1: \documentclass[11pt,draft,aps,prb,showpacs]{revtex4}
2: % \documentclass[twocolumn,aps,prb,showpacs]{revtex4}
3:
4: \usepackage{epsf}
5: \usepackage{dcolumn}
6:
7: \begin{document}
8:
9:
10: \def\mc {\multicolumn}
11: \def\tn {\tablenote}
12:
13: \title {On the observed irregular melting temperatures of
14: free sodium clusters}
15:
16: \author{S. Chacko and D. G. Kanhere}
17: \affiliation{
18: Centre for Modeling and Simulation, and Department of Physics,
19: University of Pune, Ganeshkhind, Pune 411 007, India.}
20: \author{S. A. Blundell}
21: \affiliation{
22: D\'epartement de Recherche Fondamentale sur la
23: Mati\`ere Condens\'ee, CEA--Grenoble/DSM \\
24: 17 rue des Martyrs, F--38054 Grenoble Cedex 9, France.}
25:
26: \begin{abstract}
27: Density--functional simulations have been performed on Na$_{55}$,
28: Na$_{92}$ and Na$_{142}$ clusters in order to understand the
29: experimentally observed melting properties
30: [M. Schmidt \textit{et al.}, Nature~(London) \textbf{393}, 238~(1998)].
31: The calculated melting temperatures are in excellent agreement with the
32: experimental ones. The calculations reveal a rather subtle interplay
33: between geometric and electronic shell effects, and bring out the fact
34: that the quantum mechanical description of the metallic bonding is crucial
35: for understanding quantitatively the variation in melting temperatures
36: observed experimentally.
37: \end{abstract}
38:
39: \pacs{61.46.+w, 36.40.--c, 36.40.Cg, 36.40.Ei}
40:
41: \maketitle
42:
43:
44: \section{Introduction}
45:
46: There is currently great interest in the statistical mechanics of
47: finite--sized versions of systems that display phase transitions in
48: the infinite--size limit. Recently, atomic clusters have provided an
49: example of a finite system whose caloric curve may be measured,
50: yielding evidence for a ``melting'' transition over a broad range of
51: temperatures.\cite{haberland0,haberlandnegCV,haberlandCR,jarroldGa}
52: The clusters involved in these experiments were both small~(containing
53: from a few atoms to several hundreds of atoms) and free~(not supported
54: on a surface). Now, according to old thermodynamic
55: models,\cite{pawlow,buffat} a small particle should melt at a lower
56: temperature than the bulk because of the effect of the surface, with
57: the reduction in melting temperature being roughly proportional to
58: $1/R$, where $R$ is the radius of the particle. This effect has been
59: verified quantitatively for particles of mesoscopic size~(upwards of
60: several thousand atoms) supported on a surface.\cite{buffat,boiko}
61: More recently, however, in a series of experiments on free Na clusters
62: in the size range 55 to 350, Haberland and
63: co--workers\cite{haberland0,haberlandnegCV,haberlandCR} found that the
64: simple $1/R$ scaling is lost; they did observe a substantial
65: lowering~(by about 30\%) of the melting temperature compared to bulk
66: sodium, but accompanied by rather large size--dependent fluctuations
67: in the melting temperatures. In spite of considerable theoretical
68: effort,\cite{manninen99,manninen04,cs,bb01,aguado,aguado-nalarge} the
69: precise form of the observed fluctuations remains unexplained to date.
70:
71:
72: As is well known, metallic clusters~(such as sodium) possess the same
73: first few ``magic'' sizes as atomic nuclei, $N=2$, 8, 20, \ldots,\cite
74: {knight84} corresponding to particularly stable systems in which the
75: delocalized valence electrons form closed fermionic shells. Unlike
76: nuclei, however, metallic clusters also contain positively charged
77: ions, which are much heavier than electrons and may be treated
78: classically to a good approximation, leading to the possibility that
79: geometric packing effects may enter into competition with electronic
80: shell effects in determining certain properties. The precise pattern
81: of melting points observed experimentally\cite{haberland0,haberlandCR}
82: shows maxima at sizes that correspond neither to magic numbers of
83: valence electrons nor to complete Mackay icosahedra of ions, thereby
84: suggesting a subtle interplay between geometric and quantum electronic
85: effects.\cite{haberland0,haberlandCR,manninen99,manninen04}
86:
87:
88: Reliable simulations to determine the melting properties of clusters
89: are made difficult by a combination of two crucial requirements: the
90: need to compute the ionic potential--energy surface accurately, and
91: the need for high statistics to converge the features in the caloric
92: curve. Good statistics have been obtained using a variety of
93: \emph{parametrized} interatomic potentials,\cite{manninen04,cs} as
94: well as with a treatment of the valence electrons in the extended
95: Thomas--Fermi~(ETF)\cite{bb01} and related density--based~(DB)
96: approximations.\cite{aguado,abhijat01} But these attempts have failed
97: to reproduce crucial features of the experimental results, such as the
98: precise pattern of observed melting temperatures. An interesting
99: observation is that in all earlier simulations the melting temperature
100: of Na$_{55}$ has been considerably underestimated~(by about 100~K
101: lower than the experimental value).
102:
103:
104: Clearly, what is required is a more realistic treatment of interacting
105: electrons, in particular, one that incorporates correctly electronic
106: shell effects, which are explicitly excluded from the above--mentioned
107: work using parametrized, ETF, or DB potentials. We have recently
108: demonstrated the power of \emph{ab initio} density--functional
109: theory~(DFT), in the Kohn--Sham (KS) formulation, for simulating the
110: melting properties of small Sn\cite{snmelting} and Ga\cite {gamelting}
111: clusters in the size range 10$-$20. Unfortunately, the computational
112: expense of the KS approach, combined with the relatively large sizes
113: of the Na data~($N=$ 55--350), have so far made it difficult to
114: perform KS simulations with high statistics in the size range of the
115: experiment. Recently, a KS simulation of
116: Na$_{55}{}^{+}$,\cite{manninen99,manninen04} performed with limited
117: statistics, shows encouraging results, with an estimated melting
118: temperature $T_m$ between 300 and 350~K [$T_m(\mathrm{expt})\approx
119: 290$~K for Na$_{55}{}^{+}$].
120:
121: In this paper, we report the first \emph{ab initio} KS thermodynamic
122: calculations up to size $N=$ 142, within the size range of the
123: experiment. We present simulations for Na$_{N}$ with $N=$ 55, 92, and
124: 142, each with total sampling times of the order of 2--3~ns. For each
125: size there is a pronounced peak in specific--heat curve, whose
126: position agrees with the experimental finding to better than 8\%.
127: This error is also the approximate level of statistical error for the
128: simulation times used, suggesting that a KS approach is capable of
129: quantitative predictions of melting temperatures at this level of
130: accuracy or better. Analysis of the structural and dynamical
131: information in these simulations sheds new light on the respective
132: roles of geometric and electronic effects in this intriguing problem.
133:
134:
135: \section{Computational Details}
136:
137: Our simulations have been carried out using isokinetic
138: Born--Oppenheimer molecular--dynamics with ultrasoft pseudopotentials
139: in the local density approximation, as implemented in the
140: \textsc{vasp} package.\cite{vasp} We have carried out calculations for
141: up to 17 temperatures for Na$_{55}$ and 14 temperatures for Na$_{92}$
142: and Na$_{142}$ in the temperature range 100 to 500~K. The simulation
143: times are 90~ps for temperatures below
144: $T_m$, and between 150--350~ps for temperatures near the melting
145: transition or higher. Such high statistics have to be performed in
146: order to achieve good convergence of thermodynamic indicators such as
147: the root-mean-square bond-length fluctuation
148: $\delta_{\rm rms}$.\cite{deltarms} In Fig.~\ref{fig.na55.delta_conv},
149: we show $\delta_{\rm rms}$ for Na$_{55}$ as a function of the total
150: time used to compute the time averages. We have discarded at least
151: the first 25~ps at each temperature to allow for thermalization. This
152: plot shows that, for lower temperatures (below the transition
153: temperature, that is, $T\le240$~K) simulation times of the order of
154: 60--70~ps are sufficient for convergence. (For a discussion of melting
155: temperatures $T_m$, see Sec.~\ref{results-discussion}.) For the higher
156: temperatures (300~K and above), at least 100--300~ps are required to
157: achieve proper convergence. However, near the melting transition,
158: namely, around 270 and 280~K, $\delta_{\rm rms}$ is rather poorly
159: converged even up to about 300~ps, indicating the need for higher
160: statistics at these temperatures. Further, with increase in the
161: cluster size, these averaging times may increase, making the \emph{ab
162: initio} simulations of larger clusters very difficult.
163:
164: An energy cutoff of 3.6~Ry is used
165: for the plane--wave expansion of the wavefunction, with a convergence
166: in the total energy of the order of 0.0001--0.0005~eV.
167: Increasing the energy cutoff to, say, about 22~Ry causes a change of
168: only 0.31\% and 0.28\%, respectively, in the binding energy and
169: bond--length of the Na$_{2}$ dimer. We have also checked these results
170: against those obtained by the projected augmented wave~(PAW)
171: method,\cite{paw} taking only the $1s^2$ electrons in the core and
172: with an energy cutoff of about 51~Ry. The Na$_{2}$ binding energy and
173: bond--length according to the present DFT method differ by less than
174: 3\% from those given by the PAW method (the PAW method yielding
175: smaller energies and bond--lengths). The DFT method employed for our
176: thermodynamic simulations can therefore be considered to be of
177: comparable accuracy to the (nearly) all--electron method. The lengths
178: of the simulation cells for Na$_{55}$, Na$_{92}$, and Na$_{142}$ are
179: 24, 28, and 31~\AA, respectively.
180:
181:
182: We have also performed certain tests on the Na$_8$ cluster to check
183: the convergence of the total energy and forces at high temperatures.
184: In Fig.~\ref{fig.na8.hist}, we plot the histograms of
185: potential--energy for the Na$_8$ cluster at 180, 220, 300, and 450~K
186: with a minimum of 3, 4, and 5 iterations\cite{vasp-wfn-extrapol} for
187: obtaining the self--consistent field~(SCF) solution of the KS
188: equations. Clearly, all the plots overlap, indicating that about 4--6
189: SCF iterations are sufficient to obtain a good convergence of the
190: potential--energy. The force convergence test has been performed on
191: the Na$_8$ cluster at 400~K. The forces are converged up to about
192: 0.0001~eV/{\AA} in just 4 SCF iterations. Hence, in all our
193: calculations, we have used a minimum of 4, and in some cases up to 5
194: or 6 iterations for self--consistency.
195:
196:
197: The resulting trajectory data have been used to compute the ionic
198: specific--heat, via a multihistogram fit, as well as other
199: thermodynamic indicators. Details can be found in
200: Ref.~\onlinecite{abhijat01}.
201:
202:
203: \begin{figure}
204: {\centerline {\epsfxsize=1.0\textwidth \epsffile{na55.delta_conv.eps}}}
205: {\caption{
206: \label{fig.na55.delta_conv}
207: The root--mean--square bond--length fluctuations $\delta_{\rm rms}$
208: for Na$_{55}$, as a function of the total time used to compute the
209: average value.
210: %
211: % We have discarded at least the first 25~ps at each
212: % temperature to allow for thermalization. This plot shows that, for
213: % lower temperatures (below the transition temperature, that is,
214: % $T\le240$~K), simulation times of the order of 60--70~ps are
215: % sufficient for convergence. For the higher temperatures (300~K and
216: % above), at least 100--300~ps are required to achieve proper
217: % convergence. However, near the melting transition, namely, around 270
218: % and 280~K, $\delta_{\rm rms}$ is rather poorly converged even up to
219: % about 300~ps, indicating the need for higher statistics at these
220: % temperatures. Further, with increase in the cluster size, these
221: % averaging times may increase, making the \emph{ab initio} simulations
222: % of larger clusters very difficult.
223: %
224: }}
225: \end{figure}
226:
227:
228: \begin{figure}
229: {\centerline {\epsfxsize=1.0\textwidth \epsffile{na8.hist.eps}}}
230: {\caption{
231: \label{fig.na8.hist}
232: Comparison of the potential--energy distributions of the Na$_8$
233: cluster at 180, 220, 300, and 450~K, computed with minimum $\rm N_{\rm
234: scf}=$ 3, 4, 5, iterations\cite{vasp-wfn-extrapol} used for obtaining
235: the self--consistent field solution of the Kohn--Sham equations.
236: }}
237: \end{figure}
238:
239:
240: \section{Results and Discussion}
241: \label{results-discussion}
242:
243: We have carried out a substantial search for the lowest--energy ionic
244: structures. A basin--hopping algorithm\cite{bhop} with around
245: 10$^{5}$ basin hops was employed to generate several hundred
246: structures using the second--moment approximation~(SMA) parametrized
247: interatomic potential of Li \textit{et al.}\cite{smali}. About 25 of
248: the lowest of these structures were then used as input for relaxation
249: within the KS approach. In addition, several quenches were carried
250: out for structures selected from a long high--temperature \emph{ab
251: initio} MD run, typically taken at a temperature well above the
252: melting point~(say, for $T>400$~K). In this way, we have obtained
253: about 20--25 distinct geometries for each size. Most of these
254: structures should be close to the true ground--state, although even
255: this method may not yield the correct absolute ground--state
256: structure. The resulting lowest--energy structures found for the three
257: sizes are shown in Figs.~\ref{fig.struc}(a)--\ref{fig.struc}(c). The
258: lowest--energy geometry of Na$_{55}$ is found to be a two--shell
259: Mackay icosahedron (with a very slight distortion from perfect
260: regularity), in conformity with previous theory\cite{manninen99,brack}
261: and experimental evidence.\cite{icoexpt} Na$_{142}$ is also found to
262: be icosahedral, with 5 atoms missing from the outermost shell. (A
263: three--shell Mackay icosahedron has 147 atoms.) As to Na$_{92}$, the
264: structure consists of surface growth over a slightly distorted
265: icosahedral Na$_{55}$ core. The additional 37 surface atoms are
266: accommodated so as to maintain a roughly spherical shape. Note that
267: for Na$_{92}$, the SMA basin--hopping process has an unclear
268: convergence with respect to the exact configuration of these surface
269: atoms, and so even for the SMA potential we are unlikely to have found
270: the true global ground--state structure. However, since our
271: lowest--energy SMA structure turns out to be quite spherical, and
272: since Na$_{92}$ has a closed--shell configuration of valence
273: electrons, we believe that the overall shape of our lowest--energy KS
274: structure is likely to be correct.
275:
276: \begin{figure}
277: % {\centerline {\epsfxsize=0.38\textwidth \epsffile{struc.eps}}}
278: {\centerline {\epsfxsize=0.9\textwidth \epsffile{struc.eps}}}
279: \caption{
280: \label{fig.struc}
281: Ground state geometries of (a)~Na$_{55}$, (b)~Na$_{92}$,
282: and (c)~Na$_{142}$. Structure (d)~is a representative deformed
283: excited--state structure of Na$_{55}$.
284: }
285: \end{figure}
286:
287:
288: Next we examine the specific--heat for the three clusters
289: investigated, which are plotted in
290: Figs.~\ref{fig.spheat_delta}(a)--\ref{fig.spheat_delta}(c). These
291: curves feature a single dominant peak with a width varying from about
292: 40~K for Na$_{55}$ to about 20~K for Na$_{142}$, in general agreement
293: with the experimental observation. A detailed investigation of the
294: ionic dynamics shows this peak to correspond to a ``melting--like''
295: transition, from a low--temperature solidlike state, in which ions
296: vibrate about fixed points~(and the cluster as a whole may rotate), to
297: a higher--temperature liquidlike state with a diffusive ionic
298: dynamics. As one illustration of this, we show in
299: Figs.~\ref{fig.spheat_delta}(d)--\ref{fig.spheat_delta}(f) the
300: root--mean--square bond--length fluctuation
301: $\delta_{\mathrm{rms}}$.\cite{deltarms} For each size there is a sharp
302: step in $\delta_{\mathrm{rms}}$ that correlates closely with the peak
303: in the specific--heat. The ``melting'' phenomenology found is
304: qualitatively similar to that found in earlier studies with simpler
305: potentials.\cite{cs,bb01,aguado-nalarge}
306:
307:
308: \begin{table}
309: \caption
310: {
311: \label{table.data}
312: Melting temperatures, latent heat, and $\delta E$~(see text) for
313: Na$_{N}$ given by the present Kohn--Sham~(KS) approach, the
314: second--moment approximation~(SMA) potential, and a
315: density--based~(DB) approach. The statistical error in the KS
316: melting
317: temperatures is about 8\%.
318: }
319: \begin{ruledtabular}
320: \begin{tabular}{cccccc}
321: & \mc{5}{c}{Melting Temperature~(K)} \\
322: $N$ & KS & Expt.\footnotemark[1] & SMA\footnotemark[2] &
323: SMA\footnotemark[3] & DB\footnotemark[4] \\
324: \hline
325: 55 & 280 & 290 & 175 & 162 & 190 \\
326: 92 & 195 & 210 & (170)\footnotemark[5] & (133)\footnotemark[5] & 240 \\
327: 142 & 290 & 270 & 240 & 186 & 270 \\
328: \hline
329: & \mc{4}{c}{Latent Heat (meV/atom)} & $\delta E$ (meV/atom)\\
330: $N$ & \mbox{KS} & \mbox{Expt.}\footnotemark[1] &
331: \mbox{SMA}\footnotemark[2] & \mbox{SMA}\footnotemark[3] & \mbox{KS}\\
332: \hline
333: 55 & 13.8 & 13.0 & 20.1 & 8.3 & 53.2 \\
334: 92 & 6.4 & 6.0 & (16.9)\footnotemark[5] & (4.2)\footnotemark[5] & 32.3 \\
335: 142 & 14.9 & 14.0 & (23.0)\footnotemark[6], (23.6)\footnotemark[7] & 186 & 58.0 \\
336: \end{tabular}
337: \end{ruledtabular}
338: \footnotetext[1]{M. Schmidt and H. Haberland, Ref.\ \onlinecite{haberlandCR}.}
339: \footnotetext[2]{F. Calvo and F. Spiegelmann, Ref.\ \onlinecite{cs}.}
340: \footnotetext[3]{K. Manninen \textit{et al.}, Ref.\ \onlinecite{manninen04}.}
341: \footnotetext[4]{A. Aguado \textit{et al.}, Ref.\ \onlinecite{aguado-nalarge}.}
342: \footnotetext[5]{$N = 93$}
343: \footnotetext[6]{$N = 139$}
344: \footnotetext[7]{$N = 147$}
345: \end{table}
346:
347:
348: The KS melting temperatures $T_{m}$ are given along with the
349: approximate ``latent heats'' $L$ in Table \ref{table.data}.
350: Following the convention in the experimental
351: works,\cite{haberland0,haberlandCR} we define $T_{m}$ here as the
352: maximum of the peak in the specific--heat. We have also calculated a
353: quantity $\delta E$ defined as the average potential--energy of the
354: just melted cluster with respect to the ground--state structure at
355: $T=0$~K. Schmidt \emph{et al.}\cite{schmidt-na-dE} have inferred from
356: the experimental caloric curve that the melting temperature is
357: strongly influenced by such an energy contribution. They showed
358: further that $T_m$ follows closely the variation in $\delta E$ as a
359: function of the cluster size. In Table \ref{table.data}, we have also
360: given $T_m$ and $L$ calculated by the SMA parametrized
361: potential.\cite{manninen04,cs} The striking feature of the results
362: summarized in this table is that only the first principles KS--based
363: calculations reproduce, qualitatively as well as quantitatively, the
364: experimentally observed variation in the melting temperatures and the
365: latent heat for the three clusters investigated.\cite{comment1}
366:
367:
368: A noteworthy feature of the data in Table \ref{table.data} is that
369: Na$_{92}$ melts at a significantly lower temperature than Na$_{142}$.
370: This is true both for the KS data, which include electronic shell
371: effects explicitly, and for the SMA, DB, and ETF\cite{bb01} data,
372: which do not. We are thus led to conclude that geometry plays a
373: significant role in the melting--point depression of Na$_{92}$:
374: Na$_{55}$ and Na$_{142}$ are complete, or close to complete, Mackay
375: icosahedra, while Na$_{92}$ has surface growth on a two--shell
376: icosahedral core and is less stable. However, as we shall see, the
377: electronic structure does play a subtle role in the behavior of the
378: melting temperatures of these clusters.
379:
380:
381: \begin{figure}
382: % {\centerline {\epsfxsize=0.46\textwidth \epsffile{spheat-delta.eps}}}
383: {\centerline {\epsfxsize=0.9\textwidth \epsffile{spheat-delta.eps}}}
384: {\caption{
385: \label{fig.spheat_delta}
386: Left panel:
387: Normalized canonical specific heat for
388: (a)~Na$_{55}$, (b)~Na$_{92}$, and (c)~Na$_{142}$.
389: $C_0=(3N-9/2)k_B$ is the zero--temperature classical
390: limit of the rotational plus vibrational canonical
391: specific heat.
392: Right panel:
393: Root--mean--square bond--length fluctuation
394: $\delta_{\rm rms}$\protect\cite{deltarms} for
395: (d)~Na$_{55}$, (e)~Na$_{92}$, and (f)~Na$_{142}$.
396: } }
397: \end{figure}
398:
399:
400: As mentioned earlier, previous simulations gave a melting temperature
401: for Na$_{55}$ significantly lower than the experimental one. Moreover,
402: for each of the SMA,\cite{manninen04,cs} ETF,\cite{bb01} and
403: DB\cite{aguado-nalarge} models, $T_{m}(55)$ is found to be
404: significantly less than $T_{m}(142)$ calculated within the same model,
405: while in the experiment $T_{m}(55)$ is very slightly greater than
406: $T_{m}(142)$.\cite{walesberry90} This discrepancy is largely removed
407: within the present KS model: we find that $T_{m}(55)\approx
408: T_{m}(142)$, within statistical error, and that both melting
409: temperatures agree with experiment. This suggests quite strongly that
410: the high melting point of Na$_{55}$ relative to Na$_{142}$ is due to
411: electronic shell effects, since these are the main new element in our
412: KS approach not included in previous simulations. This is
413: particularly clear in the case of the DB and ETF methods. In these
414: approaches, quantum shell effects are effectively averaged out as a
415: function of size by the use of a density--dependent electron
416: kinetic--energy functional, and the total cluster energy follows quite
417: closely a smooth dependence on cluster size given by a
418: liquid--drop--type formula.\cite{bb01} In all other respects, however,
419: such as the use of pseudopotentials and density--dependent
420: exchange--correlation functionals, the DB and ETF methods are similar
421: to the present KS method.\cite{dbpseudo} We note that both Na$_{55}$
422: and Na$_{142}$ are close to ``magic'' systems~(Na$_{58}$ and
423: Na$_{138}$, respectively), but shell effects are relatively more
424: important for smaller systems,\cite{bohrmottelson} which could lead to
425: a relative enhancement in stability for Na$_{55}$ and thus a
426: relatively higher melting temperature.
427:
428:
429: \begin{figure}
430: % {\centerline {\epsfxsize=0.28\textwidth \epsffile{eps-pro.eps}}}
431: {\centerline {\epsfxsize=0.6\textwidth \epsffile{eps-pro.eps}}}
432: {\caption{
433: \label{fig.eps-pro}
434: Time--averaged coefficient $\epsilon_{\rm def}$~(see text) describing
435: the degree of quadrupole deformation of Na$_{55}$~(continuous line)
436: and Na$_{92}$~(dotted line).
437: }}
438: \end{figure}
439:
440:
441: Further evidence for the role of quantum shell effects in melting may
442: be obtained by examining the \emph{shape} of the cluster before and
443: after melting. In Fig.~\ref{fig.eps-pro} we plot the deformation
444: parameter $\epsilon_{\mathrm{def}}$ for Na$_{55}$, defined as
445: $\epsilon_{\mathrm{def}}={2Q_{1}}/(Q_{2}+Q_{3})$, where $Q_{1} \geq
446: Q_{2} \geq Q_{3}$ are the eigenvalues, in descending order, of the
447: quadrupole tensor $Q_{ij}=\sum_{I} R_{Ii}R_{Ij}.$ Here $i$ and $j$ run
448: from 1 to 3, $I$ runs over the number of ions, and $R_{Ii}$ is the
449: $i$th coordinate of ion $I$ relative to the cluster center of mass.
450: A spherical system~($Q_{1} = Q_{2} = Q_{3}$) has
451: $\epsilon_{\mathrm{def}}=1$; a value $\epsilon_{\mathrm{def}}>1$
452: indicates a quadrupole deformation of some type. From
453: Fig.~\ref{fig.eps-pro} we see that at low temperatures
454: $\epsilon_{\mathrm{def}}\approx 1$, corresponding to the compact
455: icosahedral ground--state, but as the cluster melts, the system
456: acquires a quadrupole deformation with $\epsilon_{\mathrm{def}}\approx
457: 1.6$. A more detailed investigation of this deformation with
458: two--dimensional deformation parameters such as the Hill--Wheeler
459: parameters\cite{bohrmottelson}~(not shown) indicates the cluster to be
460: undergoing shape fluctuations around a net prolate deformation. A
461: typical deformed liquidlike structure is shown in Fig.\
462: \ref{fig.struc}(d). A related phenomenon was observed earlier by
463: Rytk\"onen \emph{et al.}\cite{rytkonen98} for Na$_{40}$, except that
464: the magic Na$_{40}$ cluster underwent only an \emph{octupole}
465: deformation rather than a quadrupole deformation as observed here for
466: the nonmagic Na$_{55}$. We do not observe statistically significant
467: quadrupole deformations for Na$_{92}$ or Na$_{142}$, for which
468: $\epsilon_{\mathrm{def}}\approx 1$ at all temperatures.
469:
470:
471: Interestingly, simulations of Na$_{55}$ carried out by us within the
472: SMA model do not show a deformation upon melting, but rather the
473: cluster remains essentially spherical at all temperatures,
474: $\epsilon_{\mathrm{def}}\approx 1$. Since the SMA model explicitly
475: excludes quantum shell effects, we believe that the deformation of
476: Na$_{55}$ is due to the quantal Jahn--Teller distortion of the
477: open--shell system of valence electrons. As further support for this
478: explanation, we note that the mean prolate deformation in the
479: liquidlike state agrees well with that found in jellium--model
480: calculations for neutral Na$_{55}$ that we have carried out, which
481: yield a uniaxial prolate deformation with a major-- to minor--axis
482: ratio of $R_{\mathrm{maj}}/R_{\mathrm{min}} \approx
483: \sqrt{\epsilon_{\mathrm{def}}} \approx 1.3$, in close agreement with
484: the liquidlike simulations. Evidently, the compact ground--state
485: structure is favored by the possibility of geometric packing into an
486: icosahedron, while in the nonrigid liquidlike state the cluster can
487: lower its free energy by undergoing a spontaneous shape deformation.
488: On this reasoning, the magic Na$_{92}$ cluster would not be expected
489: to deform upon melting, consistent with the observation. On the other
490: hand, the Na$_{142}$ cluster \emph{would} be expected to deform, at
491: least slightly; presumably, the Jahn--Teller forces here are
492: sufficiently weak that any deformation is not statistically
493: significant.
494:
495:
496: Note that, as mentioned earlier, the melting temperature is strongly
497: influenced by the potential--energy difference $\delta E$ between
498: liquidlike and solidlike states. Therefore, even if the ground--state
499: is quite spherical, as for Na$_{55}$, the melting temperature may
500: still be influenced by important quantal deformation effects entering
501: only in the liquidlike state. The KS simulations undertaken here
502: incorporate all these various effects correctly.
503:
504:
505: \section{Conclusion}
506:
507: In conclusion, the KS approach appears to be capable of making
508: quantitative predictions of melting temperatures and latent heats in
509: Na clusters. Na$_{55}$, Na$_{92}$, and Na$_{142}$ are each magic or
510: nearly magic, but only Na$_{55}$ and Na$_{142}$ are also close to
511: icosahedral shell closures. The fact that Na$_{92}$ melts at a
512: significantly lower temperature than the other two shows that
513: geometric effects are very important in determining the pattern of
514: melting temperatures observed experimentally. However, electronic
515: shell effects can play an important role too, both in influencing
516: overall binding energies and bond--lengths as a function of size, and
517: indirectly via shape deformation effects that may arise differently in
518: the solidlike and liquidlike states. For an accurate treatment of the
519: metallic bonding, and a quantitative prediction of melting
520: temperatures and latent heats, it is essential to incorporate
521: electronic shell effects appropriately, as is possible, for example,
522: within the KS approach used here. This is especially true for the
523: smaller sizes of clusters. There is a size regime up to $N \approx
524: 150$ or so where a full KS treatment is warranted.
525:
526:
527: \section{Acknowledgment}
528:
529: One of us (SC) acknowledges financial support from the Center for
530: Modeling and Simulation, University of Pune. We gratefully acknowledge
531: the support of the Indo--French Center for Promotion for Advance
532: Research. It is a pleasure to acknowledge C--DAC~(Pune) for providing
533: us with supercomputing facilities.
534:
535:
536: \begin{references}
537:
538: \bibitem{haberland0}
539: M. Schmidt, R. Kusche, W. Kronm\"{u}ller, B. v. Issendorff, and
540: H. Haberland,
541: Phys. Rev. Lett. \textbf{79}, 99 (1997);
542: M. Schmidt, R. Kusche, B. v. Issendorff, and H. Haberland,
543: Nature (London) \textbf{393}, 238 (1998);
544: R. Kusche, Th. Hippler, M. Schmidt, B. v. Issendorff, and
545: H. Haberland,
546: Eur. Phys. J. D \textbf{9}, 1 (1999).
547: % M. Schmidt \textit{et al.},
548: % Phys. Rev. Lett. \textbf{79}, 99 (1997);
549: % M. Schmidt \textit{et al.},
550: % Nature (London) \textbf{393}, 238 (1998);
551: % R. Kusche \textit{et al.},
552: % Eur. Phys. J. D \textbf{9}, 1 (1999).
553:
554: \bibitem{haberlandnegCV}
555: M. Schmidt, R. Kusche, Th. Hippler, J. Donges, W. Kronm\"{u}ller,
556: B. v. Issendorff, and H. Haberland,
557: Phys. Rev. Lett. \textbf{86}, 1191 (2001).
558: % M. Schmidt \textit{et al.},
559: % Phys. Rev. Lett. \textbf{86}, 1191 (2001).
560:
561: \bibitem{haberlandCR}
562: M. Schmidt and H. Haberland,
563: C. R. Physique \textbf{3}, 327 (2002).
564:
565: \bibitem{jarroldGa}
566: G. A. Breaux, R. C. Benirschke, T. Sugai, B. S. Kinnear, and M. F.
567: Jarrold,
568: Phys. Rev. Lett. \textbf{91}, 215508 (2003).
569: % G. A. Breaux \textit{et al.},
570: % Phys. Rev. Lett. \textbf{91}, 215508 (2003).
571:
572: \bibitem{pawlow}
573: P. Pawlow,
574: Z. Phys. Chem. (Leipzig) \textbf{65}, 1 (1909);
575: K.--J. Hanszen,
576: Z. Phys. \textbf{157}, 523 (1960).
577:
578: \bibitem{buffat}
579: Ph. Buffat and J.--P. Borel,
580: Phys. Rev. A \textbf{13}, 2287 (1976).
581:
582: %\bibitem{boiko}
583: % B. T. Boiko \textit{et al.},
584: % Tela (Leningrad) \textbf{10}, 3567 (1968)
585: % [Sov. Phys. Solid State \textbf{10}, 2832 (1969)];
586: % R. P. Berman and A. E. Curzon,
587: % Can. J. Phys. \textbf{52}, 923 (1974).
588: \bibitem{boiko}
589: B. T. Boiko, A. T. Pugachev, and V. M. Bratsykhin,
590: Fiz. Tverd. Tela (Leningrad)
591: \textbf{10}, 3567 (1968) [Sov. Phys. Solid State
592: \textbf{10}, 2832 (1969)];
593: R. P. Berman and A. E. Curzon, Can. J. Phys.
594: \textbf{52}, 923 (1974).
595:
596: %\bibitem{manninen99}
597: % A. Rytk\"{o}nen \textit{et al.},
598: % Eur. Phys. J. D \textbf{9}, 451 (1999).
599: \bibitem{manninen99}
600: A. Rytk\"{o}nen, H. H\"{a}kkinen, and M. Manninen,
601: Eur. Phys. J. D \textbf{9}, 451 (1999).
602:
603: %\bibitem{manninen04}
604: % K. Manninen \textit{et al.},
605: % Eur. Phys. J. D \textbf{29}, 39 (2004).
606: \bibitem{manninen04}
607: K. Manninen, A. Rytk\"{o}nen, and M. Manninen, Eur.
608: Phys. J. D \textbf{29}, 39 (2004).
609:
610: \bibitem{cs}
611: F. Calvo and F. Spiegelmann,
612: J. Chem. Phys. \textbf{112}, 2888 (2000);
613: \textit{ibid.} \textbf{120}, 9684 (2004).
614:
615: \bibitem{bb01}
616: P. Blaise and S. A. Blundell,
617: Phys. Rev. B \textbf{63}, 235409 (2001).
618:
619: %\bibitem{aguado}
620: % A. Aguado \textit{et al.},
621: % J. Chem. Phys. \textbf{111}, 6026 (1999).
622: \bibitem{aguado}
623: A. Aguado, J. M. L\'{o}pez, J. A. Alonso, and M. J. Stott,
624: J. Chem. Phys. \textbf{111}, 6026 (1999).
625:
626: %\bibitem{aguado-nalarge}
627: % A. Aguado \textit{et al.},
628: % J. Phys. Chem. B \textbf{105}, 2386 (2001).
629: \bibitem{aguado-nalarge}
630: A. Aguado, J. M. L\'{o}pez, J. A. Alonso, and M. J. Stott,
631: J. Phys. Chem. B \textbf{105}, 2386 (2001).
632: The $T_m$ for Na$_{92}$ noted in this reference is for the
633: icosahedral isomer, which is not the correct ground--state.
634:
635: %\bibitem{knight84}
636: % W. D. Knight \textit{et al.}
637: % Phys. Rev. Lett. \textbf{52}, 2141 (1984).
638: \bibitem{knight84}
639: W. D. Knight, K. Clemenger, W. A. de Heer, W. A.
640: Saunders, M. Y. Chou, and M. L. Cohen,
641: Phys. Rev. Lett. \textbf{52}, 2141 (1984).
642:
643: %\bibitem{abhijat01}
644: % A. Vichare \textit{et al.},
645: % Phys. Rev. B \textbf{64}, 045408 (2001).
646: \bibitem{abhijat01}
647: A. Vichare, D. G. Kanhere, and S. A. Blundell,
648: Phys. Rev. B \textbf{64}, 045408 (2001).
649:
650: % \bibitem{ks} W. Kohn and L. J. Sham, Phys. Rev. \textbf{140}, A1133 (1965).
651:
652: %\bibitem{snmelting}
653: % K. Joshi \textit{et al.},
654: % Phys. Rev. B \textbf{66}, 155329 (2002);
655: % \textit{ibid.}
656: % \textbf{67}, 235413 (2003).
657:
658: \bibitem{snmelting}
659: K. Joshi, D. G. Kanhere, and S. A. Blundell,
660: Phys. Rev. B \textbf{66}, 155329 (2002);
661: \textit{ibid.} \textbf{67}, 235413 (2003).
662:
663: %\bibitem{gamelting}
664: % S. Chacko \textit{et al.},
665: % Phys. Rev. Lett. \textbf{92}, 135506 (2004).
666: \bibitem{gamelting}
667: S. Chacko, K. Joshi, D. G. Kanhere, and S. A. Blundell,
668: Phys. Rev. Lett. \textbf{92}, 135506 (2004).
669:
670: \bibitem{vasp}
671: Vienna {\em Ab initio } Simulation Package,
672: Technische Universit\"at Wien (1999);
673: G. Kresse and J. Furthm\"uller,
674: Phys. Rev. B {\bf 54}, 11169 (1996).
675:
676: \bibitem{deltarms}
677: $\delta_{\mathrm{rms}}$ is defined in Ref.~\onlinecite{snmelting},
678: for example, and corresponds to a finite--sized version of the bulk
679: Lindemann criterion.
680:
681: \bibitem{paw}
682: G. Kresse, J. Joubert,
683: Phys. Rev. B \textbf{59}, 1758 (1999);
684: P. E. Bl\"ochl,
685: Phys. Rev. B \textbf{50}, 17953 (1994).
686:
687: \bibitem{vasp-wfn-extrapol}
688: In the \textsc{vasp} package during molecular--dynamics, a
689: second--order extrapolation of the wavefunctions and charge density is
690: performed. This reduces the number of self--consistency iterations
691: required to as few as 4--6.
692:
693: \bibitem{bhop}
694: Z. Li and H. A. Scheraga,
695: Proc. Natl. Acad. Sci. U.S.A. \textbf{84}, 6611 (1987);
696: D. J. Wales and J. P. K. Doye,
697: J. Phys. Chem. A \textbf{101}, 5111 (1997).
698:
699: \bibitem{smali}
700: Y. Li, E. Blaisten--Barojas, and D. A. Papaconstantopoulos,
701: Phys. Rev. B \textbf{57}, 15519 (1998).
702:
703: %\bibitem{brack}
704: % S. K\"ummel \textit{et al.},
705: % Phys. Rev. B \textbf{62}, 7602 (2000).
706: \bibitem{brack}
707: S. K\"ummel, M. Brack, and P.--G. Reinhard,
708: Phys. Rev. B \textbf{62}, 7602 (2000).
709:
710: %\bibitem{icoexpt}
711: % G. Wrigge \textit{et al.},
712: % Phys. Rev. A \textbf{65}, 063201 (2002).
713: \bibitem{icoexpt}
714: G. Wrigge, M. Astruc Hoffmann, and B. v. Issendorff,
715: Phys. Rev. A \textbf{65}, 063201 (2002).
716:
717: %\bibitem{schmidt-na-dE}
718: % M. Schmidt \textit{et al.},
719: % Phys. Rev. Lett. \textbf{90}, 103401 (2003).
720: \bibitem{schmidt-na-dE}
721: M. Schmidt, J. Donges, Th. Hippler, and H. Haberland,
722: Phys. Rev. Lett. \textbf{90}, 103401 (2003).
723:
724: \bibitem{comment1}
725: Although our calculations are for neutral clusters,
726: we do not expect the results to change significantly for
727: the singly charged clusters studied experimentally.
728: See Ref.~\onlinecite{cs} for a model.
729:
730: \bibitem{walesberry90}
731: For Lennard--Jones clusters also, $T_m(55) < T_m(142)$;
732: D. J. Wales and R. S. Berry,
733: J. Chem. Phys. \textbf{92}, 4473 (1990).
734:
735: \bibitem{dbpseudo}
736: There are also differences between the pseudopotentials used in the
737: DB and ETF calculations and in the present KS calculation. However,
738: the DB calculation, Ref.~\onlinecite{aguado-nalarge}, used a
739: norm--conserving pseudopotential, while the ETF calculation,
740: Ref.~\onlinecite{bb01}, used a softer, phenomenological
741: pseudopotential, and in both cases it was found that $T_m(55)$ was
742: significantly less than $T_m(142)$, suggesting that this result is
743: largely independent of the pseudopotential used. In the present KS
744: calculations, as we have mentioned, the ultrasoft pseudopotentials
745: compare well with (nearly) all--electron PAW calculations. We
746: therefore believe that differences in the pseudopotential are unlikely
747: to account for the success of the present KS method in reproducing the
748: high melting point of $T_m(55)$ relative to $T_m(142)$.
749:
750: \bibitem{bohrmottelson}
751: See, for example,
752: A. Bohr and B. R. Mottelson,
753: \textit{Nuclear Structure} (Benjamin, New York, 1974),
754: Vols. I and II.
755:
756: %\bibitem{rytkonen98}
757: % A. Rytk\"onen \textit{et al.},
758: % Phys. Rev. Lett. \textbf{80}, 3940 (1998).
759: \bibitem{rytkonen98}
760: A. Rytk\"onen, H. H\"akkinen, and M. Manninen,
761: Phys. Rev. Lett. \textbf{80}, 3940 (1998).
762:
763: \end{references}
764:
765: \end{document}
766:
767: