1: \documentclass[letter,twocolumn,showpacs,aps,prb,floatfix]{revtex4}
2: \usepackage{graphicx,subfigure,epsfig,verbatim,psfrag,amsmath,amssymb,color}
3: \newcommand{\red}[1]{{\textcolor{red}{#1}}}
4: \newcommand{\blue}[1]{{\textcolor{blue}{#1}}}
5: \newcommand{\mean}[1]{\left<#1\right>}
6: \newcommand{\abs}[1]{\left|#1\right|}
7: \makeatletter
8: \input epsf
9: \def\prb{Phys. Rev. B}
10: \def\prl{Phys. Rev. Lett.}
11: \def\pla{Phys. Lett. A}
12: \def\pr{Phys. Rev.}
13: \def\be{\begin{equation}}
14: \def\ee{\end{equation}}
15: \def\ba{\begin{eqnarray}}
16: \def\ea{\end{eqnarray}}
17: \makeatother
18:
19: \newcommand{\up}{\uparrow}
20: \newcommand{\dn}{\downarrow}
21:
22:
23: \begin{document}
24:
25: \title{Thermodynamics of Ising spins on the triangular kagome lattice: Exact analytical method and Monte Carlo simulations}
26: \author{Y.~L.~Loh, D.~X.~Yao, and E.~W.~Carlson}
27: \affiliation{Department of Physics, Purdue University, West Lafayette, IN 47907}
28: \date{November 22, 2007}
29:
30: \begin{abstract}
31:
32: A new class of two-dimensional magnetic materials
33: $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
34: (cpa=2-carboxypentonic acid; X=F,Cl,Br)
35: was recently fabricated in which Cu sites form a Triangular Kagome Lattice (TKL).
36: As the simplest model of geometric frustration in such a system,
37: we study the thermodynamics of Ising spins on the TKL
38: using exact analytic methods as well as Monte Carlo simulations.
39: We present the free energy, internal
40: energy, specific heat, entropy, sublattice magnetizations, and susceptibility. We describe the rich phase diagram of the model as a function of coupling constants, temperature, and applied magnetic field.
41: For frustrated interactions in the absence of applied field,
42: the ground state is a spin liquid phase with residual entropy per spin $s_0/k_B=\frac{1}{9} \ln 72\approx 0.4752\dots$.
43: %%YL: Removed the word "integer"
44: In
45: %EC2
46: weak
47: applied field, the system maps to the dimer model on a honeycomb lattice, with residual entropy $0.0359$ per spin and quasi-long-range order with power-law spin-spin correlations that should be detectable by neutron scattering.
48: The power-law correlations become exponential at finite temperatures, but the correlation length may still be long.
49: \end{abstract}
50: \pacs{75.30.Kz, 75.40.Mg, 75.10.Hk, 64.60.-i}
51: \maketitle
52:
53:
54: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
55: \section{Introduction}
56: Geometrically frustrated spin systems
57: %EC2 have attracted much attention for several decades. Besides giving
58: give rise to many novel classical and quantum spin liquid phases.
59: They may have technological applications as refrigerants, via
60: adiabatic demagnetization,\cite{zhitomirsky-2003} in which reducing the applied magnetic field
61: results in a cooling effect as the spins absorb entropy from other degrees of freedom.
62: Unlike paramagnetic salts which are limited by ordering or spin-glass transitions due to residual interactions between the spins, geometrically frustrated systems can remain in a disordered, cooperative paramagnetic state
63: down to the lowest temperatures.
64: Furthermore, they may exhibit an enhanced magnetocaloric effect in the vicinity of phase transitions at finite
65: applied fields.\cite{zhitomirsky-2004,derzhko-2004,isakov-2004,aoki-2004}
66:
67: %EC2 we should either include more or drop this part.
68: %Mekata\cite{mekata77} discovered a partially disordered phase with
69: %three sublattices in the TIAF with an additional ferromagnetic
70: %second-neighbor interaction. There has since been considerable
71: %interest in such partially disordered phases for triangular, kagome,
72: %and other systems.~\cite{mekata78,tonegawa82,saito85,miyashita86,miyashita91,mekata93,takagi95,mekata94,tanaka96,wolf88,azaria87,giacomini88a,azaria87b,giacomini88b,tang88,takagi93}
73:
74:
75: %\red{Besides the triangular and kagome lattices, triangular kagome Lattice
76: %(TKL, see Fig.~\ref{lattice}) has been found in
77: A new class of two-dimensional magnetic materials
78: $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
79: (cpa=2-carboxypentonic acid, a derivative of ascorbic acid; X=F,Cl,Br)
80: ~\cite{gonzalez93,maruti94, mekata98} was recently fabricated, which is an
81: experimental realization of a new type of geometrically frustrated lattice.
82: The Cu spins in these materials are interconnected in a
83: ``triangles-in-triangles'' kagome pattern, which we refer to as a triangular
84: kagome lattice (TKL, see Fig.~\ref{lattice}).
85: %EC2
86: %Compared to other triangle-containing spin systems (TIAF and KIAF),
87: %the TKL has both similarities and differences. It is therefore a new
88: %playground for the study of magnetic ordering and phase transitions.
89:
90: \begin{figure}
91: \resizebox*{0.65\columnwidth}{!}{\includegraphics{tkl_lattice.eps}}
92: \caption{(Color online) The triangular kagome lattice (TKL). Solid (open)
93: circles represent ``a'' (``b'') sublattices. Thick and thin lines
94: represent interactions $J_{aa}$ and $J_{ab}$ respectively. The shaded
95: region represents a unit cell. By shifting the cell slightly (as indicated by the dashed parallelogram) it can be seen that there are nine spins per unit cell.}
96: \label{lattice}
97: \end{figure}
98: Experiments on the $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
99: compounds show no spontaneous magnetization down to
100: at least $T=1.7$K,\cite{maruti94} consistent with a spin liquid ground state,
101: indicating that $J_{aa}$ is antiferromagnetic.
102: However, whether $J_{ab}$ is ferromagnetic or antiferromagnetic is still
103: an open question. Based on the observation of a partial rather than saturated
104: magnetization,\cite{maruti94,mekata98}
105: Maruti {\em et al.} concluded that
106: the intertrimer coupling is antiferromagnetic, $J_{ab}<0$.\cite{maruti94}
107: In a later theoretical study using variational mean field theory
108: on the quantum Heisenberg model on the TKL,
109: Stre\v{c}ka\cite{strecka07} concluded that the intertrimer coupling is ferromagnetic, $J_{ab}>0$.
110: Regardless, the lack of observed hysteresis despite the observation of
111: a magnetization plateau in finite field\cite{maruti94}
112: is consistent with a multitude of ground states which
113: can be connected by a series of local spin flips.
114:
115: %YL: rearranged things in more sensible order
116: %Motivation for this study:
117: %In general, quantum spin models can exhibit very complicated behavior and are
118: %difficult to solve.
119: In this paper we study the
120: classical TKL Ising model
121: %EC2 for which many exact results can be obtained.
122: using exact analytic methods and Monte Carlo simulations.
123: The purpose of this study is to provide explicit predictions,
124: in order to determine to what
125: extent the experiments can be explained in terms of a classical Ising model.
126: %EC4 added
127: In particular, we find significant difference between the behavior of the
128: susceptibility as a function of temperature for $J_{ab}>0$ and $J_{ab}<0$,
129: and this may be used as an experimental means of distinguishing the two
130: cases. Discrepancies between experiments and our theoretical predictions
131: will indicate the vector nature of the actual spins (XY or Heisenberg), the
132: effects of quantum fluctuations, or possibly higher order interactions. This
133: model was previously studied by one of us \cite{daoxinMSthesis}. In that
134: work, Monte Carlo simulations were used to study the phase
135: transitions and basic thermodynamics. %DY2
136: Zheng and Sun\cite{zheng05}
137: mapped the partition function to that of a kagome lattice and
138: found an analytic expression for the phase boundary in zero field.
139: %Here we first use
140: %$\Delta$-Y transformations to map the system to a honeycomb lattice
141: %with two strong bonds. then map it a honeycomb lattice with a single
142: %oupling. Using this way,
143: %In this paper we derive exact solutions for the free energy, energy, heat capacity, and entropy of the zero-field TKL Ising model.
144: %by using $\Delta$-Y transformations and series reductions to map the model exactly to a honeycomb lattice.
145: % YL: I comment this out because "all exact solutions are the same, so the methods used to obtain them are irrelevant" (unlike numerical solutions, where different methods give answers with different types of errors)..
146: %We plot these for various parameter combinations. We also perform To check the analytic results, we perform Monte Carlo simulations; these also allow us to study thermal quantities, spin configurations and phase diagrams in the presence of an external magnetic field.
147: %EC2 lots added here
148:
149: In this paper we present exact results in zero field at finite temperature,
150: and also in an applied field at zero temperature.
151: We report the full phase diagram as a
152: function of coupling constants, temperature, and applied magnetic field.
153: We find several field-induced transitions.
154: In particular, for frustrated interactions in applied field, we find a quasi-long-range ordered phase
155: which maps to hard-core dimers on the honeycomb lattice.
156: We complement these exact analytic results with Monte Carlo simulations
157: on the magnetization and susceptibility.
158: We report the temperature-dependence of the magnetic susceptibility,
159: and show how it can be used to deduce the sign of the coupling constants.
160:
161: The paper is organized as follows. The TKL Ising model is described in
162: Section~\ref{model}. In Section~\ref{zerofield}, we present exact results
163: for the TKL Ising model in zero field. In Section~\ref{zerotemperature} we
164: present exact results at zero temperature. In Section~\ref{finiteTfiniteh}
165: we present the phase diagram and describe the various phases. In
166: Section~\ref{montecarlo} we present Monte Carlo results for the
167: susceptibility, spontaneous magnetization, and magnetization curves.
168: In Section~\ref{discussion} we compare our model to
169: models of geometrically frustrated magnets on other lattices,
170: as well as to experiments on TKL systems.
171: In Section~\ref{conclusions} we present our conclusions, and in
172: Appendix
173: %EC2 , Section~\ref{meanfield},
174: we present two mean-field approximations
175: in order to illustrate their failure in the frustrated regime.
176: %EC2 \blue{We can combine this with the bulleted list above.}
177: %DY2 Is that good to use two sections
178: %to present two mean-field theories.
179:
180:
181: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
182: \section{Model \label{model}}
183: %%YL: added this sentence
184: %EC2 modified this sentence.
185:
186: The TKL (Fig.~\ref{lattice}) can be obtained by
187: inserting triangles inside the triangles of the kagome lattice,
188: for which it is sometimes referred to in the literature as the ``triangles-in-triangles''
189: kagome lattice.
190: Alternatively, it can be
191: derived from the triangular lattice by periodically deleting seven out of
192: every sixteen lattice sites. This structure has two different spin
193: sublattices ``a'' and ``b'', which correspond to small trimers and large
194: trimers, respectively. We study Ising spins on the TKL with two kinds of
195: nearest-neighbor interactions, the ``intratrimer'' couplings $J_{aa}$ and the
196: ``intertrimer'' couplings $J_{ab}$. Each spin has four nearest neighbors.
197: The Hamiltonian is
198: \begin{equation}
199: H= -J_{aa}\sum_{i,j \in a} \sigma_{i}\sigma_{j} -
200: J_{ab}\sum_{i\in a, j\in b} \sigma_{i}\sigma_{j} - h \sum_{i}
201: \sigma_{i}
202: \label{e:hamiltonian}
203: \end{equation}
204: where $\sigma_{i}=\pm 1$, summations run over the nearest spin pairs
205: and all spin sites, $h$ is an external magnetic field.
206: The shaded region in Fig.~\ref{lattice} is one unit cell, which contains 6
207: $a$-spins, 3 $b$-spins, 6 $a$--$a$ bonds, and 12 $a$--$b$ bonds. We
208: shall use $N_{a}$ and $N_{b}$ to denote the total numbers of spins on
209: the a and b sublattices, so that $N_{a}:N_{b} =2:1$.
210: The space group of the TKL is the same as that of the hexagonal lattice,
211: $p6m$, in Hermann-Mauguin notation. %DY2
212: % EC The TKL has the symmetry of the hexagonal lattice (space group p6m). %in Hermann-Mauguin notation
213: %that the TKL has $C_6$ symmetry.
214:
215:
216:
217: There are four energy scales in the problem: $J_{aa}$, $J_{ab}$, $T$, and $h$.
218: We have found it most convenient to take $|J_{ab}|$ as the unit of energy.
219: Thus, the model can be described by a three-dimensional phase diagram in the space of the three dimensionless parameters
220: $J_{aa}/|J_{ab}|$, $h/|J_{ab}|$, and $T/|J_{ab}|$.
221: The phase diagram also depends on the sign of $J_{ab}$.
222:
223: %We report results scaled by $J_{ab}$, {\em i.e.} our phase diagram is
224: %For simplicity, we use
225: %($|J_{ab}|=1$).
226:
227: %EC3 We disagree with this now.
228: %Susceptibility measurements on
229: %$\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
230: %suggest that both $J_{aa}$ and $J_{ab}$ are antiferromagnetic and that $|J_{aa}| \gg
231: %|J_{ab}|$.\cite{haar95,maruti94}
232: %*** We need to confirm this. ***
233: %Therefore the present study deals mainly with this regime.
234:
235: %, that there is no phase transition to long-range order and that there is ............spin frustration and the absence of traditional long-range order.
236: %The Curie-Weiss temperatures in the range of -$240 K$ strongly suggest
237:
238:
239:
240:
241:
242: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
243: \section{Exact results in zero field \label{zerofield}}
244: In this section we present exact analytic results for the TKL Ising model
245: in zero magnetic field ($h=0$), including the free energy,
246: %EC2 added stuff from last section
247: internal energy, specific heat, and entropy.
248: We use a sequence of $\nabla$-$Y$
249: transformations and series reductions (shown in
250: Fig.~\ref{f:tklreduction}) to transform the Ising model on a TKL into
251: one on a honeycomb lattice, and then use the known solution for the
252: honeycomb lattice. (See Ref.~\onlinecite{zheng05} for a similar
253: analysis that transforms the TKL to a kagome lattice keeping the
254: overall value of the partition function unchanged.)
255: We remark that frustrated Ising models with quenched bond disorder may
256: be studied by \emph{numerical} application of $\nabla$-$Y$ and
257: $Y$-$\nabla$ transformations\cite{loh2006,loh-jeremy2007} and/or Pfaffian
258: methods\cite{chineseremainder}.
259: Our results from this section are plotted in Figs.~\ref{f:fuctFM} and \ref{f:fuctAF}, using the analytic formulas below (as well as series approximations for extreme values of $T$). These thermodynamic quantities do not depend on the sign of $J_{ab}$.
260:
261: %EC2
262: %It should be possible to obtain analytic results for the spontaneous
263: %magnetization of the TKL Ising model, for example, by the methods of
264: %Ref.~\onlinecite{barry1995}.
265:
266: %%%%%%% F I G U R E %%%%%%%%%%%%
267: \begin{figure}[t]
268: \psfrag{J1}{$J_{aa}$}
269: \psfrag{J2}{$J_{ab}$}
270: \psfrag{J3}{$J_3$}
271: \psfrag{J4}{$J_4$}
272: \psfrag{J5}{$J_5$}
273: \psfrag{J6}{$J_6$}
274: \psfrag{J7}{$J_7$}
275: \psfrag{J8}{$J_8$}
276: \psfrag{J9}{$J_{h}$}
277: \psfrag{Nabla-Y}{$\nabla$--$Y$}
278: \centering
279: {\resizebox*{\columnwidth}{!}{\includegraphics{tklreduction.eps}}}
280: \caption{(Color online). Transformation of a triangular kagome lattice (TKL) to a honeycomb lattice. Figure (a) depicts a section of the TKL, with couplings $J_{aa}$ and $J_{ab}$. The procedure begins by applying $\nabla$-$Y$ transformations to the six downward-pointing triangles in each unit cell. This gives lattice (b). Now take the two strong bonds ($J_3$) in series to give $J_5$, and the two weak bonds ($J_4$) in series to give $J_6$, to obtain a ``3--12 lattice'' (c). Apply $\nabla$-$Y$ transformations to the triangles to obtain a decorated honeycomb lattice (d). Finally, perform series reductions to obtain the honeycomb lattice (f) with a single coupling $J_h$.
281: \label{f:tklreduction}}
282: \end{figure}
283: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
284:
285: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
286:
287: \subsection{Effective coupling on the equivalent honeycomb lattice \label{sec:mapping}}
288:
289: If an Ising model is
290: %EC2
291: on a particular lattice
292: contains a spin $\sigma_0$ connected to only two other spins $\sigma_1$ and $\sigma_2$ via couplings $J_1$ and $J_2$, then we can `integrate out' $\sigma_0$
293: while preserving the value of the partition function, in order
294: to obtain an effective coupling $J_{12}$. The transformation also produces a constant factor multiplying the partition function, $A$. That is,
295: \begin{align}
296: \sum_{\sigma_0=-1}^{+1}
297: e^{\beta (
298: J_{1} \sigma_1 \sigma_0
299: + J_{2} \sigma_2 \sigma_0
300: )}
301: &=
302: A e^{\beta
303: J_{12} \sigma_1 \sigma_2
304: }
305: \end{align}
306: for all combinations of values of $\sigma_1$ and $\sigma_2$.
307: It is most convenient to write the effective coupling in terms of $t_{1},t_2,t_{12}$ where $t_{i}=\tanh \beta J_{i}$, and the partition function changes in terms of $x_{1},x_{2},x_{12}$ where $x_{i}=e^{-2\beta J_{i}}$:
308: \begin{align}
309: t_{12} &= t_1 t_2 \\
310: A &= (1 + x_1 x_2) \sqrt{\frac{x_{12}}{x_1x_2}}
311: \end{align}
312: The $x$'s and $t$'s are related by M\"obius duality transformations, $x_i=(1-t_i)/(1+t_i)$ and $t_i=(1-x_i)/(1+x_i)$.
313:
314: Similarly, if a spin $\sigma_0$ is connected to only three other spins $\sigma_{1,2,3}$ via couplings $J_{1,2,3}$, we can `integrate out' $\sigma_0$
315: while preserving the partition function, in order
316: to obtain effective couplings $J_{23,31,12}$ together with a free energy shift. This is known as a star-triangle or $Y$-$\nabla$ transformation. The reverse transformation exists, and is known as a $\nabla$-$Y$ transformation: given a ``$\nabla$'' of couplings $J_{23,31,12}$, we can find an equivalent ``$Y$''.
317: Again, it is convenient to use the variables $t_1=\tanh \beta J_1$
318: (and similarly for $t_2$, $t_3$) and $x_1=e^{-2 \beta J_1}$:
319: %EC2 minor formatting changes
320: \begin{equation}
321: t_{1} = {\sqrt{a_1a_2a_3/a_0} \over a_1} \quad\text{(cycl.)}
322: \end{equation}
323: where
324: \begin{align}
325: a_0 &= 1 + t_{23} t_{31} t_{12} \\
326: a_1 &= t_{23} + t_{31} t_{12} \quad\text{(cycl.)} \\
327: A &=
328: \frac{1}{1 + x_1x_2x_3}
329: \sqrt{\frac{x_1x_2x_3}{x_{23}x_{31}x_{12}}}~,
330: \end{align}
331: and ``cycl.'' means that $t_{2}$, $t_3$, $a_2$, and $a_3$ are found by cyclic permutation of the indices $1,2,3$. In general the $a$'s may be negative or complex-valued, so
332: %EC2
333: that it is not correct to
334: replace $\sqrt{a_1a_2a_3/a_0}/a_1$ by $\sqrt{a_2a_3/a_0a_1}$.
335:
336: Using a sequence of $\nabla$-$Y$ transformations and series reductions, we transform the TKL Ising model (with couplings constants $J_{aa}$ and $J_{ab}$) into a honeycomb Ising model (with a single coupling constant $J_h$), as shown in Fig.~\ref{f:tklreduction}. The transformation equations (in terms of the $t_i=\tanh \beta J_i$ variables) are:
337: %[I have not been too careful with branch cuts below:]
338: \begin{align}
339: t_3 &= \sqrt{ (t_{aa} + {t_{ab}}^2) / (1 + t_{aa}{t_{ab}}^2)} \\
340: t_4 &= (t_{ab}+t_{aa}t_{ab}) / \sqrt{ (t_{aa} + {t_{ab}}^2)(1 + t_{aa}{t_{ab}}^2) } \\
341: t_5 &= {t_3}^2 \\
342: t_6 &= {t_4}^2 \\
343: t_7 &= 1 / \sqrt{ (t_5 + {t_5}^{-1} - 1)} \\
344: t_8 &= t_6 t_7 \\
345: t_h &= t_8 t_7
346: \end{align}
347: We can write $t_h$ directly in terms of $t_{aa}$ and $t_{ab}$:
348: \begin{align}
349: t_h &=
350: \frac{
351: \left(1+{t_{aa}}\right)^2 {t_{ab}}^2
352: }{
353: \left(1-{t_{aa}}+{t_{aa}}^2\right) \left(1+{t_{ab}}^4\right)
354: - \left(1-4{t_{aa}}+{t_{aa}}^2\right) {t_{ab}}^2 }
355: \label{th-taa-tab}
356: \end{align}
357: It will be convenient to
358: %EC2
359: rewrite this in terms of $x_i = e^{-2 \beta J_{i}}$, as this is simpler:
360: %use the expression for $x_h$ ($=e^{-2\beta J_h}$) in terms of $x_{aa}$ and $x_{ab}$, as this is slightly simpler:
361: \begin{align}
362: x_h &=
363: \frac{2 \left(3 {x_{aa}}^2+1\right) {x_{ab}}^2}{{x_{ab}}^4+6 {x_{aa}}^2
364: {x_{ab}}^2+1}
365: \label{xh-xaa-xab}
366: \end{align}
367:
368: %%%%%%% F I G U R E %%%%%%%%%%%%x
369: \begin{figure}
370: \psfrag{J1}{$J_1$}
371: \psfrag{J2}{$J_2$}
372: \psfrag{J3}{$J_3$}
373: \psfrag{J4}{$J_4$}
374: \psfrag{J5}{$J_5$}
375: \psfrag{J6}{$J_6$}
376: \psfrag{J7}{$J_7$}
377: \psfrag{J8}{$J_8$}
378: \psfrag{J9}{$J_h$}
379: \centering
380: {\resizebox*{0.47\textwidth}{!}{\includegraphics{effhoneycoupling.eps}}}
381: \caption{(Color online). Effective dimensionless coupling $\beta J_h$ of the honeycomb lattice Ising model as a function of the original inverse temperature $\beta$, for the triangular Ising antiferromagnet (TIAF), kagome Ising antiferromagnet (KIAF), or TKL Ising antiferromagnet (TKLIAF).
382: In the case of the TIAF, $\beta J_h$ is imaginary, and $\text{Im~} \beta J_h$ is shown as a dashed curve; for the KIAF and TKLIAF, $\beta J_h$ is real.
383: For the TKLIAF in the unfrustrated regime, $\beta J_h$ crosses the dotted line (the critical coupling of the honeycomb Ising model), indicating a phase transition. The other models do not have phase transitions.
384: \label{f:effhoneycoupling}}
385: \end{figure}
386: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
387:
388: %EC2
389: For Ising models,
390: both the triangular lattice and the kagome lattice can be transformed into
391: the honeycomb lattice by a similar procedure. Figure
392: \ref{f:effhoneycoupling} shows the effective dimensionless coupling of the
393: honeycomb lattice, $\beta J_h$, as a function of the inverse temperature for
394: %EC2 several
395: antiferromagnetic Ising models
396: on different lattices for comparison. Note that:
397: %DY2 This is a report style.
398:
399: \begin{itemize}
400: \item For the triangular Ising AF (TIAF), the effective honeycomb coupling is imaginary, and there is no long-range order.
401: \item For the kagome Ising AF (KIAF), the effective honeycomb coupling is real. As the kagome couplings are increased, the effective honeycomb coupling increases
402: %EC2
403: until it saturates at large $\beta$.
404: However, it never grows beyond the critical coupling for the honeycomb model, $\beta J_h^c = \tanh^{-1} 1/\sqrt{3} \approx 0.658$. Therefore the
405: %EC2 kagome model
406: KIAF does not have a phase transition.
407: \item For the triangular kagome lattice Ising antiferromagnet (TKLIAF) in the unfrustrated regime (e.g., $J_{aa}=-0.8, J_{ab}=-1$), the plot of $\beta J_h$ intersects the dotted line, indicating a phase transition at $\beta \approx 3.5$.
408: \item In contrast, for the TKLIAF in the frustrated regime ($J_{aa}=-0.8, J_{ab}=-1$), $\beta J_h$ grows to a maximum and then decays to zero, indicating the absence of a phase transition. Paradoxically,
409: %EC2
410: and unlike the other lattices,
411: \emph{stronger} bare couplings lead to a \emph{weaker} effective coupling! The fact that $\beta J_h\rightarrow 0$ as $\beta\rightarrow \infty$ is closely connected to the fact that the residual entropy of the TKL lattice has the simple value of $\ln 72$ per unit cell (as we will show), unlike the cases of the TIAF\cite{wannier50} and KIAF\cite{kano53} in which the residual entropies are non-trivial two-dimensional integrals.
412: %%YL: replaced "integer, irrational" by "simple, non-trivial". If referee complains, replace "non-trivial" by "...that cannot be expressed/reduced to closed form".
413: \end{itemize}
414:
415:
416:
417:
418: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
419: \subsection{Phase boundary}
420:
421: %\blue{YL says: Zheng and Sun present an extremely complicated equation for the critical temperature. That is unnecessarily complicated. It seems that they were too lazy to simplify their equation. I have verified that it simplifies to the expressions that I present here. Do we want to mention that Zheng and Sun are unnecessarily complicated?}
422: %EC: Yes
423:
424: %EC lots rewritten here
425: The phase boundary of the TKL Ising model in zero applied field was
426: calculated in Ref.~\onlinecite{zheng05}. We show an alternative exact
427: derivation of the results here for pedagogical reasons. Once we have used
428: the techniques of Sec.~\ref{sec:mapping} to map the TKL Ising model into the
429: Ising model on a honeycomb lattice, we can use known results for the
430: honeycomb lattice. The critical temperature of the honeycomb Ising model is
431: given by $t^c_h = \tanh \beta J_h = 1/\sqrt{3}$, or, equivalently, $x^c_h =
432: \exp -2\beta J_h = 2 - \sqrt{3}$. Substituting in the equivalent coupling of
433: the honeycomb lattice, Eq.~\ref{xh-xaa-xab}, leads to an implicit equation
434: for the critical temperature $1/\beta_c$ of the TKL Ising model: %DY2: I
435: %think you mean |J_{ab}| in the following.
436: \begin{align}
437: e^{-4\beta_c J_{aa}} &=
438: (\sqrt{3}-1) \cosh 4\beta_cJ_{ab}
439: -\left(\sqrt{3} + 1 \right) .
440: \label{e:phaseboundary}
441: \end{align}
442: Eqn.~7 of Ref.~\onlinecite{zheng05} is equivalent to the simpler expression reported here.
443: This critical curve is plotted in Fig.~\ref{zerofieldphasediagram}.
444: For large ferromagnetic $J_{aa}$, the critical temperature saturates at a finite value,
445: $T_c/|J_{ab}| \approx
446: 4/\ln\left( 2+\sqrt{3}+\sqrt{6+4\sqrt{3}} \right) \approx 2.00838$.
447: As $J_{aa}$ is reduced towards $-1$, the critical temperature falls to zero. Near
448: %EC2 $J_{aa}=-1$,
449: $J_{aa}=-|J_{ab}|$,
450: the critical curve is approximately linear:
451:
452:
453:
454: %%%%%%% F I G U R E %%%%%%%%%%%%
455: \begin{figure}[!t]
456: \centering
457: {\resizebox*{0.47\textwidth}{!}{\includegraphics{tklphasediagram.eps}}}
458: \caption{(Color online) Phase diagram of the TKL Ising model in the $(J_{aa},T)$ plane, for $J_{ab}=\pm 1$ and $h=0$. The thick curve is the exact solution (equivalent to that in Ref.~\onlinecite{zheng05}). The dotted and dashed curves are mean-field approximations (see Appendix, Sec.~\ref{meanfield}).
459: The ordered phase is ferromagnetic if $J_{ab}>0$ and ferrimagnetic if $J_{ab}<0$. The disordered state is paramagnetic. For $J_{aa} < -J_{ab}$, it persists down to $T=0$, where the entropy is $\ln 72$ per unit cell and the susceptibility obeys a Curie law. The arrows indicate the two TKLIAF cases discussed in Fig.~\ref{f:effhoneycoupling}.
460: \label{zerofieldphasediagram}}
461: \end{figure}
462: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
463:
464:
465:
466: Since the mapping from the TKL to the honeycomb lattice preserves the nature of the singularity in the partition function, the phase transition is a continuous second-order transition in the 2D Ising universality class.
467:
468:
469: In zero field ($h=0$) the partition function $Z$ is invariant under a change of sign of $J_{ab}$, and the topology of the phase diagram is independent of this sign, although the identification of the phases is not.
470: First consider the model for the case $J_{aa}=0$, where the TKL reduces to a decorated kagome lattice. If $J_{ab}$ is ferromagnetic, the model develops {\em ferromagnetic} order below the Curie temperature.
471: If $J_{ab}$ is antiferromagnetic, the model develops {\em ferrimagnetic} order
472: below the ordering temperature.
473: Although the decorated kagome lattice is bipartite,
474: the numbers of spins on the $a$- and $b$-sublattices are not equal: there are twice as many $a$-spins as $b$-spins. Hence we have \emph{ferrimagnetic} order, with unequal numbers of up and down spins producing a net moment.
475: %The ferrimagnetic transition breaks the Ising $Z_2$ symmetry, rather than translational symmetry as in the case of a N\'eel transition.
476: However,
477: %EC2 the thermodynamic quantities $T_c$,$f$,$u$,$c$,$s$
478: the transition temperature, free ener:gy, internal energy, specific heat, and entropy
479: are independent of the sign of $J_{ab}$ in the absence of applied field.
480: %EC2 combined two paragraphs.
481: Now introduce the coupling $J_{aa}$. If this is ferromagnetic, it has very little effect, since the $a$-spins already have a tendency to align. However, if $J_{aa}$ is antiferromagnetic, it fights against the ordering induced by $J_{ab}$. If $J_{aa}$ is %EC2 strong enough
482: sufficiently antiferromagnetic, ($J_{aa}/|J_{ab}| < -1$), the system is in a frustrated regime with no order even at zero temperature.
483:
484:
485:
486:
487: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
488: \subsection{Partition function}
489:
490: \newcommand{\shifted}{\text{shifted}} %\heartsuit}
491:
492: The partition function
493: %EC2 switched order
494: per unit cell, $z_{\rm TKL}$,
495: of the TKL Ising model is equal to that of the equivalent honeycomb lattice
496: %EC2 $f_h$, Also promoted h to "H"
497: $z_H$,
498: multiplied by the factors $z_1, \dotsc, z_6$ below
499: which are accumulated during the sequence of $\nabla$-$Y$ transformations and series reductions
500: necessary to effect the transformation.
501: %EC2 (remembering to include the number of times the transformation is applied per unit cell).
502: %\blue{Of course, when I say ``partition function per unit cell'', I mean that the partition function of the system is $Z_\text{sys} = Z^N$ where $N$ is the number of unit cells, i.e., $Z=Z_\text{sys} {}^{1/N}$. If you're worried about people being confused, we could let $Z$ be the partition function of the whole system and $z$ be the partition function per unit cell.}
503: These factors are
504: \begin{align}
505: z_1 &=
506: \frac{ 1 }{ 1 + x_4 x_3{}^2}
507: \sqrt{\frac{ x_4 x_3{}^2 }{x_{aa} x_{ab}{}^2 }}
508: \quad\text{(first $\nabla-Y$)}\\
509: z_2 &=
510: (1 + x_3{}^2)
511: \sqrt{\frac{x_5}{x_3{}^2}}
512: \qquad\text{($J_3$ in series)} \\
513: z_3 &=
514: (1 + x_4{}^2)
515: \sqrt{\frac{x_6}{x_4{}^2}}
516: \qquad\text{($J_4$ in series)} \\
517: z_4 &=
518: \frac{ 1 }{ 1 + x_7{}^3}
519: \sqrt{\frac{ x_7{}^3 }{ x_5{}^3 }}
520: \qquad\text{(second $\nabla-Y$)} \\
521: z_5 &=
522: \frac{ 1 }{ 1 + x_6 x_7 }
523: \sqrt{\frac{ x_8 }{ x_6 x_7 }}
524: \qquad\text{($J_6,J_7$ in series)} \\
525: z_6 &=
526: \frac{ 1 }{ 1 + x_8 x_7 }
527: \sqrt{\frac{ x_9 }{ x_8 x_7 }}
528: \qquad\text{($J_8,J_7$ in series)}
529: \end{align}
530: The total accumulated partition function change is therefore
531: %EC2 major changes to this presentation
532: \begin{align}
533: z_{\rm TKL}
534: &=z_1{}^6 z_2{}^6 z_3{}^3 z_4{}^2 z_5{}^3 z_6{}^3~z_H
535: \nonumber\\
536: &=
537: \frac{
538: (1+x_{ab}{}^2) ^2
539: (1 + 6x_{aa}{}^2x_{ab}{}^2 + x_{ab}{}^4) ^3
540: }{
541: 1 + 2(1 + 6x_{aa}{}^2)x_{ab}{}^2 + x_{ab}{}^4
542: }~z_H~.
543: \end{align}
544: The partition function per unit cell of the honeycomb lattice has been calculated in the literature by, {\em e.g.}, the Pfaffian method\cite{kasteleyn1963,fisher1966}. It is
545: %EC 2 changed symbol Z_h -> z_H
546: \begin{align}
547: z_H(x_h)
548: &=\frac{\sqrt{2} (1 - {x_h}^2)}{x_h}
549: \exp \left[ \tfrac{1}2 \Omega(w(x_h)) \right]
550: \end{align}
551: where
552: \begin{align}
553: \Omega(w)
554: &=\int_0^{2\pi} \frac{dp}{2\pi} \int_0^{2\pi} \frac{dq}{2\pi}
555: \ln (w - \cos p - \cos q - \cos (p+q))
556: \label{Omega-w}
557: \end{align}
558: and
559: \begin{align}
560: w(x_h)
561: &= \frac{1 - 2{x_h} + 6{x_h}^2 - 2{x_h}^3 + {x_h}^4}{2x_h(1-x_h)^2}
562: .
563: \label{w-xh}
564: \end{align}
565: %&= w_{hon} (t) = \frac{1+3t^4}{2t^2(1-t^2)} \\
566:
567: For the purposes of numerical evaluation,
568: we rewrite the function $\Omega(w)$ in the following form
569: \begin{align}
570: \Omega(w)
571: &=\frac{2}{\pi} \int_0^{\pi/2} dp
572: \ln \left[\cos p + \text{arccosh} \frac{w-\cos 2p}{2 \cos p}
573: \right]~.
574: \end{align}
575: In order to get accurate numerical results one has to further split the range
576: of integration according to the singularities of the integrand.
577: %YL: Do we really want to waste space talking about this? Plots of the free energy aren't interesting, anyway; it is the heat capacity that is interesting.
578: % note: square lattice case.. expansion of elliptic integral.. series coefficients are Binomial(2n,n)^2. ... simple
579: % for tri latt, coeffs are
580: %1, 0, 6, 12, 90, 360, 2040, 10080, 54810, 290640, 1588356, 8676360, 47977776, 266378112, 1488801600, 8355739392, 47104393050, 266482019232, 1512589408044, 8610448069080, 49144928795820, 281164160225520
581:
582: Thus the partition function of the TKL Ising model (per unit cell) is
583: \begin{align}
584: z_{\rm TKL}(x_{aa},x_{ab})
585: &=\Psi(x_{aa},x_{ab})
586: \exp \left[ \tfrac{1}2 \Omega (w (x_h (x_{aa}, x_{ab}))) \right]
587: \end{align}
588: where
589: \begin{align}
590: \Psi
591: &=
592: 2 {x_{aa}}^{-3} {x_{ab}}^{-5}
593: \left(1 - {x_{ab}}^4\right)^2
594: \times\nonumber\\&{}\quad
595: \sqrt{ \left( 1 + 3 {x_{aa}}^2 \right)
596: \left( 1 + 6 {x_{aa}}^2 {x_{ab}}^2 + {x_{ab}}^4 \right) }
597: \label{eqn:Psi}
598: \end{align}
599: and $\Omega$, $w$, and $x_h$ are defined in Eqs.~(\eqref{Omega-w}), (\eqref{w-xh}), (\eqref{xh-xaa-xab}),
600: respectively.
601: The total partition function $Z_{\rm TKL}$ is related to the partition function per unit cell $z_{\rm TKL}$
602: by $Z_{\rm TKL} \equiv z_{\rm TKL}^N$, where N is the number of unit cells.
603: We show plots of
604: %EC3
605: $-{\rm ln}z_{\rm TKL}/(\beta |J_{aa}|)$
606: in Figs.~\ref{f:fuctFM} and \ref{f:fuctAF} (red curves)
607: in the unfrustrated and frustrated regimes, respectively.
608:
609: %\end{widetext}
610:
611: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
612: \subsection{Energy}
613: The energy per unit cell of the TKL Ising model can be obtained by differentiation of the
614: %EC2 free energy:
615: partition function:
616: \begin{align}
617: u
618: &=-\frac{d\ln z}{d \beta}
619: =
620: -\frac{d x_{aa}}{d \beta} \frac{\partial \ln z}{\partial x_{aa}}
621: -\frac{d x_{ab}}{d \beta} \frac{\partial \ln z}{\partial x_{ab}}
622: \\
623: &=\sum_{i=aa,ab}
624: J_i x_i \left[2\frac{\partial \ln\Psi}{\partial x_i}
625: + \frac{\partial x_h}{\partial x_i} \frac{d w}{d x_h} \frac{d \Omega}{d w}
626: \right]~,
627: \end{align}
628: where $\Psi$ is given in Eqn.~\ref{eqn:Psi}. $\frac{d \Omega}{d w}$
629: is the Green function of a particle on a triangular lattice and can
630: be expressed in terms of the complete elliptic integral of the first
631: kind, $K$:\cite{horiguchi1992}
632: %YL3: Added citation about elliptic integral formula
633: \begin{align}
634: \frac{d \Omega}{d w}
635: &=\int_0^{2\pi} \frac{dp}{2\pi} \int_0^{2\pi} \frac{dq}{2\pi}
636: \frac{1}{ w - \cos p - \cos q - \cos (p+q) }
637: \\
638: &=-\tfrac2{\pi
639: (-w-1)^{3/4}(-w+3)^{1/4}}
640: \times\\&{}\qquad\quad
641: K
642: \left( \tfrac{1}2+
643: \tfrac{w^2-3}{2(w+1)(-w-1)^{1/2}(-w+3)^{1/2}} \right).
644: \end{align}
645: We show plots of $-u/|J_{aa}|$ in Figs.~\ref{f:fuctFM} and \ref{f:fuctAF} (green curves).
646:
647:
648: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
649: \subsection{Specific heat}
650: The heat capacity per unit cell, $c=\frac{du}{dT}$, can be obtained by further differentiation:
651: \newcommand{\dd}{\partial}
652: \begin{widetext}
653:
654: \begin{align}
655: c &= 2\Bigg\{
656: (\beta J_{aa})^2 x_{aa} \left[
657: \frac{\dd^2\ln\Psi}{\dd x_{aa} \dd x_{aa}}
658: + \left( \frac{\dd x_h}{\dd x_{aa}}
659: + \frac{\dd^2 x_h}{\dd x_{aa}{}^2} x_{aa}
660: \right)
661: w' \Omega'
662: + \left( \frac{\dd x_h}{\dd x_{aa}} \right)^2 x_{aa}
663: (w'' \Omega' + (w') ^2 \Omega'')
664: \right]
665: \nonumber\\&{}~~~
666: +
667: \left[ \text{previous term with $aa$ replaced by $ab$} \right]
668: \nonumber\\&{}~~~
669: +
670: 2\beta^2 J_{aa} J_{ab} x_{aa} x_{ab} \left[
671: \frac{\dd^2\ln\Psi}{\dd x_{aa} \dd x_{ab}}
672: + \frac{\dd^2 x_h}{\dd x_{aa} \dd x_{ab}} w' \Omega'
673: + \frac{\dd x_h}{\dd x_{aa}} \frac{\dd x_h}{\dd x_{ab}}
674: (w'' \Omega' + (w')^2 \Omega'')
675: \right]
676: \Bigg\}
677: \end{align}
678: \end{widetext}
679: where $w'$, $\Omega'$, etc., represent derivatives of the functions $w(x_h)$ and $\Omega(w)$ with respect to their arguments.
680: We show plots of $c$ in Figs.~\ref{f:fuctFM} and \ref{f:fuctAF} (blue curves).
681:
682: In the unfrustrated case ($J_{aa}>0$, Fig.~\ref{f:fuctFM}),
683: the specific heat has a broad hump just above $T = J_{aa}$,
684: and a sharp peak
685: near $T = 2 |J_{ab}|$.
686: %at $T \sim |J_{ab}|$
687: The broad hump is due to ferromagnetic alignment within each $a$-plaquette.
688: The sharp peak corresponds to the phase transition to a ferromagnetic (for $J_{ab}>0$) or ferrimagnetic (for $J_{ab}<0$) state, governed by the weakest links, $J_{ab}$.
689: The position of the sharp peak is consistent with the
690: solution of Eqn.~\ref{e:phaseboundary}. (See also Fig.~\ref{zerofieldphasediagram}.)
691: In the frustrated case,
692: %there are still features at both of these energy scales, albeit broadened.
693: broadened features remain at both of these energy scales,
694: as shown in Fig.~\ref{f:fuctAF}.
695: %there are two broad features at the energy scales defined by $|J_{ab}|$ and $|J_{aa}|$, respectively.
696:
697:
698: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
699: \subsection{Zero-temperature limit: residual entropy \label{s:entropy}}
700:
701: Far from the critical curve, results for two limits can be obtained, corresponding to the
702: ordered phase and to the disordered phase (which persists even at zero temperature).
703:
704: In the first case, $J_{aa}/|J_{ab}| > -1$, the system
705: %EC2 is unfrustrated
706: orders at low temperatures, going into either the ferromagnetic state (for $J_{ab}>0$) or ferrimagnetic state
707: (for $J_{ab}<0$).
708: In the low temperature limit, the partition function and internal energy may be expanded as
709: %EC2 We have
710: \begin{align}
711: \ln Z(\beta)
712: &= 6 \ln x_{ab} - 3 \ln x_{aa} + \frac{6 {x_{aa}}^2}{{x_{ab}}^2} + \dotso,
713: \\
714: u(\beta)
715: &= -12|J_{ab}| + 6|J_{aa}|
716: + 24 e^{-4\beta |J_{ab}-J_{aa}|} |J_{ab}-J_{aa}|
717: + \dotso .
718: \end{align}
719: As $T \rightarrow 0$, the residual entropy
720: %EC2 (obtained by taking the limits in the opposite order)
721: is zero, whether in the ferromagnetic phase or the ferrimagnetic phase.
722: %\blue{The number 24 probably indicates that the degeneracy of the first excited state is 24 times the number of unit cells.}
723:
724:
725: Suppose $J_{aa}/|J_{ab}| < -1$. In this case, the model becomes frustrated
726: when $T\rightarrow 0$, $\beta \rightarrow \infty$. We have $x_{aa}, x_{ab}
727: \rightarrow \infty$, $x_h \rightarrow 1^-$, $w \rightarrow \infty$.
728: Expanding $\Omega(w) \approx \ln w$ in a series in $w$, and then expanding
729: $\ln Z$ as a series in $x_{aa}$ and $x_{ab}$, we find:
730: \begin{align}
731: \ln Z(\beta)
732: &= \ln 72 + \ln x_{aa} + 2 \ln x_{ab}
733: \nonumber\\&{}
734: + \frac{2}{{x_{ab}}^2} + \frac{1}{3{x_{aa}}^2} + \frac{{x_{ab}}^2}{6{x_{aa}}^2} + \dotso .
735: \end{align}
736: We can thus obtain the following low-temperature approximation for the energy per unit cell:
737: \begin{align}
738: u(\beta) &= -2|J_{aa}| - 4|J_{ab}|
739: + \tfrac2{3} e^{-4\beta |J_{aa}-J_{ab}|} |J_{aa}-J_{ab}|
740: \nonumber\\&{}
741: + \tfrac2{3} e^{-4\beta |J_{aa}+J_{ab}|} |J_{aa}+J_{ab}|
742: + \tfrac{4}{3} e^{-4\beta |J_{aa}|} |J_{aa}|
743: \nonumber\\&{}
744: + 8 e^{-4\beta |J_{ab}|} |J_{ab}|
745: + \dotso .
746: \end{align}
747: The first two terms are the ground state energy.
748: The other terms represent different types of excitatiotns about the ground
749: state(s). The coefficients represent relative degeneracies of excited states and ground states, and the exponents represent excitation energies.
750: %\blue{YL: I think it is quite interesting that fractional coefficients such as $2/3$ appear here. This seems to indicate that the degeneracy of the first excited state is smaller than that of the ground state.}
751: %\red{DY: Configurations with excited energy $|J_{aa}-J_{ab}|$, $|J_1+J_{ab}|$, etc. have different degeneracy that we may enumerate}
752:
753: The first term in the series expansion of $\ln Z$ gives the residual entropy per unit cell:
754: \begin{align}
755: s_0 &= \lim_{\beta J_{ab}\rightarrow -\infty} \lim_{\beta
756: J_{aa}\rightarrow -\infty}
757: \left( \ln Z + \beta u\right)
758: = \ln 72.
759: \end{align}
760: Thus the residual entropy is exactly $\ln 72 \approx 4.2767\dots$ per unit cell, or
761: $\frac{1}{9} \ln 72 \approx 0.4752\dots$ per site. This number will be discussed in more detail in Sec.~\ref{zerotemperature}.
762:
763: %\blue{To put this in context: When $J_{aa}=J_{ab}=0$, the unit cell has nine independent spins, so the degeneracy per unit cell is $\ln 2^9$. When $J_{aa} < 0$ but $J_{ab}=0$, the three $b$-spins in each cell are independent, and the two $a$-triangles each have six ground states, so the degeneracy per unit cell is $2^3 \times 6^2=288$. We have not been able to find such an explanation for a degeneracy of 72.}
764:
765:
766:
767:
768: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
769: %\subsection{Plots of thermodynamic functions}
770: %EC2 moved stuff here to previous sections.
771:
772:
773: %YL: I removed the figure fucbeta
774: %%%%%%% F I G U R E %%%%%%%%%%%%
775: \begin{figure}[htb]
776: \centering
777: {\resizebox*{0.47\textwidth}{!}{\includegraphics{fuctFM.eps}}}
778: \caption{(Color online). Thermodynamic functions vs temperature $T$ for unfrustrated (ferromagnetic) couplings $J_{aa}=5$ and $|J_{ab}|=1$. Red: free energy $f(T)=-k_B T\ln Z$. Green: energy $u(T)$. Blue: heat capacity $c(T)$. Yellow: entropy $s(T)$. All values quoted per unit cell; each unit cell contains 9 sites.
779: \label{f:fuctFM}}
780: \end{figure}
781: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
782:
783: %%%%%%% F I G U R E %%%%%%%%%%%%
784: \begin{figure}[htb]
785: \centering
786: {\resizebox*{0.47\textwidth}{!}{\includegraphics{fuctAF.eps}}}
787: \caption{(Color online). Thermodynamic functions vs temperature $T$ in frustrated regime, $J_{aa}=-5$ and $|J_{ab}|=1$, where the intratrimer coupling $J_{aa}$ is strong and antiferromagnetic.
788: \label{f:fuctAF}}
789: \end{figure}
790: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
791:
792:
793:
794:
795:
796: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
797: \section{Exact results at zero temperature \label{zerotemperature}}
798:
799:
800:
801: In this section we show how the zero-temperature phase diagram (along with the thermodynamic properties and correlations of the various phases) can be systematically deduced
802: %EC2
803: with and without applied field
804: by considering ground states of large triangular plaquettes.
805: %EC2 I think in fact we're general below.
806: %We will focus on the more interesting case, $J_{ab} < 0$, unless otherwise stated.
807: By explicitly comparing the ground state energies, we
808: derive the phase diagram.
809: The results of this section are summarized in
810: Figs.~\ref{phase_diag_Jab-1_T0} and \ref{phase_diag_Jab1_T0}.
811:
812: %==============================================================================
813: \subsection{Zero Field (phases V and VI)\label{sec:zerofield}}
814:
815:
816: %The Hamiltonian may be decomposed into a sum of terms for each large triangle....
817: %EC2 added an introductory paragraph, and changed the title of the section.
818: The phase diagram for zero applied field is shown in
819: Fig.~\ref{zerofieldphasediagram}.
820: For $J_{aa}>-|J_{ab}|$ and low temperature, the system is in an ordered phase
821: which is ferromagnetic for $J_{ab}>0$, and ferrimagnetic if $J_{ab}<0$.
822: For $J_{aa}<-|J_{ab}|$, the system remains disordered even at zero temperature,
823: with a residual entropy of $s_0={\rm ln 72}$.
824: The zero-field disordered phase is labeled phase V in
825: Figs.~\ref{phase_diag_Jab-1_T0} and \ref{phase_diag_Jab1_T0},
826: and the zero-field ordered phase is labeled phase VI.
827:
828: The degeneracy of the ground state manifold can be understood
829: by considering the energetics of a single large plaquette, {\em i.e.},
830: an $a$-spin trimer along with its enclosing $b$-spin trimer.
831: Representative plaquette configurations within the ground state are shown in Fig.~\ref{zeroTzeroh}.
832: We enumerate all possible plaquette energies in Table~\ref{energytable}.
833: %We see that for each of the $8$ configurations of the $b$-triangle, there are exactly $3$ configurations of the $a$-triangles that have the lowest energy,
834: %$E = -\abs{J_{aa}} - 2\abs{J_{ab}} $ (see Fig.~\ref{zeroTzeroh}).
835: As can be seen from the table, for any of the $2^3=8$ possible configurations of the $b$-trimer,
836: there are three and only three configurations of the enclosed $a$-trimer
837: which are all within the ground state.\cite{fnote} This means that the $b$-spins are effectively
838: free within the ground state manifold.
839:
840: In order to count the ground state degeneracy, we now turn to the unit cell.
841: Since the $b$-spins are free, and there are $3$ $b$-spins per unit cell,
842: this contributes $2^3 = 8$ configurations per unit cell to the ground state manifold.
843: For any given configuration of the $b$-spins, each $a$-trimer in the lattice has a $3$-fold degeneracy.
844: Since there are two $a$-trimers per unit cell, these contribute a factor of $3^2 = 9$
845: to the ground state degeneracy.
846: The total degeneracy per unit cell in the ground state is therefore $8\times 9 = 72$,
847: as we showed in Sec.~\ref{s:entropy}.
848: %%YL : removed "integer-valued"
849: %Thus we can see that the physical origin of
850: %the integer-valued entropy per unit cell, $s_0 = \ln 72$,
851: %EC2 = 4.2767...$.
852: %is that the $b$ spins are effectively free.
853: %Within the ground state manifold, the b-spins are effectively free
854: %(as will be shown below).
855: %The energy per unit cell is of course $u= - 2\abs{J_{aa}} - 4\abs{J_{ab}} $.
856: %EC2This analysis shows that the effective coupling between $b$-spins on the TKL is zero. In other words, every $b$-spin is independent. The $b$-spins become completely decoupled. This can be seen by studying Eq.~(4) of Ref.~\onlinecite{zheng05}, which is the effective coupling on the kagome lattice $\beta J_{bb}$ as a function of $\beta J_{aa}$ and $\beta J_{ab}$. In the frustrated regime, $J_{aa}/|J_{ab}| < -1$, one finds that $\beta J_{bb}$ tends to zero at zero temperature -- again the paradox that ``stronger is weaker''.
857:
858: %%YL : changed a bit
859: The fact that the $b$-spins are effectively independent also means that the correlation function is ``perfectly localized'': it is exactly zero beyond a distance $r_{bb}$ (the distance between two $b$-spins). The correlation length $\xi$ is thus zero (where $\xi$ is defined as the asymptotic decay length of the correlation function at \emph{large} distances).
860:
861: For comparison, at $T=h=0$, the triangular Ising AF has power-law correlations.
862: The kagome Ising AF is more frustrated than the triangular lattice case, since its ground state
863: has exponentially decaying correlations.
864: %--this behavior is referred to in the literature as ``super-frustrated''.
865: We have shown here that the ground state of the TKL Ising AF in the frustrated regime has perfectly localized correlations,
866: making this model even more frustrated than either the triangular or kagome cases.
867: %``{\em hyper-}frustrated''.
868:
869:
870: %%%%%%%%%% TABLE OF GROUND STATE ENERGIES OF LARGE TRIANGLE
871: \begin{table}[htb]
872: %
873: \[
874: \begin{array}{cc}
875: & \sigma _a \\
876: \sigma _b &
877: \begin{array}{c|cccccccc}
878: & \uparrow \uparrow \uparrow & \downarrow \uparrow \uparrow & \uparrow \downarrow \uparrow & \uparrow \uparrow \downarrow & \uparrow \downarrow
879: \downarrow & \downarrow \uparrow \downarrow & \downarrow \downarrow \uparrow & \downarrow \downarrow \downarrow \\
880: \hline
881: \uparrow \uparrow \uparrow & \text{ 15} & \text{ -1} & \text{ -1} & \text{ -1} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \text{ 3} \\
882: \downarrow \uparrow \uparrow & \text{ 11} & \textbf{\blue{ -5}} & \text{ -1} & \text{ -1} & \text{ -1} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \text{ 7} \\
883: \uparrow \downarrow \uparrow & \text{ 11} & \text{ -1} & \textbf{\blue{ -5}} & \text{ -1} & \textbf{\blue{ -5}} & \text{ -1} & \textbf{\blue{ -5}} & \text{ 7} \\
884: \uparrow \uparrow \downarrow & \text{ 11} & \text{ -1} & \text{ -1} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \text{ -1} & \text{ 7} \\
885: \uparrow \downarrow \downarrow & \text{ 7} & \text{ -1} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \text{ -1} & \text{ -1} & \text{ 11} \\
886: \downarrow \uparrow \downarrow & \text{ 7} & \textbf{\blue{ -5}} & \text{ -1} & \textbf{\blue{ -5}} & \text{ -1} & \textbf{\blue{ -5}} & \text{ -1} & \text{ 11} \\
887: \downarrow \downarrow \uparrow & \text{ 7} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \text{ -1} & \text{ -1} & \text{ -1} & \textbf{\blue{ -5}} & \text{ 11} \\
888: \downarrow \downarrow \downarrow & \text{ 3} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \textbf{\blue{ -5}} & \text{ -1} & \text{ -1} & \text{ -1} & \text{ 15}
889: \end{array}
890: \end{array}
891: \]
892: \caption{
893: Energy of a large triangle (consisting of 3 $b$-spins and 3 $a$-spins) as a function of the configurations of $a$ and $b$ spins. For clarity of presentation, we present the results for the case $J_{aa}=-3$, $J_{ab}=-1$, $h=0$, but the form of the table is representative of the entire line $J_{aa}/|J_{ab}| < -1$. The boldfaced numbers indicate the lowest-energy configurations, which have energy $J_{aa}+2J_{ab}$. Note that each row contains exactly three boldfaced numbers. See text for discussion.
894: \label{energytable}}
895: \end{table}
896: %%%%%%%%%% END TABLE
897: %%%%%%%%%% FIGURE ILLUSTRATING DECIMATION OF a-spins AT T=0
898: \begin{figure}[htb]
899: {\centering
900: \subfigure[All three $b$-spins pointing up]
901: {\resizebox*{0.9\columnwidth}{!}{\includegraphics{zeroTzerohdecimation.eps}\label{fig:lattice}}}
902: \subfigure[Two of three $b$-spins pointing up]
903: {\resizebox*{0.9\columnwidth}{!}{\includegraphics{zeroTzerohdecimation2.eps}\label{fig:bond}}}
904: \par}
905: \caption{(Color online). The ln72 phase.
906: Given a configuration of the three $b$-spins (on the outer triangle), it can be shown, by enumerating all possibilities, that there are exactly three states of the $a$-spins (on the inner triangle). The figure illustrates this for two configurations of $b$-spins; results for the other configurations can be seen from the Ising symmetry and the local rotational symmetry
907: of the triangular plaquettes.
908: \label{zeroTzeroh}}
909: \end{figure}
910: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
911:
912:
913: %==============================================================================
914: \subsection{Saturated ferromagnetic phase (phase I)}
915: %EC2
916: %The ``decimation'' process above can be repeated for other combinations of parameters. (When calculating the energy of the large triangular unit, we count the Zeeman energy of the $b$-spins in the magnetic field as $-h\sigma_b/2$, because each $b$-spin is shared by two large triangles.)
917:
918: At very high fields, when $h/|J_{ab}| > \max(4, 2|J_{aa}|/|J_{ab}|+2)$, we find that there is a unique ground state where all the $b$-spins and $a$-spins are up. This state is easily seen to have magnetization $m=9$, entropy $s=0$, and energy $u= - 3h + 6\abs{J_{aa}} + 12\abs{J_{ab}} $ per unit cell.
919:
920: %*** How do we find it? Explicit enumeration of ground states? Direct energy comparison of ground states?
921: %How is the phase boundary line calculated?
922: %Will this all be clear if we show in Table I the full Ja, Jb, and h-dependence?
923: %***
924:
925: %==============================================================================
926: \subsection{Ferrimagnetic phase (phase II)}
927: At lower fields $0 < h/|J_{ab}| < 4$ and when $J_{aa}/|J_{ab}| > -1$, there is a ferrimagnetic phase: the $a$-spins, being more numerous, align parallel to the field (up); the $b$-spins are then induced to point down due to the antiferromagnetic $J_{ab}$ interaction. This phase has
928: $m=3$, $s=0$, and $u= - 3h + 6\abs{J_{aa}} - 12\abs{J_{ab}}$ per unit cell.
929:
930: %*** Same question.***
931:
932: %==============================================================================
933: \subsection{Log 9 phase (phase III)}
934: If $4 < h/|J_{ab}| < 2|J_{aa}|/|J_{ab}|+2$, the $b$-spins are completely polarized, but each $a$-triangle has 3 degenerate states (see Fig.~\ref{threestatepotts}).
935: Therefore, in this phase the system is equivalent to a set of non-interacting three-state Potts spins.
936: This phase has $m=5$, $s=\ln 9=2.1972\dots$, and $u= - 5h - 2\abs{J_{aa}} + 4\abs{J_{ab}}$ per unit cell. Again, the correlation function is perfectly localized, and the correlation length is $\xi=0$.
937:
938:
939: %%%%%%% F I G U R E %%%%%%%%%%%%
940: \begin{figure}[htb]
941: \centering
942: {\resizebox*{0.47\textwidth}{!}{\includegraphics{threestatepotts.eps}}}
943: \caption{(Color online). Phase III of the TKL Ising model. When the field is quite strong, the $b$-spins (outer triangle) are all polarized, and each $a$-trimer (inner triangle) has three degenerate ground states.
944: \label{threestatepotts}}
945: \end{figure}
946: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
947:
948:
949: %==============================================================================
950: \subsection{Dimer phase (phase IV)}
951: The most interesting situation occurs when $0 < h/|J_{ab}| < 4$ and $J_{aa}/|J_{ab}| < -1$.
952: %EC2
953: Table~\ref{energytable} shows that
954: the system will have the lowest energy if each
955: $b$-trimer has exactly one $b$-spin pointing down.
956: Counting the number of ways to satisfy this constraint globally is a non-trivial problem. The situation is the same as that for a kagome Ising AF at $T=0$ and $0<h<4|J_\text{kagome}|$, which has been studied before\cite{moessner2000,moessner2001,udagawa2002,moessner2003}.
957: The down-$b$-spins behave like a lattice gas on the kagome lattice with nearest-neighbor exclusion, at maximal density.
958: The ground states can be mapped to configurations of dimers occupying a honeycomb lattice (see Fig.~\ref{honeydimer}).
959: This problem has been solved exactly using the Pfaffian method\cite{kasteleyn1963}; the entropy per unit cell is
960: \begin{align}
961: & \frac{1}{8\pi^2} \int_0^{2\pi} dp \int_0^{2\pi} dq ~ \ln \left(
962: 1 - 4\cos p \cos q + 4\cos^2 q
963: % 3 + 2\cos p + 2\cos q + 2\cos (p+q)
964: \right)
965: \nonumber\\
966: &= 0.3231\dots .
967: \end{align}
968: Therefore, the entropy of the TKL phase IV is also $0.3231\dots$ per unit cell, or $0.03590\dots$ per site.
969: %YL3: A sentence here was slightly confusing and slightly wrong. I've moved it to a better place.
970:
971: %%%%%%% F I G U R E %%%%%%%%%%%%
972: \begin{figure}[htb]
973: \centering
974: {\resizebox*{0.47\textwidth}{!}{\includegraphics{honeydimer.eps}}}
975: \caption{(Color online). Phase IV of the TKL Ising model: Mapping of the spin configurations to configurations of dimers on a honeycomb lattice. The yellow patches represent defects in the spin configuration, which correspond to vacancies (or monomers) in the dimer picture.
976: \label{honeydimer}}
977: \end{figure}
978: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
979:
980: The correlation function of the $b$-spins, $C_{bb} (\mathbf{r}) = \mean{\sigma_b(\mathbf{0})\sigma_b(\mathbf{r})}$, is equivalent to the dimer-dimer correlation function. This correlation function decays as a power law, $1/r^2$, and nice visualizations in real space and reciprocal space are given in Ref.~\onlinecite{moessner2003}. Thus, the model is in a critical phase, in the ``Villain-Stephenson universality class''. The magnetization per unit cell is $m=3$ and the energy per unit cell is $u= - 3h - 2\abs{J_{aa}} - 4\abs{J_{ab}}$.
981:
982:
983:
984: %==============================================================================
985: \subsection{Phase diagram}
986:
987: We now combine the above results in order to report the full
988: phase diagram of the TKL Ising model.
989: %The partition function is invariant under a change of sign of $h$, so
990: Fig.~\ref{phase_diag_Jab-1_T0} shows the phase diagram for
991: antiferromagnetic intertrimer coupling, $J_{ab}<0$,
992: and in Fig.~\ref{phase_diag_Jab1_T0} we show the phase diagram for
993: ferromagnetic intertrimer coupling, $J_{ab}>0$.
994: In both cases, the phase diagram is symmetric under
995: for $h\rightarrow -h$, with simultaneous change of sign of all spins.
996: %Figure~\ref{phase_diag_Jab-1_T0} shows the case of $J_{ab}<0$, which is more interesting, as well as being more relevant to experiment.
997: %EC2 we may need this earlier, but it doesn't seem to fit here.
998: %At $T=0$, the equilibrium phase is the one with the lowest energy density; it may be verified that the phase boundaries are the lines at which the energies of the adjacent phases become equal.
999: The entropy and magnetization change discontinuously across every
1000: zero-temperature phase boundary. Exactly on the phase boundaries and
1001: intersections of these boundaries, the entropy will be higher than in either adjacent
1002: phase, because the system can choose from states within each set of ground
1003: states.
1004:
1005: Note that the high-field phases (I and III) are common to both phase diagrams.
1006: The more interesting case is that of
1007: Fig.~\ref{phase_diag_Jab-1_T0},
1008: which has antiferromagnetic
1009: intertrimer coupling $J_{ab}<0$, and more phases at intermediate field strength.
1010: %The $h=0$ boundary between IV and
1011: %its time-reversed pair IV' deserves special mention.
1012: %EC2 We shall call it phase V. As calculated in Sec.~\ref{zerofield}, it has an entropy of $\ln 72$ per unit cell.
1013: %%YL: removed "hyper-frustrated"
1014: Right at $h=0$ in both phase diagrams, the ground state is phase V, the spin liquid with
1015: residual entropy $s_0= {\rm ln 72}$ that we discussed in Sec.~\ref{sec:zerofield}.
1016: When $J_{ab}<0$, the application of an infinitesimal field
1017: induces a critical state with power law correlation functions (phase IV),
1018: which we have mapped to the problem of hard-core dimers on a honeycomb lattice.
1019:
1020:
1021:
1022: %EC2 Are we allowed to directly quote in a journal?
1023: %The following observation by Moessner and Sondhi\cite{moessner2000,moessner2001} for the kagome Ising AF in field is equally relevant here: ``This result is rather unusual. Starting from a disordered magnet, we obtain a critical state with nonzero moment and extensive entropy upon application of an infinitesimal field; these properties, including the criticality, persist for a finite range of field strengths.''
1024:
1025: %EC2
1026: %Figure~\ref{phase_diag_Jab1_T0} shows the phase diagram as a function of
1027: %$J_{aa}/|J_{ab}|$ and $h/|J_{ab}|$ in the less interesting case where $J_{ab}>0$. This is mostly the same as Figure~\ref{phase_diag_Jab-1_T0}, except that states II and IV are no longer present.
1028:
1029:
1030: %%%%%%% F I G U R E %%%%%%%%%%%%
1031: \begin{figure}[!t]
1032: %\psfrag{h}{\LARGE $h/\abs{J_{ab}}$}
1033: \psfrag{h}{\LARGE $\frac{h}{\abs{J_{ab}}}$}
1034: \psfrag{Jaa}{\Large $J_{aa}/\abs{J_{ab}}$}
1035: \psfrag{Hc}{$h=2\abs{J_{aa}}+2\abs{J_{ab}}$}
1036: \psfrag{+Sat}{$\begin{array}{l} m=9 \\ s=0 \\
1037: \end{array}$}
1038: \psfrag{+Ferri}{$\begin{array}{l} \text{Ferrimagnetic} \\ m=3 \\ s=0 \\
1039: \end{array}$}
1040: \psfrag{+Log9}{$\begin{array}{l}
1041: \\ m=5 \\ s=\ln 9 =2.1972 \\
1042: \end{array}$}
1043: \psfrag{+Dimer}{$\begin{array}{l} \text{Critical dimer phase} \\ m=3 \\ s=0.3231
1044: \end{array}$}
1045: \psfrag{ln72}{$\begin{array}{l} \displaystyle\mathbf{V} \\ m=0 \\ s=\ln 72=4.2767 \end{array}$}
1046: \psfrag{LRO}{$\begin{array}{l} \displaystyle\mathbf{VI} \end{array}$}
1047: \centering
1048: {\resizebox*{0.47\textwidth}{!}{\includegraphics{phase_diag_Jab-1_T0.eps}}}
1049: \caption{%EC3 *** We can say $Jab<0$, rather than $Jab=-1$, since it's scaled. Same in next figure.***
1050: (Color online). Phase diagram of the TKL Ising model in the $(J_{aa},h)$ plane, for
1051: $J_{ab}<0$ (antiferromagnetic intertrimer coupling) and $T=0$. The phase diagram is symmetric under a change of sign
1052: of $h$.
1053: The phases I', II', III', IV' are just mirror images of I, II, III, IV obtained by swapping up and down spins. The yellow region represents the critical phases with power-law correlations; note that it does not include the thin line at $h=0$, which is phase V, the $\ln 72$ phase described in the text. The thick line (VI) persists as a true ferrimagnetic phase transition at finite $T$; all other lines turn into crossovers.
1054: %YL3: changed a few things
1055: \label{phase_diag_Jab-1_T0}}
1056: \end{figure}
1057: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
1058:
1059:
1060:
1061:
1062:
1063: %%%%%%% F I G U R E %%%%%%%%%%%%
1064: \begin{figure}[htb]
1065: %\psfrag{h}{\LARGE $h/\abs{J_{ab}}$}
1066: \psfrag{h}{\LARGE $\frac{h}{\abs{J_{ab}}}$}
1067: \psfrag{Jaa}{\Large $J_{aa}/\abs{J_{ab}}$}
1068: \psfrag{Hc}{$h=2\abs{J_{aa}}-2\abs{J_{ab}}$}
1069: \psfrag{ferro}{$\begin{array}{l} \text{Ferrimagnetic} \\ m=3 \\ s=0 \\
1070: \end{array}$}
1071: \psfrag{ln9}{$\begin{array}{l}
1072: \\ m=5 \\ s=\ln 9 =2.1972 \\
1073: \end{array}$}
1074: \psfrag{ln72}{$\begin{array}{l} \displaystyle\mathbf{V} \\ m=0 \\ s=\ln 72=4.2767 \end{array}$}
1075: \psfrag{LRO}{$\begin{array}{l} \displaystyle\mathbf{VI} \end{array}$}
1076: \centering
1077: {\resizebox*{0.47\textwidth}{!}{\includegraphics{phase_diag_Jab1_T0.eps}}}
1078: \caption{(Color online). Phase diagram of the TKL Ising model in the $(J_{aa},h)$ plane, for $J_{ab}>0$ (ferromagnetic intertrimer coupling) and $T=0$.
1079: \label{phase_diag_Jab1_T0}}
1080: \end{figure}
1081: %%%%%%% E N D F I G U R E %%%%%%%%%%%%
1082:
1083:
1084:
1085:
1086:
1087:
1088: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1089: \section{Finite temperature and finite field\label{finiteTfiniteh}}
1090:
1091:
1092: %YL3: Z2 -> Z_2
1093: Most of the phase transitions in the zero-temperature phase diagrams
1094: are destroyed by thermal fluctuations.
1095: The clear exception is phase VI, which has spontaneously broken
1096: $Z_2$ symmetry. This long-range ordered phase survives at finite temperature,
1097: and has a true phase transition at a Curie temperature $T_c\ne0$.
1098: Since $h$ is a relevant perturbation, this finite-temperature phase transition
1099: is destroyed at any finite field, leaving only a crossover.
1100:
1101: %First consider the transition between states II and II' at $h=0$ (the thick line in Fig.~\ref{phase_diag_Jab-1_T0}).
1102: %This transition corresponds to the spontaneous breaking of Ising symmetry, with a ferrimagnetic order parameter. Here, $h$ is a relevant perturbation.
1103: %%\blue{ (unlike in a square antiferromagnet where $h$ is irrelevant).}
1104: %Therefore, in the three-dimensional parameter space, the first-order critical surface is restricted to the $T$--$J_{aa}$ plane, and it is bounded above by the second-order critical curve (Fig.~\ref{zerofieldphasediagram} and Eq.~\eqref{e:phaseboundary}).
1105:
1106: The other transitions in
1107: Figs.~\ref{phase_diag_Jab-1_T0} and \ref{phase_diag_Jab1_T0}
1108: are not characterized by a \emph{spontaneously} broken symmetry with an order parameter.
1109: %\blue{(unlike the case of the triangular Ising AF, where applying a field induces a spontaneous breaking of translational symmetry).}
1110: Therefore, they cannot persist at finite $T$ as traditional order-disorder transitions.
1111: However, a more subtle analysis is required to understand whether the critical phase, phase IV, persists at finite $T$ (bounded, e.g., by a Kosterlitz-Thouless transition curve).
1112: The situation is quite similar to that for the kagome Ising AF described in Ref.~\onlinecite{moessner2003}.
1113: Based on the table of ground state energies, Table~\ref{energytable},
1114: it is possible to enumerate the types of defects that can occur in phase IV.
1115: These defects correspond to breaking a dimer into two monomers (see Fig.~\ref{honeydimer}), and can only be created in pairs.
1116: Each defect has energy $\Delta U = \min(h, 4|J_{ab}|-h)$. The entropy $\Delta S$ associated with creating a defect pair is at least $O(\ln L^2)$, because one may choose to break any of the $O(L^2)$ dimers; in fact the entropy is even greater than this because the resulting monomers can be moved apart, resulting in many new configurations. Hence at any finite $T$, the density of monomers is finite. A theorem of Lieb and Heilman\cite{heilman1972} states that monomer-dimer models cannot have phase transitions at any finite density of monomers. Therefore, \emph{the critical phase only exists at $T=0$, and it is destroyed at finite $T$}.
1117:
1118: However, if the density of monomers is low, the correlation length $\xi$ may still be long; na\"ively, one might expect $\xi^2 \sim 1/n_\text{monomers} \propto \exp (E_\text{monomer}/T)$, but a more careful treatment accounting for the effective Coulomb attraction between monomers gives the result $\xi^2 \sim 1/n_\text{monomers} \propto \exp (8E_\text{monomer}/7T)$.\cite{moessner2003}
1119: This scenario is consistent with the size-dependent peaks in the susceptibility that we find from Monte Carlo simulations (see Sec.~\ref{montecarlo}).
1120:
1121:
1122:
1123: %&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
1124: \section{Monte Carlo Simulations \label{montecarlo}}
1125:
1126:
1127: %%%%%%%%%%% FIGURE %%%%%%%%%%%%%%%%%%
1128: \begin{figure}[t]
1129: {\centering
1130: \subfigure[~Heat capacity\label{fm:ct}]{\resizebox*{0.8\columnwidth}{!}{\includegraphics{fm_ct.eps}}} \par}
1131: {\centering
1132: \subfigure[~Susceptibility\label{fm:xt}]{\resizebox*{0.8\columnwidth}{!}{\includegraphics{fm_xt.eps}}}
1133: \par}
1134: \caption{(Color online).Temperature dependence of heat capacity and
1135: susceptibility from Monte Carlo simulation for $J_{aa}=5$ and $J_{ab}=+1$ with system
1136: sizes $L=12,24,36,60$. Red line (no dots) represents the exact solution of specific
1137: heat. Peaks become taller and narrower as $L$ increases.}
1138: \label{fm}
1139: \end{figure}
1140: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1141: %%%%%%%%%%% FIGURE %%%%%%%%%%%%%%%%%%
1142: \begin{figure}[t]
1143: {\centering
1144: \subfigure[~Specific heat\label{af1:ct}]{\resizebox*{0.8 \columnwidth}{!}{\includegraphics{af1_ct.eps}}} \par}
1145: {\centering
1146: \subfigure[~Inverse susceptibility ($1/\chi(T)$)\label{af1:ix}]{\resizebox*{0.8\columnwidth}{!}{\includegraphics{af1_ixt.eps}}}
1147: \par}
1148: {\centering
1149: \subfigure[~$T\chi(T)$\label{af1:tx}]{\resizebox*{0.8\columnwidth}{!}{\includegraphics{af1_tx.eps}}}
1150: \par}
1151: \caption{(Color online). Temperature dependence of heat capacity and inverse susceptibility ($1/\chi(T)$) and $T\chi(T)$ from Monte Carlo simulation for $J_{aa}=-10$ and $J_{ab}=-1$ with system sizes $L=12,24,36,60$.}
1152: \label{af1}
1153: \end{figure}
1154: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1155: %%%%%%%%%%% FIGURE %%%%%%%%%%%%%%%%%%
1156: \begin{figure}[t]
1157: {\centering
1158: \subfigure[~Specific heat\label{af2:ct}]{\resizebox*{0.8 \columnwidth}{!}{\includegraphics{af2_ct.eps}}} \par}
1159: {\centering
1160: \subfigure[~Inverse susceptibility ($1/\chi(T)$) \label{af2:ix}]{\resizebox*{0.8\columnwidth}{!}{\includegraphics{af2_ixt.eps}}}
1161: \par}
1162: {\centering
1163: \subfigure[~$T\chi(T)$ \label{af2:tx}]{\resizebox*{0.8\columnwidth}{!}{\includegraphics{af2_tx_add.eps}}}
1164: \par}
1165: \caption{(Color online). Temperature dependence of heat capacity, inverse susceptibility ($1/\chi(T)$) and $T\chi(T)$ from Monte Carlo simulation for $J_{aa}=-10$ and $J_{ab}=1$ with system sizes $L=12,24,36,60$.}
1166: %YL3: changed af2_tx.eps to af2_tx_add.eps
1167: \label{af2}
1168: \end{figure}
1169: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1170:
1171:
1172: In this section we present the results of Monte Carlo (MC) simulations of the TKL
1173: Ising model, Eq.~(\ref{e:hamiltonian}), for various combinations of parameters.
1174: The simulations corroborate our analytic predictions
1175: and also allow us to perform calculations at finite $h$ and $T$
1176: as well as to compute the magnetization and susceptibility.
1177:
1178:
1179: %DY2
1180: We use the Wolff algorithm for $J_{aa}>0$ at $h=0$, and the Metropolis
1181: algorithm for $J_{aa}<0$ at $h=0$ and $h \neq 0$. The system sizes we use are
1182: $L=12, 24, 36, 60$, where $L$ is the
1183: %EC2 size
1184: length of the underlying triangular
1185: lattice, so that the total number of spins is $N = \frac{9}{16} L^2$; for
1186: periodic boundary conditions, $L$ should be a multiple of 3 in order to avoid
1187: introducing boundary defects and additional frustration.
1188:
1189: In order to evaluate the heat capacity $C$ and magnetic susceptibility $\chi$,
1190: we use the fluctuation-dissipation theorem:
1191: \begin{align}
1192: C &=\frac{\mean{H^2} - \mean{H}^2}{NT^2},\\
1193: \chi &=\frac{\mean{M^2} - \mean{M}^2}{NT}, \label{chi}
1194: \end{align}
1195: where $\mean{H}$ and $\mean{M}$ are the Monte Carlo averages of the
1196: total energy (i.e., the Hamiltonian) and magnetization, respectively.
1197: %Note that H includes the h.M Zeeman energy
1198: %, and the Boltzmann constant is set to unity.
1199: We define the sublattice magnetizations as
1200: \begin{equation}
1201: m_a= \frac{1}{N_a}\ \sum_{i \in a} \sigma_{ai}~,
1202: \end{equation}
1203: \begin{equation}
1204: m_b= \frac{1}{N_b}\ \sum_{i \in b} \sigma_{bi}~,
1205: \end{equation}
1206: where $N_a$ is the number of $a$-spin sites,
1207: and $N_b$ is the number of $b$-spin sites.
1208:
1209: %Order parameter $A$ is defined as
1210: %\begin{equation}
1211: % A=\frac{1}{2}\ (m_a^2 + m_b^2)
1212: %\end{equation}
1213: %which has $C_3$ symmetry.
1214:
1215:
1216:
1217: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1218: \subsection{Zero magnetic field} %Antiferromagnetic $J_{aa}$}
1219: %DY2
1220:
1221:
1222: We first show Monte Carlo results at zero field.
1223: %EC2 Fig.~\ref{af}
1224: Fig.~\ref{fm} shows the
1225: temperature evolution of the heat capacity and susceptibility
1226: in a representative unfrustrated case, $J_{aa} = 5 J_{ab} >0$.
1227: %EC2 Why |M| in the algorithm? Either explain, or don't mention.
1228: %Susceptibility is calculated
1229: %using Eq.~(\ref{chi}) but with absoulte value of M in the Wolff algorithm when $J_{aa}>0$.
1230: The Monte Carlo results for the heat capacity are consistent with the exact results in
1231: Sec.~\ref{zerotemperature}; the peak in the heat capacity
1232: in our simulations becomes taller
1233: and narrower as $L$ increases, tending towards the exact solution for
1234: $L=\infty$.
1235: %EC2, see Fig.~\ref{fm:ct}.
1236:
1237: Figs.~\ref{af1} and
1238: ~\ref{af2} show results for two frustrated parameter combinations
1239: ($J_{aa}/|J_{ab}| < -1$).
1240: In both cases, the susceptibility shows a marked difference from the
1241: case of ferromagnetic intratrimer coupling, $J_{aa}>0$.
1242: For $J_{aa}>0$, the susceptibility shows a sharp peak at the Curie
1243: temperature $T_c$, whereas for $J_{aa}<0$,
1244: it tends to $\infty$ as $T\rightarrow 0$.
1245:
1246:
1247: Unlike the free energy, heat capacity, entropy, and internal energy,
1248: the susceptibility depends on the sign of $J_{ab}$.
1249: %, since it is a response to a relevant field.
1250: %YL3: more like, "since it is a response to a symmetry-breaking field".
1251: Our predictions about the
1252: susceptibility can be used to distinguish whether a physical TKL Ising
1253: system has ferromagnetic or antiferromagnetic coupling between the $a$
1254: and $b$ sublattices. For $J_{ab}>0$, the inverse susceptibility
1255: $1/\chi(T)$ shows two linear pieces, with a crossover at the lower
1256: coupling constant $J_{ab}$. However, when $J_{ab}<0$, the inverse
1257: susceptibility approaches zero much slower than the $J_{ab}>0$ case.
1258:
1259: The difference between antiferromagnetic and ferromagnetic
1260: coupling $J_{ab}$ is even more evident in the plots of $T \chi$. At low temperatures,
1261: $T\chi$ tends to a small finite constant as $T \rightarrow 0$ if the intertrimer coupling
1262: is antiferromagnetic, $J_{ab}<0$; whereas, if the intertrimer coupling
1263: is ferromagnetic, $J_{ab}>0$, $T\chi$ goes through a minimum around $T\sim J_{aa}$, increases sharply, and saturates at a finite value as $T\rightarrow 0$.
1264: %In this case,
1265: %$T \chi(T)$ also has a minimum.
1266: %EC2 we can put this in the discussion section.
1267: %From the profile of
1268: %inverse susceptibility, $J_{ab}<0$ looks more like the experimental
1269: %data on the $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$ materials.\cite{haar95} However, it does not give a minimum for $T \chi(T)$.
1270:
1271:
1272: %This is
1273: %one reason why $J_{ab}>0$ is used by some papers. However, we can not
1274: %exclude that in quantum case, antiferromagnetic $J_{ab}$ can explain $T \chi(T)$.
1275:
1276:
1277:
1278:
1279:
1280:
1281:
1282:
1283:
1284:
1285: %YL: I took this out because (i) we're not showing a graph to back this up, (ii) there doesn't seem to be any point in using our MC results to confirm other people's exact result, especially as we have the same exact result ourselves.
1286: %When $J_{aa}=J_{ab}$, we find that $\tilde{T}_c=1.88\pm 0.02$ which is
1287: %consistent with Ref.~\onlinecite{zheng05}. It is less than the
1288: %critical temperatures of the square,~\cite{onsager44}
1289: %triangular,~\cite{takagi95} and kagome~\cite{takagi93} lattices
1290: %($2.27$, $3.65$, and $2.15$ respectively), probably because of the
1291: %more dilute connectivity of the TKL.)
1292:
1293:
1294:
1295:
1296: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1297: \subsection{Finite magnetic field} %Ferromagnetic $J_{aa}$}
1298:
1299:
1300: %DY2
1301: %A magnetic field lifts degeneracy and changes the magnetic ordering.
1302: In Figs.~\ref{mh} and \ref{xh}, we show Monte Carlo results at finite field $h$ and low temperature $T=0.1$ for $J_{aa}=-2$ and $J_{ab}=-1$.
1303: The magnetization curves in Fig.~\ref{mh} have a series of steps and plateaux (as is typical of frustrated spin systems).
1304: Starting from $h=0$, as a field is applied, both the $a$ and $b$ sublattices immediately respond
1305: as the critical dimer phase is induced,
1306: developing (normalized) sublattice magnetizations of $m_a=m_b=1/3$.
1307: As field is increased, the $b$-spins are more easily polarized, while the $a$-sublattice spins only become fully polarized when the magnetic field is strong enough.
1308:
1309:
1310:
1311: The susceptibility as a function of applied field has a series of peaks, as shown in Fig.~\ref{xh}.
1312: %EC2 made this crystal clear - is it correct?
1313: %YL3: No, that was not correct. Sorry. I've edited it. ... Actually this is all very subtle and we can't make too strong statements... sigh...
1314: %Had the transitions at $h=0$ and $h=6$ persisted at finite temperature,
1315: %the susceptibility peaks at these points would not have grown with increasing
1316: %system size. %DY3 they are size independent
1317: %EC2 We need to argue this better if it stays in.
1318: %In particular, for $h=\pm 6$,
1319: %the susceptibility curves are independent of system size because
1320: %the correlation length $\xi$ for both phases is zero or very short (temperature
1321: %effect).
1322: % These results are consistent with our previous phase analysis.
1323: The peaks at $h=\pm 4$ increase with increasing system size from $L=12$ to $L=24$.
1324: This indicates that the correlation length (in the vicinity of the peaks) is comparable to, or larger than, the system size. This does not indicate a true finite-temperature phase transition. Rather, because phase IV is critical (with infinite correlation length), the correlation length diverges
1325: as temperature is lowered toward this phase. At the low temperatures we have simulated,
1326: the correlation length is comparable to our system sizes. See Sec.~\ref{finiteTfiniteh}.
1327: %YL3: (The peaks at $h=0$ and $h=6$ appear to be size-independent, but the converse of the above logic does not hold: lack of size dependence in a particular thermodynamic quantity cannot be taken as evidence that correlations are short-ranged. TIAF is a case in point --- there are true transitions at finite T, but one of them shows size dpenedence and the other does not. I have results regarding this --- probably somewhere in the literature, but I havne't foudn them.....
1328: %YL3: The peak height changes by a factor of 3 in going from L=12 to L=24... can't really call that a weak size dependence. I Removed the word "weak"
1329:
1330: %%%%%%%%%%%%%FIGURE%%%%%%%%%%%%%%%
1331: \begin{figure}[!t]
1332: \centering
1333: \includegraphics[width=8cm,clip]{mh.eps}
1334: \caption{(Color online). Magnetization $m$ vs. external field $h$ from Monte Carlo simulation for $J_{aa}=-2$, $J_{ab}=-1$ at
1335: $T=0.1$ for $L=12$. Sublattice magnetizations $m_a$ and $m_b$ are also shown.}
1336: \label{mh}
1337: \end{figure}
1338: %%%%%%%%%%%%%FIGURE%%%%%%%%%%%%%%%
1339: \begin{figure}[!t]
1340: \centering
1341: \includegraphics[width=8cm,clip]{xh.eps}
1342: \caption{\label{xh}
1343: (Color online). Susceptibility $\chi$ vs. external field $h$ from Monte Carlo simulation for $J_{aa}=-2$, $J_{ab}=-1$ at
1344: $T=0.1$ for $L=12 $. }
1345: \end{figure}
1346: %%%%%%%%%%%%%END FIGURE%%%%%%%%%%%%%%%
1347:
1348:
1349:
1350: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1351: \section{Discussion \label{discussion}}
1352: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1353: %EC2 completely rewritten -- mostly to re-order things. Cut things are at the end of this section.
1354:
1355:
1356: Phases I,II,and III can easily be simulated using the Metropolis algorithm. However, phase IV has extremely slow dynamics when simulated using the Metropolis algorithm at low temperature -- this could be described as glassy dynamics.
1357: %EC2 wrote: We have used cluster algorithms in order to hasten the approach to ergodicity.
1358: %YL3: No, we did not use cluster algorithms for phase IV, because Wolff is not valid at h>0, and we would have to use the geometric cluster algorithm. My implementation of the GCA is buggy and I haven't fixed it. I've edited this paragraph. Sorry.
1359: %YL3: P.S.: Sondhi actually has a remark about glassy dynamics in the physical spin systems, and why it is ok to use equilibrium thermodynamics -- he argues that most quantities are self-averaging.
1360: We believe that the simulations may be made more efficient by using geometric cluster methods\cite{heringa1998} or by augmenting the Metropolis algorithm with directed-loop updates tailored to the honeycomb dimer state.\cite{krauth-2003,sandvik-2006}
1361: The slow dynamics under the Metropolis algorithm may, however, be representative of the true dynamics in a physical realization of the TKL Ising antiferromagnet.
1362: %However, the slow time dynamics may lead to glassy behavior as
1363: %temperature is lowered toward the critical phase IV in a physical realization of the
1364: %TKL Ising antiferromagnet.
1365: %EC2 On the other hand, it is possible that the glassy dynamics have a real physical significance.
1366:
1367:
1368: It may be possible to stabilize more phases at finite temperature by introducing
1369: an appropriate perturbation.
1370: For example, introducing coupling in the third direction may be sufficient to
1371: stabilize the critical phase it finite T.
1372: In addition, it may be possible to induce a Kasteleyn transition in phase IV by applying some kind of orienting field ({\em e.g.}, uniaxial strain) to the TKL, similar to what was suggested by Moessner and Sondhi\cite{moessner2003} on the Kagome lattice.
1373: %EC2 *** Yen: IS IT KAGOME? ***
1374: In light of existing studies on triangular and kagome lattices, it
1375: is reasonable to expect that adding next-nearest-neighbor interactions
1376: to the TKL will produce an even richer phase diagram, like Kagome lattice.~\cite{takagi93}
1377: %, possibly stabilizing the dimer phase at finite T.
1378: %EC2 containing BKT transitions.
1379:
1380:
1381:
1382: %EC2 I don't mind including the next paragraph, as long as we conclude something about what we say -- do we or do we not expect the resonating dimer phase?
1383: %Theoretically, it is interesting to study quantum spin models and higher
1384: %order interactions on TKL. With continuous quantum spins, it is possible to have other types of emergent degrees of freedom and ordered states. For example, a small transverse coupling would lead to a resonating dimer state, analogous to a RVB state. However, unlike the case of the KIAF in field\cite{moessner2000} where two ground states are connected by quantum processes at sixth order in perturbation theory, getting between two ground states in the TKL (in phase IV) requires a minimum of nine spin flips -- the quantum processes are of order $(J_{aa}^\perp)^3 (J_{ab}^\perp)^6$ or $(J_{ab}^\perp)^{12}$.
1385:
1386: %\red{Regarding experiments:}
1387: %The experiments suggest that $J_{aa}$ and $J_{ab}$ are both negative and that $|J_{aa}|/|J_{ab}| \gg 1$.
1388: %This corresponds to the interesting regime of our phase diagram, with
1389: %up to five field-driven transitions.
1390:
1391: \subsection{Comparison with Other Frustrated Ising Models}
1392:
1393:
1394:
1395: In Table~\ref{comparison-table}, we present a comparison of the
1396: spin-spin correlations and the residual entropy among the frustrated
1397: TKL, triangular, and kagome lattices. The simplest example of a
1398: geometrically frustrated system is the triangular Ising
1399: antiferromagnet (TIAF), {i.e.}, a set of Ising spins on a triangular
1400: lattice with antiferromagnetic pairwise couplings. Due to the
1401: presence of odd cycles in the lattice graph, it is impossible for all
1402: pairs of nearest neighbor spins to be simultaneously antiparallel. As
1403: a result, the antiferromagnetic interactions are unable to produce
1404: long-range order even at zero temperature. Instead, at zero
1405: temperature, the TIAF has a quasi-long-range-ordered state in which
1406: the correlations decay with distance as $r^{-1/2}$.
1407: \cite{stephenson64,stephenson70} This ground state is macroscopically
1408: degenerate, with a zero-temperature residual entropy of $0.3231k_B$ per spin. \cite{wannier50,syozi50a,syozi50b,foot1} This number is the same as the entropy per unit cell of the random dimer model on a honeycomb lattice, and it crops up in many other places.
1409: %DY %YL: foot1 is in bib file
1410: Although the TIAF has no phase transition in zero field, the
1411: application of a finite field produces a surprisingly rich phase
1412: diagram. A small field induces a Berezinskii-Kosterlitz-Thouless
1413: (BKT) transition to a `spin crystal' phase that breaks translational
1414: symmetry and has long-range order -- the correlation function
1415: oscillates with distance but does not decay. At larger fields the
1416: crystalline order is destroyed via a transition in the 3-state Potts
1417: universality class. In this limit the TIAF is related to Baxter's
1418: exactly soluble hard hexagon lattice gas model.\cite{baxter-1980}
1419:
1420:
1421:
1422: Another frustrated spin system is the antiferromagnetic Ising model on
1423: a kagome lattice, formed by periodic removal of a quarter of the sites
1424: from the triangular lattice. Unlike the TIAF, the KIAF in the absence
1425: of field has pair correlations that decay
1426: exponentially at all temperatures including
1427: $T=0$. \cite{syozi,suto1981} Its ground state entropy is $1.5055k_B$ per unit cell or $0.5018 k_B$
1428: per spin,\cite{kano53} higher than that of the TIAF because the
1429: smaller coordination number allows more freedom. At finite $h$, there is a different spin
1430: liquid state that can be mapped to random dimers on a honeycomb
1431: lattice. In this state the spin-spin correlation function decays as a
1432: power law, $1/r^2$. The residual entropy is $0.3231k_B$ per unit cell, or $0.1077k_B$ per spin.
1433: %YL3: emphasizing that the 0.3231 appears everywhere (the number which Wannier made a mistake with in 1950, haha)
1434:
1435: The ground state of the TKL Ising AF without applied field is
1436: even more frustrated than that of the kagome lattice:
1437: the correlation function becomes exactly zero beyond a certain cutoff radius.
1438: The residual entropy per spin is $\frac{1}{9}\ln 72 = 0.4752\dots$.
1439: At finite $h$, there is a correlated, critical spin liquid state
1440: which we have mapped to hard core dimers on a honeycomb lattice.
1441: %honeycomb dimer spin liquid state that is basically the same as the KIAF+h;
1442: This state has a residual entropy of $0.3231k_B$ per unit cell, or $0.03590$ per spin.
1443: (See Table~\ref{comparison-table}.)
1444: %YL3: Originally 0.0358962.
1445: %\cite{barry1997}
1446:
1447: \begin{table}
1448: \begin{ruledtabular}
1449: \begin{tabular}{ccc}
1450: %\hline
1451: Lattice & Entropy & Spin-spin correlation \\ \hline
1452: Triangular & $0.3231\dots$ & $r^{-1/2}$ \\
1453: Triangular in field & $0$ & long-range-ordered \\
1454: Kagome & $1.5055\dots$ & $e^{-r/\xi}$ %EC2 , $\xi=\red{???}$
1455: \\
1456: Kagome in weak field & $0.3231\dots$ & $r^{-2}$ \\
1457: TKL & $\ln 72$ & $0$ for $r\geq r_{bb}$ \\
1458: TKL in weak field & $0.3231\dots$ & $r^{-2}$ \\
1459: TKL in medium field & $\ln 9$ & $0$ for $r\geq r_{bb}$
1460: %\hline
1461: \end{tabular}
1462: \end{ruledtabular}
1463: \caption{Comparison of various frustrated Ising models at $T=0$. Residual entropies are quoted per unit cell; they are the logarithms of irrational numbers, unless otherwise stated.
1464: %%YL: keeping this "irrational"
1465: %YL3: Making it clear that this is at T=0.
1466: \label{comparison-table}}
1467: \end{table}
1468:
1469:
1470:
1471: \subsection{Comparison with Experiment}
1472:
1473:
1474: As discussed in the introduction, the recently fabricated family of compounds
1475: $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
1476: have an arrangement of the copper sites which forms a triangular kagome lattice.
1477: There is no evidence of
1478: spontaneous magnetization down to
1479: at least $T=1.7$K,\cite{maruti94} consistent with a spin liquid ground state, which
1480: indicates that $J_{aa}$ is antiferromagnetic.
1481: However, there is no agreement yet as to whether $J_{ab}$ is
1482: ferromagnetic or antiferromagnetic.\cite{maruti94,strecka07}
1483:
1484: In studying the magnetic susceptibility, several groups find that the
1485: slope of $1/\chi$ versus $T$ is roughly linear at high
1486: temperature, but as $T$ is lowered, the slope increases.\cite{maruti94,haar95,mekata98}.
1487: This is consistent with either $J_{ab}$ ferromagnetic or antiferromagnetic,
1488: as seen in Figs.~\ref{af1:ix} and~\ref{af2:ix}. %EC figures corrected
1489: A distinction can be made, though, by the behavior of $T \chi$.
1490: Whereas $T \chi$ saturates to a finite value at low $T$ for antiferromagnetic
1491: intertrimer coupling $J_{ab}<0$, $T \chi$ reaches a minimum at intermediate $T$
1492: before saturating at a finite value as $T \rightarrow 0$ if %DY3 Consistent with Sec. VI. A. statement
1493: the intertrimer coupling is {\em ferromagnetic},
1494: $J_{ab}>0.$
1495: To the extent that the
1496: $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
1497: materials can be described by an Ising TKL model like the one in this paper,
1498: our calculations indicate that the intertrimer coupling
1499: must be ferromagnetic, $J_{ab}>0$,
1500: so that upon application of a field, the system should be in
1501: Phase III, rather than Phase IV.
1502: Since there is a striking difference in the residual entropies
1503: in these two phases ($s_0 = \ln 9=2.1972\cdots$ in Phase III
1504: {\em vs.} $s_0 = 0.3231\cdots$ in Phase IV),
1505: heat capacity measurements will also be useful in distinguishing these phases.
1506: Other future experiments, including neutron scattering, NMR, and $\mu$SR, can also provide useful data for comparing to models of frustrated
1507: magnetism on the TKL.
1508: %YL3: fixxed some typos and redudnacies ("including...among others")
1509:
1510:
1511: %EC2 %%%%%%%%% ***CUT*** %%%%%%%%%%%
1512: %EC2
1513: %In the light of existing studies on triangular and kagome lattices, it
1514: %is reasonable to expect that adding next-nearest-neighbor interactions
1515: %to the TKL will produce an even richer phase diagram, possibly containing BKT
1516: %transitions.
1517:
1518:
1519: % EC2 Saturation magnetizations: Maruti/TerHaar have $m = 1.1 \mu_B$, whereas Mekata has $m=1/3 \mu_B$
1520: %per site. Seems contradictory.
1521: %But susceptibility diverges as $T \rightarrow 0$, which is
1522: %found in experiments.\cite{haar95,maruti94}
1523: %EC2
1524: %In the frustrated regime of the TKL Ising model ($J_{aa}/|J_{ab}| < -1$) at low $T$, it is interesting that increasing the magnetic field tunes the system between disordered states with high entropy ($4.2767\dots$, $2.1972\dots$ per unit cell) and localized correlations ($\xi=0$) and a peculiar critical state with low entropy ($0.3231\dots$) and power-law correlations.
1525:
1526: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1527: \section{Conclusions \label{conclusions}}
1528:
1529:
1530: In conclusion, we have studied an Ising model on the triangular kagome
1531: lattice (TKL) with two interactions $J_{aa}$ and $J_{ab}$, temperature
1532: $T$, and external field $h$. We have calculated the complete phase
1533: diagram in $(J_{aa}, J_{ab}, h, T)$ parameter space \emph{exactly}.
1534: Furthermore, we have obtained exact results for thermodynamic
1535: quantities (free energy, energy, heat capacity, and entropy) at all
1536: $T$ for $h=0$, and at all $h$ for $T=0$, and plotted them for
1537: representative cases.
1538:
1539: %EC2 \blue{Specific heat:}
1540: In the experimentally relevant regime, $\abs{J_{aa}} \gg
1541: \abs{J_{ab}}$, if $J_{aa}$ is ferromagnetic, the specific heat shows a
1542: broad hump corresponding to intra-trimer ordering, as well as a sharp
1543: peak at lower temperatures due to the onset of true long-range order.
1544: If $J_{aa}$ is antiferromagnetic there are two broad features.
1545:
1546: %EC2 \blue{Susceptibility:}
1547: We have computed the magnetization $M(T,h)$ and susceptibility $\chi(T,h)$ in various regimes using Monte Carlo simulations. %DY3 we used both metropolis and cluster
1548: To the extent that experiments on the $\mbox{Cu}_{9}\mbox{X}_2(\mbox{cpa})_{6}\cdot x\mbox{H}_2\mbox{O}$
1549: materials can be compared with an Ising TKL model, our calculations
1550: indicate that $J_{ab}>0$.
1551:
1552:
1553: %%YL: removed "integer", explained more carefully
1554: %%YL: irrational-> closed form
1555: %%YL: removed "super" and "hyper"
1556: We find that at strong frustration and zero field, as temperature is
1557: reduced, the model enters a spin liquid phase with residual entropy $s_0 = \ln 72$ per unit cell, with ``perfectly localized'' correlations. This stands in contrast with the triangular
1558: and kagome Ising antiferromagnets, whose residual entropies cannot be expressed in closed form.
1559: %It may be said that the TKL is ``hyper-frustrated'', {\em i.e.}, that it is even more frustrated than the ``super-frustrated'' kagome lattice.
1560:
1561: %EC2 \blue{We could emphasize that WE have predicted this in the case of the TKL...}
1562: The most interesting feature of the model is a correlated critical
1563: spin liquid phase (with power-law correlations) that appears at strong
1564: frustration, weak fields, and zero temperature.
1565: %EC2 I don't mind keeping this comparison to moesner, but we need to say how it is similar. \blue{The situation is very similar to that of the kagome Ising AF in a field\cite{moessner2001}.}
1566: We have mapped this phase to hard core dimer coverings of a honeycomb
1567: lattice. The critical power-law correlations are reduced to
1568: exponential correlations at finite $T$, but the correlation length may
1569: still be large.
1570: %EC2 %This is likely to be an explanation of the size-dependent peaks in the magnetic susceptibility seen in our Monte Carlo simulations.
1571: Such a phenomenon would be detectable by neutron scattering measurements.
1572: %EC2 \blue{ (if a physical realization of the TKL Ising AF exists)}.
1573:
1574:
1575:
1576:
1577: %@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
1578: \acknowledgments
1579: %The authors would like to thank Takeo Takagi and Mamoru Mekata at Fukui University.
1580: It is a pleasure to thank T.~Takagi, M.~Mekata, G.~Ortiz, N.~Sandler, and M.~Ma for helpful discussions.
1581: This work was supported by Purdue University and Research Corporation (Y.~L.~L. and D.~X.~Y.).
1582: E.~W.~C. is a Cottrell Scholar of Research Corporation.
1583:
1584:
1585:
1586:
1587:
1588:
1589:
1590:
1591: %&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
1592: \appendix
1593: \section{Mean-Field Approximations \label{meanfield}}
1594:
1595: For pedagogical purposes, we examine the zero-field TKL Ising model using two mean-field approaches. Compared with the exact solution (see Fig.~\ref{zerofieldphasediagram}), we see that such approaches can be quite misleading in the frustrated regime $J_{aa}/|J_{ab}|<-1$.
1596:
1597: In the simplest mean-field approximation (MF1), every spin is assumed to fluctuate thermally in a mean field determined by the average magnetizations of its neighbors. This leads to a pair of simultaneous equations for the spontaneous magnetizations of the $a$- and $b$-sublattices, $m_a=\mean{\sigma_a}$ and $m_b=\mean{\sigma_b}$,
1598: \begin{align}
1599: m_a &= \tanh \beta(2J_{aa}m_a+2J_{ab}m_b),
1600: \nonumber\\
1601: m_b &= \tanh \beta(4J_{ab}m_a).
1602: \label{mft}
1603: \end{align}
1604: where $\beta=1/T$.
1605: Linearizing the tanh function in $\mean{\sigma_a}$ and $\mean{\sigma_b}$ gives the critical temperature,
1606: \begin{align}
1607: %T_c{}^2 - 2J_{aa} T_c - 8J_{ab} {}^2 = 0, \\
1608: %\therefore\quad
1609: T_c^\text{MF1} = J_{aa} + \sqrt {J_{aa} {}^2 + 8 J_{ab} {}^2}.
1610: \end{align}
1611: %keeping only the positive solution.
1612: %As expected, $T_c$ does not depend on the sign of $J_{ab}$.
1613: %It is natural that the mean-field $T_c$ is more than twice as high as the true $T_c$. However, we need to point out that the mean-field treatment works bad in the regime $J_{aa}\gg J_{ab}>0$, where $T_c$ is dominated by the weakest link in the system ($T_c \rightarrow 2.00838 J_{ab}$) but the mean-field theory does not know about the connectivity ($T_c \rightarrow 2J_{aa}$).
1614:
1615: We also present a more advanced mean-field approximation (MF2), similar to the method used by Stre\v{c}ka\cite{strecka07} for the quantum Heisenberg TKL model, in which we sum over all eight states of the $a$-trimers with appropriate Boltzmann weights instead of treating each $a$-spin independently. The mean-field equations are then
1616: \begin{align}
1617: m_a &= \frac{
1618: e^{4\beta J_{aa}} \sinh 6\beta J_{ab}m_b
1619: + \sinh 2\beta J_{ab}m_b
1620: }{
1621: e^{4\beta J_{aa}} \cosh 6\beta J_{ab}m_b
1622: + 3 \cosh 2\beta J_{ab}m_b
1623: },
1624: \nonumber\\
1625: m_b &= \tanh \beta(4J_{ab}m_a),
1626: \label{e:mf2}
1627: \end{align}
1628: and the critical temperature is given by
1629: \begin{align}
1630: \frac{T_c^\text{MF2} }{ \left|J_{ab} \right| }
1631: = \sqrt{\frac{8( 1+3e^{4J_{aa}/T_c^\text{MF2}} )}{ (3+e^{4J_{aa}/T_c^\text{MF2}}) }}.
1632: \end{align}
1633:
1634: $T_c^\text{MF1}$ and $T_c^\text{MF2}$ are the dotted and dashed curves, respectively, in Fig.~\ref{zerofieldphasediagram}. The mean-field approximations overestimate $T_c$ by a factor of two or more. In the regime $J_{aa}/|J_{ab}| \gg 1$, $T_c$ is dominated by the weak links ($J_{ab}$), so $T_c/|J_{ab}|$ tends to a constant, a fact which is captured by MF2. However, in the frustrated regime $J_{aa}/|J_{ab}| < -1$, both MF1 and MF2 predict the wrong behavior of $T_c$ (see Fig.~\ref{zerofieldphasediagram}). Also, if an external field is included in the analysis, these mean-field approximations predict an induced ferromagnetic or ferrimagnetic moment, but by construction, they are unable to capture phases III, IV, and the zero-field ``$\ln 72$'' phase in the rich phase diagram presented in Sec.~\ref{zerotemperature}.
1635:
1636:
1637:
1638:
1639:
1640:
1641:
1642:
1643:
1644:
1645:
1646: %\bibliography{tkl}
1647: \bibliographystyle{forprb}
1648: \begin{thebibliography}{10}
1649: \newcommand{\enquote}[1]{``#1''}
1650:
1651:
1652: \bibitem{zhitomirsky-2003}
1653: M.~E. Zhitomirsky, Phys. Rev. B {\bf 67}, 104421 (2003).
1654:
1655: \bibitem{zhitomirsky-2004}
1656: M.~E. Zhitomirsky and H.~Tsunetsugu, Phys. Rev. B {\bf 70}, 100403(R) (2004).
1657:
1658: \bibitem{derzhko-2004}
1659: O.~Derzhko and J.~Richter, Phys. Rev. B {\bf 70}, 104415 (2004).
1660:
1661: \bibitem{isakov-2004}
1662: S.~V. Isakov, K.~S. Raman, R.~Moessner, and S.~L. Sondhi, Phys. Rev. B {\bf
1663: 70}, 104418 (2004).
1664:
1665: \bibitem{aoki-2004}
1666: H.~Aoki, T.~Sakakibara, K.~Matsuhira, and Z.~Hiroi, J. Phys. Soc. Japn. {\bf
1667: 73}, 2851 (2004).
1668:
1669: \bibitem{gonzalez93}
1670: M.~Gonzalez, F.~Cervantes-Lee, and L.~W. ter Haar, Mol. Cryst. Liq. Cryst. {\bf
1671: 233}, 317 (1993).
1672:
1673: \bibitem{maruti94}
1674: S.~Maruti and L.~W. ter Haar, J. Appl. Phys {\bf 75}, 5949 (1993).
1675:
1676: \bibitem{mekata98}
1677: M.~Mekata, M.~Abdulla, T.~Asano, H.~Kikuchi, T.~Goto, T.~Morishita, and
1678: H.~Hori, J. Magn. Magn. Matt. {\bf 177}, 731 (1998).
1679:
1680: \bibitem{strecka07}
1681: J.~Stre\v{c}ka, J. Magn. Magn. Matt. {\bf 316}, e346 (2007).
1682:
1683: \bibitem{daoxinMSthesis}
1684: D.~X. Yao, {\em Invariant Theory of Time-Dependent Quantum Systems and Ising
1685: Model on Triangular Kagome Lattice\/}, Master's thesis, Zhejiang University,
1686: Hangzhou, China (1998).
1687:
1688: \bibitem{zheng05}
1689: J.~Zheng and G.~Sun, Phys. Rev. B {\bf 71}, 052408 (2005).
1690:
1691: \bibitem{loh2006}
1692: Y.~L. Loh and E.~W. Carlson, Phys. Rev. Lett. {\bf 97}, 227205 (2006).
1693:
1694: \bibitem{loh-jeremy2007}
1695: Y.~L. Loh, E.~W. Carlson, and M.~Y.~J. Tan, Phys. Rev. B {\bf 76}, 014404
1696: (2007).
1697:
1698: \bibitem{chineseremainder}
1699: A.~Galluccio, M.~Loebl, and J.~Vondrak, Phys. Rev. Lett. {\bf 84}, 5924 (2000).
1700:
1701: \bibitem{wannier50}
1702: G.~H. Wannier, Phys. Rev {\bf 79}, 357 (1950).
1703:
1704: \bibitem{kano53}
1705: K.~Kano and S.~Naya, Prog. Theor. Phys. {\bf 10}, 158 (1953).
1706:
1707: \bibitem{kasteleyn1963}
1708: P.~W. Kasteleyn, J. Math. Phys. {\bf 4}, 287 (1963).
1709:
1710: \bibitem{fisher1966}
1711: M.~E. Fisher, J. Mod. Phys. {\bf 7}, 10 (1966).
1712:
1713: \bibitem{horiguchi1992}
1714: T.~Horiguchi, K.~Tanaka, and T.~Morita, J. Phys. Soc. Jpn. {\bf 61}, 64 (1992).
1715:
1716: \bibitem{moessner2000}
1717: R.~Moessner, S.~L. Sondhi, and P.~Chandra, Phys. Rev. Lett. {\bf 84}, 4457
1718: (2000).
1719:
1720: \bibitem{moessner2001}
1721: R.~Moessner and S.~L. Sondhi, Phys. Rev. B {\bf 63}, 224401 (2001).
1722:
1723: \bibitem{udagawa2002}
1724: M.~Udagawa, M.~Ogata, and Z.~Hiroi, J. Phys. Soc. Jpn. {\bf 71}, 2365 (2002).
1725:
1726: \bibitem{moessner2003}
1727: R.~Moessner and S.~L. Sondhi, Phys. Rev. B {\bf 68}, 064411 (2003).
1728:
1729: \bibitem{heilman1972}
1730: O.~J. Heilman and E.~H. Lieb, Commun. Math. Phys. {\bf 25}, 190 (1972).
1731:
1732: \bibitem{heringa1998}
1733: J.~R. Heringa and H.~W.~J. Bl\"ote, Phys. Rev. E {\bf 57}, 4976 (1998).
1734:
1735: \bibitem{krauth-2003}
1736: W.~Krauth and R.~Moessner, Phys. Rev. B {\bf 67}, 064503 (2003).
1737:
1738: \bibitem{sandvik-2006}
1739: A.~W. Sandvik and R.~Moessner, Phys. Rev. B {\bf 73}, 144504 (2006).
1740:
1741: \bibitem{takagi93}
1742: T.~Takagi and M.~Mekata, J. Phys. Soc. Jpn. {\bf 62}, 3943 (1993).
1743:
1744: \bibitem{stephenson64}
1745: J.~Stephenson, J. Math. Phys. {\bf 5}, 1009 (1964).
1746:
1747: \bibitem{stephenson70}
1748: J.~Stephenson, J. Math. Phys. {\bf 11}, 413 (1970).
1749:
1750: \bibitem{syozi50a}
1751: K.~Husimi and I.~Syozi, Prog. Theor. Phys. {\bf 5}, 177 (1950).
1752:
1753: \bibitem{syozi50b}
1754: K.~Husimi and I.~Syozi, Prog. Theor. Phys. {\bf 5}, 341 (1950).
1755:
1756: \bibitem{foot1}
1757: The erroneous numerical result in Wannier's original paper ($0.3383 k_B$) is
1758: often quoted, although it was corrected in an erratum 23 years
1759: later.~\cite{wanniererratum}.
1760:
1761: \bibitem{baxter-1980}
1762: R.~J.~Baxter, J. Phys. A {\bf 13}, L61 (1980).
1763:
1764: \bibitem{syozi}
1765: I.~Syozi, \enquote{in {\em Phase Transitions and Critical Phenomena}, Domb and
1766: Green (eds), vol. 1,} (1972).
1767:
1768: \bibitem{suto1981}
1769: A.~S\"ut\"o, Z. Phys. B {\bf 44}, 121 (1981).
1770:
1771: \bibitem{haar95}
1772: S.~M. S.~Ateca and W.~ter Haar, J. Magn. Magn. Matt. {\bf 147}, 398 (1995).
1773:
1774: \bibitem{wanniererratum}
1775: G.~H. Wannier, Phys. Rev. B {\bf 7}, 5017 (1973).
1776:
1777: \bibitem{fnote}
1778: Note that the enclosed $a$-trimer is {\em not} restricted to a particular $S_z$ subspace, as shown in Fig.~\ref{zeroTzeroh}(b).
1779:
1780: \end{thebibliography}
1781:
1782:
1783: \end{document}
1784:
1785: %Onsager's exact solution of the square lattice Ising model\cite{onsager44} stimulated interest in exact solutions on other lattices.
1786:
1787: