1: % From Finite to Infinite Range Order via Annealing ...
2: % by DPV and JPC
3: % Started 09Mar03
4: % dpv: 6/26/03
5: % jpc: 7/4/03, 7/10/03
6:
7: \documentclass[prb,twocolumn,showpacs,superscriptaddress,preprintnumbers,amsmath,amssymb,floatfix]{revtex4}
8:
9: % Required packages
10: \usepackage{dcolumn}
11: \usepackage{graphicx}
12: \usepackage{bm} % bold math
13:
14: \input{cmechabbrev}
15: \input{spectraabbrev}
16:
17: \begin{document}
18:
19: \title{From Finite to Infinite Range Order via Annealing:\\
20: The Causal Architecture of Deformation Faulting in\\
21: Annealed Close-Packed Crystals
22: }
23:
24: \author{D. P. Varn}
25: \affiliation{Santa Fe Institute, 1399 Hyde Park Road, Santa Fe,
26: New Mexico 87501}
27:
28: \author{J. P. Crutchfield}
29: \affiliation{Santa Fe Institute, 1399 Hyde Park Road, Santa Fe,
30: New Mexico 87501}
31:
32: \date{\today}
33: \begin{abstract}
34: We analyze solid-state phase transformations that occur in zinc-sulfide
35: crystals during annealing using a random deformation-faulting mechanism with
36: a very simple interaction between adjacent close-packed double layers. We show that, through
37: annealing, infinite-range structures emerge from initially short-range crystal
38: order. That is, widely separated layers carry structurally significant
39: information and so layer stacking cannot be completely described by any
40: finite-range Markov process. We compare our results to two experimental
41: diffraction spectra, finding excellent agreement.
42: \end{abstract}
43:
44: \pacs{
45: 61.72.Dd, % Experimental determination of defects by diff/scattering
46: 61.10.Nz, % Single crystal/powder diffraction
47: 61.43.-j, % Disordered Solids
48: 81.30.Hd % Constant-Composition Solid State Transformations\\
49: \\
50: \begin{center}
51: }
52:
53: \maketitle
54:
55: There has been considerable interest in understanding planar disorder in
56: crystals for some time.~\cite{hendricks42,jagodzinski49a,pandey80a,pandey80b,pandey80c,berliner86,gosk01}
57: There are several reasons for this emphasis.
58: Since planar defects correspond to a shift of an entire layer of atoms that
59: nonetheless preserves the crystallinity within the layer, the disorder
60: is confined along the stacking direction and results in a structure that
61: may be treated as one-dimensional, making theoretical analysis tractable.~\cite{varn01a}
62: Additionally, different stacking configurations can affect the physical
63: properties of the material. For example, the band gap in SiC changes as the
64: stacking structure is changed~\cite{sebastian94} and an anomalous photovoltaic effect
65: has been observed in ZnS crystals that have disordered stacking sequences.~\cite{ellis58}
66:
67: While many crystals exhibit planar defects, they are especially common in a
68: class of materials known as {\em polytypes}.~\cite{sebastian94} Polytypism is
69: the phenomenon where a crystal is built up from the stacking of identical
70: two-dimensional layers, called {\em modular layers}
71: (MLs).~\cite{varn01a,varn02,varn03a,varn03b} Typically there is a small, finite set of possible
72: ways a ML may be stacked upon another and, since these different stacking
73: orientations often preserve the atomic coordination number for nearest and
74: next-nearest neighbors, the energy difference between different stackings can
75: be quite small. It is perhaps not surprising then that polytypic materials have many different
76: crystalline structures. There are, for instance, $185$ known crystalline
77: stacking structures in ZnS, some with unit cells extending over $100$
78: MLs.~\cite{sebastian94,trigunyat91} Other highly polytypic materials include
79: SiC, CdI$_2$, and AgI with about $150$, $200$, and $50$ known crystalline
80: structures each, respectively.~\cite{sebastian94}
81:
82: Understanding the variety and origin of spatial organization in crystalline
83: ZnS polytypes on length scales clearly in excess of the calculated inter-ML
84: interactions---$\sim$ 1 ML for ZnS~\cite{engel90b} and $\sim$ 3 MLs for
85: SiC~\cite{cheng87}---has been a puzzle for some time and numerous theories
86: have been proposed.~\cite{price84,yeomens88,frank51b,trigunyat91} Recently,
87: we extended this by demonstrating that {\em disordered} ZnS polytypes also
88: possess a long-range spatial organization in excess of the calculated
89: inter-ML interaction.~\cite{varn02,varn03b}
90:
91: In this paper we simulate transformations between ordered and disordered
92: polytypes, in particular, proposing a simple model to analyze the solid-state
93: transformation of annealed ZnS crystals from a 2H to a 3C structure and
94: finding a novel structural description---in the form of an
95: \eM~\cite{crutchfield89, Crut98d}---of the resulting disordered twinned
96: 3C crystal. Our model is decidedly simpler than previous ones, designed
97: with the goal of determining the minimal complication necessary to produce
98: the experimentally observed long-range spatial order. For example, from
99: the \eM\ we show that infinite-range spatial memories arise even though the
100: interaction length between MLs is restricted to nearest neighbors.
101:
102: Disordered crystals are often formed by stressing a perfect crystal thermally,
103: mechanically, or through irradiation so that stacking faults are introduced.
104: Various models have been proposed to explain these transformations. Typically,
105: the faulting process is assumed to proceed slowly, so that a ML is randomly
106: chosen and a fault is introduced if the local stacking configuration meets
107: some criteria, usually derived from a (theoretically or empirically determined)
108: Hamiltonian that describes the energetics of the inter-ML interactions. Further,
109: a particular faulting mechanism is assumed.
110:
111: Many authors have previously performed simulation studies on transformations in
112: polytypes to understand the structure of faulted crystals.
113: For example, by assuming interactions between MLs of up to a distance of three,
114: Kabra and Pandey~\cite{kabra88} were able to deduce that layer-displacement
115: faulting was the primary defect found in the 2H $\rightarrow$ 6H~\cite{note1}
116: transformation in SiC. They also discovered that this faulting mechanism
117: can result in long-range correlations between MLs without short-range order.
118: Engel~\cite{engel90a} applied a similar model to treat the 2H $\rightarrow$ 3C
119: transformation in ZnS via deformation faulting, with the assumption of
120: next-nearest neighbor interactions. These authors were able find stacking
121: configurations that gave x-ray diffractograms qualitatively similar to
122: experimental ones.
123: Shrestha {\it et al.}~\cite{shrestha96a} treated the 2H $\rightarrow$ 3C
124: martensitic transformation
125: in hypothetical close-packed structures by introducing the infinitely
126: strong repulsion model. They assumed that the presence of a fault inhibited
127: the introduction of another fault on adjacent MLs. That is, that there was some
128: coordination between faults, called {\em nonrandom faulting}.~\cite{sebastian87a,sebastian87b}
129: Gosk~\cite{gosk00,gosk01} developed a model to describe disordered 2H crystals
130: that used a probability function for faulting on the next ML as one scans the
131: crystal that depended on the distance from the last observed fault.
132: By tuning various model parameters one could introduce nonrandom faulting into
133: the stacking sequence.
134:
135: ANNNI models~\cite{yeomens88} have also been used to explain polytypism since
136: they are known to give long-range spatial organization with only short-range
137: interactions (up to next-nearest neighbor). However, ANNNI models require
138: fine tuning the coupling parameters in an interaction Hamiltonian and, thus,
139: may be of limited applicability to polytypes.
140:
141: ZnS has two stable phases: the hexagonal close-packed (or
142: 2H)~\cite{varn03a,varn03b} for temperatures above 1024 C and the cubic
143: closed-packed (or 3C) for temperatures below 1024 C.~\cite{sebastian94}
144: If 2H ZnS crystals are cooled sufficiently fast to room temperature, they will retain
145: this 2H structure even though it is not the stable phase; presumably due to
146: the difficulty of shifting MLs to form the cubic (3C) structures. These
147: crystals are often further annealed for an hour at temperatures $500-800$ C,
148: thus perhaps providing the necessary activation energy to induce a solid-state
149: transformation from the 2H to a twinned 3C structure. This transformation has
150: been studied before, and many authors have concluded that deformation faulting
151: is the main mechanism responsible,~\cite{roth60,varn03b} at least for the case
152: of weak faulting.
153:
154: Engel and Needs~\cite{engel90b} performed a first-principles pseudopotential
155: energy calculation at $T=0$ to determine the coupling constants between MLs
156: up to separation three. They found an expression for the energy in the form
157: \begin{eqnarray}
158: {\mathcal H} = -\sum_{m=1}^{2} J_m \sum_{i=1}^{N}
159: \Bigl[ \bigl( s_{i}-\frac{1}{2}\bigr) \bigl( s_{i+m}-\frac{1}{2}\bigr) \Bigr]~,
160: \label{eq:hexagonality}
161: \end{eqnarray}
162: where the $s_i \in\{0,1\}$ are \emph{inter-ML spins}~\cite{varn02} and $J_m$
163: the inter-ML coupling constants at separation $m$. They found $J_1 > 0$ and
164: $J_2 < 0$, but much smaller in magnitude than $J_1$. All other couplings
165: between MLs were found to be negligible.
166:
167: Our approach is to assume the following very simple model describes the
168: 2H $\rightarrow$ 3C transformation in ZnS.
169:
170: \begin{trivlist}
171:
172: \item (i) {\it The energetics of ML stacking in ZnS is describable by a
173: one-dimensional Ising chain with a nearest-neighbor interaction Hamiltonian
174: ${\mathcal H}$ between MLs}. Further, we assume that entropic effects are small
175: at all temperatures of interest, so that the free energy $F = {\mathcal H}$.
176:
177: \item (ii) {\it The effective coupling $J_1(T)$ between adjacent MLs is
178: temperature dependent, being positive at temperatures below the transition,
179: $T_c \approx 1000$ C, and negative above.} While little is known about the
180: temperature dependence of the effective couplings $J_i(T)$,~\cite{engel90a}
181: this assumption gives the correct stable phases for ZnS (2H and 3C) above and below the
182: the transition temperature.
183:
184: \item (iii) {\it There is only one faulting mechanism---deformation
185: faulting---and it is driven by Glauber dynamics.} Deformation
186: faulting~\cite{varn03a} occurs in close-packed crystals when there is a slip
187: by a non-Bravais lattice vector between adjacent MLs. In terms of the spin
188: sequence, this corresponds to flipping a single spin~\cite{varn03a} and so
189: the faulting reduces to basic Glauber dynamics.~\cite{glauber63}
190:
191: \item (iv) {\it The transformation is slow, sluggish, and random.} By
192: \emph{slow}, we mean that the time for a slip between MLs to occur is much
193: shorter than the interval between slips. By \emph{sluggish}, we mean that a
194: slip will only occur if it is energetically favorable. And by \emph{random},
195: we mean that each ML is equally likely to be selected for possible faulting,
196: without any coordination between MLs.
197: Since a deformation fault occurs when one portion of the crystal slips relative
198: to the other, it is reasonable to assume that this happens infrequently and
199: only when the slip results in a reduction of the energy.
200: While the {\em mechanism} of faulting assumed here is expressly random, we
201: will show nonetheless that this leads to a crystal {\em structure} that has a nonrandom
202: fault distribution.
203:
204: \end{trivlist}
205:
206: Our model is most similar to that introduced by Engel.~\cite{engel90a} He
207: assumed two kinds of faults, fast and slow, depending on the configuration
208: of the five-spin neighborhood centered about the candidate spin to be flipped.
209: Restricting the range of interaction to nearest-neighbors, as we do, only
210: has the effect of suppressing the slow fault mechanism.
211:
212: We simulate the solid-state transformation in ZnS by assuming an initial
213: configuration of a pure 2H crystal with $N = 1048577$ MLs, which is quickly brought
214: into a temperature range ($500-800$ C) so that the 3C structure becomes the
215: stable phase, but there is sufficient thermal energy for MLs to slip.
216: We randomly choose a spin in the configuration and apply the \emph{update rule}
217: listed in Table~\ref{update}. We visit the spins in a round-robin fashion,
218: so that each is updated once and only once, until all have been. It is easy
219: to see that the update rule permits a spin to flip only once, if at all, and
220: so we lose no generality in this model by visiting each spin only once.
221:
222: \begin{table}
223: \begin{center}
224: \caption{
225: The deformation-faulting update rule for ZnS crystals suggested by the Hamiltonian
226: $\mathcal{H}$ with $J_1(T) > 0$ and $J_2(T) \approx 0$. Only those spin flips
227: that lower the energy of the spin configuration are allowed. This is equivalent to
228: applying elementary cellular automaton (ECA) rule 232 to the spin neighborhood
229: $\eta_i^t = ( s_{i-1}^t, s_i^t , s_{i+1}^t )$. $s_i^{t+1}$ gives the value assigned to
230: the center spin when it is selected for updating. The rule is applied
231: asynchronously to randomly selected sites.
232: }
233: \label{update}
234: \vspace{5 mm}
235: \begin{tabular}{l|cccccccc}
236: \hline
237: \hline
238: $\eta_i^t$ & $111$ & $110$ & $101$ & $100$ & $011$ & $010$ & $001$ & $000$ \\
239: \hline
240: $s_i^{t+1}$ & $1$ & $1$ & $1$ & $0$ & $1$ & $0$ & $0$ & $0$ \\
241: \hline
242: \hline
243: \end{tabular}
244: \end{center}
245: \end{table}
246:
247: Let us define the \emph{faulting parameter} $f$ as the ratio of number $n_{app}$
248: of times the update rule has been applied to a crystal of size $N$:
249: $f = n_{app} / N$. The faulting parameter provides a useful
250: index of the amount of faulting the crystal has been subjected to or,
251: equivalently, of the stage of the transformation. That is, one can
252: determine the crystal structure at various stages along the transformation
253: by selecting different values for $f \in [0,1]$.
254: This algorithm is formally equivalent to applying elementary cellular automaton
255: rule $232$ asynchronously to a fraction $f$ of randomly chosen spins
256: (but no spin visited more than once). We then
257: call this model {\em asynchronous cellular automaton 232} (\ACA). Note that
258: the update rule is spin-inversion symmetric, in accord with the Hamiltonian.
259: Models of this basic flavor have been used to study systems other than
260: polytypism, such as cluster growth and phase separation in one
261: dimension,~\cite{privman92a,privman92b} diffusion-reaction problems in one
262: dimension,~\cite{spouge88,racz85} and the voter model.~\cite{bennaim96}
263: The transformation under \ACA\ of spatial configurations as a function of
264: $f$ is shown in Fig. \ref{fig:aca232sptmdiag},
265: starting from the period-two (2H) configuration.
266:
267: \begin{figure}
268: \begin{center}
269: \resizebox{!}{8.0cm}{\includegraphics{rrrca232.fsweep.eps}}
270: \end{center}
271: \caption{
272: Configurations during the 2H $\rightarrow$ 3C transformation under
273: asynchronous cellular automaton 232 on a $100$-site lattice with periodic
274: boundary conditions. The vertical axis shows the degree $f$ of faulting and
275: the horizontal axis is the spatial configuration, with black squares
276: indicating $s_i=1$ inter-ML spins and white squares $s_i=0$ inter-ML spins.
277: Initially, the crystal is in the 2H configuration as indicated by the
278: alternating black and white squares at the top of the diagram ($f=0$).
279: The faulting proceeds in a round-robin fashion until all sites are visited
280: once. At the bottom of the diagram, ($f=1$), the crystal is fully faulted
281: and there are only odd-spin domains (see text) with each domain having at
282: least three spins.
283: }
284: \label{fig:aca232sptmdiag}
285: \end{figure}
286:
287: \begin{figure}
288: \begin{center}
289: \resizebox{!}{12.0cm}{\includegraphics{emachine.anneal.eps}}
290: \end{center}
291: \caption{The reconstructed \eM\ that describes the architecture of the
292: stacking process of a faulted 2H crystal across the range of possible fault parameters
293: $f \in [0,1]$. The nodes represent causal states and the directed arcs,
294: transitions between them. The edge labels $s|p$ indicate that a transition
295: occurs from one causal state to another on symbol $s$ with probability $p$.
296: }
297: \label{fig:emachine.def.10}
298: \end{figure}
299:
300: We then extract a structural description of a spin configuration in the form of
301: an \eM, as a function of $f$, using the causal-state splitting reconstruction
302: (CSSR)~\cite{Shal02a} algorithm.
303: Specifically, we considered crystals with various amounts of faulting
304: $f \in [0,1]$ in increments of $\Delta f = 0.10$
305: and performed \eM\ reconstruction for each $f$.
306: We limit the number of causal states in the
307: reconstructed \eM\ by requiring that the addition of a new causal state reduce
308: the entropy rate by at least $1\%$.
309:
310: \begin{figure}
311: \begin{center}
312: \resizebox{!}{5.2cm}{\includegraphics{fault.probs.eps}}
313: \end{center}
314: \caption[.]{The causal-state transition probabilities $p_1, p_2$, and $p_3$
315: (see Fig. \ref{fig:emachine.def.10}) as a function of the faulting parameter
316: $f$ for a crystal ($N = 1048577$ MLs) initially in a periodic 2H stacking
317: sequence. The $p_i$ are estimated from reconstructed \eMs.
318: }
319: \label{fig:fault.probs.2H}
320: \end{figure}
321:
322: The \eM\ for a faulted 2H crystal appears in Fig.~\ref{fig:emachine.def.10}.
323: It has $10$ causal states and is a function of $3$ transition
324: probabilities---$p_1$, $p_2$, and $p_3$. These transition probabilities
325: are themselves a function of the faulting parameter $f$ and this dependence
326: is shown in Fig.~\ref{fig:fault.probs.2H}.
327: For small values of the faulting parameter, $p_i \approx f$. This implies that
328: for small $f$, the \eM\ largely cycles between the causal states $\textsf{A}$
329: and $\textsf{B}$. This produces the spin sequence $\ldots 101010 \ldots$,
330: which we recognize as the 2H stacking sequence. So, for small faulting, the
331: original 2H crystal remains largely intact; as one expects. At the other
332: extreme, when $f=1$ (a fully faulted crystal) we have $p_1=1$, and there is
333: no longer a transition between the causal states $\textsf{A}$ and $\textsf{B}$.
334: Since there are no other \emph{causal-state cycles}~\cite{varn03a} that generate
335: the $\ldots 101010 \ldots$ spin sequence, we see that the original 2H structure
336: is eliminated.
337:
338: Since the \eM\ has spin-inversion symmetry---i.e., it is invariant
339: under $1 \Leftrightarrow 0$ exchange, we only need to examine half
340: of it to get an intuitive understanding of the structures it captures.
341: Let us consider only the causal states $\textsf{A}$, $\textsf{B}$,
342: $\textsf{D}$, $\textsf{F}$, $\textsf{I}$, and $\textsf{J}$.
343: As stated before, the causal state cycle [$\textsf{A}\textsf{B}$]
344: generates the 2H structure. At $\textsf{B}$, however, it is possible to
345: emit a $0$ and make a transition to $\textsf{D}$. This implies that
346: the spin history upon entering $\textsf{D}$ is 100 and, conversely,
347: having seen the spin sequence $100$ uniquely sets the \eM\ in
348: $\textsf{D}$. At this point, we see that $\textsf{D}$ must make a
349: transition to
350: $\textsf{F}$ on spin $0$. That is, the sequence $1001$ does not occur.
351: At $\textsf{F}$, the stacking process may either enter $\textsf{A}$
352: on a $1$ giving the cumulative spin sequence $10001$ or may proceed to
353: $\textsf{I}$ on a $0$. At $\textsf{I}$, the process must see another
354: $0$ and transition to $\textsf{J}$, giving the cumulative spin sequence
355: of $100000$. This implies that the spin sequence $100001$ likewise does
356: not occur. In fact, we see that the causal states $\textsf{D}$ and
357: $\textsf{I}$ force generating a $0$ as the next spin and, thus,
358: prevent sequences where two or four $0$s are sandwiched between two
359: $1$s. Similar reasoning shows that {\em any} even number of $0$s
360: ($1$s) cannot be trapped between two $1$s ($0$s). In other words,
361: the sets of sequences $\lbrace 10^{2n}1 , n = 1, 2, 3, \ldots \rbrace$
362: and $\lbrace 01^{2n}0 , n = 1, 2, 3, \ldots \rbrace$, and any
363: configurations that contain them, are forbidden. Thus, starting from a
364: short-range crystalline structure (2H or $\ldots 010101 \ldots$),
365: annealing has produced complex spatial structures of infinite range.
366:
367: We can understand how annealing accomplished this increase in structural
368: complexity in the following way. Define a {\em spin domain} as a
369: sequence of $k$ consecutive like spins sandwiched between two unlike
370: spins, where $k$ is a positive integer. In the original 2H structure,
371: there are no even-$k$ spin domains. The update rule has the effect of
372: joining two spin domains of the same spin by eliminating the $k=1$
373: spin domain separating them. Therefore, the resulting domain is of
374: size $k_{new} = k_l + k_r + 1$, where $k_l$ ($k_r$) is the $k$ value
375: of the spin domain to the left (right) of the flipped spin. The spin
376: that is flipped also contributes a site to the new spin domain. Thus,
377: if $k_l$ and $k_r$ are initially odd, $k_{new}$ is also. Since an even
378: spin domain cannot appear, as one scans the crystal one must remember
379: the number of consecutive like spins in order to determine the
380: admissibility of the next spin value. In general, this can require
381: remembering an indefinite number of previous spins. In this way, the process
382: has an infinite memory, so we say that the \emph{memory length} \Range\ is
383: infinite.\cite{varn03a} Put another way, spins arbitrarily far apart may
384: contain information about each other not contained by the intervening spins.
385: Symbolic dynamics calls such a process
386: {\em strictly sofic}.~\cite{Weis73} The most significant consequence of
387: being sofic is that no finite-order Markov process can describe the
388: stacking sequence. Notably, this suggests that previous efforts to
389: describe annealed, disordered crystals that assume some sort of Markov
390: model~\cite{jagodzinski49a,jagodzinski49b,roth60,pandey80a,pandey80b,pandey80c,varn02,varn03a,varn03b}
391: are likely inadequate.
392:
393: As we noted earlier, the structure for a completely transformed crystal
394: occurs for $f=1$. Then the shortest
395: causal-state cycle is [$\textsf{FACEBD}$] and generates
396: the spin configuration $\ldots 111000 \ldots$. There can be forays
397: into the causal states $\textsf{I}$ and $\textsf{J}$ or $\textsf{G}$
398: and $\textsf{H}$, but these only serve to increase the spin domains
399: by an even number. Thus, the final crystal is a
400: {\em disordered, twinned 3C crystal with only odd spin domains.}
401: Notably, some time ago, Mardix~\cite{mardix86} performed a statistical
402: analysis of the known
403: crystalline polytypes in ZnS and found a significant bias toward odd number
404: Zhdanov elements in the configurations. These correspond to regions of
405: odd spin domains, in agreement with odd spin domains predicted by \ACA\
406: for disordered ZnS crystals.
407:
408: It is useful to inquire into the origin of the infinite memory length.
409: Clearly it does not derive directly from the interaction Hamiltonian,
410: which has an interaction range of one. In fact, it comes from the
411: repeated interaction of the stochastic (asynchronous) dynamics and the
412: structural constraints imposed by the allowed (deformation) faulting
413: mechanism. When a spin is flipped, information is lost. That is, when
414: scanning the spin configuration, the process must remember the last
415: $01$ or $10$ pair. The local update rule acts to eliminate pairs of
416: this form and so requires the process to remember an increasingly
417: longer ``history'' of spins, as annealing progresses.
418:
419: \begin{figure}
420: \begin{center}
421: \resizebox{!}{5.2cm}{\includegraphics{diff.patterns.eps}}
422: \end{center}
423: \caption[.]{The diffraction patterns along the $10.l$ row for partially
424: annealed 2H crystals. All spectra, here and elsewhere (except for
425: those from the fault model), are normalized to unity.
426: }
427: \label{fig:diff.patterns}
428: \end{figure}
429:
430: We now address the issue of nonrandom faulting. By examining the update rule
431: in Table~\ref{update}, we see that when a spin flips in the
432: \ACA\ model, the two adjacent spins are necessarily prevented from also
433: flipping. This therefore precludes the possibility of two adjacent faults.
434: This implies that faults are not random, since there must be at least a two
435: ML separation between them.
436: This is easily seen by examining a sequence of spins and the conditional
437: probabilities for faulting as implied by the \eM.
438:
439: Take a segment of an unfaulted 2H crystal as
440: $10101010$. Now introduce a fault (i.e. spin flip) in the spin sequence to get
441: $10\underline{0}01010$, where the underline indicates the flipped spin.
442: Now neither spin adjacent to the flipped spin can itself flip. However, the
443: spins at a distance two or greater can. Let us consider additional
444: spin flips at a separation of two, three, and four. We then have the
445: sequences $10\underline{0}0\underline{0}010$,
446: $10\underline{0}01\underline{1}10$, and
447: $10\underline{0}010\underline{0}0$ as possible stacking sequences, representing
448: deformation stacking faults at separations of two, three, and four.
449: As before, the history $10\underline{0}$ requires the \eM\ to be in causal state
450: $\textsf{D}$. The next spin must be 0, advancing the \eM\ to causal state
451: $\textsf{F}$. Thus the causal state architecture of the \eM\ has
452: prevent two adjacent faults, as required, and the spin sequence is now
453: $10\underline{0}0$. If the stacking sequence is to resume unfaulted, we expect
454: the next spin to be 1. From the \eM, however, the probability of 0, implying
455: a deformation fault separated from the first by two MLs, is $p_2$.
456: That is, the conditional probability of a second fault occurring two MLs after the first
457: is $p_2$.
458: Similar reasoning
459: shows the \eM\ specifies that the conditional probability of a deformation fault
460: occurring three and four MLs after the first is $p_1$. We find, therefore, that
461: {\em the conditional probabilities of observing deformation faults at a
462: separation of one, two, and three MLs are all different, even though
463: \ACA\ manifestly assumes only interactions between nearest neighbors.}
464: For small faulting, however, Fig.~\ref{fig:fault.probs.2H} shows that
465: $p_1 \approx p_2$, so that this distinction is minor for faults at a separation of two and three.
466:
467: The above exercise demonstrates explicitly that the \eM\ can represent
468: a nonrandom distribution of stacking faults
469: and this distribution can arise from an interaction range of one combined with
470: a simple mechanism for annealing. \ACA\ then provides some theoretical
471: justification for models that assume structures with a nonrandom fault
472: distribution without appealing to any long-range interactions or
473: coordination between faults.
474:
475: To see how well \ACA\ describes phase transitions in experimentally
476: observed in ZnS crystals, we calculated the diffraction spectrum along
477: the $10.l$ row for the faulted crystal for various values of the fault
478: parameter as shown in Fig.~\ref{fig:diff.patterns}.
479: (We employ the same procedure, definitions, and assumptions here as we
480: used previously.~\cite{varn03a})
481: Let $l$ be a dimensionless variable that indexes the magnitude of the
482: perpendicular component of the diffracted wave.
483: For small values of the faulting, the 2H Bragg peaks at $l=1/2$ and
484: $l = 1$ are widened considerably. As $f$ approaches $0.50$, the
485: enhancement in diffracted intensity at these positions has vanished
486: and there is instead significantly increased scattering near the
487: $l \approx 1/3$ and $l \approx 2/3$ positions, which is normally
488: associated with twinned 3C structure. There are also small rises in
489: the diffracted intensity at the $l \approx 1/6$ and $l \approx 5/6$
490: positions, which is often considered a sign of 6H structure. As $f$
491: approaches $1.0$, the peaks at $l \approx 1/3$ and $l \approx 2/3$
492: sharpen and the diffracted intensity at the $l \approx 1/6$ and
493: $l \approx 5/6$ positions weaken. For $f=1$, the crystal has found
494: a local minimum of the Hamiltonian, and no further transformation
495: occurs. The crystal is now a disordered, twinned 3C structure.
496:
497: Notice also that it had been assumed previously that the small rises
498: in the intensity at $l \approx 1/6$ and $l \approx 5/6$ signaled that
499: parts of the crystal were faulting into the 6H structure, but we see
500: that no special assumption of this kind is needed. Rather, the spin
501: sequence $000111$ (i.e. the 6H structure) occurs quite naturally as
502: the 2H-crystal deformation faults, giving the enhanced diffracted
503: intensity at these positions.
504:
505: \begin{figure}
506: \begin{center}
507: \resizebox{!}{5.2cm}{\includegraphics{SK134.1b.cor.ext.3.new.eps}}
508: \end{center}
509: \caption[.]{A comparison of an experimental
510: diffractogram~\cite{sebastian94,varn02,varn03b} along the $10.l$ row
511: (triangles) with the
512: fault model (solid line)~\cite{varn02} and \ACA\ (dotted line) at
513: $f=0.06$. Except for some minor differences, \ACA\ gives the same
514: description as the fault model. We find the profile
515: ${\cal R}$-factor~\cite{varn03b} between experiment and \ACA\ to be
516: $29\%$. We see that \ACA\ describes the Bragg peaks at
517: $l \approx 0.5$ and $l \approx 1$ well, but misses the small rise in
518: intensity at $l \approx 0.67$. This likely results from 3C structure,
519: so it is not surprising that it is absent the \ACA\ spectrum at
520: low $f$.
521: }
522: \label{fig:SK134.1b.cor.ext.3.new}
523: \end{figure}
524:
525: We now compare the predictions from our simple model with two experimental
526: ZnS diffraction spectra we have previously treated~\cite{varn02,varn03b}:
527: the first from an unannealed sample; the second, from an annealed
528: sample. Figure~\ref{fig:SK134.1b.cor.ext.3.new} shows the spectrum
529: from an as-grown, disordered 2H crystal. Our previous analysis showed
530: that, while there were displacement faults, growth faults, and some 3C
531: structure, the dominant faulting was due to deformation. It is
532: reasonable, then, to ask if \ACA\ can reproduce this spectrum. For
533: $f=0.06$ \ACA\ does reproduce the Bragg-like peaks at $l \approx 1/2$ and
534: $l \approx 1$ rather well. The small rise in diffuse scattering at
535: $l \approx 2/3$ is not recovered, nor should one expect it to be. This
536: feature comes from the small amount of 3C structure present and, as we
537: saw above, \ACA\ is not able to reproduce this. That it does not lends
538: further strength to our earlier interpretation~\cite{varn02} that a small
539: amount of 3C structure is present. In fact, we find that for small faulting,
540: \ACA\ gives nearly the same results as the fault model, as can
541: be seen in Fig.~\ref{fig:SK134.1b.cor.ext.3.new}.
542:
543: Figure~\ref{fig:SK135.1b.phtrans} compares an experimental ZnS
544: diffraction spectrum from a 2H crystal annealed at $500$ C for 1 h
545: with an $f = 85\%$ deformation-faulted 2H simulated diffraction spectra.
546: The agreement is excellent. The deformation-faulted crystal
547: reproduces the two disordered 3C peaks in the experimental spectrum
548: at $l \approx -2/3$ and $l \approx -1/3$, as well as the diffuse
549: scattering between these diffraction maxima. Even the slight
550: enhancement in diffraction at $l \approx -1/6$ and $l \approx 1/6$
551: is captured (see inset). Our deformation-faulting model gives a
552: stacking configuration that is spin symmetric and so it is unable to
553: account for the relative difference in intensity between the two peaks
554: at $l \approx -2/3$ and $l \approx -1/3$. There is a question here
555: concerning the quality of the data,~\cite{varn03b} and so the difference
556: in intensities may be an experimental artifact. Note that the
557: diffracted intensity between the two Bragg-like peaks
558: at $l \approx -2/3$ and $l \approx -1/3$
559: is especially well represented by
560: \ACA. This contrasts with the fault model which simply cannot recover
561: the diffuse scattering.
562:
563: \begin{figure}
564: \begin{center}
565: \resizebox{!}{5.2cm}{\includegraphics{SK135.1b.cor.ext.3.new.eps}}
566: \end{center}
567: \caption[.]{A comparison of an experimental
568: diffractogram~\cite{sebastian94,varn02} along the $10.l$ row
569: (triangles) with the
570: fault model (solid line)~\cite{sebastian94,varn02} and \ACA\ (dashed line) at
571: $f=0.85$. With a profile ${\cal R}$-factor of $18\%$
572: (compare $13\%$ for \eMSR\ ~\cite{varn03b} and $33\%$ for the fault model),
573: \ACA\ captures a surprisingly large amount of the stacking structure.
574: Notice particularly that the inset shows that \ACA\ nicely reproduces
575: the small rise in diffracted intensity at $l \approx -0.16$.
576: }
577: \label{fig:SK135.1b.phtrans}
578: \end{figure}
579:
580: There are however, features seen in diffractograms of partially
581: annealed ZnS crystals that are not observed in the simulated
582: diffractograms of Fig.~\ref{fig:diff.patterns}. Namely, the
583: diffractograms of some partially transformed ZnS crystals, as in
584: Fig. 9 of Ref.~\citep{varn03b}, show enhancement simultaneously at
585: values of $l$ associated with the 2H structure, namely $l \approx 1/2$
586: and $l \approx 1$, as well as at $l$ values associated with twinned 3C
587: structure ($l \approx 1/3$ and $l \approx 2/3$). In our simulation
588: studies, the strong 2H reflections at $l \approx 1/2$ and $l \approx 1$
589: disappear before the 3C reflections at $l \approx 1/3$ and
590: $l \approx 2/3$ start becoming prominent. \ACA\ therefore has
591: difficulty even qualitatively explaining such spectra and one is
592: led to entertain models more complicated than \ACA.
593:
594: We introduced a very simple model, called \ACA\ due to its equivalence
595: with elementary cellular automaton rule 232, to study the 2H
596: $\rightarrow$ 3C phase transition in polytypic ZnS crystals. We assumed
597: only nearest-neighbor interactions between MLs and thus a local,
598: though asynchronously applied, faulting mechanism. We simulated a
599: solid-state transition during annealing of the crystal and found a
600: description of the stacking structure for all amounts of
601: faulting in the form of an \eM.
602: The resulting \eM\ implied that, despite interactions only between nearest neighbors,
603: the structure of the crystal had a nonrandom distribution of faults.
604: The \eM\ also shows an infinite
605: memory length, which means that arbitrarily separated spins share
606: nonredundant information. The corresponding symbolic dynamics
607: is strictly sofic and this implies that no finite-order Markov process
608: can completely specify the stacking in these faulted crystals.
609: Intuitively, this arises because only odd-length spin domains are
610: allowed, if the crystal starts from a perfect 2H specimen.
611: The infinite memory length process discovered here contrasts with the $r=3$ memory length
612: we found by analyzing disordered ZnS crystals using \eMSR.~\cite{varn03b} This
613: suggests that while the $r=3$ \eM\ captures much of the stacking structure
614: present, it is nonetheless only an approximation to a more structurally
615: complex stacking process. The relation between the \eM\ found in
616: these simulation studies and that determined by \eMSR\ from experimental
617: spectra is currently a subject of investigation. Since even simple,
618: annealed crystals probably have an infinite memory length, this
619: further suggests that the Markov approximations used in the \eMSR\ algorithm
620: should be circumvented and a more general spectral reconstruction technique is
621: warranted. This, too, is a current subject of research.
622: Finally, we compared
623: diffraction spectra calculated from \ACA\ to previously analyzed ZnS
624: spectra and found good quantitative agreement.
625:
626: While not included here, our studies also considered different initial
627: stacking sequences. We found that often the memory length was infinite
628: after annealing and that the structure of the final faulted crystal was
629: highly dependent on the initial stacking structure. Although phase
630: transitions in actual ZnS crystals undoubtedly are more complicated,
631: our simple model captures much of the noncritical organization they
632: undergo. Also, considering its simplicity, and the fact that similar
633: models have been used to study other systems, the results presented
634: here should have applicability far beyond the realm of polytypism.
635:
636: \begin{acknowledgments}
637:
638: We thank Geoff Canright and Eric Smith for useful conversations.
639: This work was supported at the Santa Fe Institute under the Networks
640: Dynamics Program funded by the Intel Corporation and under the
641: Computation, Dynamics, and Inference Program via SFI's core grants from
642: the National Science and MacArthur Foundations. Direct support was
643: provided by DARPA Agreement F30602-00-2-0583.
644:
645: \end{acknowledgments}
646:
647: \bibliography{master.refs}
648:
649: \end{document}
650: