1: \documentclass[prb,aps,twocolumn,floatfix,showpacs]{revtex4}
2: \usepackage{bm}
3: \usepackage{amssymb}
4: \usepackage{graphicx}
5:
6: \def\horparallel{
7: \begin{picture}(8,7)
8: \thicklines
9: \put(1,-0.5){\line(1,0){6}}
10: \put(1,5.5){\line(1,0){6}}
11: \end{picture}}
12:
13: \def\vertparallel{
14: \begin{picture}(8,7)
15: \thicklines
16: \put(1,-0.8){\line(0,1){7}}
17: \put(6.5,-0.8){\line(0,1){7}}
18: \end{picture}}
19:
20: \begin{document}
21:
22: \title{Effective quantum dimer model for trimerized kagom\'e antiferromagnet}
23:
24: \author{M. E. Zhitomirsky}
25:
26: \affiliation{
27: Commissariat \`a l'Energie Atomique, DSM/DRFMC/SPSMS, 38054 Grenoble,
28: France}
29: \date{\today}
30:
31:
32: \begin{abstract}
33: An effective spin-orbit Hamiltonian is derived for a
34: spin-1/2 trimerized kagom\'e antiferromagnet in the
35: second-order of perturbation theory in the ratio of two coupling
36: constants. Low-energy singlet states of the obtained model
37: are mapped to a quantum dimer model on a
38: triangular lattice. The quantum dimer model is
39: dominated by dimer resonances
40: on a few shortest loops of the triangular lattice.
41: Characteristic energy scale for the dimer model constitutes only a small
42: fraction of the weaker exchange coupling constant.
43: \end{abstract}
44: \pacs{75.10.Jm, % Quantized spin models
45: 75.50.Ee % Antiferromagnetics
46: }
47:
48: \maketitle
49:
50: \section{Introduction}
51:
52: Resonating valence bond (RVB) state \cite{anderson}
53: is nowadays a popular
54: paradigm in condensed matter physics. Short-range RVB states are
55: considered to be probable candidates for an elusive spin-liquid
56: phase of magnetic insulators. The idea of short-range RVB states is
57: quantitatively formulated by so called quantum dimer models (QDM).
58: \cite{sutherland,rokhsar,moessner00}
59: In a QDM each dimer represents a singlet state (valence bond)
60: between a pair of neighboring spins.
61: The QD Hamiltonian is defined in the Hilbert space
62: of close-packed dimer coverings of a lattice. The dimer states
63: are assumed
64: to be properly orthogonalized. Local dynamics of an RVB state
65: is typically described on the smallest
66: plaquettes ($\square$), which are squares for a square lattice
67: or rhombi for a triangular lattice:
68: \begin{equation}
69: \hat{\cal H}_{\rm QD}=\sum_\square \Bigr[-t \bigl(
70: |\horparallel\rangle\langle \vertparallel|+
71: |\vertparallel\rangle\langle\horparallel|\bigr)
72: + V \bigl(
73: \left| \horparallel \right\rangle\left\langle \horparallel \right|+
74: \left| \vertparallel \right\rangle\left\langle \vertparallel \right| \bigr)
75: \Bigr].
76: \label{Hqdm}
77: \end{equation}
78: The first term is a dimer kinetic energy, which flips
79: a pair of parallel dimers around an arbitrary plaquette;
80: the second term is a potential
81: energy between such pairs.
82:
83: Rokhsar and Kivelson \cite{rokhsar} have shown that a short-range
84: RVB state given
85: by a superposition of all dimer coverings of
86: a square lattice is an exact eigenstate of the QD Hamiltonian
87: for a special choice of the parameters $t=V$.
88: On a bipartite square lattice, the RVB state
89: at the Rokhsar-Kivelson (RK) point has long-range power low correlations
90: and describes, consequently, a gapless spin-liquid state.
91: \cite{rokhsar,moessner01} Small
92: perturbations away from the RK point drive the system into one
93: of the ordered crystalline dimer states.
94: The QDM on a triangular lattice exhibits quite a different
95: behavior at the RK point. \cite{moessner01,ioselevich}
96: The short-range RVB state has exponentially decaying correlators and
97: is fully gapped. It exists, therefore, in a finite range
98: of parameters around the RK point
99: and is stable with respect to weak perturbations
100: to the QD Hamiltonian.
101: Still, question whether such states or Hamiltonians can
102: describe realistic quantum spin systems
103: remains unsettled.
104: In the present work we propose
105: a realization of QDM on a triangular lattice for a nearest-neighbor
106: Heisenberg spin model.
107:
108: The most probable candidates for a singlet spin-liquid ground state
109: are frustrated quantum antiferromagnets. \cite{book}
110: Numerical exact diagonalization studies of a spin-1/2 Heisenberg
111: kagom\'e antiferromagnet have shown that this spin model has
112: a nonmagnetic ground state with a large number of low-lying
113: singlet excitations. \cite{lecheminant,waldtmann}
114: Accessible cluster sizes do not allow to draw
115: a definite conclusion on the possible nature of the magnetically disordered
116: (singlet) ground state.
117:
118: \begin{figure}
119: \centerline{
120: \includegraphics[width=0.9\columnwidth]{trimerized.eps}}
121: \caption{Trimerized kagom\'e lattice with two exchange
122: constants: $J_1$ in $\bigtriangleup$-triangles and
123: $J_2$ in $\bigtriangledown$-triangles.}
124: \label{trimerized}
125: \end{figure}
126:
127: One of a very few analytic approaches to such strongly-correlated
128: spin systems is an
129: expansion from small clusters. The main motif
130: of a kagom\'e lattice is triangle. It is, therefore, natural to
131: start from a trimerized kagom\'e lattice shown in Fig.~\ref{trimerized}.
132: Such a strong-coupling approach has been pursued in relation
133: to kagom\'e antiferromagnet in several theoretical works.
134: \cite{subrah,mila,raghu,auerbach}
135: Recently, an experimental scheme to create a trimerized kagom\'e lattice
136: has been suggested for ultracold atomic gases in optical traps.
137: \cite{santos} This opens a way for an experimental
138: probe of RVB physics in the corresponding spin model.
139:
140: The Heisenberg model on a trimerized lattice
141: \begin{equation}
142: \hat{\cal H} = \sum_{\langle ij\rangle} J_{ij} {\bf S}_i\cdot{\bf S}_j
143: \label{Hheis}
144: \end{equation}
145: has two coupling constants:
146: $J_1$ for a stronger interaction between spins in
147: $\bigtriangleup$-triangles and $J_2$ for a weaker interaction
148: inside $\bigtriangledown$-triangles.
149: An array of isolated $\bigtriangleup$-blocks is a zeroth order
150: Hamiltonian, which has a highly degenerate ground state.
151: Interblock interaction lifts
152: such a degeneracy. In section II we derive an effective
153: Hamiltonian up to the second-order in a small parameter $J=J_2/J_1\ll 1$.
154: This Hamiltonian is mapped to a QDM in section III.
155: The obtained results and their implication for the ground state
156: properties of the trimerized kagom\'e model are discussed in
157: section IV.
158:
159: \section{Strong-coupling expansion}
160:
161: \subsection{First-order Hamiltonian}
162:
163: Let us in the beginning rederive the previous results
164: on the effective first-order Hamiltonian \cite{subrah,mila,raghu}
165: using somewhat different notations. Below
166: we normalize all energies to $J_1$ such that $J_2\rightarrow J$.
167: In a strong-coupling
168: expansion one
169: starts with an isolated triangle described by
170: \begin{eqnarray}
171: \hat{\cal H}_\bigtriangleup & = & {\bf S}_1\cdot{\bf S}_2
172: + {\bf S}_2\cdot{\bf S}_3 +
173: {\bf S}_3\cdot{\bf S}_1 \nonumber \\
174: & = & \frac{1}{2}\bigl({\bf S}_1 +{\bf S}_2+{\bf S}_3\bigr)^2
175: - \frac{3}{2} S(S+1) \ .
176: \end{eqnarray}
177: The energy levels of $\hat{\cal H}_\bigtriangleup$ are determined by
178: the total spin $S_{\rm tot}$. For on-site $S=1/2$,
179: which is always assumed below, the levels are
180: two degenerate doublets with $S_{\rm tot}=1/2$ and $E=-\frac{3}{4}$ and
181: one quartet with $S_{\rm tot}=3/2$ and $E=\frac{3}{4}$.
182: The doublet states with $S^z_{\rm tot}=1/2$ are
183: \begin{equation}
184: |d_\uparrow\rangle=
185: \frac{1}{\sqrt{2}}\bigl(\uparrow\uparrow\downarrow\!-\!
186: \uparrow\downarrow\uparrow \bigr),\ \
187: |p_\uparrow\rangle=\frac{1}{\sqrt{6}}\bigl(2\!\downarrow\uparrow\uparrow
188: \!-\!\uparrow\uparrow\downarrow\!-\!
189: \uparrow\downarrow\uparrow \bigr),
190: \label{doublet}
191: \end{equation}
192: where spin numbering in an individual triangle
193: follows Fig.~\ref{strong}a. The former state
194: $|d_\uparrow\rangle$ is a combination of the spin-up apex spin and
195: a singlet bond between the two base spins and has odd parity under
196: the permutation $\hat{P}_{23}$. The other state $|p_\uparrow\rangle$
197: is even under $\hat{P}_{23}$.
198: The two members of a quartet with $S_{\rm tot}^z=+3$
199: and $+1$ are
200: \begin{equation}
201: |q_{+3}\rangle=
202: |\!\uparrow\uparrow\uparrow\rangle \ , \ \ \
203: |q_{+1}\rangle=\frac{1}{\sqrt{3}}\bigl(
204: \downarrow\uparrow\uparrow +
205: \uparrow\downarrow\uparrow +
206: \uparrow\uparrow\downarrow \bigr)\ .
207: \label{quartet}
208: \end{equation}
209: All other states are obtained from (\ref{doublet}) and (\ref{quartet})
210: by applying
211: $S^-_{\rm tot}$ operator.
212:
213: The choice of the basis in the doublet subspace is not unique.
214: The apex spin can be put into
215: a singlet state either with its right or left
216: neighbor. The two alternative basis states obtained
217: by rotating $|d_\uparrow\rangle$
218: about a center of triangle counterclockwise
219: on $2\pi/3$ and $4\pi/3$ are
220: \begin{equation}
221: |d'_\uparrow\rangle=
222: \frac{1}{\sqrt{2}}\bigl(\downarrow\uparrow\uparrow-\uparrow\uparrow\downarrow
223: \bigr) \ , \ \
224: |d''_\uparrow\rangle=
225: \frac{1}{\sqrt{2}}\bigl(\uparrow\downarrow\uparrow-\downarrow\uparrow\uparrow\bigr)
226: \end{equation}
227: with their orthogonal partners
228: $|p'_\uparrow\rangle$ and $|p''_\uparrow\rangle$.
229: Transformation from the old
230: basis (\ref{doublet}) to the new states is
231: \begin{eqnarray}
232: |d_{\alpha}'\rangle =
233: -\frac{1}{2}|d_{\alpha}\rangle+\frac{\sqrt{3}}{2}|p_{\alpha}\rangle,\
234: |p_{\alpha}'\rangle =
235: -\frac{\sqrt{3}}{2}|d_{\alpha}\rangle-\frac{1}{2}|p_{\alpha}\rangle, &&
236: \nonumber \\
237: |d_{\alpha}''\rangle =
238: -\frac{1}{2}|d_{\alpha}\rangle-\frac{\sqrt{3}}{2}|p_{\alpha}\rangle,\ \ \
239: |p_{\alpha}''\rangle =
240: \frac{\sqrt{3}}{2}|d_{\alpha}\rangle-\frac{1}{2}|p_{\alpha}\rangle, &&
241: \label{doublet2}
242: \end{eqnarray}
243: where $\alpha=\uparrow,\downarrow$ or 1,2 is a spinor index.
244: The main difference with the previous works \cite{subrah,mila}
245: is that real basis states (\ref{doublet}) or (\ref{doublet2})
246: are used instead of complex chiral states. This yields a more
247: transparent form of the effective Hamiltonian and simplifies
248: subsequent derivation of a QDM.
249:
250: \begin{figure}
251: \centerline{
252: \includegraphics[width=0.9\columnwidth]{strong.eps}}
253: \caption{Three different geometries of $\bigtriangleup$-blocks
254: contributing to the second-order energy correction
255: in the interblock coupling.
256: The labeling of axes and sites inside triangles is shown in
257: the upper panel (a). }
258: \label{strong}
259: \end{figure}
260:
261: At this point we introduce two sets of the Pauli matrices:
262: $\sigma^i$, which act between spin-up and
263: spin-down states, and $\tau^i$, which act in the orbital subspace
264: $(d,p)$ preserving the total spin. A convenient
265: choice of orbital axes shown in Fig.~\ref{strong}a
266: corresponds to
267: \begin{equation}
268: \tau^{z_1} |d_{\alpha}\rangle = |d_{\alpha}\rangle \ ,\ \ \
269: \tau^{z_1} |p_{\alpha}\rangle =-|p_{\alpha}\rangle \ .
270: \end{equation}
271: Then, the orbital operators projected onto the rotated axes
272: \begin{equation}
273: \tau^{z_2}=-\frac{1}{2}\tau^{z_1}
274: -\frac{\sqrt{3}}{2}\tau^{x_1}, \ \
275: \tau^{z_3} = -\frac{1}{2}\tau^{z_1}+\frac{\sqrt{3}}{2}\tau^{x_1}
276: \end{equation}
277: have the following eigenstates:
278: \begin{eqnarray}
279: && \tau^{z_2}|d_{\alpha}'\rangle=|d_{\alpha}'\rangle \ , \ \ \
280: \tau^{z_2}|p_{\alpha}'\rangle=-|p_{\alpha}'\rangle \ ,
281: \nonumber \\
282: && \tau^{z_3}|d_{\alpha}''\rangle=|d_{\alpha}''\rangle \ , \ \ \
283: \tau^{z_2}|p_{\alpha}''\rangle=-|p_{\alpha}''\rangle \ .
284: \end{eqnarray}
285:
286: In order to find the effect of interblock coupling
287: in the first order of perturbation theory in $J=J_2/J_1$
288: one should neglect $S_{\rm tot}=3/2$ states separated by a finite gap
289: $\Delta E = \frac{3}{2}$ and calculate matrix elements
290: of the on-site spin operators between the low-energy doublet states
291: $|d_{\alpha}\rangle$ and $|p_{\alpha}\rangle$. This problem is greatly
292: simplified once all symmetries are taken into account.
293: Introducing operators $d^\dagger_\alpha |0\rangle =
294: |d_\alpha\rangle$ and $p^\dagger_\alpha |0\rangle=|p_\alpha\rangle$,
295: where $|0\rangle$ is a fictitious vacuum, the Hubbard-type representation
296: of on-site spins is written as
297: \begin{eqnarray}
298: && {\bf S}_1 = \frac{1}{2}\,d^\dagger_{\alpha}
299: \mbox{\boldmath$\sigma$}_{\alpha\beta}d_{\beta} -
300: \frac{1}{6}\,p^\dagger_{\alpha}\mbox{\boldmath$\sigma$}_{\alpha\beta}p_{\beta} \ ,
301: \label{rep_spinor} \\
302: &&
303: {\bf S}_{2,3} = \frac{1}{3}\,p^\dagger_{\alpha}
304: \mbox{\boldmath$\sigma$}_{\alpha\beta} p_{\beta}
305: \pm\frac{1}{2\sqrt{3}}
306: \left(p^\dagger_{\alpha}\mbox{\boldmath$\sigma$}_{\alpha\beta}d_{\beta}+{\rm h.\,c.}\right).
307: \nonumber
308: \end{eqnarray}
309: Spinor structure is a consequence of the spin-rotation symmetry,
310: while permutation of the base spins
311: $\hat{P}_{23}$ fixes the orbital part in (\ref{rep_spinor}).
312: The above representation is further simplified once the total spin
313: of a triangle ${\bf S}=\frac{1}{2}\mbox{\boldmath $\sigma$}$ is defined
314: and the orbital operators $\tau^{z_k}$ are used. Then,
315: the $n^{\rm th}$ spin ($n=1$--$3$) of the $i^{\rm th}$
316: triangular block is represented by
317: \begin{equation}
318: {\bf S}_{ni} = \frac{1}{3}\,{\bf S}_i\left( 1+2\tau^{z_n}_i\right),
319: \label{rep_orbit}
320: \end{equation}
321: where $\hat{\bf z}_n$ goes from the center of a triangle in
322: the direction of the corresponding spin, see Fig.~\ref{strong}a.
323:
324: The effective first-order Hamiltonian in the interblock coupling
325: is found by substituting Eq.~(\ref{rep_orbit})
326: into the Hamiltonian (\ref{Hheis}):
327: \begin{equation}
328: \hat{\cal H}_1 = \frac{J}{9} \sum_{\langle ij\rangle}
329: {\bf S}_i\cdot{\bf S}_j(1+2\tau_i^{z_{n}})
330: (1+2\tau_j^{z_{m}}) \ ,
331: \label{H1st}
332: \end{equation}
333: where a trivial constant term $-\frac{3}{4}N_\bigtriangleup$
334: is omitted for convenience.
335: The derived spin-orbital Hamiltonian $\hat{\cal H}_1$ is defined on
336: a triangular lattice, such that every site corresponds to
337: one $\bigtriangleup$-block of the trimerized kagom\'e model
338: and is attributed with spin and orbital operators.
339: The bond orientation uniquely determines the orbital
340: axes for two participating sites.
341:
342: \subsection{Second-order Hamiltonian}
343:
344: The second-order energy corrections
345: for weakly-coupled spin triangles have been obtained by
346: Raghu and co-workers.\cite{raghu} These authors have mostly been
347: interested in a one-dimensional model, therefore, their analysis misses
348: several terms relevant for a two-dimensional array of triangles
349: in the trimerized kagom\'e model. In order to calculate
350: the second-order result in $J_2$ one needs to determine matrix
351: elements of on-site spins between doublet (\ref{doublet})
352: and quartet (\ref{quartet}) states.
353: Introducing symmetric third-rank spinor tensor $q_{\alpha\beta\gamma}$
354: such that $q_{111}=|q_{+3}\rangle$, $q_{112}=\frac{1}{\sqrt{3}}|q_{+1}\rangle$
355: $q_{122}=\frac{1}{\sqrt{3}}|q_{-1}\rangle$, and $q_{222}=|q_{-3}\rangle$,
356: and utilizing spin-rotation
357: symmetry we find by analogy with Eq.~(\ref{rep_spinor})
358: \begin{eqnarray}
359: && {\bf S}_1 = -\frac{i}{\sqrt{6}}\,q^\dagger_{\alpha\beta\gamma}
360: \bigl(\mbox{\boldmath$\sigma$}\sigma^y\bigr)_{\beta\gamma}p_{\alpha}
361: + {\rm h.\,c.}\ ,
362: \label{rep_spinor2} \\
363: &&
364: {\bf S}_{2,3} = \frac{i}{\sqrt{6}}\,q^\dagger_{\alpha\beta\gamma}
365: \bigl(\mbox{\boldmath$\sigma$}\sigma^y\bigr)_{\beta\gamma}
366: \Bigl(\pm\frac{\sqrt{3}}{2}d_\alpha + \frac{1}{2}p_{\alpha}\Bigr)
367: + {\rm h.\,c.}
368: \nonumber
369: \end{eqnarray}
370:
371: The second-order energy correction in the interblock
372: coupling is, generally, given by
373: \begin{equation}
374: \hat{\cal H}_2(G,G') = \sum_X \frac{\langle G|\hat{\cal H}|X\rangle
375: \langle X|\hat{\cal H}|G'\rangle}{E_G-E_X} \ ,
376: \label{2nd}
377: \end{equation}
378: where $G$ and $G'$ denote combinations of lowest doublet
379: states on $\bigtriangleup$-blocks and $X$ are excited states.
380: The nonzero second-order terms appear if either
381: (i) one $J_2$-bond acts twice in the numerator of Eq.~(\ref{2nd})
382: or (ii) two adjacent $J_2$-bond emerging from the same
383: $\bigtriangleup$-block are used subsequently in the
384: matrix elements $\langle G|\hat{\cal H}|X\rangle$
385: and $\langle X|\hat{\cal H}|G'\rangle$. This determines
386: three different geometries for two- and three-block
387: interaction terms shown in Fig.~\ref{strong}.
388: In the first case of two-block interaction, Fig.~\ref{strong}a,
389: either one or both
390: triangular blocks have quartets in the intermediate states $X$.
391: For the three block interactions (Fig.~\ref{strong}b,c),
392: only a middle block has excited quartets in the intermediate states.
393: Every pair of free (uncoupled) spins in a $\bigtriangleup$-block imposes
394: an extra permutation symmetry on the second-order Hamiltonian
395: $\hat{\cal H}_2(G,G')$.
396: For example, the two-block cluster in Fig.~\ref{strong}a has extra
397: $\hat{P}_{12}$ symmetry for the left $i$-block
398: and $\hat{P}_{13}$ symmetry for the right $k$-block.
399: Therefore, the orbital states, $|d''_\alpha\rangle$
400: or $|p''_\alpha\rangle$ for the left block and
401: $|d'_\alpha\rangle$
402: or $|p'_\alpha\rangle$ for the right block, remain unchanged
403: during the second-order perturbation process (\ref{2nd}).
404: In other words $\hat{\cal H}_{2a}$ commutes with $\tau_i^{z_3}$
405: and $\tau_k^{z_2}$.
406: The conservation of orbital state is also fulfilled for all
407: triangles in the three-block term in Fig.~\ref{strong}b
408: and for the side triangles in Fig.~\ref{strong}c, whereas the middle
409: block does change its orbital state during the second-order
410: process. The above conservation laws significantly simplify summation
411: over intermediate states in Eq.~(\ref{2nd}). The final results are
412: \begin{eqnarray}
413: \hat{\cal H}_{2a} & = & -\frac{J^2}{54} \sum_{\langle ik\rangle}
414: \Bigl[(3+4\,{\bf S}_i\cdot{\bf S}_k)(1-\tau_i^{z_l}\tau_k^{z_m})
415: \nonumber \\ & & \mbox{}\ \ \ \ \ \ \ \ \ \ +
416: (1-\tau_i^{z_{n}})(1-\tau_k^{z_{m}})\Bigr]
417: \label{H2a}
418: \end{eqnarray}
419: for the two-block interaction;
420: \begin{equation}
421: \hat{\cal H}_{2b}= -\frac{4J^2}{243}\!\sum_{\langle ijk\rangle}
422: {\bf S}_i\cdot{\bf S}_k(1+2\tau_i^{z_l})(1+2\tau_k^{z_m})
423: (1-\tau_j^{z_n})
424: \label{H2b}
425: \end{equation}
426: for the three-block interaction shown in Fig.~\ref{strong}b
427: with corresponding labeling of blocks;
428: and
429: \begin{eqnarray}
430: \hat{\cal H}_{2c} & = & \frac{2J^2}{243}\sum_{\langle ijk\rangle}
431: (1+2\tau_i^{z_n})(1+2\tau_k^{z_n})\Bigl[{\bf S}_i\cdot{\bf S}_k
432: (1+2\tau_j^{z_n})
433: \nonumber \\ & & \mbox{}\ \ \ \ \ \ \ \ \ \ \ + \sqrt{3}\,\tau_j^y\:
434: {\bf S}_i\cdot({\bf S}_j\times{\bf S}_k)\Bigr]
435: \label{H2c}
436: \end{eqnarray}
437: for the three-block interaction in geometry of Fig.~\ref{strong}c.
438: Polarization of orbital operators $\tau_i^{z_n}$ in
439: the above equations is again found by simple inspection
440: of the arrangement of corresponding blocks
441: on a trimerized kagom\'e lattice. Three-block terms in
442: $\hat{\cal H}_{2b}$ exist on $\bigtriangledown$-plaquettes
443: of an effective triangular lattice with three different
444: $z$-axes in Eq.~(\ref{H2b}). Three-body terms in $\hat{\cal H}_{2c}$
445: appear on $\bigtriangleup$-plaquettes of the triangular
446: lattice with one polarization of $\tau^z$ operators
447: in Eq.~(\ref{H2c}), which changes under permutation of $(ijk)$.
448:
449: The interactions $\hat{\cal H}_{2a}$ and $\hat{\cal H}_{2c}$ coincide
450: up to a trivial change of notations with the previously derived terms
451: for a one-dimensional array of triangles,\cite{raghu} while the term
452: $\hat{\cal H}_{2b}$ is a novel one.
453: Note, that $\hat{\cal H}_{2c}$ contains a remarkable three-body spin-chiral
454: interaction term. By deriving the effective spin-orbital Hamiltonians
455: $\hat{\cal H}_1$ and $\hat{\cal H}_2$,
456: we have substantially restricted the Hilbert space and
457: simplified the original problem of finding the ground and the lowest
458: energy states of the spin model (\ref{Hheis}). The remaining problem
459: of solving $\hat{\cal H}_1+\hat{\cal H}_2$
460: is still highly nontrivial.
461: The spin-orbital Hamiltonian $\hat{\cal H}_1$ has been studied so far in
462: the mean-field approach. \cite{mila}
463: An effective Hamiltonian derived by the contractor renormalization group,
464: which partially resembles $\hat{\cal H}_1+\hat{\cal H}_2$,
465: has also been analyzed in the mean-field approximation. \cite{auerbach}
466: Below we discuss a mapping of the derived Hamiltonians (\ref{H1st})
467: and (\ref{H2a})--(\ref{H2c}) to an effective QDM.
468: The obtained QD Hamiltonian is dominated by the kinetic
469: energy for dimer tunneling. The mean-field approximation, which assumes
470: a frozen pattern
471: of dimers, is, therefore, a poor approximation in the present problem.
472:
473: \section{Quantum dimer model}
474:
475: \subsection{General remarks}
476:
477: Search for the low-energy states of the first-order Hamiltonian (\ref{H1st})
478: can be started by considering first a two-site problem (two adjacent
479: $\bigtriangleup$-blocks of the original kagom\'e lattice). \cite{mila}
480: This problem is solved exactly and its ground state
481: corresponds to a spin singlet with
482: the orbital degrees fully polarized along the bond:
483: $\langle \tau_i^{z_n}\rangle =\langle \tau_j^{z_m}\rangle =1$.
484: The ground-state energy is $-\frac{3}{4}J$.
485: A variational singlet ground state for the lattice problem (\ref{H1st})
486: is constructed by splitting the whole lattice into a close-packed
487: structure of dimers between nearest-neighbor sites, such that the dimer
488: wave-function is given by the ground-state solution of the two-site problem.
489: A remarkable feature of these variational states is that
490: at the mean-field level with respect to orbital degrees
491: of freedom the total energy is just a sum of energies of individual
492: dimers and does not depend on a particular dimer covering of
493: a triangular lattice. \cite{mila}
494: Indeed, once $\langle\tau_i^{z_n}\rangle=1$,
495: then for the two other axes $\langle\tau_i^{z_m}\rangle
496: =\langle \tau_i^{z_k}\rangle \equiv -\frac{1}{2}$.
497: Therefore, the expectation value of any empty
498: bond, {\it i.e.\/}, a bond without a dimer, identically vanishes
499: over the variational wave-function: either one or both sites
500: of the bond have $\langle 1+2\tau_i^{z_m}\rangle\equiv 0$.
501:
502: The degenerate set of variational mean-field states has
503: been identified with low energy states of spin-1/2 antiferromagnets
504: on trimerized and isotropic kagom\'e lattices.
505: \cite{mila,mambrini} The number of low-lying singlets of
506: the kagom\'e model scales, then, as $1.15^N$ in good agreement
507: with the full exact diagonalization study. \cite{waldtmann}
508: The previous works leave, however, without answer
509: question about validity of the mean-field approximation
510: and further lifting of degeneracy by quantum fluctuations.
511:
512: \begin{figure}
513: \centerline{
514: \includegraphics[width=0.9\columnwidth]{triangular.eps}}
515: \caption{Effective triangular lattice with five
516: shortest loops. The arrow directions indicate the sign convention
517: for singlet wave-functions on each bond}
518: \label{triangular}
519: \end{figure}
520:
521: In order to beyond the mean-field approximation, one has to consider
522: off-diagonal matrix elements of the Hamiltonian (\ref{H1st}) between
523: various dimer configurations as well as the corresponding overlap
524: matrix. The general rule to compute the overlap matrix for models,
525: where every dimer represents a singlet pair, is to construct
526: transition or overlap graph by drawing two dimer configurations on the
527: same lattice. \cite{sutherland} Every closed nonintersecting
528: loop of dimers contributes $2/2^{l/2}$
529: to the overlap matrix, $l$ being the length of the loop.
530: The sign of the overlap matrix element depends on a sign convention
531: for singlet wave-functions. We adopt the standard convention
532: \cite{sutherland,moessner01,moessner02}
533: such that the singlet bond wave-function is
534: $[ab]=\frac{1}{\sqrt{2}}(\uparrow_a\downarrow_b-\downarrow_a\uparrow_b)$,
535: where $b$ is an upper site in the pair
536: or is directly to the right from $a$, see Fig.~\ref{triangular}.
537:
538: Local dynamics of singlet bonds
539: in trimerized kagom\'e model is determined by
540: a few shortest loops on an
541: effective triangular lattice, which include two- and three-dimer
542: moves, see Fig.~\ref{triangular}.
543: Taking into account the orbital part of the wave-functions
544: the overlaps of two dimer configurations on each loop are
545: calculated as $c_1=-1/2^4$, $c_2=c_3=-1/2^7$, $c_4=-1/2^8$, and
546: $c_5=1/2^5$.
547: These overlap matrix elements are significantly smaller than for singlet
548: bond configurations on the original triangular lattice. In the latter case
549: the corresponding loops have $c_1=1/2$,
550: $c_2=c_3=1/2^2$, $c_4=c_5=-1/2^2$.
551: The difference reflects the fact that loops on an effective
552: triangular lattice correspond to significantly longer loops
553: on the original trimerized kagom\'e lattice.
554: For example, the shortest $C_1$ loop corresponds to a loop
555: of length $l=10$
556: on a kagom\'e lattice.
557: Loops $C_4$ and $C_5$ are different for the
558: considered model because kagom\'e lattice has only
559: a three-fold rotation axis in the center of every triangle.
560: Significant difference of the overlap matrix elements explains
561: why a QDM description is a poor
562: approximation for a spin-1/2 Heisenberg antiferromagnet on a triangular
563: lattice,\cite{anderson} but may be a good one for the
564: trimerized kagom\'e antiferromagnet.
565: Below in this section we assume that the ground states of
566: the first and second-order effective Hamiltonians
567: are given variationally by close-packed dimer states
568: and compute a new effective QD Hamiltonian.
569: The above assumption is supported by numerical
570: treatment of the trimerized kagom\'e antiferromagnet.\cite{mambrini}
571:
572: Derivation of a QDM from a particular spin Hamiltonian
573: has been formulated via calculation of the
574: inverse square root of the overlap matrix. \cite{rokhsar}
575: We find that actual calculations become more transparent
576: by operating with the wave-functions. The final result
577: are, of course, equivalent in both approaches.
578: Specifically, let us consider two linearly independent, normalized
579: states $|\psi_1\rangle$ and $|\psi_2\rangle$, which have a small
580: overlap $\langle\psi_1|\psi_2\rangle=\langle\psi_2|\psi_1\rangle=c$.
581: Matrix elements of the Hamiltonian between the
582: two states are assumed to be known
583: $E_{11}=\langle\psi_1|\hat{H}|\psi_1\rangle$,
584: $E_{21}=E_{12}=\langle\psi_1|\hat{H}|\psi_2\rangle$,
585: and $E_{22}=\langle\psi_2|\hat{H}|\psi_2\rangle$. The aim is to
586: compute matrix elements in a new properly
587: orthogonalized basis $|\varphi_n\rangle$. Transformation to the new
588: basis is given by a symmetric matrix:
589: \begin{equation}
590: |\varphi_1\rangle = \lambda\bigl(|\psi_1\rangle -
591: \mu|\psi_2\rangle\bigr), \
592: |\varphi_2\rangle = \lambda\bigl(|\psi_2\rangle -
593: \mu|\psi_1\rangle\bigr)\ .
594: \label{phi2}
595: \end{equation}
596: Conditions $\langle\varphi_1|\varphi_2\rangle=0$,
597: $\langle\varphi_1|\varphi_1\rangle=
598: \langle\varphi_2|\varphi_2\rangle=1$ determine $\mu$ and $\lambda$.
599: Assuming $E_{11}=E_{22}$, for simplicity, and calculating matrix
600: elements of $\hat{\cal H}$ between the new states one finds
601: \begin{eqnarray}
602: &&\tilde{E}_{11} = E_{11}+\frac{c}{1-c^2}\bigl( cE_{11}-E_{12}\bigr)\ ,
603: \nonumber \\
604: &&
605: \tilde E_{12}=E_{12}+\frac{c}{1-c^2}\bigl( cE_{12}-E_{11}\bigr)\ .
606: \label{Eortho}
607: \end{eqnarray}
608:
609: In the following we shall subtract from the effective Hamiltonians
610: $\hat{\cal H}_1$ and $\hat{\cal H}_1+\hat{\cal H}_2$ the corresponding mean-field
611: energies $E_{\rm MF}$, which are the same for all dimer coverings.
612: Then, $E_{11}=E_{22}=0$ and the matrix elements (\ref{Eortho})
613: are directly related to the parameters of a QDM:
614: \begin{equation}
615: t = -\tilde{E}_{12} \approx - E_{12}\ , \ \ \
616: V = \tilde{E}_{11} \approx -c E_{12} \ ,
617: \label{Vt2}
618: \end{equation}
619: where $|c|\ll 1$ is used.
620:
621: \subsection{First-order mapping}
622:
623: Let us apply the outlined procedure to the first-order Hamiltonian
624: (\ref{H1st}). The mean-field energy $E_{\rm MF}=-\frac{3}{4}JN_d$ of
625: an arbitrary configuration of $N_d$ dimers
626: is always subtracted from Eq.~(\ref{H1st}).
627: The wave-functions for two dimer states
628: on the shortest loop $C_1$ shown in Fig.~\ref{triangular}
629: are written explicitly as
630: \begin{equation}
631: |\psi_1\rangle=[12][43]|d_1d'_2d'_3d_4\rangle, \
632: |\psi_2\rangle=[32][41]|d'_1d'_2d''_3d''_4\rangle,
633: \end{equation}
634: where sites are numbered counter-clockwise beginning with the
635: lower right corner of the rhombus.
636: The first part of $|\psi_{1,2}\rangle$ is given by a product
637: of two spin singlet states, whereas the second part is
638: an orbital wave-function represented as a product of states
639: (\ref{doublet}) or (\ref{doublet2}).
640: Explicit calculation of the off-diagonal matrix element
641: $E_{12} = \langle\psi_2|\hat{\cal H}_1|\psi_1\rangle$
642: yields
643: \begin{equation}
644: t_1 = - \frac{3}{2^6} J \ , \ \ \ \ \
645: V_1 =\frac{3}{2^{10}}J \ .
646: \label{c1}
647: \end{equation}
648: The largest kinetic energy term in the QD Hamiltonian (\ref{Hqdm})
649: amounts to less
650: than 5\% of the weaker coupling constant.
651: The ratio of
652: the potential energy to the kinetic term constant is also very small
653: $V/|t|=1/16$.
654: Note, that $t<0$ from the above calculation.
655: Remaining freedom in the choice of sign of tunneling matrix elements
656: is discussed in the next section.
657:
658: Since the potential energy $V_1$ is an order of magnitude
659: smaller than the dimer hopping $t_1$,
660: the next relevant interactions besides the kinetic energy
661: of two-dimer moves around $C_1$
662: may be three-dimer resonances along longer loops $C_2$, $C_3$, $C_4$,
663: and $C_5$.
664: We find no tunneling for loops
665: $C_4$ and $C_5$, {\it i.e.}, $V_{4,5}=t_{4,5}\equiv 0$.
666: For loop $C_5$, vanishing of the off-diagonal matrix element
667: can be understood
668: by drawing two dimer configurations on a corresponding
669: cluster of a kagom\'e lattice, which is a six-point star.
670: Two dimer states around perimeter of such
671: a star are exact degenerate eigenstates \cite{syromyat}
672: and, hence, $t_5\equiv 0$.
673: In the former case, loop $C_4$, the
674: $E_{12}=0$ result is valid only in the first order in $J$, see the
675: next subsection.
676:
677: Coherent motion of three dimers along composite loops $C_2$ and $C_3$
678: is not described by Eqs.~(\ref{Eortho}) and (\ref{Vt2}) because of
679: resonances around small rhombi.
680: Let the additional state with three parallel dimers on the parallelogram
681: $C_2$ be denoted by $|\psi_1\rangle$ and
682: the two dimer states on the perimeter
683: of $C_2$ be
684: $|\psi_2\rangle$ and $|\psi_3\rangle$, $\langle\psi_{2,3}|\psi_1\rangle=c$,
685: while $\langle\psi_2|\psi_1\rangle=c'$.
686: Then, the matrix elements $E_{12}=E_{13}$ describe short-loop resonances,
687: while $E_{23}$ corresponds to a tunneling along the composite loop.
688: Generalizing transformation (\ref{phi2}) to three
689: states we finally obtain:
690: \begin{eqnarray}
691: &&\tilde{E}_{11} \approx - c E_{12} - \bigl(c'-\frac{3}{4}c^2\bigr)E_{23}
692: \ ,
693: \nonumber \\
694: &&
695: \tilde E_{12} \approx E_{12} \ , \ \
696: \tilde E_{23} \approx E_{23}- c E_{12}
697: \label{Eortho3}
698: \end{eqnarray}
699: in the relevant case $|c'|\ll|c|\ll 1$.
700: Tunneling of dimers along a longer loop $E_{23}$ is renormalized
701: by short-loop hopping.
702: Using the above expressions to calculate resonance
703: of singlet bonds on $C_2$ and $C_3$ one obtains that in both cases
704: \begin{equation}
705: t_2 = t_3 = - \tilde{E}_{23} = - \frac{15}{2^{10}} J \ .
706: \end{equation}
707: The potential energy given
708: by the second term in $\tilde{E}_{11}$ in (\ref{Eortho3})
709: is again extremely small $V_2/|t_2|\approx 0.02$ and can be
710: completely neglected.\cite{milaXZ}
711: Further extension of the above calculations to longer
712: loops show that
713: tunneling matrix elements of four-dimer moves are rather small
714: $\sim 0.07t_1$ and should also be neglected.
715:
716:
717: \subsection{Second-order mapping}
718:
719: Analysis starts again with calculation of the mean-field contribution
720: from $\hat{\cal H}_2$ to the ground state energy (diagonal matrix elements)
721: for an arbitrary dimer state. Every
722: dimer has a finite energy contribution
723: from $\hat{\cal H}_{2b}$: $\frac{1}{6}J^2$, while all nondimer
724: bonds receive mean-field contributions from $\hat{\cal H}_{2a}$:
725: $-\frac{1}{12}J^2$. The total mean-field energy does not, therefore,
726: depend on a chosen dimer covering of a triangular lattice
727: and is equal to
728: \begin{equation}
729: E_{\rm MF} = \Bigl( - \frac{3}{4} - \frac{3}{8}J - \frac{1}{8}J
730: \Bigr) \frac{N}{3} \ ,
731: \label{MF}
732: \end{equation}
733: where $N$ is number of sites on a kagom\'e lattice.
734: The mean-field energy (\ref{MF}) is subtracted in the following from
735: $\hat{\cal H}_1+\hat{\cal H}_2$, such that all diagonal matrix elements
736: vanish.
737:
738: The off-diagonal matrix element of $\hat{\cal H}_2$
739: for the shortest loop $C_1$ has nonzero contributions
740: from $\hat{\cal H}_{2a}$: $-\frac{1}{64}J^2$, and from
741: $\hat{\cal H}_{2c}$: $-\frac{1}{48}J^2$. Combining them with
742: the first-order result Eq.~(\ref{c1}) we obtain
743: \begin{equation}
744: t_1 =-E_{12}=-\frac{3}{2^6}J \Bigl(1-\frac{7}{9}J\Bigr)\ .
745: \label{t1}
746: \end{equation}
747: Similar calculation for the loops $C_2$ and $C_3$ yields
748: \begin{equation}
749: E_{23}=\frac{3}{2^8}J\Bigl(1+\frac{1}{9}J\Bigr)\ .
750: \label{c23}
751: \end{equation}
752: Taking into account Eq.~(\ref{Eortho}) the tunneling matrix element
753: between orthogonal dimer states along the loop $C_2$ ($C_3)$ becomes
754: \begin{equation}
755: t_2 =t_3 = -\tilde{E}_{23}=-\frac{15}{2^{10}}J
756: \Bigl(1-\frac{1}{15}J\Bigr)\ .
757: \label{t2}
758: \end{equation}
759: Loop $C_4$ also acquires a finite tunneling rate
760: between the two dimer
761: states in the second-order. The corresponding matrix element
762: is, however, small $E_{12} \approx 0.003J^2$.
763:
764:
765: \section{Discussion}
766:
767: Signs of the tunneling matrix elements calculated
768: in the previous section have certain arbitrariness.
769: \cite{rokhsar,moessner01}
770: The negative sign of the resonance matrix element for
771: the shortest loop $C_1$ can be changed to positive by a gauge
772: transformation. For this, real singlet wave-functions
773: have to be multiplied by complex factors
774: $i^{n_r + n_{l,e} -n_{l,o}}$, where for a given dimer configuration
775: $n_r$ counts the number of dimers on links pointing
776: upwards and right, $n_{l,e}$ ($n_{l,o}$) counts the number
777: of dimers on links pointing upwards and left from sites with
778: even (odd) vertical coordinates. Dimers on strictly horizontal
779: bonds do not contribute to the phase factor.
780: By this operation resonance moves along every $C_1$ loop
781: pick up an extra $(-1)$ factor, changing $t_1\rightarrow -t_1$.
782: At the same time, amplitudes of dimer tunneling along $C_2$ and $C_3$ loops
783: do not change sign by the above gauge transformation:
784: $t_{2,3}\rightarrow t_{2,3}$.
785: An effective QD Hamiltonian for the trimerized kagom\'e
786: antiferromagnet is, therefore, dominated by the kinetic energy terms
787: for resonance moves
788: between orthogonal dimer configurations
789: $|\varphi_{c_n}\rangle$ and $|\varphi'_{c_n}\rangle$
790: for every loop $C_n$ of three different
791: types $n=1$--3 on an effective triangular
792: lattice:
793: \begin{eqnarray}
794: && \hat{\cal H}_{\rm QD} = \sum_{l_n} -t_n |\varphi_{c_n}\rangle
795: \langle\varphi'_{c_n}| \ ,
796: \label{QDTri}
797: \\ &&
798: t_1 = \frac{3}{64}J\Bigl(1-\frac{7}{9}J\Bigr), \ \
799: t_2=t_3=-\frac{15}{1024}J\Bigl(1-\frac{1}{15}J\Bigr).
800: \nonumber
801: \end{eqnarray}
802: Amplitudes for
803: three-dimer tunneling processes
804: have no significant
805: smallness compared to the strongest resonance move:
806: $t_{2,3}/t_1\approx -0.31$ for $J\ll 1$.
807: The kinetic coefficients $t_n$ are differently renormalized
808: by the second-order processes such that importance
809: of three-dimer moves is further increased towards the isotropic limit:
810: $t_{2,3}/t_1\approx -0.5$ for $J=0.5$ and $t_{2,3}/t_1\approx -1$ as
811: $J\rightarrow 1$.
812:
813: The QDM (\ref{QDTri}) with only two-dimer resonances
814: has been studied via mapping to a frustrated
815: Ising model in transverse field.\cite{moessner01,moessner02} The ground
816: state of this dimer model is believed to be a crystalline
817: $\sqrt{12}\times\sqrt{12}$
818: state, which consists of locally resonating dimer pairs and breaks
819: translational symmetry of the lattice.
820: Such a state should have a fully gapped excitation spectrum.
821: Properties of the QDM (\ref{QDTri}) with several competing dimer
822: resonances have not been studied so far.
823: Dimer resonances along longer loops, though not very small,
824: frustrate each other. The ground state of the realistic model
825: (\ref{QDTri}) should not be, therefore, very far from the idealized
826: model with $t_1$ terms only. In particular, we expect that
827: the ground state breaks certain lattice
828: symmetries. The excitation spectrum is also
829: expected to be gapped unless a fine tuning of $t_n$ drives the system
830: towards an Ising-type transition point between two crystalline states.
831:
832: In conclusion, the presented derivation of the QDM for a realistic
833: Heisenberg spin model on trimerized kagom\'e lattice illustrates a generation
834: of small energy scales in frustrated quantum magnets.
835: The dimer resonance matrix elements in (\ref{QDTri}) are
836: given by small fractions of a weaker exchange constant, {\it e.g.},
837: $t_1\approx 0.047J$. In a wide temperature
838: interval $t_1\ll T\ll J$ the quantum spin systems
839: is described by an RVB liquid of singlet pairs.
840: At very low temperatures $T<t_1$ a valence bond crystal
841: probably replaces an RVB state. The gap between the ground state
842: and the first excited singlet levels is a fraction of $t_1$.
843: It is extremely difficult to resolve such a tiny energy scale
844: in the exact numerical diagonalization of small clusters.
845: Similarly,
846: the low-temperature
847: regime $T\lesssim t_1$ might be beyond experimental reach
848: for possible realizations
849: of the spin-1/2 trimerized kagom\'e antiferromagnet.
850: The dimer crystallization at $T=0$ is driven by local
851: resonances. Therefore, variational mean-field type approaches
852: \cite{mila,auerbach} are not capable to describe the precise
853: nature of the corresponding ground states.
854:
855: \acknowledgments
856:
857: I thank to D. A. Ivanov and G. Jackeli for valuable
858: discussions. The hospitality of
859: the Condensed Matter Theory Institute of Brookhaven
860: National Laboratory during the course
861: of present work is gratefully acknowledged.
862:
863: \begin{thebibliography}{99}
864:
865: \bibitem{anderson}
866: P. Fazekas and P. W. Anderson,
867: Philos. Mag. {\bf 30}, 423 (1974);
868: P. W. Anderson, Science {\bf 235}, 1196 (1987).
869:
870: \bibitem{sutherland}
871: B. Sutherland,
872: Phys. Rev. B {\bf 37}, R3786 (1988).
873:
874: \bibitem{rokhsar}
875: D. S. Rokhsar and S. A. Kivelson,
876: Phys. Rev. Lett. {\bf 61}, 2376 (1988).
877:
878: \bibitem{moessner00}
879: R. Moessner, S. L. Sondhi, and P. Chandra,
880: Phys. Rev. Lett. {\bf 84}, 4457 (2000).
881:
882: \bibitem{moessner01}
883: R. Moessner and S. L. Sondhi,
884: Phys. Rev. Lett. {\bf 86}, 1881 (2001).
885:
886: \bibitem{moessner02}
887: R. Moessner and S. L. Sondhi,
888: Prog. Theor. Phys. Suppl. {\bf 145}, 37 (2002).
889:
890: \bibitem{ioselevich}
891: A. Ioselevich, D. A. Ivanov, and M. V. Feigelman,
892: Phys. Rev. B {\bf 66}, 174405 (2002).
893:
894: \bibitem{book}
895: A. Auerbach, {\it Interacting Electrons and Quantum Magnetism}
896: (Springer, New York, 1994).
897:
898: \bibitem{lecheminant}
899: P. Lecheminant, B. Bernu, C. Lhuillier, L. Pierre,
900: and P. Sindzingre,
901: Phys. Rev. B {\bf 56}, 2521 (1997).
902:
903: \bibitem{waldtmann}
904: C. Waldtmann, H.-U. Everts, B. Bernu, C. Lhuillier,
905: P. Sindzingre, P. Lecheminant, and L. Pierre,
906: Eur. Phys. J. B {\bf 2}, 501 (1998).
907:
908: \bibitem{subrah}
909: V. Subrahmanyam,
910: Phys. Rev. B {\bf 52}, 1133 (1995).
911:
912: \bibitem{mila}
913: F. Mila,
914: Phys. Rev. Lett. {\bf 81}, 2356 (1998).
915:
916: \bibitem{raghu}
917: C. Raghu, I. Rudra, S. Ramasesha, and D. Sen,
918: Phys. Rev. B {\bf 62}, 9484 (2000).
919:
920: \bibitem{auerbach}
921: R. Budnik and A. Auerbach,
922: Phys. Rev. Lett. {\bf 93}, 187205 (2004).
923:
924: \bibitem{santos}
925: L. Santos, M. A. Baranov, J. I. Cirac, H.-U. Everts, H. Fehrmann,
926: and M. Lewenstein,
927: Phys. Rev. Lett. {\bf 93}, 030601 (2004).
928:
929: \bibitem{mambrini}
930: M. Mambrini and F. Mila,
931: Eur. Phys. J. B {\bf 17}, 651 (2000).
932:
933: \bibitem{syromyat}
934: A. V. Syromyatnikov and S. V. Maleyev,
935: Phys. Rev. B {\bf 66}, 132408 (2002).
936:
937: \bibitem{milaXZ}
938: similar results have recently been obtianed by
939: F. Mila et al., private communication (2004).
940:
941: \end{thebibliography}
942:
943: \end{document}
944:
945:
946: