168: simulations\cite{BML92,Naga93-anisotropy,FukuiIshi93,TadaKiku94,TadaKiku95,greenwave96,MartCuesta95,ChauWanYan98}.
169: Theoretical analysis has been limited to mean-field
170: approaches\cite{Naga93-anisotropy,IshiFukui94,Cuesta-LB-95,WangWooHui96},
171: and attempts to start with continuum hydrodynamic equations and
172: formulate an equivalent discrete model\cite{Naga99}. There are general
173: beliefs about this model, that the transition is first order and that
174: the critical density, $\rho_c$, decreases with increasing system size,
175: possibly reaching the value $\rho_c=0$ as the system size approaches
176: infinite\cite{BML92}. The BML model has been simulated extensively,
177: but there are inconsistencies in the literature and lack of detail of
178: numerical implementations (such as the size of the ensemble being
179: averaged over). Details of numerical studies have been published only
180: for small systems, on the order of
181: $L=10-50$\cite{FukuiIshi93,Naga93-anisotropy}. Larger systems have
182: only been studied coarsely, or in the context of self-organized versus
183: random jams\cite{TadaKiku94,TadaKiku95}.
184:
185: Despite extensive numerical simulation, the existence of stable
186: intermediate phases (with $0 < v < 1$) has not been explicitly
187: reported previously. Fukui and Ishibashi\cite{FukuiIshi93} do show
188: evidence of the existence of an intermediate phase in one plot. They
189: note that for intermediate values of $\rho$, $v$ ``fluctuates around a
190: certain value for a long time''. The value shown in their plot is
191: extremely close to the values we observe (plotted in
192: Sec.~\ref{sec:results}). Aside from this comment, they do not pursue
193: the issue further. A careful study by T\"{o}r\"{o}k and
194: Kert\'{e}sz\cite{greenwave96} contains precise details of their
195: numerical simulations. They are studying a variant of BML with faster
196: convergence times (called the green wave model). Since it is not
197: possible to theoretically predict the convergence time, they estimate
198: it, and apply the following, very reasonable, empirical heuristic. If
199: a realization has not reached a state, with $v\approx 0$ or $v \approx
200: 1$, within an allotted time (taken to be 5 times the estimated
201: convergence time), that realization is discarded. We can only assume
202: some of the studies of the BML model may have used a similar criteria
203: of discarding ``non-converged'' states. Note that for continuum
204: models, intermediate phases of jammed wavefronts moving through
205: otherwise freely flowing traffic have been reported. See for instance
206: Ref.~\cite{KimaSasa95}.
207:
208: Sensitivity of the BML model to boundary conditions has been reported
209: previously. Mart\'{i}nez, {\em et al.}\cite{MartCuesta95} study the
210: dilute limit ($\rho \rightarrow 0$), and show that different results
211: are obtained for an ``entangled'' torus versus a conventional one.
212: They raise interesting questions about how to get at the bulk
213: properties using only finite size simulations, but do not quantify nor
214: pursue the effects further. Chau, Wan, and Yan\cite{ChauWanYan98}
215: study the BML model on the torus with random boundary conditions (BCs)
216: (meaning, particles moving off the right (top) edge reappear at some
217: randomly selected site on the left (bottom) edge). They claim $v>0$
218: whenever $\rho<1$, and hence dismiss such systems as being ``not very
219: interesting''. They also note that the velocity and critical density
220: depend sensitively on the choice of BCs, but they also do not pursue
221: the effects further.
222:
223:
224:
225: %\input{results}
226: \section{Simulation Results}\label{sec:results}
227: We implement the BML model on square lattices of finite size $L\times
228: L'$, for a range of sizes and varying aspect ratios. For square
229: aspect ratios ({\em i.e.}, $L=L'$), the lengths studied range from
230: $L=64$ to $L=512$. We also study rectangular aspect ratios where the
231: width, $L$, is an integer multiple of the height, $L'$, and where $L$
232: and $L'$ are relatively prime. On each lattice, we implement a range
233: of densities $\rho$, studying at least ten realizations for each
234: density. All simulations are implemented on a special purpose
235: cellular automata machine, the CAM8\cite{cam8-waterloo}. The CAM8
236: performs approximately one-billion site updates per second, comparable
237: to a modern high-performance desktop computer. The main advantage of
238: the CAM8 is that it allows for excellent visualization of the system,
239: with no overhead in the rendering, giving us live video output of the
240: dynamics of the system. As discussed in Sec.~\ref{sec:geom},
241: visualizing the kinetic pathway gives crucial insight into the
242: formation of the intermediate states.
243:
244: The main subtlety involved with simulating this model is determining
245: the convergence time ({\em i.e.}, the time is takes to reach the
246: asymptotic behavior). All realizations were simulated until converged
247: ($v=0$, $v=1$, or the periodic limit cycle was reached) or for times
248: out to at least $t=2\times 10^6$ time steps. Of the realizations
249: that had not yet converged, many were simulated for orders of
250: magnitude beyond. We find this a reasonable compromise, since the
251: compute power to simulate all samples to times greater than $10^8$ is
252: beyond our current capacity. Throughout the remainder of the
253: manuscript, we refer interchangeably to the east-bound cars as ``red''
254: and the north-bound as ``blue''.
255:
256: \begin{figure}[t]
257: {\hfill(a)\resizebox{2in}{!}{\includegraphics{64rawVelo.ps}}
258: \hfill(b)\resizebox{2in}{!}{\includegraphics{128rawVelo+line.ps}}\hfill\hfill\hfill\\
259: \hfill(c)\resizebox{2in}{!}{\includegraphics{256rawVelo+line.ps}}
260: \hfill(d)\resizebox{2in}{!}{\includegraphics{512rawVelo+line.ps}}\hfill}
261: %\vspace{-0.25in}
262: \caption{Plot of the average velocity for each individual realization,
263: $v$, versus the density, $\rho$, for that realization, for an $L\times
264: L$ lattice: (a) $L=64$, (b) $L=128$, (c) $L=256$, (d) $L=512$. Note
265: the emergence of an intermediate phase for $L>64$. The value of $v$
266: for the intermediate state becomes more crisply defined with
267: increasing system size, and that the window of coexistence between the
268: phases broadens. For the $L=512$ system, the average value of $v$ in
269: the intermediate phase is $\left<v_s\right>=0.673 \pm 0.005$. The
270: dotted line shown in (b)-(d) is the prediction from Eq.(\ref{eqn:v_s}).}
271: \label{fig:bml-64-512}
272: \end{figure}
273:
274: \subsection{Square aspect ratios}
275: For small size systems we actually observe the predicted behavior of a
276: sharp transition from freely flowing to total gridlock.
277: Figure~(\ref{fig:bml-64-512}a) is for an $L \times L$ system with
278: $L=64$. It is a plot of the final average velocity observed for each
279: realization, $v$, versus the density for that particular realization.
280: Note we are plotting the average velocity for each individual
281: realization, not an average over all realizations initialized with the
282: same $\rho$ (hence error bars are on the order of the size of the
283: plotting symbol used). Surprisingly, when we implement systems with
284: $L>64$, we observe a bifurcation where two phases start to coexist, as
285: we go from low to intermediate values of $\rho$. The second phase
286: that emerges has average velocity $v\sim 2/3$, as shown in
287: Fig.(\ref{fig:bml-64-512}). In addition, these intermediate states
288: have a very well defined geometry, of bands of red stripes with slope
289: one-half, criss-crossing bands of blue stripes with slope two. An
290: example of the geometry is shown in Fig.(\ref{fig:intermediate}).
291: Jammed wavefronts are located at the intersections of the bands, and,
292: as the systems evolves in time, move as solid structures uniformly
293: down towards the southwest with unit velocity. Particles are freed
294: from the head of each jam, but a like number of new jammed particles
295: aggregate at the tail. As discussed in detail below, the underlying
296: lattice imposes constraints on the allowed topologies of the
297: configurations. In Sec.~\ref{sec:geom}, we derive a simple
298: formulation of the constraints.
299:
300: \begin{figure}[tb]
301: {\hfill(a)\resizebox{2.25in}{!}{\includegraphics{inter512-2.ps}}\hfill\hfill
302: (b)\resizebox{4.15in}{!}{\includegraphics{typGeom2.ps}}\hfill}
303: %\vspace{-0.25in}
304: \caption{Examples of typical intermediate geometry. (a) A system
305: with square aspect ratio where $L=512$, $\wr=1/2$ and $\wb=2$. (b) A
306: system with rectangular aspect ratio where $L=377$, $L'=233$, $\wr=1$,
307: and $\wb=3$. Note that in (a) there are many disordered, random cars
308: in the space between the bands, yet in (b) all cars are ordered. We
309: find this crisp order shown by the latter example for all realizations
310: studied on rectangular aspect ratios with $L$ and $L'$ relatively
311: prime. If $L$ and $L'$ are not relatively prime, random disordered
312: cars located between the bands persist. Note that
313: Fig.(\ref{fig:inter-closeup}) is a close up of the region that has
314: just shed from the jams in (b).}
315: \label{fig:intermediate}
316: \end{figure}
317:
318: \begin{figure}[tb]
319: \rotatebox{270}{\resizebox{3in}{!}{\includegraphics{converge-64-512.ps}}}
320: \vspace{-0.1in}
321: \caption{Median convergence times for the different square aspect ratios
322: simulated. These results are consistent with previous studies. Note a
323: value of $t=2\times 10^6$ should be interpreted as $t>2\times 10^6$.}
324: \label{fig:converge}
325: \end{figure}
326:
327: All realizations contributing to Fig.(\ref{fig:bml-64-512}) were
328: simulated to at least $t=2\times 10^6$ full updates of the entire
329: space, with various realizations simulated for $t>10^8$. We do
330: occasionally observe a realization persist in the $v\sim 2/3$ state
331: for orders of magnitude, then suddenly jump to either $v=0$ or $v=1$,
332: with the latter being more common. Regardless, the intermediate state
333: is at least metastable, persisting for longer than we could simulate
334: most realizations. Furthermore it is ``universal'' in the sense that
335: the value of $v\sim 2/3$ is independent of system size and density,
336: and all systems that do not go to $v=0$ or $v=1$ go to the same
337: intermediate state ({\em i.e.}, same geometric structure and
338: approximate value of $v$). The reason we do not average over the
339: individual realizations, is that it would obscure the
340: behavior. Instead of displaying the three distinct quantized states,
341: averaging would produce a deceitfully smoothly decaying curve.
342:
343: From our data, it is difficult to determine the exact bifurcation
344: point where the phases begin to coexist, and the point where they
345: cease to. We attempted to identify the factors that distinguished
346: realizations which converge to $v=0$ from those, with the same
347: density, that converge to $v\sim 2/3$. We first investigated
348: connections to anisotropy, such as an imbalance between the total
349: number of red versus blue particles. But we found no correlation.
350: The probability a realization would jam or go to the intermediate
351: state is independent of this asymmetry. We also looked at a more
352: fine-grained measure: the line density of red particles versus the
353: blue. Again we found no correlation between this asymmetry and the
354: likelihood of jamming.
355:
356: In Sec.~\ref{sec:geom} we discuss the kinetic mechanism observed,
357: which gives rise to the interleaved band structure exhibited by the
358: intermediate states. As mentioned above, the jam interface moves
359: ballistically, with unit speed, towards the southwest. The width of
360: the jam interface can fluctuate. It seems for the $L=64$ system, the
361: fluctuations are large enough that eventually the head of one jam
362: meets the tail of the previous, continuing until eventually one global
363: jam forms. Kinetic mechanisms aside, it could be that the convergence
364: times for $L=64$ are short enough that the intermediate state cannot
365: be considered metastable. Fig.(\ref{fig:converge}) shows the median
366: convergence times observed for the systems with square aspect ratios.
367: Note that the a value of $t=2\times 10^{6}$ really means $t>2\times
368: 10^{6}$.
369:
370:
371: \subsection{Rectangular aspect ratios}
372: We also implement the BML model on systems with varying rectangular
373: aspect ratios. In particular we study lattices where the two lattice
374: lengths are subsequent Fibonacci numbers.
375: Figure~(\ref{fig:fibo-55x89}a) is the plot analogous to those shown in
376: Fig.(\ref{fig:bml-64-512}), for a system with $L=89$ and $L'=55$. Note
377: we see the same intermediate velocity of $v\sim 2/3$, but we also see
378: the emergence of one more possible phase with $v\sim 2/5$. All the
379: configurations with $v \sim 2/3$ resemble the one shown in
380: Fig.(\ref{fig:fibo-55x89}b), with one red band wrapping around the
381: $\hat{x}$-axis, and three blue bands. The jam points are at the
382: intersections of the bands. Note the crisp, regular geometry.
383: Figure~(\ref{fig:intermediate}b) is for an equivalent, but larger,
384: system with the same Fibonnaci aspect ratio but with $L=377$ and
385: $L'=233$. This figure more clearly illustrates the highly ordered
386: geometry. It also includes the definitions of several of the
387: parameters used in the analysis in the subsequent sections. All
388: realizations with $v\sim 2/5$ resemble the one shown in
389: Fig.(\ref{fig:fibo-55x89}c), with one red band and approximately two
390: blue bands (though the latter are not so clearly defined).
391:
392: \begin{figure}[tb]
393: {\hfill\resizebox{2.5in}{!}{\includegraphics{55x89rawVelo+line.ps}}\hfill\hfill
394: \resizebox{2.5in}{!}{\includegraphics{fibo-type1and2.ps}}\hfill}
395: %\vspace{-0.25in}
396: \caption{(a) The average velocity for each individual realization,
397: $v$, versus the density, $\rho$, for that realization, for an $L\times
398: L'$ lattice, with $L=89$ and $L'=55$. Note the appearance of yet
399: another well defined intermediate state. Note also, the bifurcation
400: point, where phase-coexistence ceases, has not yet been reached,
401: despite the range of $\rho>0.44$. The dotted lines are the predicted
402: velocities from Eqs.(\ref{eqn:vr1}) and (\ref{eqn:vr2}),
403: $v_{r1}\approx 0.7430$ and $v_{r2}\approx 0.3707$. The empirically
404: determined average values are respectively, $\left<v_{r1}\right> =
405: 0.700 \pm 0.002$ and $\left<v_{r2}\right> = 0.364 \pm 0.004$. The
406: geometries of the two types of intermediate states are distinct: (b) A
407: typical configuration with $v\sim 2/3$. Interface slope $s=1$; (c) A
408: typical configuration with $v\sim 2/5$. Interface slope $s\approx
409: 2/3$.}
410: \label{fig:fibo-55x89}
411: \end{figure}
412:
413: \begin{figure}[b]
414: {\hfill(a)\rotatebox{270}{\resizebox{1.75in}{!}{\includegraphics{convTime-55x89.ps}}}\hfill\hfill(b)\rotatebox{270}{\resizebox{1.75in}{!}{\includegraphics{convTau-55x89.ps}}}\hfill}
415: %\vspace{-0.25in}
416: \caption{For simulations with $L=89$ and $L'=55$ and $\rho=0.38$, we
417: plot the sample number versus, (a) the time to reach the periodic
418: limit cycle; (b) the period of the cycle, $\tau$. Eight of the ten
419: realizations simulated reach final configurations of type I, as shown
420: in Fig.(\ref{fig:fibo-55x89}b). They all have the shorter values of
421: $\tau$. Surprisingly six of these eight realizations have the same
422: period, $\tau=5114$, though their microscopic configurations differ.
423: The other two realizations, with significantly larger values of
424: $\tau$, are of type II, as shown in Fig.(\ref{fig:fibo-55x89}c).}
425: \label{fig:convTau}
426: \end{figure}
427:
428: One of the most striking differences when comparing these rectangular
429: aspect ratios to the square, is that for the rectangular, the
430: intermediate configurations are exactly periodic with a short
431: recurrence time: the exact microscopic configuration of particles
432: repeats every $\tau$ updates. We observe systems of size $L=89$ and
433: $L'=55$, settling into the periodic behavior typically in a time less
434: than $t \sim 100,000$ updates, with a period on the order of $\tau
435: \sim 6000$ updates. Figure~\ref{fig:convTau} is a plot illustrating
436: such typical behavior. Another striking difference, is that not one
437: realization has jammed fully ($v=0$), despite the fact that the
438: largest density simulated is $\rho = 0.45$. Note that for the square
439: aspect ratios, the bifurcation point where the phases cease to coexist
440: is at $\rho \approx 0.40$. Another striking difference is the lack
441: of disorder for the relatively prime systems with rectangular aspect
442: ratios. In Fig.(\ref{fig:intermediate}a) there are isolated particles
443: (``dislocations'') moving in the area between the bands. However for
444: the rectangular case, all the particles manage to join the ordered
445: bands.
446:
447: %\input{geom}
448: \section{Geometric considerations}\label{sec:geom}
449: \subsection{Kinetic pathways}\label{subsec:kinetics}
450: By watching the dynamical evolution of the system, starting from the
451: initial configuration, the mechanism by which the intermediate phases
452: form can be observed. Often one global jam initially begins to form,
453: yet the head of the jam just fails to meet up with the tail, leaving a
454: few lattice sites of distance between them. Particles shed from the
455: head as soon as allowed by the local environment (since all particles
456: move whenever possible), leaving with a well defined order. See
457: Ref.~\cite{bml-webpage}, for video images of the dynamical evolution.
458:
459: To understand the pattern formed by the shedding, consider first one
460: row of a solid isolated block of red particles. Since the particles
461: only advance provided the site they wish to occupy is empty, a
462: particle would leave the head of the jam only every other timestep.
463: However, we can now consider a diagonal interface formed by a
464: triangular block of red particles in contact with a triangular block
465: of blue particles. See Fig.(\ref{fig:shed}). Note red (eastbound)
466: particles are represented by ``x'' and blue (northbound) particles by
467: ``o''. Each subsequent time step illustrated corresponds to one
468: complete update of the space ({\em i.e.}, one north-step followed by
469: one east-step). Recall, all particles of the same species update
470: synchronously. Step 1: No ``o's'' are able to move. But, during the
471: east-phase of the timestep, the first ``x'' moves away from the jam,
472: opening up a space. Step 2: The first ``o'' advances. This blocks
473: all other ``x's'' in the original row, yet opens up a space for an
474: ``x'', one site from the original ``x'' along the southwest diagonal,
475: to advance. The original ``x'' to move, also continues advancing.
476: Step 3: That first ``o'' continues advancing and opens up a space for
477: an ``x'' in the original row to advance---two time steps delayed from
478: the first ``x'' to move in that row. The ``o'' below this original
479: ``o'' is currently blocked. However an ``o'' one site from the
480: original along the southwest diagonal is now free to advance. Note
481: this ``o'' blocks the ``x'' in its newly occupied row from advancing,
482: yet opens up a space for the ``x'' one row southward to advance.
483: Hence a pattern emerges: a red particle sheds from within the same row
484: only every third timestep, yet from a site further southwest every
485: timestep, yielding red bands of density $\rho_r = 1/3$ with slope
486: $s_r=1/2$. Likewise for the blue particles, a blue particle sheds
487: from within the same column only every third timestep, yet from a
488: successive site along the southwest every timestep, yielding blue
489: bands of density $\rho_b = 1/3$ with slope $s_b = 2$. The jams occur
490: at the intersections of the bands, and the interface of a jam has
491: slope $s$. Typically, $s=1$. In Fig.(\ref{fig:shed}) one can see the
492: order beginning to emerge.
493: \begin{figure}[t]
494: \resizebox{5in}{!}{\includegraphics{stripes2-invert.eps}}
495: \vspace{-0.15in}
496: \caption{Shedding from a jam.}
497: \label{fig:shed}
498: \end{figure}
499:
500: More illuminating is to view a closeup near the jammed regions.
501: Fig.(\ref{fig:inter-closeup}) is a zoomed in view of the region near
502: one of the jams shown in Fig.(\ref{fig:intermediate}b). Note the
503: order that exists in the region above the jam interface, which has
504: shed from the jam in the manner described above. In this region there
505: are alternating diagonal stripes of blue, red, and empty. The stripes
506: have ``phase locked''---on the next update the blue stripes move into
507: the empty stripe regions, leaving room for the red stripes to move on
508: the subsequent update, and so on. Hence the system has organized
509: itself into the highest density packing that still allows all
510: particles to move with $v=1$.
511: % say something about non-interacting lattices????!!!!
512: \begin{figure}[t]
513: \resizebox{3.5in}{!}{\includegraphics{geom-closeup.ps}}
514: \caption{A zoomed in view of a jam shown in Fig.(\ref{fig:intermediate}b).}
515: \label{fig:inter-closeup}
516: \end{figure}
517:
518:
519: %\input{winding}
520: \subsection{Winding number}\label{subsec:winding}
521: Since the system lives on a torus, the bands must wrap seamlessly
522: around it. Noting these facts, we can develop a mathematical
523: expression for the number of red bands, $\omega_r$ and the number of
524: blue, $\omega_b$, (referred to respectively as the ``winding'' number
525: for red and blue), that must be present. Essentially, we can
526: calculate the length of the regions of slope-$1/2$, slope-$s$, and
527: slope-$2$ that must be present for the bands to wrap around the torus.
528: (The slopes are respectively those of the red band, the interface, and
529: the blue band). Consider an $L\times L'$ lattice, and a red band
530: starting in the lower left hand corner. Moving out along the
531: $\hat{x}$-direction, there are $k$ sites with slope-$1/2$ and $k'$
532: sites with slope-$s$. Similarly for a blue band starting in the lower
533: left hand corner, there are $m$ sites with slope-$2$ and $m'$ sites
534: with slope-$s$. The constraints are:
535: \begin{eqnarray}
536: k + k' & = & L \label{eqn:k1} \\
537: \frac{1}{2} k + sk' & = & \omega_r L' \label{eqn:k2} \\
538: m + m' & = & L \\
539: 2 m + sm' & = & \omega_b L' \\ \label{eqn:m2}
540: k' & = & m', \label{eqn:interface}
541: \end{eqnarray}
542: where, due to the lattice, $\{k, k', m, m' \}$ are positive integers,
543: and $\{ \omega_r,\omega_b \}$ are either positive integers or equal to
544: $1/g$, where $g$ is a positive integer ({\em e.g.}, for the
545: configuration shown in Fig.(\ref{fig:intermediate}a), $\wr=1/2$). The
546: final equation, Eq.~(\ref{eqn:interface}), expresses that the length
547: of the jam interface must be the same for the red and blue cars.
548: Combining this system (of five equations in seven unknowns), we can
549: solve some combination of variables in terms of the others. Solving
550: first for $k'$:
551: \begin{equation}\label{eqn:kprime}
552: k' = \left(2\wr L' - L\right)/(2s-1).
553: \end{equation}
554: Using this, we can solve for $\omega_b$ in terms of $\omega_r$, $s$,
555: $L$, and $L'$, and obtain:
556: \begin{equation}\label{eqn:winding}
557: \wb = \frac{2L}{L'} - \frac{(2-s)}{(2s-1)}\left(2\wr-\frac{L}{L'}\right).
558: \end{equation}
559:
560: Knowing that the interface slope $s\sim 1$, we can tabulate the
561: allowed values of $\omega_r$ and $\omega_b$ in terms of the aspect
562: ratio of the space, $L/L'$. The allowed values for various aspect
563: ratios are listed in Table~\ref{table:winding}. Note, the value
564: $\omega_r=1/2$ means the red band has only reached height $L'/2$ in
565: traversing distance $L$ (as in Fig.(\ref{fig:intermediate}a)). We
566: implement systems with the various aspect ratios shown in
567: Table~\ref{table:winding}, and find the experimental configurations
568: observed all match the predicted behavior. See in addition to the
569: previous figures, Fig.(\ref{fig:bml-aspect3}). The final lines in
570: Table~\ref{table:winding} are for systems where we observe empirically
571: the values of $\wr$ and $\wb$, and using Eq.(\ref{eqn:winding}) can
572: predict the value of $s$. Recall $\wr$ and $\wb$ must be integers.
573: The system seems to tune the value of $s$ to allows this. For
574: instance, if $L$ and $L'$ are successive Fibonacci numbers ({\em
575: i.e.}, $L/L' = (1+\sqrt{5})/2$), a rearrangement of
576: Eq.(\ref{eqn:winding}) predicts $s=1.17$, which matches our empirical
577: observations. Note in Fig.(\ref{fig:intermediate}b), the upper jam
578: has small glitches where $s>1$.
579: \begin{table}
580: \begin{tabular}{ccccc}
581: $L/L'$ & \ \ \ \ \ \ $s$ \ \ \ \ & \ \ \ $\omega_r$ \ \
582: \ & \ \
583: \ $\omega_b$ \ \ \ & \ \ \ \ \ \ $v $ \ \ \ \ \ \ \\ \hline
584: 1 & 1 & 1/2 & 2 & $0.673 \pm 0.005$ \\
585: 1 & 1 & 1 & 1 & 0 \\
586: 2 & 1 & 1 & 4 & $\sim$ 2/3\\
587: 2 & 1 & 2 & 2 & 0 \\
588: 5/3 & 1 & 1 & 3 & $\sim$ 2/3\\
589: 3 & 1 & 2 & 5 & $\sim$ 2/3\\
590: {\scriptsize{(1+$\sqrt{5}$)/2}} & 1.17 & 1 & 3 & $0.700 \pm 0.002$\\
591: {\scriptsize{(1+$\sqrt{5}$)/2}} & 7/10 & 1 & 2 & $0.364 \pm 0.004$\\
592: \end{tabular}
593: \caption{Allowed winding numbers, $\wr$ and $\wb$, for lattices with
594: different aspect ratios, $L/L'$, and interface slope $s$. The values
595: of $v$ which include error bars are from our numerical simulations.}
596: \label{table:winding}
597: \end{table}
598: \begin{figure}
599: \resizebox{4.5in}{!}{\includegraphics{bml-769x256.ps}}
600: \caption{A typical realization with $L=769$ and $L'=256$ ({\em i.e.}. $L =
601: 3L'+1$, where the additional lattice site is to make the lengths
602: relatively prime). Note, that as predicted in
603: Table~\ref{table:winding}, the configuration is consistent with $s=1$,
604: $\wr = 2$, and $\wb=5$. Realizations with $L=768$ and $L'=256$ ({\em
605: i.e.}, $L=3L'$) have a similar structure, but surprisingly lack the
606: crisp order, and instead have disordered cars at random locations
607: between the bands. For images of the latter, see
608: Ref.~\cite{bml-webpage}.}
609: \label{fig:bml-aspect3}
610: \end{figure}
611:
612: %\input{aveVelo}
613: \subsection{Average velocity}
614: Figure~\ref{fig:intermediate}, illustrates the typical geometry for
615: realizations with $v\sim 2/3$. In this example, $\wr=1$ and
616: $\wb=3$. We label on the figure the lengths $a$, $b$, $c$, and $d$,
617: denoting respectively the width of the red band, the blue band, the
618: blue jam and the red jam. We define two discontinuous functions which
619: will simplify notation later on:
620: \begin{equation}
621: \Theta_r(\wr) = \left\{ \begin{array}{ll}
622: 1 & {\rm \ \ if \ \ } \wr \ge 1 \\
623: 1/\wr & {\rm \ \ if \ \ } \wr < 1,
624: \end{array}
625: \right.
626: \end{equation}
627: \begin{equation}
628: \Theta_b(\wb) = \left\{ \begin{array}{ll}
629: \wb & {\rm \ \ if \ \ } \wb \ge 1 \\
630: 1 & {\rm \ \ if \ \ } \wb < 1,
631: \end{array}
632: \right.
633: \end{equation}
634: In other words, $\Theta_r$ is the number of independent red bands in
635: one column of the lattice. $\Theta_b$ is the number of independent
636: blue bands in one row of the lattice. If we denote the density of the
637: red band as $\rho_a$, we can express the average number of red
638: particles in a column of the space:
639: \begin{equation}\label{eqn:a}
640: \Theta_r(\wr) \rho_a a = \rho L'/2.
641: \end{equation}
642: Likewise, the average number of blue particles in a row:
643: \begin{equation}\label{eqn:b}
644: \Theta_b(\wb) \rho_b b = \rho L/2.
645: \end{equation}
646: Recall $\rho$ is the overall particle density (red plus blue).
647: Empirically we determined that $\rho_a = \rho_b = 1/3$, which is a
648: consequence of the dynamics described in Sec.~\ref{subsec:kinetics}.
649: Using the basic equations described in Sec.~\ref{subsec:winding} and
650: knowledge of the ``typical'' geometry, we can solve for the velocity
651: of the intermediate state. Unfortunately, we have to consider the
652: square and rectangular aspect ratios independently, and do not have
653: one equation that describes all cases. The assumptions described
654: above are valid for the rectangular aspect ratios, but fail to capture
655: the full behavior of the square ones.
656:
657: \subsubsection{Rectangular aspect ratios, type I:} Configurations such as
658: the one shown in Fig.~(\ref{fig:intermediate}) have a regular, highly
659: structured, geometry that is well described by the formalism in
660: Sec.~\ref{subsec:winding}. We find that the values of $k'$ predicted
661: by Eq.~\ref{eqn:kprime} exactly match those obtained experimentally.
662: Typically there are $n_j$ jams, with the particles equally divided
663: amongst them. The interface width per jam is $k'_i = k'/n_k$, where
664: $n_k$ is the total number of jams one red band is involved in, as it
665: wraps once around the $\hat{x}$-axis. See
666: Fig.(\ref{fig:intermediate}a), for example, where $n_k\approx 2$, and
667: $n_j=3$.
668:
669: The structure of the jams and the relevant geometric factors are shown
670: in Fig.~\ref{fig:geom}(a). The jams form trapezoidal shapes of width
671: $\delta$. From simple geometric considerations, we can show that
672: $\delta = k'_i \sin(\theta-\phi) \equiv \kappa k'_i$. The jam width
673: per line, $d$, and per column, $c$ is
674: \begin{equation}
675: c = d = \frac{\delta}{\cos \xi} = \frac{\sin(\theta-\phi)}{\cos \xi}
676: k'_i \equiv \Gamma k'_i.
677: \end{equation}
678: However, note this is only true provided that there are enough
679: particles available in each column and in each row. Otherwise,
680: \begin{equation}
681: d=d_{\rm max}=\rho L/2\wr \Theta_r(\wr), {\rm \ \ \ and \ \ \ }
682: c=c_{\rm max}=\rho L' \Theta_b(\wb)/2\wb.
683: \end{equation}
684:
685:
686: From similar geometric considerations we can show the length of the
687: blue jam, $l_b$, is approximately
688: \begin{equation}
689: l_b \approx b \left\{\cos \phi +
690: \frac{\sin \phi }{\tan(\sigma - \phi)} \right\}
691: \equiv b \gamma.
692: \end{equation}
693: Similarly, the length of the red jam, $l_r$, is approximately
694: \begin{equation}
695: l_r \approx a \left\{\cos \xi +
696: \frac{\sin \xi}{\tan\left(\sigma -
697: \phi\right)}\right\} \equiv a \varphi.
698: \end{equation}
699: The total number of particles involved in the jams, $J$, is the number
700: of jams multiplied by the width and length:
701: \begin{eqnarray}\label{eqn:j-rect}
702: J & \approx & n_j \delta \left(l_b + l_r\right) \nonumber \\
703: & = & \left(2\wr L' - L\right) \kappa \left[\left(\frac{3}{2} \rho
704: L/\Theta_b(\wb)\right)\gamma + \left(\frac{3}{2} \rho
705: L'/\Theta_r(\wr)\right)\varphi \right] \frac{n_j}{n_k} .
706: \end{eqnarray}
707: For the data plotted in Fig.~\ref{fig:fibo-55x89}, $L'$ and
708: $L$ are two successive Fibonacci numbers, hence $L/L' = (1+\sqrt{5})/2
709: \equiv {\cal{F}}$. In agreement with expected values shown in
710: Table~\ref{table:winding}, $\wr=1$ (thus $\Theta_r(\wr)=1$), and
711: $\wb=3$ (thus $\Theta_b(\wb)=3$). For this geometry, $n_j=n_k=2$.
712: Plugging in these values into Eq.~\ref{eqn:j-rect},
713: \begin{equation}
714: J = (2 - {\cal{F}}) \kappa \left[\frac{1}{2} \gamma + \frac{3}{2{\cal{F}}}
715: \varphi\right] \rho L L'
716: \approx 0.2570 \cdot \rho L L'.
717: \end{equation}
718: Solving for the velocity, noting that the overall number of particles,
719: $N=\rho L L'$,
720: \begin{equation}\label{eqn:vr1}
721: v_{r1} = 1 - \frac{J}{N} \approx 0.7430.
722: \end{equation}
723: Note, this is independent of $\rho$ and independent of $L$ and $L'$.
724: This predicted value for $v_{r1}$ is included in the plot of the
725: experimentally determined velocities, shown in
726: Fig.~\ref{fig:fibo-55x89}. Note the calculated value slightly
727: underestimates the number of particles involved in the jam (hence
728: slightly overestimates $v_{r1}$).
729: \begin{figure}[tb]
730: \hfill\resizebox{2in}{!}{\includegraphics{type1.eps}}\hfill\hfill
731: \resizebox{1.5in}{!}{\includegraphics{type2.eps}}\hfill
732: \vspace{-0.25in}
733: \caption{Typical jam configurations. For type I jams, shown in (a), $c=d=\Gamma
734: k'$. For type II jams, shown in (b), $d=d_{\rm max}$ and $c=c_{\rm
735: max}$. Knowing the slopes of the lines in (a), $s=1$,$s_r=2$, and
736: $s_b=1/2$, we can determine all of the angles: $\phi =
737: \tan^{-1}(1/2)$, $\theta=\pi/4$, $\sigma = \tan^{-1}2$, and $\xi =
738: (\pi/2 - \tan^{-1}2)$.}
739: \label{fig:geom}
740: \end{figure}
741:
742:
743: \subsubsection{Rectangular aspect ratios, type II:} As mentioned,
744: rectangular lattices are well described by the formalism in
745: Sec.~\ref{sec:geom}. We observe the ``rich'' jam described above, but
746: also a second type of ``depleted'' jam (not enough
747: particles). Empirically, all observations of this type have $s\approx
748: 2/3$, and one large jam. See for instance
749: Fig.~(\ref{fig:fibo-55x89}c). The per row and per column width of the
750: jams are the maximum, $d_{\rm max}$ and $c_{\rm max}$ respectively,
751: since $\Gamma k'$ is greater than the number of particles available in
752: a row or column. A jam interface of length $k'$ and slope $s$ (with
753: $s>1$) involves $k'$ columns, but only $sk'$ rows. Hence the total
754: number of particles involved in the jams:
755: \begin{eqnarray}
756: J & = & k' \left(s\cdot d_{\rm max} + c_{\rm max}\right) \nonumber \\
757: & = & \left[(2\wr L' - L)/(2s-1)\right]\left[s \rho L/2\wr\Theta_r(\wr) +
758: \rho L' \Theta_b(\wb)/2\wb\right].
759: \end{eqnarray}
760: For the realizations contributing to the plot in Fig.~\ref{fig:fibo-55x89},
761: $L/L' = (1+\sqrt{5})/2 \equiv {\cal{F}}$, $\wr=1$ (thus
762: $\Theta_r(\wr)=1$), $\wb=2$ (thus $\Theta_b(\wb)=2$), and $s=2/3$
763: (which is empirically determined). Plugging these values in we find:
764: \begin{equation}
765: J = \frac{5}{2}\left(\frac{2}{{\cal{F}}} -
766: 1\right)\cdot\left(\frac{7}{20}{\cal{F}} + \frac{1}{2}\right)\cdot\rho
767: L L' \approx 0.6293 \cdot \rho L L'.
768: \end{equation}
769: Thus the average velocity,
770: \begin{equation}\label{eqn:vr2}
771: v_{r2} = 1 - \frac{J}{N} \approx 0.3707.
772: \end{equation}
773: We include this predicted value in the plot of
774: Fig.~\ref{fig:fibo-55x89}. Note the agreement with the experimental
775: data.
776:
777: \subsubsection{Square aspect ratios:} The typical geometry of an
778: intermediate state for a square lattice is shown in
779: Fig.(\ref{fig:intermediate}a). The interfaces and especially the
780: edges of the bands are disordered and jagged. The slope assumptions
781: only hold approximately: $s_r \approx 1/2$ and $s_b\approx 2$.
782: Furthermore, there are several particles moving freely in the low
783: density regions, unlike for the rectangular lattices, where all
784: particles eventually order into the bands and jams. Also unlike for
785: the rectangular lattices (see for instance
786: Fig.(\ref{fig:intermediate}b)), the red and blue bands cross through
787: each other without the pronounced shifting upwards.
788:
789: We cannot use the formalism developed above in
790: Sec.~\ref{subsec:winding} for this situation, since that formalism is
791: based solely on geometric constraints of winding seamlessly around the
792: lattice. A configuration on a square lattice with $s_r=1/2$, $s_b=2$,
793: $s=1$, would have $w_r=1/2$. Plugging into Eqn.~(\ref{eqn:kprime}) we
794: find the required length of the overall interface, $k'=0$. Instead,
795: empirically we find the blue jams form a trapezoidal shape of
796: approximate length $b$ and height $a/4$. Likewise the red jams form a
797: trapezoid of approximate length $a$ and width $b/4$. Each jam has
798: this shape and there are $n_j=3$ jams altogether ({\em i.e.}, three
799: distinct intersections of the bands). The number of particles involved
800: in jams, $J$ is:
801: \begin{equation}
802: J \approx n_j (a b/4 + a b/4) = 3 a^2/2.
803: \end{equation}
804: Using Eq.~\ref{eqn:a} to solve for $a$ and the fact that the overall
805: number of particles, $N=\rho L L'$, we can solve for the fraction of
806: particles in the jammed state:
807: \begin{equation}
808: \frac{J}{N} \approx \frac{3}{2} \left( \frac{3}{4} \rho L
809: \right)^2 \frac{1}{\rho L^2} = \frac{27}{32}\rho.
810: \end{equation}
811: Hence the velocity,
812: \begin{equation}\label{eqn:v_s}
813: v_s \approx 1 - \frac{J}{N} = 1 - \frac{27}{32}\rho.
814: \end{equation}
815: This predicted value $v_s$ is included in the plots of
816: Fig.~\ref{fig:bml-64-512}. It captures the features of the
817: experimental data, including the slight decrease in $v_s$ with
818: increasing $\rho$.
819:
820:
821: %\input{discussion}
822: \section{Discussion}\label{sec:disc}
823:
824: The BML traffic model is a simple model of a jamming transition with
825: self organization. In our study, instead of agreement with
826: conventional beliefs, we find stable intermediate configurations with
827: phase coexistence of jammed and free-flowing traffic. Such
828: configurations have not been previously reported in the literature,
829: despite the extensive amount of past work on the BML model.
830: Furthermore, these intermediate configurations have interesting
831: geometric and topological properties, with different behaviors
832: resulting as a consequence of different aspect ratios of the
833: underlying lattice. We develop a formalism, based on geometric
834: constraints imposed by the lattice, to predict the asymptotic
835: velocities of the coexisting phases. Visualizing the kinetic pathways
836: of the evolving configurations was a key element in uncovering the
837: existence of the intermediate phases and, moreover, their periodic
838: nature on lattices with relatively prime aspect ratios. The
839: observations described in this manuscript open up a range of new
840: questions about the BML model.
841:
842: As mentioned, instead of a phase transition as a function of density,
843: we observe a bifurcation point where the intermediate states first
844: begin appearing, and a second bifurcation point, where they completely
845: cease to appear. Perhaps more interesting than predicting the
846: asymptotic velocities, would be to calculate the locations of the
847: bifurcation points. From our experimental data, the exact location of
848: the bifurcation points are difficult to determine, and moreover, also
849: depend on the aspect ratio of the underlying lattice.
850:
851: It is possible that there is a sharp phase transition. However, in
852: such a case, the density $\rho$ would not be the appropriate order
853: parameter. Perhaps a more appropriate order parameter would be an
854: interaction energy between north-bound and east-bound particles. Note
855: that when in the free-flowing state, the north and east particles have
856: moved onto non-interacting lattices. It may be possible that one can
857: define an initial energy based on the overlap or interaction between
858: two lattices, and use that as an order parameter.
859:
860: A complication which makes theoretical treatment of the BML model
861: difficult, is that it is not strictly monotonic. Adding particles to a
862: configuration that is known to jam ({\em i.e.}, increasing $\rho$),
863: can actually change the sequence of particle interactions and result
864: in that configuration going to free-flowing instead of jamming.
865: Furthermore, it is known that certain discrete models with the same
866: property as BML---namely that the randomness is in the initial
867: condition, yet the dynamics fully deterministic---can be notoriously
868: difficult to deal with analytically. Examples include bootstrap
869: percolation\cite{Holroyd-bootstrap} and the Lorentz lattice
870: gas\cite{grimmett-pinball}. We modified the BML model to include a
871: small probability for particles to flip species-types at each update.
872: Our preliminary studies, adding this small amount of randomness to the
873: dynamics, suggest that the model with randomness has extremely
874: different geometric properties than the original BML model. In
875: addition, for the model with randomness, we did not observe the
876: intermediate configurations described herein.
877:
878:
879: \acknowledgments{This work has benefited greatly from discussions with
880: L\'{a}szl\'{o} Lov\'{a}sz, Alexander Holroyd, and Roman Kotecky.}
881:
882: \bibliographystyle{unsrt}
883: \bibliography{/home/raissa/Bibl/percolation,%
884: /home/raissa/Bibl/cars,%
885: /home/raissa/Bibl/ca,%
886: /home/raissa/Bibl/markov,%
887: /home/raissa/Bibl/statistics,%
888: /home/raissa/Bibl/rwalks}
889:
890: \end{document}
891: