1: \documentstyle[12pt,graphicx]{mrsproc}
2: \def\baselinestretch{0.9}
3:
4: \begin{document}
5:
6: \begin{center}
7: {\bf Equilibrium structure of decagonal AlNiCo}
8: \end{center}
9:
10: \raggedright
11: \vspace{0.2cm}
12:
13: \noindent S. Naidu, M. Mihalkovi\v{c}$^1$ and M. Widom,\\
14: Department of Physics, Carnegie Mellon University, Pittsburgh, PA 15213\\
15: $^1$also at: Institute of Physics, Slovak Academy of Sciences, 84228 Bratislava, Slovakia
16:
17: \parindent0.3in
18:
19: \begin{abstract}
20: We investigate the high temperature decagonal quasicrystalline phase
21: of Al$_{72}$Ni$_{20}$Co$_8$ using a quasilattice gas Monte-Carlo
22: simulation. To avoid biasing towards a specific model we use an
23: over-dense site list with a large fraction of free sites, permitting
24: the simulation to explore an extended region of perpendicular space.
25: Representing the atomic surface occupancy in a basis of harmonic
26: functions directly reveals the 5-fold symmetric
27: component of our data. Occupancy is examined in
28: physical and perpendicular space.
29: \end{abstract}
30:
31: \section{INTRODUCTION}
32:
33: AlNiCo exhibits quasicrystalline phases over a range of compositions
34: and temperatures\cite{Ritsch}. Of special interest is the Ni-rich
35: quasicrystalline phase around the composition
36: Al$_{72}$Ni$_{20}$Co$_8$. This is a decagonal phase with a period of
37: 4.08~\AA~ along the periodic axis, making it a simple phase relative
38: to other members of the Al-Ni-Co family. Additionally, it appears to
39: be most perfect structurally, even though it is stable only at high
40: temperatures around T=1000-1200K. Its structure has been extensively
41: studied experimentally by X-ray diffraction\cite{Takakura,Cervellino}
42: and electron microscopy\cite{Abe,Pennycook}. Finally, since
43: qualitatively accurate pair potentials are available\cite{MGPT},
44: structural predictions can be made based on total energy
45: considerations\cite{Mihalkovic}.
46:
47: An idealized deterministic structure for this phase has been proposed
48: by studying the total energy\cite{Mihalkovic}. This prediction
49: employed a multi-scale simulation method, in which small system sizes
50: were simulated starting with very limited experimental input, then
51: rules discovered through the small scale simulations were imposed to
52: accelerate simulations of larger-scale models. Although efficient,
53: this approach leaves open the question of how strongly the final model
54: was biased by the chosen method.
55:
56: We adopt a different approach here, starting from slightly different
57: experimental input and working directly at the relevant high
58: temperatures. The experimental input is: (1) the
59: density~\cite{Takakura}, composition and temperature at which the
60: phase exists; (2) the hyperspace positions of atomic surfaces (these
61: are simply the positions for a Penrose tiling, with AS1 at $\nu=\pm 1$
62: and AS2 at $\nu=\pm 2$); (3) the fact that the quasicrystal is
63: layered, with space group $10_5/mmc$, and its lattice constants (we
64: take $a_q=6.427$, $c=4.08$\AA~). The chief unknowns to be determined
65: are the sizes, shapes and chemical occupancies of the atomic surfaces.
66:
67: Like the prior study~\cite{Mihalkovic}, we employ Monte Carlo
68: simulation using the same electronic-structure derived pair
69: potentials\cite{MGPT}. Also, like the prior study, we employ a
70: discrete list of allowed atomic positions. However, instead of using
71: sparse decorations of fundamental tiles, where the configurational
72: freedom arose largely from flipping of the decorated tiles, in the
73: present study we employ a fixed site list based on a rich decoration
74: of fixed tiles. For a given tiling the density of allowed sites in
75: our new simulation is much greater than the actual atomic density. The
76: resulting atomic surface (see Fig.~\ref{fig:model-perp}a) corresponds
77: to that of a Penrose tiling with tile edge length
78: $a_q/\tau^3=1.52$~\AA~ (plus a few additional sites such as inside some
79: fat rhombi). These atomic surfaces include within them the atomic
80: surfaces previously proposed on the basis of total energy
81: calculations~\cite{Mihalkovic}, and analysis of experimental
82: data~\cite{Takakura,Cervellino}. This larger atomic surface avoids
83: any bias towards particular atomic surface shapes.
84:
85: \begin{figure}
86: \includegraphics*[width=6in]{combo}
87: \caption{\label{fig:model-perp}(a) Oversized atomic surface used in
88: the present simulation. AS1 (left) at $\nu=\pm1$, AS2 (right) at
89: $\nu=\pm2$. For comparison the ideal model proposed in
90: Ref.~\cite{Mihalkovic} is inscribed inside. Color scheme: black=Co,
91: gray=Ni, light-gray=Al, white=partial Al occupancy. (b) Total
92: occupancy of AS1 (left) and AS2 (right) at T=1160K.}
93: \end{figure}
94:
95: The following section describes the site list and other
96: simulation details more precisely. Then we present site occupancy in real
97: space and on the atomic surfaces. We represent atomic surface occupancies
98: using Fourier-Bessel series, allowing us
99: to isolate the five-fold symmetric component and filter sampling
100: noise. Such a representation could prove fruitful in
101: crystallographic structure determination as well. Refs.~\cite{Perez-Mato,JQ}
102: pursued related goals.
103:
104: We compare our results with a recent experimental study. Briefly, we
105: find broad agreement with the sizes, shapes and chemical occupancies
106: of the atomic surfaces. A few specific points of disagreement can be
107: addressed by relaxing our fixed site positions. Finally, we conclude
108: with a summary of our results and outlook for future work.
109:
110: \section{SIMULATION METHOD}
111:
112: We should perform simulations in the continuum, with no predefined
113: site list to bias our results. Unfortunately, strong binding of Al to
114: transition metal atoms (Co and Ni) leads to phase separation into a
115: CsCl-type crystal plus regions of excess pure Al. This reflects a
116: deficiency of our interatomic potentials, which should not be applied
117: at a transition-metal concentration as high as the 50\% found in
118: CsCl-type crystals. Restricting atomic positions a set that does
119: not include the CsCl crystal inhibits phase separation.
120:
121: Rather than specify these sites in a complex manner whose details may
122: fail to allow some especially favorable structure, we take a
123: relatively neutral assignment that still allows quasicrystalline
124: structures to form. Specifically we restrict the possible atomic
125: positions to the vertices of a Penrose lattice whose edge length is
126: small compared to the spacing between atoms. The possible atomic
127: sites fill space with a density 2.55 times greater than the actual
128: atomic density, so the majority of sites are empty in any particular
129: configuration. Our prior study of prefered structures proposed an
130: idealized arrangement of atoms on an HBS (hexagon-boat-star) tiling of
131: edge length $a_q$, as illustrated in Fig.~\ref{fig:model-para}.
132:
133: \begin{figure}
134: \includegraphics*[width = 2 in]{ideal}
135: \includegraphics*[width = 2 in]{sample}
136: \includegraphics*[width = 2 in]{orbit_occ}
137: \caption{\label{fig:model-para}This figure illustrates our fixed site
138: list in parallel space. Ideal tile decoration (left) and typical
139: T=1160K configuration (center). Color scheme as in
140: Fig~\ref{fig:model-perp}a. Large/small atoms indicate upper/lower
141: layer. Very small white circles indicate vacancies. Occupancy data
142: (right) at T=1160K with atomic size indicating total occupancy and Al
143: fraction indicated by gray scale (black=100\% TM).}
144: \end{figure}
145:
146: Our Monte Carlo simulation distributes atoms among these sites in a
147: manner consistent with thermal equilibrium at a given temperature. We
148: focus on the temperature T=1160K, which lies within the stability
149: range for the Ni-rich decagonal phase~\cite{Ritsch}. We simulate a
150: low phason strain quasicrystal approximant, with lattice parameters
151: $159.38\times 83.79\times 4.08$~\AA$^3$, containing 9776 lattice sites
152: and holding 3838 atoms. We initially decorated the structure randomly
153: at the desired density and composition. We then lowered the
154: temperature in steps from an initial T=4000K, annealing at each
155: temperature, until we reached our data collection temperature of
156: T=1160K.
157:
158: Our data analysis differs from conventional approaches to quasicrystal
159: atomic surface modeling. It is common to model atomic surface
160: occupancy by breaking the atomic surface up into polygonal domains
161: piecewise constant occupancy in each domain. This approach highlights
162: the manner in which occupancy depends on the local real space
163: environment, which changes discontinuously as a function in
164: perpendicular space. Unfortunately such a piecewise continuous
165: representation does not properly capture the effects of chemical
166: disorder and phason fluctuations characteristic of a quasicrystal at
167: high temperatures.
168:
169: We show below that the mean occupancy statistics actually vary quite
170: smoothly over the atomic surface. To represent smoothly varying
171: atomic surface occupancy functions it is helpful to introduce a
172: complete basis set of analytic functions and expand the occupancy in
173: this basis. Owing to the axial symmetry of the atomic surfaces it is
174: handy to introduce polar coordinates $(r_{\perp},\theta)$. The
175: natural function basis set for polar coordinates are combinations of
176: cylindrical Bessel functions multiplied by complex exponentials.
177: Thus, on a given atomic surface we express the occupancy by chemical
178: species $\alpha$ as
179: \begin{equation}
180: \rho_{\alpha}(r_{\perp},\theta) =
181: \sum_{m,n} A^{\alpha}_{m,n} J_m(k_{m,n} r_{\perp}) e^{i m \theta}.
182: \label{eq:bessel}
183: \end{equation}
184: The $m$ index represents the angular mode frequency. For a given
185: angular frequency $m$, the sum over $n$ allows the representation of
186: an arbitrary radial variation, with radial frequencies $k_{m,n}$ such
187: that all Bessel functions share a common zero outside our atomic
188: surface. To analyze our data we invert eq.~(\ref{eq:bessel}) to
189: obtain the coefficients $A^{\alpha}_{m,n}$ using orthonormality of the
190: basis set. We then filter out the components whose angular variation
191: lacks 5-fold symmetry by restricting $m$ to multiples of 5.
192:
193: \section{RESULTS}
194:
195: Each Monte Carlo step consisted of many attempted jumps (of up to
196: $6.5$~\AA~ in length) per atom. An attempted jump to an occupied site
197: means an attempt to swap the two atoms. We recorded the detailed
198: atomic site occupancy for 9000 configurations separated by
199: sufficiently many Monte Carlo steps that the configurations become
200: uncorrelated.
201:
202: Phason fluctuations such as bowtie flips~\cite{Mihalkovic}
203: cause the center of gravity of the
204: structure to drift in perpendicular space. Note that the lattice
205: sites do not move during a phason flip, only their atomic occupancy
206: changes. A bowtie flip in our ideal model interchanges
207: two NiNi pairs with two AlCo pairs. To avoid smearing caused by this drift,
208: which could obscure small-scale atomic surface details,
209: we recentered the perp space center of gravity for each configuration
210: before projecting onto the Fourier-Bessel coefficients. Without
211: recentering the atomic surface occupancy
212: functions would be weaker and broader.
213:
214: \begin{figure}
215: %\begin{center}
216: \hspace{1.5in}
217: \includegraphics*[width = 1.18 in, angle=90]{AS1Al2}
218: \includegraphics*[width = 1.18 in, angle=90]{AS1Ni2}
219: \includegraphics*[width = 1.18 in, angle=90]{AS1Co2}
220: %\end{center}
221: \caption{\label{fig:AS1AlNiCo}This figure illustrates the chemical
222: occupancy of the $\nu=\pm 1$ atomic surfaces for Al, Ni and Co atoms
223: (left to right). The gray scales are proportional to occupancy, scaled
224: to the maximum for each species.}
225: \end{figure}
226:
227:
228: As in our prior~\cite{Mihalkovic} low temperature study, we find that
229: the HBS tiling effectively describes the quasiperiodic layers, but the
230: prefered chemical decoration of the tiles now becomes highly variable
231: at certain sites. Fig.~\ref{fig:model-perp}(b) displays the average
232: occupancy distribution on each atomic surface. The AS1 occupancy is a
233: combination of all three chemical species. The AS2 occupancy is
234: virtually 100\% Al. The breakdown of AS1 occupancy among Al, Ni and
235: Co is shown in Fig.~\ref{fig:AS1AlNiCo}. A striking feature of all
236: these occupancy plots is the smooth, continuous variation of
237: occupancy. It is clearly not appropriate to model these densities
238: using piecewise constant functions with polygonal boundaries, even
239: though five-fold symmetry is present. Ni and Co are strongly mixed
240: with very similar distributions, concentrated at the center of
241: AS1. Because the amount of Ni present exceeds the amount of Co, the Co
242: occupancy vanishes more quickly than the Ni occupancy away from the
243: center, while the Ni occupancy remains more nearly constant. In the
244: region where the TM occupancy drops off it is largely replaced with Al
245: so the total occupancy remains close to 100\% in this crossover
246: region.
247:
248: In parallel space atoms in the TM to Al crossover region
249: correspond to tile edges. Effects of this gradient from TM to
250: Al occupancy can be seen in Fig.~\ref{fig:model-para} (right), near tiling
251: vertices at which two or more edges join in a $\pm72^{\circ}$ angle.
252: The pair of occupied sites on these tile edges show unequal Al/TM
253: occupancy. The site closer to the vertex contains relatively less TM
254: and more Al than the site further away. The site closer to the vertex
255: in parallel space lies in the atomic surface region that is further
256: from the center in perp space. Note also that bowtie phason
257: flips that replace TM--TM pairs with Al--TM pairs on tile edges are one
258: type of excitation that replace the TM site close to a 72$^{\circ}$
259: vertex with an Al atom. We understand these effects energetically
260: because the inner edge site is in a TM-rich environment. Al--TM bonds
261: are strongly favorable so this is a likely site for Al occupancy.
262:
263: Far from the atomic surface centers, where the occupancy falls
264: smoothly to zero, the atoms are mainly Al. These highly mobile atoms
265: are in locations in parallel space where there are multiple ideal
266: sites, too close to be simultaneously occupied, but each one with a
267: rather similar local environment so that there is no strong energetic
268: preference among the sites, for example the pair of symmetry-related
269: sites near the center of the hexagon tile. Other examples are rings
270: of 10 sites surrounding the ideal Co position at the centers of boat
271: and star tiles. These rings can hold at most three Al atoms. Jumps of
272: Al atoms among these sites correspond to phason tile flips of the
273: underlying very small 1.52~\AA~ rhombus tiling.
274:
275: \section{COMPARISON WITH EXPERIMENT}
276:
277: We compare primarily with the refinement by Takakura~\cite{Takakura}.
278: We first point areas of general agreement between our results and the
279: experimental data, then discuss the main points of disagreement.
280: Our simulated occupation probabilities generally agree well
281: with experiment. The approximate sizes and shapes of our atomic
282: surfaces agree with Takakura. We agree that AS1 contains transition
283: metal atoms within a central region (Takakura orbit numbers 1-5),
284: separated by a fully occupied region of mixed Al/TM (orbits 6 and 7)
285: and finally radial spokes in which Al occupancy diminishes from full
286: to partial (orbits 8-10). We agree that AS2 (orbit numbers 11-23)
287: contains primarily Al atoms. Takakura finds full occupancy of orbit
288: numbers 14 and 16, but only 82\% occupancy of orbit numbers 15 and 17,
289: which are locally fairly similar (these are the Al sites adjacent to
290: tile edge TM atoms). Our results qualitatively support this, with
291: occupancy about 90\% corresponding to orbit numbers 14 and 16,
292: dropping to around 70-80\% for orbit numbers 15 and 17.
293:
294: Now consider discrepancies between our results and
295: experiment. Orbits 19 and 23, localized around special points
296: on the fringes of AS2 are fully occupied in the refinement, while we
297: find partial occupancy. In parallel space these correspond to
298: pairs of ideal sites midway between TM atoms. Partial occupancy is
299: forced because the spatial separation of these ideal points
300: (2.25~\AA) is too small to allow both to be simultaneously occupied.
301: However, if we consider multilayer strucures, with $c=4\times
302: 4.08$~\AA, and allow structural relaxation, we know~\cite{HMW} it
303: is possible to occupy 3 out of every 4 such sites per 8.16~\AA~.
304: The lack of structural relaxation thus causes an error in our
305: occupancy for this orbit.
306:
307: Takakura's fully occupied orbits numbers 20-22 are located on the
308: corners of AS2. We find 60-80\% occupancy instead of full occupancy,
309: consistently with our smoothly decreasing occupation probability.
310: Takakura also finds a small fraction (about 20\%) of TM atoms on
311: orbits 20 and 21, while we find only Al. There is no strong energetic
312: reason that TM atoms should not occupy those sites. Perhaps the TM
313: occupancy of orbit number 20 is related to the large atomic
314: displacement of this orbit Takakura found during the refinement.
315:
316: Surprisingly, Takakura finds 50\% occupancy on orbit number 11, at the
317: very center of AS2. In parallel space this site corresponds to HBS
318: tile vertices at which five edges meet. We find it is fully occupied,
319: similar to all other HBS tile vertices.
320:
321: \section{CONCLUSION}
322:
323: We presented a method to compute atomic surface occupancy statistics,
324: and applied it to the basic Ni-rich decagonal phase of AlNiCo. This
325: work launches a promising set of possible future research directions.
326: The method can be applied to other decagonal phases and also to
327: icosahedral phases. The main limitation is the availability of
328: interatomic potentials. Further, a great deal remains to be done even
329: within AlNiCo, as we now discuss.
330:
331: Our simulations generate a complete set of equilibrium configurations,
332: in which the precise location of every atom is known. The analysis
333: above only presents the atomic surface occupation probabilities, but
334: does not address correlations among the partially occupied sites.
335: Such correlations are contained in the Patterson function, which is
336: also known experimentally~\cite{Steurer} and can be computed from our
337: existing data. Actual atomic configurations contain more information than the
338: atomic surface occupancies, because all correlations are reflected in
339: the actual configurations. Fig.~\ref{fig:model-para} (center)
340: illustrates an atomic arrangement from our
341: simulation. Comparing with the ideal model (left) reveals a great
342: deal of disorder, including phason flips, chemical disorder and
343: vacancies.
344:
345: Among the prominent correlations are strong anticorrelation of
346: too-close sites. The deep clefts between arms of AS1
347: correspond to sites that are close in parallel space to sites
348: near the corners of AS2. When these AS2 sites are occupied sites
349: within the clefts of AS1 are empty. When these AS2 sites are empty
350: then we find partial occupancy inside the clefts of AS1.
351: Another anticorrelation occurs because the pairs of ideal
352: sites midway between pairs of TM atoms (e.g. Takakura orbit number 19
353: and 23) are too close for simultaneous occupancy. This
354: anticorrelation arises because we used only ``ideal'' positions
355: projected into parallel space from flat atomic surfaces. In reality
356: the atomic positions are displaced somewhat away from the ideal sites,
357: and we showed above that such displacements can have a substantial
358: influence on occupancy statistics in cases where the ideal sites are
359: just slightly too close together. Indeed, relaxations of our ideal
360: structures, using either pair potentials~\cite{HMW} or full
361: {\em ab-initio} calculations result in displacements similar
362: to those reported experimentally~\cite{Takakura}.
363:
364: Finally, our simulation used only a single 4.08~\AA~ repeat unit along
365: the 10-fold symmetry axis. Owing to our periodic boundary conditions,
366: swapping a pair of atoms in the single repeat implies a simultaneous
367: swap of all pairs of atoms in infinite columns extending above and
368: below the original pair in our single repeat. But when we calculate
369: the energy for the swap, we only include the energies of the original
370: atoms. This type of finite-size error needs to be eliminated, as it
371: miscalculates the energies of phason flips.
372:
373: Simply stacking two repeats above each other to make an 8.16~\AA~
374: unit, however, usually overestimates the energy cost of atom swaps
375: corresponding to bowtie phason flips. This is because uncorrelated
376: flips between repeat units create phason stacking faults whose energy
377: cost is large in the absence of atomic relaxation but is small, and
378: can even be negative, when relaxation is included. We must include
379: relaxed atomic positions to achieve a realistic model of
380: three-dimensional phason disorder.
381:
382: \section{ACKNOWLEDGEMENTS}
383:
384: We acknowledge discussions with C.L. Henley support by NSF grant DMR-0111198.
385:
386: \begin{thebibliography}{0}
387: \bibitem{Ritsch} S. Ritsch, C. Beeli {\it et al.}, Phil. Mag. Lett, {\bf 78}, 67 (1998).
388: \bibitem{Takakura} H. Takakura, A. Yamamoto and A.-P. Tsai, Acta Cryst. A, {\bf 57}, 576 (2001).
389: \bibitem{Cervellino} A. Cervellino, T. Haibach and W. Steurer, Acta Cryst. B, {\bf 58}, 8 (2002).
390: \bibitem{Abe} E. Abe, K. Saitoh {\it et al.}, Phys. Rev. Lett., {\bf 84}, 4609 (2000).
391: \bibitem{Pennycook} Y. Yan, S. J. Pennycook and A.-P. Tsai, Phys. Rev. Lett., {\bf 81}, 5145 (1998).
392:
393: \bibitem{MGPT} I. Al-Lehyani {\it et al.}, Phys. Rev. B, {\bf 64}, 075109 (2001).
394: \bibitem{Mihalkovic} M. Mihalkovi\v{c} {\it et al.}, Phys. Rev. B, {\bf 65}, 104205 (2002).
395: %\bibitem{Gahler} F. G\"ahler and S. Hocker, Proceedings of Eighth International Conference on Quasicrystals, to appear in J. Non-Cryst. Solids (2003)
396: \bibitem{Perez-Mato} L. Elcoro and J. M. Perez-Mato and G. Madaraiga, Acta Cryst. A, {\bf 50}, 182 (1994).
397: \bibitem{JQ} M. V. Jaric and S.-Y. Qiu, Phys. Rev. B, {\bf 49}, 6614 (1994).
398: \bibitem{HMW} C. L. Henley, M. Mihalkovi\v{c} and M. Widom, J. Alloys and Comp., {\bf 342}, 221 (2002).
399: \bibitem{Steurer} W. Steurer, T. Haibach, B. Zhang, S. Kek and R. Luck, Acta Cryst. B, {\bf 49}, 66, (1993).
400: \end{thebibliography}
401:
402: \end{document}
403: