1: %\documentstyle[aps]{revtex}
2:
3:
4: \documentstyle[aps,twocolumn]{revtex}
5: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6: %TCIDATA{Created=Fri May 22 12:11:54 1998}
7: %TCIDATA{LastRevised=Wed Mar 29 09:34:55 2000}
8: %TCIDATA{Language=American English}
9:
10: \tighten
11: \tolerance = 1000
12: \draft
13:
14: \begin{document}
15: \title{Kondo spin liquid and magnetically long-range ordered states in the Kondo
16: necklace model }
17: \author{Guang-Ming Zhang$^1$, Qiang Gu$^1$, and Lu Yu$^{2,3}$}
18: \address{$^1$Center for Advanced Study, Tsinghua University, Beijing 100084, P. R.
19: China.\\
20: $^2$International Center for Theoretical Physics, P. O. Box 586, Trieste
21: 34100, Italy.\\
22: $^3$ Institute for Theoretical Physics, Academic Sinica, Beijing 100080, P.
23: R. China}
24: \date{\today}
25: \maketitle
26:
27: \begin{abstract}
28: A simplified version of the symmetric Kondo lattice model, the Kondo
29: necklace model, is studied by using a representation of impurity and
30: conduction electron spins in terms of local Kondo singlet and triplet
31: operators. Within a mean field theory, a spin gap always appears in the spin
32: triplet excitation spectrum in 1D , leading to a Kondo spin liquid state for
33: any finite values of coupling strength $t/J$ (with $t$ as hopping and $J$ as
34: exchange); in 2D and 3D cubic lattices the spin gaps are found to vanish
35: continuously around $(t/J)_c\approx 0.70$ and $(t/J)_c\approx 0.38$,
36: respectively, where quantum phase transitions occur and the Kondo spin
37: liquid state changes into an antiferromagnetically long-range ordered state.
38: These results are in agreement with variational Monte Carlo, higher-order
39: series expansion, and recent quantum Monte Carlo calculations for the
40: symmetric Kondo lattice model. \newline
41: {PACS numberes: 71.27.+a, 71.10.Hf, 75.10.Jm, 75.30.Hx}\newline
42: \end{abstract}
43:
44: \smallskip
45:
46: Since the discovery of the class of stoichiometric insulating compounds
47: known as Kondo insulators \cite{aeppli}, there has been revived interest in
48: the symmetric Kondo lattice Hamiltonian,
49: \begin{equation}
50: H=-t\sum_{{\bf \langle }i,j{\bf \rangle }}(C_{i,\sigma }^{\dagger
51: }C_{j,\sigma }+h.c.)+J\sum_i{\bf S}_i\cdot C_{i,\alpha }^{\dagger }{\bf %
52: \sigma }_{\alpha \beta }C_{i,\beta }, \nonumber
53: \end{equation}
54: as a model of concentrated magnetic impurity spins coupled to conduction
55: electrons. One of the important issues is the interplay between the Kondo
56: screening and the magnetic interactions among localized spins mediated by
57: the conduction electrons. The former effect favors a nonmagnetic Kondo spin
58: liquid (singlet) state, while the latter interactions tend to stabilize an
59: antiferromagnetically (AF) long-range ordered state. The character of such a
60: transition between these two distinct phases has been a long standing issue
61: since it was first pointed out by Doniach \cite{doniach}. There have been a
62: lot of investigations for the symmetric one-dimensional model, showing that
63: its ground state is a disordered Kondo spin liquid state for {\it any}
64: finite values of the coupling strength $t/J$ \cite{review}. For two and
65: three dimensional models, however, various approximate approaches such as
66: variational Monte Carlo calculations \cite{wang}, higher-order series
67: expansions \cite{shi}, quantum Monte Carlo simulations \cite{assaad,vekic},
68: and infinite dimensional calculations \cite{jarrell}, suggest that the Kondo
69: spin liquid state may change into an AF long-range ordered state at certain
70: value of the coupling strength at low temperatures.
71:
72: Since there are a lot of difficulties in directly attacking the symmetric
73: Kondo lattice model even in the 1D case, a simplified version called Kondo
74: necklace model was introduced by Doniach \cite{doniach},
75: \begin{equation}
76: H=t\sum_{\langle i,j\rangle }(\tau _i^x\tau _j^x+\tau _i^y\tau _j^y)+J\sum_i%
77: {\bf S}_i\cdot {\bf \tau }_i,
78: \end{equation}
79: where both ${\bf \tau }_i$ and ${\bf S}_i$ are spin 1/2 Pauli operators,
80: denoting the conduction electron spin and impurity spin operators,
81: respectively, and $\langle i,j\rangle $ means summation over the nearest
82: neighbor conduction electron sites. Actually this simplified model is
83: meaningful in general D dimensions ($D=1,2,3$) in its own right. Due to the
84: suppression of charge fluctuations in the symmetric model, the charge
85: degrees of freedom are frozen out, so the first term of Eq.(2) represents
86: the spin degrees of freedom imitating the propagation of the conduction
87: electrons. This can be clearly seen in the 1D case, where the first term is
88: equivalent after a Jordan-Wigner transformation to a band of spinless
89: fermions, which interact with localized spins via an AF spin-spin exchange
90: coupling \cite{doniach}.
91:
92: Although the simplified model has only U(1) spin symmetry, lower than SU(2)
93: for the Kondo lattice model, the essential feature of these two models is
94: kept. Thus, one would expect that the main physical properties of the
95: original symmetric Kondo lattice model should be maintained in the Kondo
96: necklace model. However, most of approaches used to treat the 1D Kondo
97: necklace model, including the variational mean-field calculation \cite
98: {doniach}, approximate real-space renormalization group theory \cite{jullien}%
99: , and recent finite size scaling analysis \cite{solyom}, have found a finite
100: critical value of coupling strength $(J/t)_c\sim 0.24-0.38$, below which an
101: AF quasi-long-range order state appears, {\it in contrast to} $J_c=0$, the
102: result of quantum Monte Carlo simulation for the 1D Kondo necklace model
103: \cite{scalettar} and the numerical result for the 1D symmetric Kondo lattice
104: model \cite{pfeuty,tsunetsugu}. It is thus controversial whether the
105: simplified spin model can be used to approximate the original symmetric
106: Kondo lattice model. In this paper, we try to resolve this issue, starting
107: from the Kondo necklace model, using the Kondo spin singlet and triplet
108: representation, to reproduce correct ground states of the symmetric Kondo
109: lattice model. In the 1D case, the system is found to be in a Kondo spin
110: liquid state with a finite spin gap for any finite $t/J$, while on 2D and 3D
111: cubic lattices a quantum phase transition occurs around $(t/J)_c\sim 0.70$
112: and $(t/J)_c\sim 0.38$, respectively, where the Kondo spin liquid state
113: changes into an AF long-range ordered state, in excellent agreement with the
114: variational Monte Carlo calculation \cite{wang}, higher-order series
115: expansion \cite{shi}, and recent quantum Monte Carlo simulation \cite{assaad}%
116: , on the corresponding symmetric Kondo lattice model.
117:
118: Our starting point is the strong coupling limit $t=0$, where the lowest
119: energy state of the model Hamiltonian Eq.(2) reduces to a sum over
120: contributions from independent local Kondo spin singlet states at each
121: lattice site. When $t\neq 0$, interactions between these independent local
122: Kondo spin singlets are switched on. It will be seen later that this leads
123: to very reasonable results even for $t\geq J$, which is of interest here.
124: Usually, for two $s=1/2$ spins ${\bf \tau }_i$ and ${\bf S}_i$ placed on a
125: lattice site, the local Hilbert space is spanned by four states consisting
126: of one singlet and three triplet states defined as being created out of the
127: vacuum $|0\rangle $ by the singlet and triplet creation operators: $%
128: |s\rangle =s^{\dagger }|0\rangle $ and $|t_\alpha \rangle $ $=t_\alpha
129: ^{\dagger }|0\rangle $, ($\alpha =x,$ $y,$ $z$). A representation of the
130: impurity spins and conduction electron spins in terms of these singlet and
131: triplet operators is given by
132:
133: \begin{eqnarray}
134: S_{n,\alpha } &=&\frac 12(s_n^{\dagger }t_{n,\alpha }+t_{n,\alpha }^{\dagger
135: }s_n-i\epsilon _{\alpha \beta \gamma }t_{n,\beta }^{\dagger }t_{n,\gamma }),
136: \nonumber \\
137: \tau _{n,\alpha } &=&\frac 12(-s_n^{\dagger }t_{n,\alpha }-t_{n,\alpha
138: }^{\dagger }s_n-i\epsilon _{\alpha \beta \gamma }t_{n,\beta }^{\dagger
139: }t_{n,\gamma }),
140: \end{eqnarray}
141: where $\alpha $, $\beta $, and $\gamma $ represent components along the $x$,
142: $y$, and $z$ axes, respectively, and $\epsilon $ is the antisymmetric
143: Levi-Civita tensor. This type of spin representation in terms of singlet and
144: triplet (bond) operators was first proposed by Sachdev and Bhatt to study
145: the properties of dimerized phases \cite{sachdev} and then it was
146: successfully used to consider the spin ladders\cite{rice} and $s=1$
147: antiferromagnetic Heisenberg spin chains\cite{hanting}. As shown later, this
148: representation faithfully describes the low temperature physics in the
149: symmetric Kondo lattice model. In order to restrict the physical states to
150: either singlets or triplets, a local constraint is introduced: $s_n^{\dagger
151: }s_n+\sum\limits_\alpha t_{n,\alpha }^{\dagger }t_{n,\alpha }=1.$ Taking the
152: singlet and triple operators at each site to satisfy bosonic commutation
153: relations: $[s_n,s_n^{\dagger }]=1$, $[t_{n,\alpha },t_{n,\beta }^{\dagger
154: }]=\delta _{\alpha ,\beta }$, and $[s_n,t_{n,\alpha }^{\dagger }]=0$, the
155: SU(2) algebra of the spins ${\bf \tau }_n$ and ${\bf S}_n$ can be reproduced,
156:
157: \begin{eqnarray}
158: \lbrack S_{n,\alpha },S_{n,\beta }] &=&i\epsilon _{\alpha \beta \gamma
159: }S_{n,\gamma },\text{ \quad }[\tau _{n,\alpha },\tau _{n,\beta }]=i\epsilon
160: _{\alpha \beta \gamma }\tau _{n,\gamma }, \nonumber \\
161: \lbrack S_{n,\alpha },\tau _{n,\beta }] &=&0,\text{ \quad }{\bf S}_n^2={\bf %
162: \tau }_n^2=\frac 34.
163: \end{eqnarray}
164: Substituting the operator representation of the impurity and conduction
165: electron spins, we obtain the following form of the model Hamiltonian,
166:
167: \begin{eqnarray}
168: H &=&H_0+H_1+H_2+H_3, \nonumber \\
169: H_0 &=&\frac J4\sum_i(-3s_i^{\dagger }s_i+\sum_\alpha t_{i,\alpha }^{\dagger
170: }t_{i,\alpha }) \nonumber \\
171: &&\text{ \quad }+\sum_i\mu _i(s_i^{\dagger }s_i+\sum_\alpha t_{i,\alpha
172: }^{\dagger }t_{i,\alpha }-1), \nonumber \\
173: H_1 &=&\frac t4\sum_{\langle ij\rangle }\left[ s_i^{\dagger }s_j^{\dagger
174: }\left( t_{i,x}t_{j,x}+t_{i,y}t_{j,y}\right) \right. \nonumber \\
175: &&\text{ \quad }+\left. s_i^{\dagger }s_j\left( t_{i,x}t_{j,x}^{\dagger
176: }+t_{i,y}t_{j,y}^{\dagger }\right) +h.c.\right] , \nonumber \\
177: H_2 &=&-\frac t4\sum_{\langle ij\rangle }\left[ t_{i,z}^{\dagger
178: }t_{j,z}^{\dagger }\left( t_{i,x}t_{j,x}+t_{i,y}t_{j,y}\right) \right.
179: \nonumber \\
180: &&\text{ \quad }-\left. t_{i,z}^{\dagger }t_{j,z}\left(
181: t_{i,x}t_{j,x}^{\dagger }+t_{i,y}t_{j,y}^{\dagger }\right) +h.c.\right] ,
182: \nonumber \\
183: H_3 &=&\frac{it}4\sum_{\langle ij\rangle }\sum_{\alpha ,\beta ,\gamma
184: }\epsilon _{\alpha \beta \gamma \text{ }}\left[ s_i^{\dagger }t_{i,\alpha
185: }t_{j,\beta }^{\dagger }t_{j,\gamma }+s_j^{\dagger }t_{j,\alpha }t_{i,\beta
186: }^{\dagger }t_{i,\gamma }+h.c.\right],
187: \end{eqnarray}
188: where a site-dependent chemical potential $\mu _i$ has been introduced to
189: impose the local constraint. Here the local spin triplet states are split
190: into two parallel spin states with $m_s=\pm 1$ and an anti-parallel spin
191: state with $m_s=0$. $H_1$ describes the couplings between the singlet state
192: and the parallel spin triplet states, while $H_2$ corresponds to the
193: couplings of the parallel spin and the anti-parallel spin triplet states. $%
194: H_3$ describes an interaction of one singlet boson and three different
195: components of triplet bosons.
196:
197: The above Hamiltonian can be solved by a mean field decoupling of the
198: quartic terms. It yields an effective Hamiltonian $H_{mf}$ with only
199: quadratic operators. We take $\langle s_i^{\dagger }\rangle =\langle
200: s_i\rangle =\overline{s}$, which corresponds to {\it a condensation of the
201: local Kondo spin singlets on each site }in accordance with the configuration
202: of the ground state in the strong coupling limit, and the local chemical
203: potential is replaced by a global one. We will consider here only the terms $%
204: H_0$ and $H_1$, as it can be shown that inclusion of $H_2$ changes the
205: results only slightly \cite{rice,gmzhang} and all the decouplings of $H_3$
206: identically vanish within the present mean field theory. After performing a
207: Fourier transformation of the boson operators, $t_{i,\alpha }=\frac 1{\sqrt{N%
208: }}\sum\limits_{{\bf k}}t_{{\bf k},\alpha }$ $e^{-i{\bf k\cdot r}_i}$, the
209: mean field effective Hamiltonian is given by
210: \begin{eqnarray}
211: && H_{mf}=N\left( -\frac 34J\text{ }\overline{s}^2+\mu \overline{s}^2-\mu
212: \right) +\left( \frac J4+\mu \right) \sum_{{\bf k}}t_{{\bf k},z}^{\dagger
213: }t_{{\bf k},z} \nonumber \\
214: && \hspace{0.5cm} +\sum_{{\bf k},\beta =x,y}\left[ \Lambda _{{\bf k}}t_{{\bf %
215: k},\beta }^{\dagger }t_{{\bf k},\beta }+\Delta _{{\bf k}}\left( t_{{\bf k}%
216: ,\beta }^{\dagger }t_{-{\bf k},\beta }^{\dagger }+t_{{\bf k},\beta }t_{-{\bf %
217: k},\beta }\right) \right] ,
218: \end{eqnarray}
219: with $\Lambda _{{\bf k}}=\left( \frac J4+\mu \right) +\frac 12t\overline{s}%
220: ^2\lambda ({\bf k)}$, $\Delta _{{\bf k}}=\frac 14t\overline{s}^2\lambda (%
221: {\bf k)}$, and $\lambda ({\bf k)=}\sum\limits_{a=1}^d\cos k_a$. The lattice
222: spacing has been taken to be unity. This mean field Hamiltonian can be
223: diagonalized by a Bogoliubov transformation into new boson operators: $%
224: \widetilde{t}_{{\bf k},\beta }=u_{{\bf k}}t_{{\bf k},\beta }+v_{{\bf k}}t_{-%
225: {\bf k},\beta }^{\dagger }$, where the coefficients $u_{{\bf k}}$ and $v_{%
226: {\bf k}}$ are even functions of ${\bf k}$, and are determined to be: $u_{%
227: {\bf k}}^2+v_{{\bf k}}^2=\cosh 2\theta _{{\bf k}}=\frac{\Lambda _{{\bf k}}}{%
228: \sqrt{\Lambda _{{\bf k}}^2-(2\Delta _{{\bf k}})^2}}$ and $2u_{{\bf k}}v_{%
229: {\bf k}}=\sinh 2\theta _{{\bf k}}=-\frac{2\Delta _{{\bf k}}}{\sqrt{\Lambda _{%
230: {\bf k}}^2-(2\Delta _{{\bf k}})^2}}.$ Then we obtain
231: \begin{equation}
232: H_{mf}=\omega _0\sum_{{\bf k}}t_{{\bf k},z}^{\dagger }t_{{\bf k},z}+\sum_{%
233: {\bf k},\beta =x,y}\omega _{{\bf k}}\widetilde{t}_{{\bf k},\beta }^{\dagger }%
234: \widetilde{t}_{{\bf k},\beta }+E_g,
235: \end{equation}
236: where $\omega _0=\left( \frac J4+\mu \right) $ is the dispersionless energy
237: level of the anti-parallel spin triplet excited state, $\omega _{{\bf k}}=%
238: \sqrt{\Lambda _{{\bf k}}^2-(2\Delta _{{\bf k}})^2}$ corresponds to the
239: excitation spectrum of the parallel spin triplet excited states, and the
240: ground state energy of the system is $E_g=N\left( -\frac 34J\text{ }%
241: \overline{s}^2+\mu \overline{s}^2-\mu \right) +\sum\limits_{{\bf k}}\left(
242: \omega _{{\bf k}}-\Lambda _{{\bf k}}\right) $. By minimizing the ground
243: state energy with respect to $\mu $ and $\overline{s}$, we derive the
244: following saddle-point equations,
245:
246: \begin{eqnarray}
247: \frac 1N\sum_{{\bf k}}\frac{\Lambda _{{\bf k}}}{\sqrt{\Lambda _{{\bf k}%
248: }^2-(2\Delta _{{\bf k}})^2}} &=&(2-\overline{s}^2), \nonumber \\
249: \frac tN\sum_{{\bf k}}\sqrt{\frac{\Lambda _{{\bf k}}-2\Delta _{{\bf k}}}{%
250: \Lambda _{{\bf k}}+2\Delta _{{\bf k}}}}\lambda ({\bf k)} &=&2J\left( \frac 34%
251: -\frac \mu J\right) .
252: \end{eqnarray}
253: When a dimensionless parameter $d=\frac tJ\frac{\overline{s}^2}{(\frac 14+%
254: \frac uJ)}$ is introduced, a self-consistent equation for $d$ can be obtained
255:
256: \begin{equation}
257: d=\frac{2t}J\left[ 1-\frac 1{2N}\sum_{{\bf k}}\frac 1{\sqrt{1+d\lambda ({\bf %
258: k)}}}\right] ,
259: \end{equation}
260: to determine the variational parameters $\overline{s}$ and $\mu $ and the
261: spin triplet excitation spectra: $\omega _0=J\left( \frac 14+\frac \mu J%
262: \right) $ and $\omega _k=J\left( \frac 14+\frac \mu J\right) \sqrt{%
263: 1+d\lambda ({\bf k)}}$. There is a minimum spin gap in the parallel spin
264: triplet spectrum at the AF reciprocal vector momentum ${\bf k}={\bf Q}$: $%
265: \Delta _{sp}=J\left( \frac 14+\frac \mu J\right) \sqrt{1-Zd/2}$, where $Z$
266: is the total number of the nearest neighbors on the cubic lattice.
267:
268: In the 1D case, we first numerically calculate the parameters $d$, $%
269: \overline{s}^2$, and $\mu /J$ for a range of the coupling strength $0<t/J<$ $%
270: 5$, and the minimum spin gap $\Delta _{sp}=J\left( \frac 14+\frac \mu J%
271: \right) \sqrt{1-d}$ is evaluated in the range of $0<t/J<$ $5$, which has
272: been delineated in Fig.1. The dispersive band can also be parameterized by a
273: spin density wave with a velocity given by $v_s=J\left( \frac 14+\frac \mu J%
274: \right) \sqrt{\frac d2}$. A linear drop of the spin gap is seen for small
275: values of $t/J$. As $t/J$ gets larger, the spin gap deviates considerably
276: from the linear behavior and there is no indication at all suggesting a
277: critical value for $t/J$ where the gap would vanish. Since the excitation
278: spectra are real and positive everywhere in the Brillouin zone, the system
279: will be in a quantum disordered --- Kondo spin liquid state for {\it finite}
280: values of the coupling strength $t/J$, and the spin-spin correlation
281: function decays exponentially at large distances with a correlation length $%
282: \xi =\frac{v_s}{\Delta _{sp}}$. This is indeed consistent with both quantum
283: Monte Carlo simulations for the 1D Kondo necklace \cite{scalletar} and
284: numerical results for the 1D symmetric Kondo lattice model \cite
285: {pfeuty,tsunetsugu}. Thus, starting from the limit $t/J=0$ with localized
286: Kondo spin singlets on each site, we see that any finite coupling strength
287: delocalizes the local Kondo singlets, reducing the magnitude of the gap but
288: not closing it completely.
289:
290: Having secured the correct ground state for the 1D symmetric Kondo lattice
291: model, we now turn to two and three dimensional ''Kondo necklace'' models on
292: a cubic lattice. In 2D, the variational parameters $d$, $\overline{s}^2$,
293: and $\mu /J$ can also be calculated from the saddle-point equations. The
294: minimum spin gap appears in the parallel spin triplet excitation at ${\bf k}%
295: =(\pi ,\pi )$: $\Delta _{sp}=J\left( \frac 14+\frac \mu J\right) \sqrt{1-2d}$%
296: , displayed in Fig.2. The most important feature here is that as the
297: coupling parameter $t/J$ increases, the drop of the spin gap in the small
298: values of $t/J$ continues down to the point $(t/J)_c\approx 0.70$ where the
299: spin gap actually vanishes. The critical coupling $(t/J)_c\approx 0.70$
300: corresponds to a quantum critical point for a phase transition from the
301: quantum disordered Kondo spin liquid to a magnetically long-range ordered
302: state. Surprisingly, the location of the critical point for the 2D Kondo
303: necklace model is {\it precisely} the value obtained from the variational
304: Monte Carlo calculation \cite{wang}, the higher-order series expansion \cite
305: {shi}, and recent quantum Monte Carlo simulation \cite{assaad} for the 2D
306: symmetric Kondo lattice model. When a similar calculation is carried out in
307: the 3D Kondo necklace model, the minimum spin gap appears at ${\bf k}=(\pi
308: ,\pi ,\pi )$ and $\Delta _{sp}=J\left( \frac 14+\frac \mu J\right) \sqrt{1-3d%
309: }$, shown in Fig.3. As $t/J$ grows, the spin gap decreases and exhibits a
310: critical value $(t/J)_c\approx 0.38$, where the spin gap disappears
311: completely, showing a quantum phase transition from the quantum disordered
312: Kondo spin liquid to a magnetically long-range ordered state as well. This
313: transition point is in the same range as the higher-order series expansion
314: \cite{shi} for the 3D symmetric Kondo lattice model: $(t/J)_c\approx 0.50$.
315:
316: Moreover, the present mean field theory can also be applied to the
317: magnetically long-range ordered phase in the 2D and 3D Kondo necklace
318: models. If we assume that, not only the local Kondo spin singlets ($s$
319: bosons) condenses, one of the local Kondo spin triplets ($t_{{\bf k},x}$
320: bosons) condenses as well on the AF reciprocal vector $t_{_{{\bf k},x}}=%
321: \sqrt{N}\overline{t}\delta _{{\bf k,Q}}+\eta _{{\bf k},x}$, corresponding to
322: {\it fixing the orientation of the localized spins along }$x$-{\it direction}%
323: , it will lead to another mean field effective Hamiltonian
324:
325: \begin{eqnarray}
326: &&H_{mf}^{\prime }=E_g^{\prime }+\omega _0\sum_{{\bf k}}t_{{\bf k}%
327: ,z}^{\dagger }t_{{\bf k},z}+\sum\limits_{{\bf k}}\omega _{{\bf k}}(%
328: \widetilde{t}_{{\bf k},y}^{\dagger }\widetilde{t}_{{\bf k},y}+\widetilde{%
329: \eta }_{{\bf k},x}^{\dagger }\widetilde{\eta }_{{\bf k},x}), \nonumber \\
330: &&E_g^{\prime }=N\left[ -\frac 34J\overline{s}^2+\mu \overline{s}^2-\mu
331: +\left( \frac J4+\mu -\frac 12Zt\overline{s}^2\right) \overline{t}^2\right]
332: \nonumber \\
333: &&\text{ \quad \qquad }+\sum\limits_{{\bf k}}\left( \omega _{{\bf k}%
334: }-\Lambda _{{\bf k}}\right) ,
335: \end{eqnarray}
336: where $\omega _{{\bf k}}$ has the same form as in the Kondo spin liquid
337: phase, and $\widetilde{\eta }_{{\bf k},x}^{\dagger }$ and $\widetilde{\eta }%
338: _{{\bf k},x}$ are the transverse spin triplet excitation mode. When the
339: order parameter $\overline{t}$ is nonzero, the saddle point equation for $%
340: \overline{t}$ yields $\mu =\frac 12Zt\overline{s}^2-\frac J4$, which makes
341: the parallel spin triplet excitation spectrum {\it gapless}: $\omega _{{\bf k%
342: }}=\frac 12Zt\overline{s}^2\sqrt{1+2\lambda ({\bf k)}/Z}.$ The ground state
343: corresponds to a magnetically long-range ordering state with a maximum
344: momentum ${\bf q=Q}$, and the mean field $\overline{t}$ represents the AF
345: order parameter. It has been suggested that a very appealing physical
346: picture of forming AF long-range order in the Kondo necklace or the
347: symmetric Kondo lattice models: when $t/J$ is small, the conduction electron
348: spins are locked and the impurity spins are screened completely, and the
349: ground state is a product of the local Kondo spin singlets -- quantum
350: disordered phase \cite{review}. As $t/J$ becomes larger and larger, the
351: conduction electrons (the spin degrees of freedom) have more possibility to
352: propagate to the nearest neighbor sites, and the localized magnetic impurity
353: spins is only {\it partially } screened{\it \ (}$\overline{s}\neq 0${\it )},
354: then the remaining part of the magnetic impurities on different lattice
355: sites start to develop long-range correlations ($\overline{t}\neq 0$)
356: mediated by the conduction electron spins \cite{sachdev}. Such a
357: magnetically long-range ordered state might be related to the ground states
358: of the U-based heavy fermion compounds (URu$_2$Si$_2$ and UPt$_3$) with a
359: very small magnitude of induced staggered magnetic moments. In order to
360: determine the parameters $\overline{s}$ and $\overline{t}$, we minimize the
361: ground state energy, derive the saddle point equations, and finally obtain
362: \begin{eqnarray}
363: \overline{s}^2 &=&1+\frac J{Zt}-\frac 1{2N}\sum_{{\bf k}}\sqrt{1+2\lambda (%
364: {\bf k)/}Z}, \nonumber \\
365: \overline{t}^2 &=&1-\frac J{Zt}-\frac 1{2N}\sum_{{\bf k}}\frac 1{\sqrt{%
366: 1+2\lambda ({\bf k)/}Z}}.
367: \end{eqnarray}
368: The AF order parameter is defined by $m_s=\overline{s}$ $\overline{t}$,
369: leading to the following expressions:
370:
371: \begin{eqnarray*}
372: m_s &=&\sqrt{(0.35712-\frac J{4t})(0.52095+\frac J{4t})},\text{ \quad \quad
373: for 2D;} \\
374: m_s &=&\sqrt{(0.44234-\frac J{6t})(0.51263+\frac J{6t})},\text{ \quad \quad
375: for 3D.}
376: \end{eqnarray*}
377: These results have also been displayed in Fig.2 and Fig.3, respectively. In
378: Fig.2, our results are also compared with the numerical results for the spin
379: gap and staggered moment of the magnetic impurity spins in the recent
380: quantum Monte Carlo simulation on the 2D symmetric Kondo lattice model at
381: zero temperature \cite{assaad}.
382:
383: In summary, we have presented a mean field theory for the Kondo necklace
384: model in 1D, 2D and 3D and have obtained their correct ground states
385: corresponding to the respective Kondo lattice model. A long standing
386: controversial issue has been thus resolved regarding the relationship
387: between these two models. As far as the spin part of the ground state
388: properties is concerned, the Kondo necklace model can reproduce the correct
389: phase diagrams of the symmetric Kondo lattice model at zero temperature.
390:
391: \begin{references}
392: \bibitem{aeppli} G. Aeppli and Z. Fisk, Comments on Condens. Matter Phys.
393: {\bf 16}, 155 (1987).
394:
395: \bibitem{doniach} S. Doniach, Physica B {\bf 91}, 231 (1977).
396:
397: \bibitem{review} H. Tsunetsugu, M. Sigrist, and K. Ueda, Rev. Mod. Phys.
398: {\bf 69}, 809 (1997), and references therein.
399:
400: \bibitem{shi} Z. P. Shi, R. R. Singh, M. P. Gelfand, and Z. Wang, Phys.
401: Rev. B {\bf 51}, 15630 (1995).
402:
403: \bibitem{wang} Z. Wang, X. P. Li, and D. H. Lee, Physica B {\bf 199-200},
404: 463 (1984).
405:
406: \bibitem{assaad} F. F. Assaad, Phys. Rev. Lett. {\bf 83}, 796 (1999).
407:
408: \bibitem{vekic} M. Vekic, J. W. Cannon, D. J. Scalapino, and R. T.
409: Scalettar, and R. L. Sugar, Phys. Rev. Lett. {\bf 74}, 2367 (1995).
410:
411: \bibitem{jarrell} M. J. Rozenberg, Phys. Rev. B {\bf 52}, 7369 (1995).
412:
413: \bibitem{jullien} R. Jullien, J. N. Fields, and S. Doniach, Phys. Rev. B
414: {\bf 16}, 4889 (1977); W. Hanke and J. E. Hirsch, {\it ibid}. {\bf 25}, 6748
415: (1982).
416:
417: \bibitem{solyom} P. Santini and J. Solyom, Phys. Rev. B {\bf 46}, 7422
418: (1992).
419:
420: \bibitem{scalettar} R. T. Scalettar, D. J. Scalapino. and R. J. Sugar,
421: Phys. Rev. B {\bf 31}, 7316 (1985).
422:
423: \bibitem{pfeuty} R. Jullien and P. Pfeuty, J. Phys. F 11, 353 (1981).
424:
425: \bibitem{tsunetsugu} H. Tsunetsugu, Y. Hatsugai, K. Ueda, and M. Sigrist,
426: Phys. Rev. B {\bf 46}, 3175 (1992); N. Shibata, T. Nishino, K. Ueda, and C.
427: Ishii,{\it \ ibid}. {\bf 53}, 8828 (1996).
428:
429: \bibitem{sachdev} S. Sachdev and R. N. Bhatt, Phys. Rev. B {\bf 41}, 9323
430: (1990).
431:
432: \bibitem{rice} S. Gopalan, T. M. Rice, and M. Sigrist, Phys. Rev. B {\bf 49}%
433: , 8901 (1994); B. Normand and T. M. Rice, {\it ibid}. {\bf 54}, 7180 (1996).
434:
435: \bibitem{hanting} Han-Ting Wang, Jue-Lian Shen, and Zhao-Bin Su, Phys. Rev.
436: B {\bf 56}, 14435 (1997).
437:
438: \bibitem{gmzhang} Guang-Ming Zhang, Qiang Gu, and Lu Yu, unpublished.
439: \end{references}
440:
441: {\bf Figure Captions}
442:
443: \smallskip
444:
445: Fig.1. The variation of the spin gap upon increasing of the coupling
446: parameter $t/J$ of the 1D model at $T=0$.
447:
448: \smallskip
449:
450: Fig.2. The spin gap and the staggered magnetic moment at zero temperature of
451: the 2D Kondo necklace model (bold line) in comparison with results of recent
452: quantum Monte Carlo simulation\cite{assaad} for the 2D Kondo lattice model.
453:
454: \smallskip
455:
456: Fig.3. The spin gap and the staggered magnetic moment at zero temperature
457: for 3D Kondo necklace model.
458:
459: \end{document}
460: