1: %\documentclass[preprint,prb,showpacs]{revtex4}
2: \documentclass[twocolumn,prb,showpacs]{revtex4}
3: \usepackage{graphicx}
4: \usepackage{bm}
5:
6:
7: \begin{document}
8: \title{Lattice gas description of pyrochlore and checkerboard
9: antiferromagnets in a strong magnetic field}
10:
11: \author{M. E. Zhitomirsky}
12: \affiliation{
13: Commissariat \`a l'Energie Atomique, DSM/DRFMC/SPSMS, 38054 Grenoble,
14: France}
15: \author{Hirokazu Tsunetsugu}
16: \affiliation{
17: Institute for Solid State Physics, University of Tokyo, Kashiwa,
18: Chiba 277-8581, Japan}
19:
20: \date{08 March, 2007}
21: %\date{\today}
22:
23: \begin{abstract}
24: Quantum Heisenberg antiferromagnets on pyrochlore and checkerboard lattices
25: in a strong external magnetic field are mapped onto hard-core lattice gases
26: with an extended exclusion region.
27: The effective models are studied by the exchange Monte Carlo simulations
28: and by the transfer matrix method.
29: The transition point and the critical exponents are obtained numerically
30: for a square-lattice gas of particles with the second-neighbor exclusion,
31: which describes a checkerboard antiferromagnet.
32: The exact structure of the magnon crystal state is determined for a pyrochlore
33: antiferromagnet.
34: \end{abstract}
35: \pacs{
36: 75.50.Ee, %Antiferromagnetics
37: 75.45.+j, %Macroscopic quantum phenomena in magnetic systems
38: 75.10.Jm, %Quantized spin models
39: 75.30.Sg} %Magnetocaloric Effect
40:
41:
42: \maketitle
43:
44: \section{Introduction}
45:
46: A characteristic feature of geometrically frustrated spin models
47: is an infinite degeneracy of the classical ground state. Such a
48: degeneracy may be lifted by quantum or thermal fluctuations via
49: the `order from disorder' effect. \cite{villain,shender,henley89}
50: A closely related property is emergence of small energy scales,
51: which are described by the effective Hamiltonians acting in the subspace
52: of nearly degenerate low-energy quantum states. To date, several types of
53: effective models have been developed for frustrated
54: antiferromagnets in different regimes. The first example is
55: a celebrated quantum dimer model introduced phenomenologically
56: by Rokhsar and Kivelson to describe a magnetically disordered
57: resontaing valence bond phase of $S=1/2$ antiferromagnets. \cite{RK}
58: Recently, effective quantum dimer models have been derived
59: for several realistic spin Hamiltonians in zero and in
60: a finite magnetic field. \cite{cabra,mzh05,bergman}
61: The second example is the effective Hamiltonian for the collinear
62: states, which applies to a semiclassical ($S\gg 1$) pyrochlore
63: antiferromagnet in zero magnetic field. \cite{hizi} The third
64: type of effective models describes geometrically frustrated
65: magnets at high magnetic fields. \cite{hard1,derzhko04,hard2,richter06}
66: These are the so called lattice-gas models considered further
67: in the present work.
68:
69: Previously, the effective lattice-gas models have been derived and
70: discussed for the sawtooth chain and a kagom\'e antiferromagnet.
71: \cite{hard1,derzhko04,hard2} These two quantum antiferromagnets
72: are mapped onto lattice gases of hard-core classical particles
73: with the nearest-neighbor exclusion on a one-dimensional chain
74: and a triangular lattice, which are
75: respectively called the hard-dimer and the hard-hexagon model.
76: In both cases, the thermodynamics of the effective models has been
77: calculated exactly. \cite{baxter}
78: In the present article, we shall focus on the high-field behavior
79: of quantum antiferromagnets on two- and three-dimensional
80: pyrochlore lattices. The corresponding effective models do not allow
81: exact solution for their thermodynamic properties, therefore,
82: we shall study them via a combination of numerical techniques.
83:
84: \begin{figure}%[t]
85: \begin{center}
86: \includegraphics[width=0.95\columnwidth]{lattices.eps}
87: \end{center}
88: \caption{
89: Pyrochlore lattice of corner-sharing tetrahedra (a)
90: and its two-dimensional analog,
91: checkerboard lattice (b). Solid lines show
92: the nearest-neighbor bonds of strength $J$.
93: Dashed lines indicate possible further neighbor exchanges.
94: \label{lattices}}
95: \end{figure}
96:
97: The pyrochlore and the checkerboard lattices (Fig.~\ref{lattices})
98: consist of 3D and 2D networks of corner-sharing tetrahedra
99: (four-spin blocks).
100: For the most part of the present work, we disregard weak further neighbor
101: exchanges and consider the nearest-neighbor Heisenberg model
102: with an arbitrary spin $S$ in an external magnetic field:
103: \begin{equation}
104: \hat{\cal H} = J \sum_{\langle ij\rangle}
105: {\bf S}_i\cdot {\bf S}_j
106: - {\bf H}\cdot \sum_i {\bf S}_i \ .
107: \label{Hamiltonian}
108: \end{equation}
109: Above the saturation field $H_s$ the ground state
110: of a Heisenberg antiferromagnet is a fully polarized vacuum
111: $|0\rangle = |\uparrow\uparrow\uparrow...\rangle$,
112: where all spins are in a state with the maximum possible value
113: of $S_i^z =S$.
114: The low-lying excitations are single spin flips
115: $|i\rangle = S_i^-|0\rangle$, whose dispersion can be straightforwardly
116: calculated.
117: The characteristic feature of many geometrically frustrated models
118: including the two considered models is
119: presence of the low-energy branch of localized magnons
120: with the energy $\varepsilon_0 = H-H_s$.
121: \cite{schnack,schulenburg,schmidt,mzh03,richter}
122: The details for the two discussed models are presented
123: in the subsequent sections.
124:
125: Localized nature of magnons from the lowest branch allows to
126: construct a subclass of exact multimagnon states: \cite{schulenburg}
127: a quantum state with $m$ localized magnons ($m$-LMs)
128: occupying spatially separated regions of the lattice, which
129: are not directly connected by the exchange bonds, is an exact eigenstate
130: of the quantum Hamiltonian (\ref{Hamiltonian}) and its energy
131: is given by $m\varepsilon_0$. It has been proven exactly that
132: isolated LM states correspond
133: to the lowest energy states in every magnetization subsector.
134: \cite{schmidt,hard2}
135: In addition, the renormalization arguments suggest
136: that delocalized (propagating) states are separated by a finite gap from
137: LMs.\cite{hard1,hard2}
138: Hence, at low temperatures the partition function
139: of a quantum frustrated antiferromagnet in the vicinity
140: of the saturation field is determined entirely by LMs:
141: \begin{equation}
142: {\cal Z} = \sum_{m=0}^{N_{\rm max}} w(m,N)\,e^{m\mu/T} \ ,\ \ \
143: \mu = H_s-H \ ,
144: \label{Z}
145: \end{equation}
146: with $z=e^{\mu/T}$ being the activity. Combinatorial factors
147: $w(m,N)$ denote a number of linearly independent $m$-LM states,
148: while $N_{\rm max}$ is the maximal number of LMs
149: for a lattice with $N$ sites ($N\rightarrow\infty$).
150: Mapping to a lattice gas of hard-core objects is used as
151: an {\it approximate} way to calculate $w(m,N)$ and $\cal Z$.
152: The general consequence of Eq.~(\ref{Z}) is the following
153: scaling of the total entropy: ${\cal S}=f[(H-H_s)/T]$.
154: At $H=H_s$ the entropy is temperature independent and, consequently,
155: even at $T=0$ a frustrated quantum antiferromagnet has
156: a finite macroscopic entropy, which is determined purely by the lattice
157: geometry. The quantum order from disorder mechanism becomes ineffective
158: for this special value of applied field due to the localized
159: nature of (exact) quantum states with zero energy.
160: In Secs.~II and III we describe derivation of lattice gas mapping
161: and obtain conclusions separately for the checkerboard and the pyrochlore
162: antiferromagnet. Role of extra exchanges beyond the nearest-neighbor
163: pairs, see Fig.~\ref{lattices}, is briefly discussed in Sec.~IV.
164:
165:
166: \section{Checkerboard antiferromagnet}
167:
168: \subsection{Localized magnon states}
169:
170: A checkerboard lattice contains
171: two spins in the primitive unit cell.
172: The corresponding Bravais lattice is formed by
173: elementary translations on ${\bf a}_1=(1,1)$ and
174: ${\bf a}_2 =(-1,1)$ and has a square shape.
175: In the saturated phase at high fields,
176: the one-magnon spectrum has, accordingly,
177: two branches:
178: \begin{equation}
179: \omega_{1{\bf k}}=H-8JS \ ,\ \ \ \ \ \omega_{2{\bf k}}=H -
180: 4JS(1-\gamma_{\bf k}) \ ,
181: \label{wchek}
182: \end{equation}
183: where $\gamma_{\bf k} = \cos k_x \cos k_y$.
184: The saturation field $H_s=8JS$ corresponds to the vanishing
185: energy of magnons from
186: the lowest dispersionless branch $\omega_{1{\bf k}}$.
187: Localization of excitations
188: from the lowest branch is determined by the lattice topology.
189: A simple localized state can be constructed as a
190: spin-flip trapped on a square void of a checkerboard
191: lattice:
192: \begin{equation}
193: |\varphi_i\rangle = \frac{1}{\sqrt{8S}} %{\textstyle }
194: \sum_{n=1}^4 (-1)^{n-1}S^-_{ni}|0\rangle \ ,
195: \label{localC}
196: \end{equation}
197: where the numbering of sites goes counterclockwise starting
198: from the lowest right corner, see Fig.~\ref{checker}.
199: The probability to find spin-flip on a site adjacent
200: to the void
201: vanishes due to the destructive interference.
202: This property crucially depends on equal strength of
203: all bonds of the checkerboard lattice.
204:
205: \begin{figure}%[t]
206: \begin{center}
207: \includegraphics[width=0.9\columnwidth]{locmag_checker.eps}
208: \end{center}
209: \caption{(color online). Localized magnons on a checkerboard lattice.
210: Thick lines indicate positions of one-magnon states with corresponding
211: phases shown by $+$ and $-$.
212: The shaded area shows excluded volume around a central square-void
213: state, which has to be respected for construction of multiparticle
214: states. Four LMs in the upper-left part satisfy a close-packed condition
215: (see the text).
216: \label{checker}}
217: \end{figure}
218:
219: A linear combination of two square-void states sharing one
220: vertex encompasses two square voids and so on, see Fig.~\ref{checker}.
221: An arbitrary 1-LM state can be constructed by drawing
222: a closed graph on the original lattice, which passes through
223: two vertices of every crossed four-site block,
224: and assigning $+$ and $-$£ signs in alternate order
225: for a spin-flip propagating around it.
226: The smallest square-void states play
227: an important role by forming a complete nonorthogonal
228: basis in the subspace of dispersionless one-magnon
229: states.\cite{hard2}
230: The multi-particle LM states are, then, constructed by
231: respecting topology of the exchange bonds:
232: LMs cannot occupy square voids, which are contiguous to the same
233: crossed square as shown in Fig.~\ref{checker} by shaded area.
234:
235: There is an upper limit on the density of isolated localized
236: magnons. For a checkerboard lattice, a close-packed structure
237: is constructed by putting LMs on every second square void
238: in horizontal rows with an alternating shift between the rows,
239: see the left part in Fig.~\ref{checker}.
240: This pattern corresponds
241: to a magnon crystal, which breaks translational symmetry
242: and has the density of LMs equal to $n_0=N_{\rm max}/N=1/8$.
243: The breaking of the translational symmetry is, however,
244: incomplete: {\it diagonal} rows of LMs can freely slide
245: without affecting magnons in adjacent rows.
246: The degeneracy of the magnon crystal state is
247: $N_{\rm deg}\sim 2^{L+1}$, where $L$ is a linear size of the system,
248: see also Sec.~IIc.
249: Note, that the magnon crystal for a quantum kagom\'e
250: antiferromagnet is only three-fold degenerate.
251: In that case there is a well-defined finite temperature
252: transition associated with the translational
253: symmetry breaking.
254:
255: The purpose of the present study is to investigate the
256: low-temperature behavior of the checkerboard antiferromagnet
257: and nature of a phase transition into the magnon crystal
258: state. In order to proceed, we map a quantum checkerboard
259: antiferromagnet in the vicinity of the saturation onto a gas of
260: hard-core particles. Such an effective model is defined on
261: a $\sqrt{2}\times\sqrt{2}$ square lattice formed
262: by centers of square voids of the original checkerboard lattice.
263: The dual square lattice is rotated by $45^\circ$ and contains $N/2$ sites.
264: Localized magnons are represented by hard-core classical particles
265: obeying the nearest- and the next-nearest neighbor exclusion principle.
266: Below, we call them `hard-polygon states.' One such
267: polygon for a checkerboard lattice is shown in Fig.~\ref{checker}
268: by a shaded square. In reality a state with two localized magnons
269: occupying adjacent sites is separated by only a finite gap from
270: the low-lying LM states. An estimate of such gap for a kagom\'e
271: lattice antiferromagnet is given in Ref.~\onlinecite{hard1}.
272: In the following only $T\rightarrow 0$ regime is considered, where
273: contribution of such higher energy states can be neglected.
274:
275:
276: \subsection{Topological classes of localized magnons}
277:
278: \begin{figure}[t]
279: \begin{center}
280: \includegraphics[width=0.95\columnwidth]{nonhard.eps}
281: \end{center}
282: \caption{(color online).
283: Examples of topologically different states of localized magnons:
284: 2-magnon subsector (a) and (b); 3-magnon subsector (c)--(e);
285: (f) shows an elementary defect state from the class (b).
286: \label{defect}}
287: \end{figure}
288:
289: The hard-polygon states being linearly independent \cite{schmidt06}
290: do not exhaust all possible
291: localized magnon states. The missing `defect' states belong to
292: different topological classes of LMs. \cite{hard1,hard2}
293: Let us consider an infinite
294: plane with open boundary conditions, which has topology of a sphere
295: without the North pole. As was explained before, one-particle
296: localized states correspond to closed lines. Successive expansion
297: of the wave-function of a one-magnon state in terms of basis plaquette
298: states can be represented as a gradual deformation of a long loop with
299: subsequent contraction into a point (center of the last plaquette).
300: Two closed loops for a 2-LM state may be either contractible
301: to two distinct points on a plane or lie inside each other
302: and, hence, be contractible into a single point,
303: see Figs.~\ref{defect}a and \ref{defect}b.
304: In the first case the two magnon state belongs to a subset
305: of two-particle hard-polygon states, whereas in the second case
306: an expansion in the hard-polygon states is impossible.
307: The two-magnon states shown in Figs.~\ref{defect}a and \ref{defect}b
308: are, therefore, {\it linearly independent}.
309:
310: Situation is somewhat different for a cluster with periodic
311: boundary conditions, which has a torus topology.
312: In that case one can continuously deform the outer line
313: in Fig.~\ref{defect}b, split it into two loops with finite winding
314: around the torus, move them around and close
315: again into one contour on the opposite side
316: such that the two-magnon state in Fig.~\ref{defect}b
317: transforms into a state of type Fig.~\ref{defect}a.
318: The linear dependence of two-magnon states in Figs.~\ref{defect}a and
319: \ref{defect}b for the torus topology is explained by presence
320: in this case of one linear relation between 1-LM square-void (hexagon)
321: states. \cite{hard2} Still, there are additional 2-LM states
322: in this topology, which are constructed
323: by putting one or two LMs on contours with nontrivial winding around
324: the torus. Topologically nontrivial classes of
325: LM states in the three-magnon subsector for a lattice with open
326: boundary conditions are shown in Figs.~\ref{defect}c--e.
327: For periodic boundary conditions the state in Fig.~\ref{defect}c
328: becomes topologically equivalent to the state in Fig.~\ref{defect}e,
329: while the state Fig.~\ref{defect}d can be transformed into
330: the hard-polygon state.
331: The macroscopic limit, however, does not depend on the boundary
332: conditions. Equivalence of the two approaches in
333: the $N\rightarrow\infty$ limit
334: is recovered by observing that the largest number of topologically nontrivial
335: LM states in $n$-magnon subsectors with $n>2$ is given by states of
336: the type shown in Fig.~\ref{defect}e,
337: which are present for both choices of boundary conditions.
338: Disentanglement of two, three, etc.\ enclosed loops in the torus topology
339: is impossible once other LMs are present. On the other hand,
340: the contribution from closed loops with
341: a nontrivial winding around the torus
342: corresponds in the thermodynamic limit only to a surface effect.
343:
344:
345: Topological origin of the additional localized magnon
346: states determines their presence for all two-dimensional
347: frustrated models including
348: antiferromagnets on kagom\'e, checkerboard, and star lattices.
349: There are no such states in one-dimensional models
350: as, for example, the sawtooth chain,
351: where the hard-particle representation is asymptotically
352: exact. \cite{hard1,hard2}
353: In order to estimate the contribution of additional
354: LMs, one has to define the basis
355: states in the topologically nontrivial classes.
356: The elementary defect state in the topological
357: class of Fig.~\ref{defect}b is shown in Fig.~\ref{defect}f
358: (open boundary conditions are assumed). Its wave-function is given by
359: \begin{equation}
360: |\textrm{2-defect}\rangle \simeq
361: \Bigl(\sum_{i=1}^{8} (-1)^{i-1}|\varphi_i\rangle + |\varphi_0\rangle\Bigr)
362: |\varphi_0\rangle\ ,
363: \label{2defect}
364: \end{equation}
365: where the numbering of plaquette states follows Fig.~\ref{defect}f.
366: This wave-function includes two LMs on adjacent voids and on
367: the central plaquette and violates, therefore, the hard-polygon
368: constraint. The two-magnon states (\ref{2defect}) residing on
369: different central squares are linearly independent and form a
370: nonorthogonal basis in the subspace of topologically nontrivial graphs
371: of Fig.~\ref{defect}b. The 2-LM defect state (\ref{2defect})
372: can be identified with a new classical particle, which has energy
373: $2\varepsilon_0$ and a longer-range repulsion.
374: Such a mapping suggests that the topologically nontrivial states
375: yield an additional macroscopic contribution to the partition
376: function $\cal Z$. Their share is, however,
377: suppressed compared to the hard-polygon contribution by a large
378: entropic factor: region
379: occupied by the basis two-magnon state, Fig.~\ref{defect}f,
380: can be occupied by 16 different 2-LM states residing
381: on small empty squares. In addition, by counting the total number of excluded
382: square voids we conclude that
383: the states (\ref{2defect}) do not contribute appreciably to
384: the magnetization subsectors with the average density of LMs
385: larger than $n\agt 2/(25\cdot 2)=0.04$. \cite{hard2}
386: Consequently, their role at high densities, {\it e.g.}, in the
387: vicinity of the transition into a magnon crystal state is strongly
388: suppressed. Alternatively, the lattice gas mapping can be improved
389: by including extra types of particles,
390: which describe LMs from different topological classes.
391: Below we consider the simplest version of the lattice gas mapping
392: with only one type of particles.
393:
394:
395: \subsection{Effective lattice-gas model}
396:
397:
398: Two-dimensional lattice gas models were originally suggested
399: to study atomic adsorption on various substrates.
400: \cite{runnels,ree,bellemans,domany,binder80,kinzel81,schick,kaski,baxter98}
401: They also describe low-$T$ behavior of Ising antiferromagnets
402: in a longitudinal field.\cite{binder80,metcalf,racz}
403: Despite the long-lasting interest there are only few
404: well-established facts for a square lattice
405: gas with the nearest- and the next-nearest-neighbor exclusion.
406: It is generally accepted that this model exhibits a continuous phase
407: transition into a partially ordered ($2\times 1$)
408: phase. \cite{ree,kinzel81}
409: This phase transition belongs to the
410: universality class of an $XY$ model with a four-fold
411: anisotropy, \cite{domany} which is a special case of
412: a more general $Z(4)$ discrete planar model. \cite{elitzur,rujan,wu}
413: Depending on the values of two coupling constants
414: the $Z(4)$ model exhibits either two Ising-like transitions
415: between a paramagnetic and an ordered state or a single
416: critical point,\cite{rujan} which has nonuniversal
417: critical exponents.\cite{jose}
418: Estimates for the transition point of a hard-square lattice
419: gas with the second-neighbor repulsion vary
420: from $\ln z_c = (\mu/T)_c=5.3$ in the early work\cite{ree}
421: to $\ln z_c = 4.7$ in the later study.\cite{kinzel81}
422: There is even less certainty about values of the critical exponents.
423: The entropy of the system at $\mu=0$, which is an interesting
424: property of a quantum antiferromagnet, has not been
425: investigated in the context of previous applications.
426:
427:
428: First, we investigate numerically the entropy of
429: the lattice gas model at zero chemical potential.
430: This can be done by the transfer matrix method and
431: Monte Carlo simulations. The former method
432: has been a standard technique since the early studies
433: of two-dimensional lattice gas models.\cite{ree,metcalf,kinzel81}
434: In this scheme a two dimensional lattice
435: is represented by a semi-infinite strip of width $M$
436: and the thermodynamic properties are
437: derived from the largest eigenvalue $\lambda_1$ of the transfer matrix.
438: In particular, the entropy normalized per one site is equal to
439: ${\cal S} = (\ln\lambda_1)/M$.
440: The actual calculations become quite simple for $\mu=0$,
441: when the elements of the transfer matrix take only values
442: 0 or 1.
443: The results for a few values of $M$ are presented in table I.
444: The convergence is very rapid and already the $M=10$ strip gives
445: four significant digits for the entropy.
446:
447: \begin{table}[t]
448: \caption{The entropy per site $\cal S$ and the density $n$ of
449: a hard-square lattice gas with second-neighbor exclusion
450: at $\mu=0$ obtained from the transfer matrix calculation
451: on a semi-infinite strip $M\times\infty$.
452: }
453: \vspace*{2mm}
454: \begin{ruledtabular}
455: \begin{tabular}{rcc}
456: $M$ & $\displaystyle {\cal S} = \frac{\ln\lambda_1}{M}$
457: & $n$
458: \\[3mm]
459: \hline
460: $8$ & $0.294795$ & $0.13713$ \\
461: $10$ & $0.294671$ & $0.13686$ \\
462: $12$ & $0.294647$ & $0.13679$ \\
463: $14$ & $0.294642$ & $0.13677$
464: \end{tabular}
465: \end{ruledtabular}
466: \end{table}
467:
468: To determine entropy from Monte Carlo (MC) simulations
469: we adopt the following procedure.
470: The standard Metropolis algorithm is used for gradual annealing
471: from the low-activity regime $\ln z=-20$,
472: where the density of particles is vanishingly small and ${\cal S}=0$,
473: to the point $\ln z=0$. The step for the chemical potential
474: is chosen to be small enough $\Delta(\ln z)=0.05$. At every value
475: of $\mu$ $10^4$ MC steps (lattice sweeps) are performed for equilibration and
476: after that $10^5$ MC steps are used to measure
477: $(\partial{\cal S}/\partial\mu)_T$, which is calculated
478: from the cumulant of the energy $E$ and the number of particles $\cal N$:
479: \begin{equation}
480: \left(\frac{\partial {\cal S}}{\partial \mu}\right)_T =
481: \frac{1}{T^2}(\langle E{\cal N}\rangle - \langle E\rangle \langle {\cal N}
482: \rangle) \ .
483: \label{dSdm}
484: \end{equation}
485: In our case $E=-\mu {\cal N}$ and the cumulant in Eq.~(\ref{dSdm})
486: is proportional to the variance of the total number of particles.
487: The error bars are estimated by performing up
488: to 100 independent runs with different sequences of random numbers.
489: Afterwords, the data for $(\partial{\cal S}/\partial\mu)_T$ are numerically
490: integrated to find ${\cal S}|_{\mu=0}$. Simulations have been
491: performed on square clusters with periodic boundary conditions
492: and $N=L^2$ sites, $L=16$, 32, and 64. The obtained results agree
493: with each other within numerical accuracy and yield ${\cal S}=0.2946(1)$
494: for the entropy per site and $n=0.13676(1)$
495: for the average density of particles. The found values are in good
496: correspondence with the transfer-matrix results included in table I
497: confirming the
498: accuracy of both methods. The particle density at $\mu=0$ is
499: quite substantial and only two times smaller than the density
500: of the close-packed structure $n_0=0.25$.
501:
502: \begin{figure}[t]
503: \begin{center}
504: \includegraphics[width=0.8\columnwidth]{crystal_checker.eps}
505: \end{center}
506: \caption{(color online). Effective lattice gas model for
507: a checkerboard antiferromagnet. The dual square lattice is
508: $45^\circ$ rotated compared to the original checkerboard lattice
509: in Fig.~2 and has twice less sites. Hard particles are shown by
510: large light circles; small dark circles indicate the excluded sites
511: around one particle.
512: The presented close-packed structure with half-filled rows randomly
513: shifted in the horizontal direction has a nonzero Fourier harmonic
514: at ${\bf Q}_1=(0,\pi)$.
515: }
516: \label{magnonC}
517: \end{figure}
518:
519: Once the chemical potential further increases ($z>1$) more and more
520: particles become condensed. Eventually the lattice gas transforms
521: into an ordered close-packed structure. The order parameter of
522: such a crystalline state are certain
523: Fourier harmonic(s) of the particle density,
524: $n_{\bf q}=(1/N) \sum_i n_i e^{-i{\bf q}{\bf r}_i}$.
525: A close-packed structure with half-filled rows, which are randomly
526: shifted in the $x$-direction has a nonzero harmonic at
527: ${\bf Q}_1=(0,\pi)$ with $n_{{\bf Q}_1}=1/4$,
528: see Fig.~\ref{magnonC}.
529: A similar state with random shifts of columns along the $y$ direction
530: is described by ${\bf Q}_2=(\pi,0)$ and $n_{{\bf Q}_2}=1/4$.
531: These two phases are called $(2\times 1)$ states in a standard
532: nomenclature adopted for lattice gas models.\cite{schick}
533: A fully symmetric structure with all rows (columns)
534: in phase with each other has equal amplitudes $n_{{\bf Q}_i}=1/4$
535: for the three wave-vectors ${\bf Q}_1$, ${\bf Q}_2$, and
536: ${\bf Q}_3=(\pi,\pi)$ and is called a $(2\times 2)$ state.
537: The wave-vectors ${\bf Q}_1$ and ${\bf Q}_2$ belong to
538: the same irreducible representation. Consequently, we consider
539: two order parameters:
540: \begin{equation}
541: M_1 = (n_{{\bf Q}_1}^2 + n_{{\bf Q}_2}^2)^{1/2} \ , \ \ \
542: M_2 = |n_{{\bf Q}_3}| \ .
543: \label{op12}
544: \end{equation}
545:
546: The infinitely degenerate close-packed structure for a
547: square-lattice gas with the second-neighbor exclusion
548: leads to equilibration problems in simple
549: MC simulations. This was evident since the early MC study
550: of a frustrated Ising antiferromagnet, \cite{binder80}
551: which investigated clusters up to only $40^2$ sites
552: using a single spin-flip relaxation method.
553: We employ instead an exchange MC algorithm proposed some time ago
554: to tackle systems with extremely long relaxation times, as, for example,
555: spin glasses.\cite{exchange}
556: With this modification we have been able to study clusters with up
557: to $120^2$ sites.
558: To ensure a substantial replica exchange rate $\sim0.7$--0.9,
559: we have used
560: 120 replicas in the range $0\leq\ln z\leq 8$
561: for all system sizes.
562: The simulation runs included $10^5$ exchange MC steps for equilibration and
563: up to $10^7$ MC steps for measurements.
564: Statistical errors were estimated from bining the MC series
565: for each value of $z$.
566:
567: \begin{figure}
568: \begin{center}
569: \includegraphics[width=0.9\columnwidth]{op.eps}
570: \end{center}
571: \caption{(color online). Dependence of the two order parameters
572: Eq.~(\ref{op12})
573: on logarithm of the fugacity $\ln z=\mu/T$ for three system sizes.
574: }
575: \label{op}
576: \end{figure}
577:
578: The specific heat exhibits a broad and rounded maximum near to
579: $\ln z\sim 4.4$.
580: The height of the peak grows very slowly with increasing $L$,
581: such that a preliminary determination of the transition point
582: is impossible from the specific heat data.
583: The ensemble averages of the two order parameters
584: $\langle M_1\rangle$ and $\langle M_2\rangle$ are shown in
585: Fig.~\ref{op}.
586: The square of the second order parameter
587: $\langle M_2^2\rangle$ goes down to zero as
588: $1/N$ with the system size, whereas $\langle M_1\rangle$
589: scales to a finite constant at $\ln z \geq 5$. The limiting value
590: for the largest cluster is very close $1/4$ in accordance
591: with the previous analysis for the $(2\times 1)$ type of ordering.
592: The precise location of the transition point is obtained
593: by measuring the Binder cumulant
594: $U_4=\langle M_1^4\rangle/\langle M_1^2\rangle^2$.
595: Its dependence on $z$ is presented in Fig.~\ref{binder}.
596: The transition point can be estimated from the crossing points of
597: Binder cumulants for different clusters. \cite{binder81}
598: This yields $\ln z_c = 4.56\pm 0.02$ for the critical activity.
599: The density of particles at the transition point is $n_c=0.2325(3)$.
600:
601: In principle, the measurement of $M_1 \neq 0$ cannot discriminate
602: between a single-$k$ structure corresponding to the $(2\times 1)$
603: order and a double-$k$ structure with two nonzero Fourier
604: harmonics $n_{{\bf Q}_1},n_{{\bf Q}_2}\neq 0$.
605: The two structures can be distinguished with the help of
606: an order parameter, which probes breaking of the $C_4$
607: rotational symmetry:
608: \begin{equation}
609: \Delta = \frac{1}{N} \sum_i n_i(n_{i+2{\bf x}}-n_{i+2{\bf y}}) \ .
610: \end{equation}
611: The order parameter $\Delta$ vanishes for the $(2\times 2)$ state and
612: all other states symmetric under $90^\circ$ rotations, whereas it has
613: a finite value in the $(2\times 1)$ phase. We have calculated
614: $\langle|\Delta(z)|\rangle$ by MC simulations and found that
615: at $\ln z>5$ within statistical errors it reaches a finite value:
616: $\langle|\Delta|\rangle=1/8$. This not only demonstrates an absence
617: of the tetragonal symmetry in the ordered
618: state, but also proves {\it random} shifts of half-filled rows (columns).
619: For example, if shifts occur in a regular alternating
620: order, than one would find $\langle|\Delta|\rangle=1/4$,
621: which is definitely excluded by our Monte Carlo results.
622:
623: \begin{figure}[t]
624: \begin{center}
625: \includegraphics[width=0.85\columnwidth]{binder.eps}
626: \end{center}
627: \caption{(color online). Binder cumulant $U_4$ as a function
628: of $\ln z=\mu/T$ for different cluster sizes $L$. The inset shows
629: scaling of the derivative $dU_4/d(\ln z)$ at the transition point
630: $\ln z_c =4.56$ versus $L$.
631: }
632: \label{binder}
633: \end{figure}
634:
635: In order to determine the critical exponents, we performed
636: finite-size scaling analysis of various thermodynamic
637: quantities measured at the estimated critical point $\ln z_c$.
638: The correlation length exponent $\nu$ is extracted from
639: the behavior of the derivative of the Binder cumulant
640: $dU_4/d(\ln z) \sim L^{1/\nu}$ at $z=z_c$. The fitting shown in the inset
641: of Fig.~\ref{binder} yields $1/\nu = 1.16\pm 0.02$. The order parameter
642: at criticality scales as $M(z_c)\sim L^{-\beta/\mu}$.
643: A fit for a few largest $L$ gives $\beta/\nu=0.15(1)$.
644: As a result, we obtain the following estimates $\nu=0.86(2)$
645: and $\beta=0.13(1)$. The largest clusters employed
646: in the present study are still not sufficient
647: to independently extract the critical exponent $\alpha$,
648: presumably because of a large regular contribution
649: to the specific heat compared to a universal singular part.
650: The obtained values for $\beta$ and $\nu$ place a hard-square gas
651: with the second-neighbor exclusion on a line of critical
652: points of the $Z(4)$ model between the Potts model, which has
653: $\beta=1/12$ and $\nu=2/3$,\cite{wu} and the vector
654: Potts model, which belongs to the Ising universality class
655: with $\beta=1/8$ and $\nu=1$. \cite{elitzur}
656: Our result for $\ln z_c$ coincides within the error bars
657: with the corresponding value from the recent independent
658: MC study of the same model, \cite{fernandes} though we do
659: not share its claim of the Ising universality class
660: for the transition.
661:
662: Properties of a quantum spin-1/2 checkerboard antiferromagnet
663: in the vicinity of the saturation field are obtained from the above
664: results by a straightforward rescaling
665: to the number of sites of checkerboard lattice, which is twice
666: larger than the number of sites for the lattice gas.
667: At $H=H_s$, the entropy and the magnetization normalized per one site
668: have universal $T$-independent values:
669: \begin{equation}
670: {\cal S}_s = 0.1473\ , \ \ \ \ M_s = 0.4316 \ .
671: \label{Scheck}
672: \end{equation}
673: Note, that the entropy of a checkerboard antiferromagnet is 30\%
674: larger than the corresponding result for a kagom\'e antiferromagnet.
675: \cite{hard1} The transition field into the magnon crystal phase
676: is given by
677: \begin{equation}
678: H_c(T) = H_s - T\ln z_c = 4J - 4.56T \ .
679: \label{Hccheck}
680: \end{equation}
681: Let us emphasize again that the entropy found within the
682: lattice-gas (hard-polygon) description of LMs is only
683: approximate. Localized magnons from different topological
684: classes will increase the above value of ${\cal S}_s$.
685: Such corrections should not be very large in view of a large
686: size of basis defect states (\ref{defect}),
687: though it would be interesting to estimate
688: $\Delta{\cal S}_s$ from an improved lattice gas mapping,
689: as discussed in the end of Sec.~IIb,
690: or from the exact-diagonalization data.
691: Propagating $n$-magnon states can be separated by rather
692: small gaps from $n$-LM states. Therefore,
693: Eq.~(\ref{Hccheck}) describes the slope
694: of the actual transition line $H_c(T)$ in
695: the frustrated quantum antiferromagnet at $T\rightarrow 0$.
696:
697:
698:
699: \section{Pyrochlore Antiferromagnet}
700:
701: \subsection{Localized magnons}
702:
703: \begin{figure}[t]
704: \begin{center}
705: \includegraphics[width=0.9\columnwidth]{locmag_pyro.eps}
706: \end{center}
707: \caption{(color online). Cross-section of a pyrochlore
708: lattice perpendicular to the $[111]$ axis.
709: Tetrahedra from the same kagom\'e layer are drawn by solid lines.
710: Thick line shows a localized magnon on one of
711: the hexagons in the kagom\'e plane. Dashed lines indicate
712: tetrahedra from the upper kagom\'e layer (shown only partially).
713: Dots with numbers denote centers of hexagons in horizontal and
714: tilted planes, which form a dual pyrochlore lattice.
715: \label{locmag_pyro}}
716: \end{figure}
717:
718: The primitive unit cell of a pyrochlore lattice contains
719: 4 spins belonging to one tetrahedron. These tetrahedra
720: are arranged into an fcc lattice, which is formed by
721: the elementary translations on ${\bf a}_1=(0,1/2,1/2)$,
722: ${\bf a}_2=(1/2,0,1/2)$, and ${\bf a}_3=(1/2,1/2,0)$.
723: (Linear size of the standard cubic cell with 16 spins is chosen
724: as the unit of length. \cite{kennedy})
725: In a strong magnetic field, where the saturated phase
726: is stable, the four one-magnon excitation branches
727: are given by
728: \begin{eqnarray}
729: \omega_{1,2}=H - 8JS,\ \omega_{3,4} = H -
730: 4JS \pm 2JS \sqrt{1\! +\! \eta_{\bf k}}, && \nonumber \\
731: \eta_{\bf k}\!=\cos\frac{k_x}{2}\cos\frac{k_y}{2}\!
732: +\!\cos\frac{k_y}{2}\cos\frac{k_z}{2}\!+\!\cos\frac{k_z}{2}\cos\frac{k_x}{2}.
733: &&
734: \label{wpyro}
735: \end{eqnarray}
736: The saturation field of an antiferromagnet on a pyrochlore lattice
737: is the same as for a checkerboard lattice: $H_s=8JS$.
738: Two dispersionless branches $\omega_{1,2}$ contain
739: $2\times(N/4-1)=N/2-2$ linearly independent localized modes,
740: $N$ being the number of pyrochlore lattice sites.
741: Geometric interpretation of these modes is essentially the same
742: as for two-dimensional frustrated lattices.
743: The smallest LMs are located on hexagon voids of kagom\'e layers,
744: which alternate with triangular layers along the $[111]$ and
745: three other equivalent directions, see Fig.~\ref{locmag_pyro}.
746: An $N$-site pyrochlore cluster contains $N$ hexagons, which
747: for $N \rightarrow \infty$ is twice more
748: than the number of localized modes in the two lowest branches
749: (\ref{wpyro}). The `elementary' hexagon modes are, therefore, not
750: only nonorthogonal but also form an overcomplete basis.
751: This can be explained by observing that hexagon loops on
752: a pyrochlore lattice obey exactly $N/2$ linear relations,
753: which leaves only $N/2$ linearly independent states. \cite{mzh03}
754: Despite this fact we shall use below the hexagon states
755: as a basis in a mapping of a pyrochlore antiferromagnet
756: near $H_s$ on a lattice gas model. Such a representation
757: gives an incorrect (larger) number of states in the magnetization
758: subsectors with nonmacroscopic numbers of LMs.
759: The mapping becomes not so bad for the multi-magnon subsectors
760: in the vicinity of close-packed structures, when the allowed
761: configurations do not include linearly dependent states.
762: In particular, using the lattice gas representation we shall be
763: able to describe an {\it exact} structure of the magnon crystal
764: for a pyrochlore antiferromagnet.
765:
766: The lowest-energy multi-particle states are constructed using
767: the same prescription as for two-dimensional frustrated models:
768: LMs are represented by closed even-sites graphs, which are not
769: directly connected by the exchange bonds.
770: For example, if a LM occupies a hexagon void
771: denoted by 1 in Fig.~\ref{locmag_pyro}, then no LMs can be placed
772: on hexagons denoted by numbers 2, 3 and 4, since they either share
773: sites with the hexagon 1 or are connected to it by the nearest-neighbor
774: bonds. The lattice gas model is constructed on a dual lattice
775: formed by centers of hexagons of the original pyrochlore lattice,
776: such that an exclusion principle for neighboring sites reproduces
777: the above rules for LMs.
778: Remarkably, the dual lattice is again a pyrochlore lattice.
779: (This explains, in particular, why the number of hexagons
780: is equal to the number of sites.)
781: If the nearest-neighbor distance is denoted by $d$
782: ($d\equiv\sqrt{2}/4$), then presence of a particle on a given site
783: excludes for occupation (i) 6 nearest-neighbor sites at the distance $d$ as,
784: for example, a pair of sites (hexagons) 2 and 4 in Fig.~\ref{locmag_pyro},
785: (ii) 12 second-neighbor sites at the distance $\sqrt{3}d$,
786: {\it e.g.}, a pair of sites 1 and 2, (iii) 6 third-neighbor sites at
787: the distance $2d$, {\it e.g.}, sites 1 and 3, and (iv) 12 fourth-neighbor
788: sites at the distance $\sqrt{5}d$, {\it e.g.}, sites 1 and 4.
789: Note, that on a pyrochlore lattice there are two types of third-neighbor
790: sites at the distance $2d$. The first type used above corresponds to
791: pairs on opposite vertices of hexagons. The second type of third
792: neighbors correspond to pairs with a third site in between.
793: There are again 6 third-neighbor sites of this type
794: around a given site. In terms of the original lattice
795: they correspond to hexagon pairs (2,5) and (2,6) in
796: Fig.~\ref{locmag_pyro}, which allow simultaneous occupation by LMs.
797: These hexagons belong to two different parallel kagom\'e planes.
798:
799: Let us now consider possible magnon crystal states obtained
800: by a close-packing of the smallest LMs on hexagon voids. The
801: initial estimate for the maximal density of LMs has been obtained from
802: the known result for a single kagom\'e plane: \cite{schulenburg}
803: filling one-third of hexagons in every kagom\'e plane, which is
804: perpendicular to one of the four cubic diagonals,
805: gives $n_0 = (1/3)\times(1/4) = 1/12$
806: for the density. No arguments were given yet that this is indeed
807: the maximal possible density.
808: In terms of the effective lattice gas model, which operates
809: on a dual pyrochlore lattice, a close-packed structure is constructed
810: by filling every second site in one of the six chains
811: formed by six edges of one tetrahedron, see Fig.~\ref{filling}.
812: Particles at the distance $2d$ along the chains correspond to the second
813: type of third-neighbors and are allowed by the exclusion principle.
814: All chains parallel to a chosen direction form a triangular
815: lattice in the perpendicular plane. Half-filling of the nearest-neighbor
816: chains is prohibited by the exclusion principle,
817: whereas second-neighbor chains in the triangular plane
818: allow for simultaneous occupation.
819: In this way it is possible to put particles only on one-third of
820: the parallel chains, which again yields
821: $n_0= (1/2)\times(1/3)\times(1/2)=1/12$ for the particle
822: density. (The last factor 1/2 corresponds to the fact that
823: parallel chains contain only a half of all lattice sites, the
824: other half belongs to the perpendicular chains.)
825: An example of the constructed structure is shown in Fig.~\ref{filling}.
826: Particles in every chain can be independently shifted
827: by half a period, which corresponds to the degeneracy $N_{\rm deg}\sim 2^{L^2}$.
828: Once all chains are in phase, {\it i.e.}, all particles occupy same vertex
829: of the unit cell tetrahedron, we recover the initial
830: structure constructed by filling hexagons in parallel kagom\'e planes.
831:
832: \begin{figure}[t]
833: \begin{center}
834: \includegraphics[width=0.95\columnwidth]{pyro_fill.eps}
835: \end{center}
836: \caption{(color online). Close-packed structure of particles
837: (large light circles) on a dual pyrochlore lattice.
838: Thick (thin) lines represent tetrahedra from the lower (upper)
839: kagom\'e layer. Small dark circles denote vertices
840: of one truncated tetrahedron.
841: }
842: \label{filling}
843: \end{figure}
844:
845: To prove that the estimate $n_0=1/12$ gives the maximal possible
846: density of localized magnons we consider a polyhedron formed by 4 hexagons
847: and 4 triangles, which has 12 vertices marked by small circles in
848: Fig.~\ref{filling}. This polyhedron is called truncated tetrahedron
849: and belongs to the thirteen Archimedean solids.
850: A special role of this polyhedron for our problem
851: is determined by its geometrical
852: structure: distance between
853: two arbitrary vertices of a truncated tetrahedron is equal to $d$,
854: $\sqrt{3}d$, $2d$, or $\sqrt{5}d$. If one vertex of a truncated
855: tetrahedron is occupied by a particle, then, according to the above
856: exclusion rules all other vertices must be empty.
857: Simple consideration yields that an $N$-site pyrochlore
858: lattice contains $N/2$ truncated tetrahedra such that every
859: lattice site is shared between six of them. Counting
860: now becomes straightforward: the number of hard-core particles
861: in the effective lattice-gas model cannot exceed one particle
862: per six truncated tetrahedra $(1/6)\times(N/2)$, which yields
863: $1/12$ as the upper bound on $n$. Above, we have constructed
864: an explicit example of the particle arrangement with the density $n_0=1/12$,
865: therefore, this is indeed the highest density of the close-packed
866: structure.
867:
868:
869: \subsection{Effective lattice-gas model}
870:
871: Finite-temperature properties of the effective lattice-gas model
872: have been studied by Monte Carlo simulations. In the low-density
873: disordered regime $z<1$ we use the same annealing protocol
874: as in the 2D case and investigate cubic clusters with periodic
875: boundary conditions and $N=4L^3$ sites, $L=6$--12. At $\mu=0$
876: ($z=1$) the entropy obtained by numerical integration of MC data and
877: the density of particles are equal to
878: \begin{equation}
879: {\cal S}/N=0.1329(1)\ , \ \ \ n=0.04718(1)\ ,
880: \label{Spyro}
881: \end{equation}
882: which yields $M=1/2-n\approx 0.4528$ for the magnetization.
883:
884:
885: \begin{figure}[b]
886: \begin{center}
887: \includegraphics[width=0.9\columnwidth]{density.eps}
888: \end{center}
889: \caption{(color online). Density of a lattice gas
890: on a pyrochlore lattice as a function of $\ln z=\mu/T$.
891: Arrow indicates the density of a close-packed structure
892: $n_0=1/12$.
893: Inset: probability distribution for the particle density
894: obtained for $N=1728$ cluster at $\ln z=5.2$. }
895: \label{density}
896: \end{figure}
897:
898: In order to study the high-density regime $z>1$ we employ
899: the exchange MC algorithm with 80 replicas distributed in the
900: range $0<\ln z<7$. In contrast with the 2D case
901: we still find significant equilibration problem for large lattices,
902: which exhibit extremely long relaxation times $t \agt 10^7$ MC steps
903: despite of a high exchange rate $\sim 0.8$ between replicas.
904: Sufficient statistics has been obtained only for three system sizes:
905: two cubic clusters with $L=4,6$ and a tetragonal prism with
906: $6\times 6\times 12$ unit cells (1728 sites). They were equilibrated
907: for $5\times 10^5$ exchange MC steps and further
908: $10^7$ MC steps were used for measurements.
909: Evolution of the particle density with increasing $z$ is presented in
910: Fig.~\ref{density} for the two biggest systems.
911: In the vicinity of $\ln z \approx 5.5$ there is an abrupt jump of $n$
912: to the value, which is very close to the density of the close-packed
913: structure $n_0=1/12$.
914:
915:
916: We expect that the ordered state at $\ln z\agt 5.5$ exhibits only a
917: partial breaking of the translational symmetry.
918: Similar to the studied 2D model, the half-filled
919: chains remain uncorrelated. This tendency should be even more pronounced
920: for the pyrochlore lattice gas in view of an increased entropic effect
921: from disordering determined by a larger $N_{\rm deg}$.
922: As a result, the symmetry breaking
923: includes only (i) selection of a chain direction and (ii)
924: formation of $\sqrt{3}\times\sqrt{3}$ structure on
925: a triangular lattice in the perpendicular plane.
926: In view of small available system sizes we have not performed
927: investigation of the corresponding order parameters.
928: Instead we have tried to clarify
929: the nature of a possible phase transition into a close-packed
930: structure by collecting histograms for the particle density.
931: Results for $N=1728$ lattice at $\ln z=5.2$ obtained
932: with $10^7$ MC steps are presented in the inset
933: of Fig.~\ref{density}.
934: A clear double peak structure with approximately equal weights
935: in each peak points at a first-order transition.
936: This should be, of course, confirmed by a similar study of larger clusters.
937: We, therefore, put a tentative estimate $\ln z_c=5.2\pm 0.2$
938: for the first-order transition into the magnon crystal state for
939: a pyrochlore antiferromagnet.
940:
941: Since the original and the dual pyrochlore lattices have the same
942: number of sites, the properties of a spin-1/2
943: pyrochlore antiferromagnet are obtained from the above results
944: without any rescaling. In particular, the estimate for the
945: entropy at the saturation field is given by Eq.~(\ref{Spyro}).
946: At zero temperature and $H=H_s$ the magnetization jumps from
947: $M=5/12$ in the plateau region to the full saturation with $M=1/2$.
948: At finite temperatures the jump in the magnetization is much smaller
949: $\Delta M \approx 0.015$, see Fig.~\ref{density}.
950: Afterward, $M(H)$ increases smoothly and asymptotically
951: saturates. The first-order transition into the magnon crystal state
952: takes place at
953: \begin{equation}
954: H_c(T) = H_s -T\ln z_c\approx 4J - 5.2 T
955: \end{equation}
956: for a spin-1/2 model.
957:
958:
959: \section{Effect of extra exchanges}
960:
961: In the previous two sections, we have found that the magnon crystal states
962: in checkerboard and pyrochlore antiferromagnets exhibit massive degeneracy
963: at $T=0$. The degeneracy is not lifted
964: at finite temperatures by fluctuations within the LM subspace. This leads
965: to interesting types of broken symmetries in 2D and 3D cases. In principle,
966: fluctuations to higher energy states outside the LM ensemble at $T>0$ can favor one
967: or the other type of close-packed magnon structures. Such a scenario
968: cannot be excluded on general grounds. The higher energy
969: states are, however, separated by finite gaps from LMs. Therefore,
970: there must be a region near $T\rightarrow 0$, $H\rightarrow H_s-0$, where
971: such a selection is ineffective and only a partial breaking
972: of translational symmetry takes place. Another mechanism of degeneracy
973: lifting can be provided by the magnetoelastic coupling, \cite{richter04}
974: though the corresponding study has considered only one possible
975: pattern of LMs for each of the two lattices.
976: In the following we investigate the third possibility: effect of further
977: neighbor exchanges, \cite{reimers} which play significant role in
978: some magnetic materials with pyrochlore lattices.
979: \cite{crawford,tsune,cepas,motome,wills}
980:
981: We begin with the checkerboard antiferromagnet, considering an additional
982: weak diagonal exchange of strength $J'$, see Fig.~\ref{lattices}b.
983: The magnon crystal state becomes completely unstable at a certain critical
984: value of $J'$. \cite{hard2}
985: We are interested in the subcritical regime $|J'|\rightarrow 0$.
986: Let us consider two localized magnons (\ref{localC}) on adjacent square
987: plaquettes along the $[11]$ direction:
988: $|\varphi_i\rangle$ and $|\varphi_{i+2{\bf a}_1}\rangle$, see
989: Fig.~\ref{checker} and Sec.~IIa for the notations. These LMs are connected by
990: an additional exchange bond
991: $\hat{V}=J'{\bf S}_{2,i}\cdot{\bf S}_{4,i+2{\bf a}_1}$.
992: To measure energy shifts relative to the energy
993: of the fully polarized state, we shall always subtract
994: a constant from the exchange bond operators: $\hat{V} \rightarrow \hat{V} - J'S^2$.
995: The potential energy $U$ of two LMs on adjacent voids is
996: found by calculating the
997: expectation value of $\hat{V}$ over the two-magnon state
998: $|\psi\rangle=|\varphi_i\varphi_{i+2{\bf a}_1}\rangle$
999: and correcting it by the self-energy of LMs without neighbors:
1000: \begin{equation}
1001: U = \langle\psi|\hat{V}|\psi\rangle -
1002: 2 \langle\varphi_i|\hat{V}|\varphi_i\rangle
1003: = J'/16 \ .
1004: \label{Echecker}
1005: \end{equation}
1006: This result applies to an arbitrary spin value $S$.
1007: For $J'>0$ presence of LMs on adjacent square voids
1008: is energetically unfavorable, because
1009: two spin flips with a finite probability
1010: occupy the same $J'$ bond. Degeneracy of the magnon crystal is,
1011: consequently, lifted in favor of an orthorhombic structure:
1012: half-filled rows of LMs alternate in phase in order
1013: to minimize the contribution (\ref{Echecker}) between the rows.
1014: For $J'<0$ localized magnons attract each other
1015: and the $(2\times 2)$ tetragonal close-packed structure is stabilized.
1016:
1017: The interaction energy between two LMs residing on hexagon voids of
1018: pyrochlore lattice is obtained in the same way as Eq.~(\ref{Echecker})
1019: with the result $U=J'/36$. For a pyrochlore antiferromagnet
1020: three additional exchanges may be present:\cite{reimers,wills}
1021: the second-neighbor exchange $J_2$
1022: for spin pairs at distance
1023: $\sqrt{3}d$, the third-neighbor exchange $J_{31}$ for spin
1024: pairs at distance $2d$ on opposite vertices of hexagons,
1025: and the third-neighbor exchange $J_{32}$ for spin pairs
1026: at distance $2d$ along chains, see Fig.~\ref{lattices}a.
1027: (Details about the crystal structure of pyrochlore oxides
1028: can be found, {\it e.g.}, in Ref.~\onlinecite{kennedy}.)
1029: Similar to the 2D case the weak additional exchanges
1030: select relative shift between adjacent half-filled chains of LMs.
1031: If two chains are not shifted relative to each other,
1032: LMs occupy hexagons in parallel kagom\'e planes,
1033: whereas if the shift is present, LMs form a nonplanar structure.
1034: Analysis of further neighbor exchange links between hexagons
1035: on a pyrochlore lattice is straightforward but cumbersome.
1036: Below we only summarize the conclusions.
1037: Once the two chains of LMs form a planar structure,
1038: one LM from the first chain is coupled to two
1039: hexagons in the second chain: to a hexagon in the same kagom\'e plane
1040: by two $J_2$ and two $J_{32}$ bonds and to a hexagon
1041: in a parallel kagom\'e plane by a $J_{31}$ bond.
1042: Then, the interaction energy between chains
1043: normalized per one LM of the first chain is
1044: $U_1 = (2J_2+J_{31}+2J_{32})/36$.
1045: In the nonplanar structure
1046: a given LM from the first chain is coupled to only one
1047: non-planar hexagon by three $J_2$ bonds and one $J_{32}$ bond,
1048: which yields $U_2=(3J_2+J_{32})/36$.
1049:
1050: Comparing the interaction energies $U_1$ and $U_2$ for the two
1051: structures, we conclude that for $(J_{31}+J_{32}-J_2)<0$
1052: localized magnons occupy hexagons in parallel kagom\'e planes. Between
1053: the planes LMs follow the ABCABC... structure
1054: of close-packed hard spheres.
1055: If the above expression changes sign, then LMs form a more
1056: complicated nonplanar structure. Amazingly enough, it is again
1057: frustrated and degenerate: half-filled chains form a triangular lattice,
1058: therefore, antiphase shifts would correspond to an effective Ising
1059: antiferromagnet on a triangular lattice.
1060: The residual degeneracy should be finally lifted by zero-point fluctuations
1061: induced by the next-neighbor exchanges. The corresponding energy
1062: scale is of the order of $J'^2/J$.
1063:
1064:
1065: \section{Summary}
1066:
1067: Knowledge of the exact ground state and the single-particle excitation
1068: spectrum of frustrated quantum antiferromagnets above the saturation
1069: field $H_s$ allows to develop a quantitative description
1070: of their thermodynamic properties in a finite range of
1071: fields $H\alt H_s$ and temperatures $T\rightarrow 0$.
1072: Localized nature of the lowest energy excitations leads to a mapping
1073: onto lattice gas models of hard-core classical particles with
1074: an appropriate exclusion principle.
1075: As a result, frustrated quantum antiferromagnets preserve
1076: a macroscopic degeneracy (finite entropy) at $H=H_s$ and $T=0$.
1077: The close-packed structure of
1078: particles (localized magnons) corresponds to the magnon crystal state,
1079: which breaks only translational symmetry and preserves
1080: continuous rotations about the field direction.
1081:
1082: The main finding of our study is high degeneracy of
1083: the magnon crystal states in pyrochlore and checkerboard
1084: antiferromagnets at zero temperature.
1085: This, however, does not exclude presence of finite-temperature
1086: phase transitions.
1087: The corresponding lattice gas models have been studied by the
1088: exchange Monte Carlo method.
1089: Specifically, we have found numerically the location
1090: $\ln z_c = 4.56(2)$ and two critical exponents $\beta = 0.13(1)$
1091: and $\nu=0.86(2)$ for a square-lattice gas with the second-neighbor
1092: exclusion, which describes a checkerboard antiferromagnet.
1093: Further numerical work is needed to clarify small differences
1094: with the recent independent MC study of the same model.\cite{fernandes}
1095: Monte Carlo simulations for the pyrochlore-lattice gas show
1096: a sharp jump in the particle density, which suggests a first-order
1097: transition into a partially ordered crystal phase.
1098:
1099:
1100:
1101: \acknowledgments
1102:
1103: We are grateful to L. Balents and A. Honecker for useful discussions.
1104: We also thank to H.-J. Schmidt for a helpful remark
1105: on topological classes of localized magnons.
1106: This work was partly supported by a Grant-in-Aid
1107: for Scientific Research on Priority Areas (No.~17071011)
1108: and Scientific Research (No.~16540313), and also by
1109: the Next Generation Super Computing Project, Nanoscience
1110: Program, from the Ministry of Education, Culture,
1111: Sports, Science and Technology of Japan.
1112:
1113:
1114:
1115: \begin{thebibliography}{99}
1116:
1117: \bibitem{villain}
1118: J. Villain, R. Bidaux, J. P. Carton, and R. J. Conte,
1119: J. de Phys. {\bf 41}, 1263 (1980).
1120:
1121: \bibitem{shender}
1122: E. F. Shender,
1123: Sov. Phys. JETP {\bf 56}, 178 (1982).
1124:
1125: \bibitem{henley89}
1126: C. L. Henley,
1127: Phys. Rev. Lett. {\bf 62} 2056 (1989).
1128:
1129: \bibitem{RK}
1130: D. S. Rokhsar and S. A. Kivelson,
1131: Phys. Rev. Lett. {\bf 61}, 2376 (1988).
1132:
1133: \bibitem{cabra}
1134: D. C. Cabra, M. D. Grynberg, P. C. W. Holdsworth, A. Honecker,
1135: P. Pujol, J. Richter, D. Schmalfuss, and J. Schulenburg,
1136: Phys. Rev. B {\bf 71}, 144420 (2005)
1137:
1138: \bibitem{mzh05}
1139: M. E. Zhitomirsky,
1140: Phys. Rev. B {\bf 71}, 214413 (2005).
1141:
1142: \bibitem{bergman}
1143: D. L. Bergman, R. Shindou, G. A. Fiete, and L. Balents,
1144: Phys. Rev. Lett. {\bf 96}, 097207 (2006).
1145:
1146: \bibitem{hizi}
1147: U. Hizi and C. L. Henley,
1148: Phys. Rev. B {\bf 73}, 054403 (2006).
1149:
1150: \bibitem{hard1}
1151: M. E. Zhitomirsky and H. Tsunetsugu,
1152: Phys. Rev. B {\bf 70}, 100403(R) (2004).
1153:
1154: \bibitem{hard2}
1155: M. E. Zhitomirsky and H. Tsunetsugu,
1156: Prog. Theor. Phys. Suppl. {\bf 160}, 361 (2005).
1157:
1158: \bibitem{derzhko04}
1159: O. Derzhko and J. Richter,
1160: Phys. Rev. B {\bf 70}, 104415 (2004).
1161:
1162: \bibitem{richter06}
1163: J. Richter, O. Derzhko, T. Krokhmalskii,
1164: Phys. Rev. B {\bf 74}, 144430 (2006).
1165:
1166:
1167: \bibitem{baxter}
1168: R. J. Baxter, {\it Exactly Solved Models in Statistical Mechanics}
1169: (Academic Press, London, 1982).
1170:
1171: \bibitem{schnack}
1172: J. Schnack, H.-J. Schmidt, J. Richter, and J. Schulenburg,
1173: Eur. Phys. J. B {\bf 24}, 475 (2001).
1174:
1175: \bibitem{schulenburg}
1176: J. Schulenburg, A. Honecker, J. Schnack, J. Richter, and H.-J. Schmidt,
1177: Phys. Rev. Lett. {\bf 88}, 167207 (2002).
1178:
1179: \bibitem{schmidt}
1180: H.-J. Schmidt,
1181: J. Phys. A {\bf 35}, 6545 (2002).
1182:
1183: \bibitem{mzh03}
1184: M. E. Zhitomirsky,
1185: Phys. Rev. B {\bf 67}, 104421 (2003).
1186:
1187: \bibitem{richter}
1188: J. Richter, J. Schulenburg, A. Honecker, J. Schnack, and H.-J. Schmidt,
1189: J. Phys.:\ Condens.\ Matter {\bf 16}, S779 (2004).
1190:
1191: \bibitem{schmidt06}
1192: H.-J. Schmidt, J. Richter, and R. Moessner,
1193: J. Phys. A {\bf 39}, 10673 (2006).
1194:
1195: \bibitem{runnels}
1196: L. K. Runnels,
1197: Phys. Rev. Lett. {\bf 15}, 581 (1965).
1198:
1199: \bibitem{ree}
1200: F. H. Ree and D. A. Chestnut,
1201: Phys. Rev. Lett. {\bf 18}, 5 (1967).
1202:
1203: \bibitem{bellemans}
1204: A. Bellemans and R. K. Nigam,
1205: J. Chem. Phys. {\bf 46}, 2922 (1967).
1206:
1207: \bibitem{domany}
1208: E. Domany and E. K. Riedel,
1209: Phys. Rev. Lett. {\bf 40}, 561 (1978).
1210:
1211: \bibitem{binder80}
1212: K. Binder and D. P. Landau,
1213: Phys. Rev. B {\bf 21}, 1941 (1980).
1214:
1215: \bibitem{kinzel81}
1216: W. Kinzel and M. Schick,
1217: Phys. Rev. B {\bf 24}, 324 (1981).
1218:
1219: \bibitem{schick}
1220: M. Schick,
1221: Physica B {\bf 109\&110}, 1811 (1982).
1222:
1223: \bibitem{kaski}
1224: K. Kaski, W. Kinzel and J. D. Gunton,
1225: Phys. Rev. B {\bf 27}, 6777 (1983).
1226:
1227: \bibitem{baxter98}
1228: R. J. Baxter,
1229: Ann. Comb. {\bf 3}, 191 (1999).
1230:
1231: \bibitem{metcalf}
1232: B. D. Metcalf and C. P. Yang, Phys. Rev. B {\bf 18}, 2304 (1978).
1233:
1234: \bibitem{racz}
1235: Z. R\`acz,
1236: Phys. Rev. B {\bf 21}, 4012 (1980).
1237:
1238: \bibitem{elitzur}
1239: S. Elitzur, R. B. Pearson, and J. Shigemitsu,
1240: Phys. Rev. D {\bf 19}, 3698 (1979).
1241:
1242: \bibitem{rujan}
1243: P. Ruj\'an, G. O. Williams, H. L. Frisch, and G. Forg\'acs,
1244: Phys. Rev. B {\bf 23}, 1362 (1981).
1245:
1246: \bibitem{wu}
1247: F. Y. Wu, Rev. Mod. Phys. {\bf 54}, 235 (1982).
1248:
1249: \bibitem{jose}
1250: J. V. Jos\'e, L. P. Kadanoff, S. Kirkpatrick, and D. R. Nelson,
1251: Phys. Rev. B {\bf 16}, 1217 (1977).
1252:
1253: \bibitem{exchange}
1254: K. Hukushima and K. Nemoto,
1255: J. Phys. Soc. Jpn. {\bf 65}, 1604 (1996).
1256:
1257: \bibitem{binder81}
1258: K. Binder,
1259: Z. Phys. B {\bf 43}, 119 (1981).
1260:
1261: \bibitem{fernandes}
1262: H. C. M. Fernandes, J. J. Arenzon, and Y. Levin,
1263: {\tt arXiv:cond-mat/0612372v1}.
1264:
1265: \bibitem{kennedy}
1266: B. J. Kennedy, B. A. Hunter, and C. J. Howard,
1267: J. Solid State Chem. {\bf 130}, 58 (1997).
1268:
1269: \bibitem{richter04}
1270: J. Richter, O. Derzhko, and J. Schulenburg,
1271: Phys. Rev. Lett. {\bf 93}, 107206 (2004);
1272: O. Derzhko and J. Richter,
1273: Phys. Rev. B {\bf 72}, 094437 (2005).
1274:
1275: \bibitem{reimers}
1276: J. N. Reimers, A. J. Berlinsky, and A.-C. Shi,
1277: Phys. Rev. B {\bf 43}, 865 (1991).
1278:
1279: \bibitem{crawford}
1280: M. K. Crawford, R. L. Harlow, P. L. Lee, Y. Zhang, J. Hormadaly, R. Flippen,
1281: Q. Huang, J. W. Lynn, R. Stevens, B. F. Woodfield, J. Boerio-Goates, and R. A. Fisher,
1282: Phys. Rev. B {\bf 68}, 220408(R) (2003).
1283:
1284: \bibitem{tsune}
1285: H. Tsunetsugu and Y. Motome,
1286: Phys. Rev. B {\bf 68}, 060405(R) (2003).
1287:
1288: \bibitem{cepas}
1289: O. Cepas and B. S. Shastry,
1290: Phys. Rev. B {\bf 69}, 184402 (2004).
1291:
1292: \bibitem{motome}
1293: Y. Motome and H. Tsunetsugu,
1294: Phys. Rev. B {\bf 70}, 184427 (2004);
1295: Prog. Theor. Phys. Suppl. {\bf 160}, 203 (2005).
1296:
1297: \bibitem{wills}
1298: A. S. Wills, M. E. Zhitomirsky, B. Canals, J. P. Sanchez,
1299: P. Bonville, P. Dalmas de Reotier, and A. Yaouanc,
1300: J. Phys.: Condens. Matter {\bf 18}, L37 (2006).
1301:
1302:
1303: \end{thebibliography}
1304:
1305: \end{document}
1306:
1307: