1: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb,prl,superscriptaddress]{revtex4}
2: \usepackage{graphicx}
3:
4: \begin{document}
5:
6: \title{\bf Correlated adatom trimer on metal surface:\\ A
7: continuous time quantum Monte Carlo study}
8:
9: \author{V. V. Savkin}
10: \email{savkin@sci.kun.nl} \affiliation{Institute of Molecules and
11: Materials, University of Nijmegen, 6525 ED Nijmegen, The
12: Netherlands}
13: \author{A. N. Rubtsov}
14: %\email{alex@shg.ru}
15: \affiliation{Department of Physics, Moscow State University,
16: 119992 Moscow, Russia }
17: \author{M. I. Katsnelson}
18: %\email{katsnel@sci.kun.nl}Molecules
19: \affiliation{Institute of Molecules and Materials, University of
20: Nijmegen, 6525 ED Nijmegen, The Netherlands}
21: \author{A. I. Lichtenstein}
22: %\email{Alexander.Lichtenstein@physnet.uni-hamburg.de}
23: \affiliation{Institute of Theoretical Physics, University of
24: Hamburg, 20355 Hamburg, Germany} \pacs{71.10.-w, 71.27.+a,
25: 73.22.-f}
26:
27: \begin{abstract}
28: The problem of three interacting Kondo impurities is solved within
29: a numerically exact continuous time quantum Monte Carlo scheme. A
30: suppression of the Kondo resonance by interatomic exchange
31: interactions for different cluster geometries is investigated. It is
32: shown that a drastic difference between the Heisenberg and Ising
33: cases appears for antiferromagnetically coupled adatoms. The
34: effects of magnetic frustrations in the adatom trimer are
35: investigated, and possible connections with available experimental
36: data are discussed.
37: \end{abstract}
38:
39: \maketitle
40:
41: The electronic structure of adatoms and clusters on surfaces
42: constitutes one of the most fascinating subjects in condensed
43: matter physics and modern nanotechnology \cite{plummer}. The
44: scanning tunneling microscopy (STM) or spectroscopy technique
45: allows the study of atomic structure, the electronic energy
46: spectrum, and magnetic properties of different surfaces at an
47: atomic scale. In particular, STM gives the unique opportunity of
48: directly investigating an essentially many-body phenomenon, namely
49: the Kondo effect \cite{STM,coral,kol}. Earlier only indirect
50: methods such as analysis of temperature and magnetic field
51: dependencies of thermodynamic and transport properties were
52: available \cite{gruner,hewson}. Recently STM studies of small
53: transition metal nanoclusters on different surfaces have been
54: performed, including Co dimers \cite{CoAu} and Cr trimers
55: \cite{CrAu} on a Au surface, and Co clusters on carbon nanotubes
56: \cite{nanotube}. The electron spectrum of these nanosystems, in
57: particular the existence of the Kondo resonance, turns out to be
58: very sensitive to the geometry of the clusters as well as to the
59: type of magnetic adatoms. The later can be important for
60: nanotechnological fine tuning of surface electronic structure.
61:
62: The ``quantum-corral'' type of STM-experiments provides an unique
63: opportunity to investigate in detail an interplay between the
64: single-impurity Kondo effect and interatomic magnetic interactions
65: in nanoclusters which is a key phenomenon in Kondo lattice physics
66: \cite{doniach,irkhin,si,castro,zhu}. The interaction between
67: itinerant electrons and localized ones leads to the screening of
68: the magnetic impurity moment which is the Kondo effect; on the
69: other hand, the RKKY exchange interaction between localized spins
70: suppresses the Kondo resonance at the Fermi level. As a result, a
71: very complicated phase diagram can be obtained with regions
72: described by a strong coupling regime, ``normal'' magnetic
73: behavior with logarithmic corrections, and non-Fermi-liquid
74: behavior \cite{irkhin}. Quantum critical points at the boundary of
75: different phases is a subject of special interest \cite{si,zhu}.
76: There exists a general belief that anomalous features of many
77: $f$-electron systems such as heavy-fermion or non-Fermi-liquid
78: behavior can be treated in terms of the Kondo lattice picture
79: \cite{hewson,stewart,amato}. This is why non-perturbative
80: investigations of the basic physical features of few-atom magnetic
81: clusters in a metallic medium is of primary interest. At the same
82: time, due to the extreme complexity of the problem, theoretical
83: investigations of electronic structure for several Kondo centers
84: usually involve some uncontrollable approximations, such as a
85: replacement of the Heisenberg interatomic exchange interactions by
86: the Ising ones \cite{zhu} or a variational approach based on a
87: simple trial function \cite{uzdin}.
88:
89: In this Letter we present results of a numerically exact solution
90: of the three Kondo impurity problem within the recently developed
91: continuous time quantum Monte Carlo (CT-QMC) method
92: \cite{rubtsov}. For the antiferromagnetic (AFM) exchange
93: interatomic interaction, in contrast to the ferromagnetic (FM)
94: one, the results for the Heisenberg and Ising systems differ
95: essentially. Based on our theoretical analysis, the recent
96: paradoxical experimental results \cite{CrAu} where the Kondo
97: resonance is observable for an isosceles magnetic triangle but not
98: for the perfect Cr-trimer or individual Cr adatom will be
99: discussed.
100:
101: We start with the system of three impurity correlated sites with
102: Hubbard repulsion $U$ in a metallic bath and with an effective
103: exchange interaction $J_{ij}$ between them, a minimal model which
104: however includes all relevant interactions necessary to describe
105: magnetic nanoclusters on a metallic surface. The effective action
106: for such cluster in a metallic medium has the following form:
107: \begin{eqnarray}\label{S}
108: &S=S_0+W, \\
109: &S_0=-\int _0^{\beta} \int _0^{\beta} d\tau d\tau'
110: \sum_{i,j;\sigma} c^\dag_{i\sigma}(\tau) {\cal{G}}
111: ^{-1}_{ij}(\tau-\tau')c_{j\sigma}(\tau'), \nonumber \\
112: &W=\int _0^{\beta} d\tau \left( U\sum_{i}n_{i\uparrow}(\tau)
113: n_{i\downarrow}(\tau)+\sum_{i,j} J_{ij} \mathbf{S}_i(\tau)
114: \mathbf{S}_j(\tau)\right). \nonumber
115: \end{eqnarray}
116: The last term in the right-hand-side of Eq.(\ref{S}) allows us to
117: consider the most important ``Kondo lattice'' feature, that is,
118: the mutual suppression of the Kondo screening and intersite
119: exchange interactions \cite{doniach,irkhin}. Another factor, the
120: coherence of the resonant Kondo scattering, is taken into account
121: by the introduction of inter-impurity hopping terms $t_{ij}$ to
122: the bath Green function which is supposed to be
123: ${\cal{G}}^{-1}_{ij}={\cal{G}}^{-1}_{i}\delta_{ij}-t_{ij}$. Here
124: ${\cal{G}}^{-1}_{i}(i\omega_n)=\mu + i (\omega_n +
125: \sqrt{\omega_n^2+1})/2$ corresponds to the semicircular density of
126: states (DOS) with band-width 2 and $t_{ij}$ are inter-impurity
127: hopping integrals. For real adatom clusters the exchange
128: interactions are mediated by conduction electrons (RKKY
129: interactions) which are dependent on the specific electronic
130: structure of both adatoms and host metal. To simulate this effect
131: we will consider $J_{ij}$ as independent parameters which is a
132: common practice in the Kondo lattice problem
133: \cite{doniach,irkhin,zhu}; otherwise for the half-filled
134: non-degenerate Hubbard model used in our calculations the exchange
135: is always antiferromagnetic. In the model (\ref{S}) the geometry
136: of the problem is specified by the values of exchange integrals
137: $J_{ij}$ and hopping parameters $t_{ij}$\cite{georges}. We will
138: concentrate on the case of equilateral triangle when $J_{ij}=J$
139: and $t_{ij}=t$; to compare with the experimental situation in
140: Ref.\onlinecite{CrAu} also an isosceles triangle will be
141: considered. To check an approximation used in Ref.\onlinecite{zhu}
142: we will investigate the case when all spin-flip exchange terms are
143: ignored and the Heisenberg ($\mathbf{S}\mathbf{S}$) form of
144: interaction $J_{ij} \mathbf{S}_i(\tau) \mathbf{S}_j(\tau)$ is
145: transformed into the Ising ($S_zS_z$) one $J_{ij} S^z_i(\tau)
146: S^z_j(\tau)$.
147:
148: We use the numerically exact CT-QMC method \cite{rubtsov} for our
149: computer simulations. Unlike the Hirsh-Fye discrete-time scheme
150: \cite{hirsch} it does not involve auxiliary Ising spins, but
151: performs a random walk in the space of terms of the perturbation
152: expansion for the Green function. One of the advantages of this
153: novel approach is the opportunity it provides to study systems
154: with non-local (in space and in time) interactions, which in the
155: usual Hirsh-Fye approach would involve a huge increase in the
156: required number of auxiliary fields and time slices
157: \cite{rubtsov}.
158:
159: In brief, numerical simulations make use of the division of the
160: action into to Gaussian and interaction parts. The later can be
161: presented in the form $\int d \tau \sum w_{iji'j'}(\tau)$, where
162: $w=u_{iji'j'} (c^{\dag}_{i, \uparrow} c_{j, \uparrow} -
163: \alpha_{\uparrow}) (c^{\dag}_{i', \downarrow} c_{j', \downarrow} -
164: \alpha_{\downarrow})$, where the $\alpha$'s are $c$-numbers chosen
165: in a special way to minimize the sign problem in the QMC
166: simulation. It is worth noting that only the presence of these
167: $\alpha's$ makes possible the simulation of systems with repulsive
168: forces \cite{rubtsov}.
169:
170: The formal interaction-representation expansion in powers of
171: the interaction for the Green function $G_{ij}(\tau,\tau')$ reads
172: \begin{equation}\label{G}
173: \sum_k \sum_{I_1, ... , I_k} \int d\tau_1 ... d\tau_k \frac{{\rm
174: Tr} \left(T c^\dag_{i \tau'} c_{j \tau} w_{I_1}(\tau_1) ...
175: w_{I_k}(\tau_k) \right)}{Z k!}.
176: \end{equation}
177: Here $Z$ is the partition function and $I$ denotes the set $i, j,
178: i', j'$. Our algorithm performs the random-walk in a space of all
179: possible values for $k; \tau_1 ... \tau_k; I_1 ... I_k$. The
180: numerical averaging of Eq.(\ref{G}) over this random walk gives
181: the desired Green function for the interacting system with the
182: action (\ref{S}). Typical value for $k$ in our calculation is
183: $k\approx \int d\tau \sum ||w_I(\tau)||\approx N \beta U+ N^2
184: \beta J$, where $\beta$ is the inverse temperature and $N$ is
185: number of atoms in the cluster. One can see that for the case $U
186: \gg N J$ the exchange interaction indeed does not slow the
187: calculation down, its complexity is determined by the local
188: (Coulomb) interaction.
189:
190:
191: In order to check the CT-QMC algorithm for a system with
192: complicated Heisenberg interactions we apply this method to a
193: simple Hamiltonian analogue of the model (\ref{S}), {\it i.e.} for
194: ${\cal{G}}^{-1}_{i}(i\omega_n)=\mu + i \omega_n$. We compare our
195: CT-QMC approach with the solution obtained using the exact
196: diagonalization method. Results for the system with AFM $J$
197: ($J>0$) are shown in Fig.1 for $S_zS_z$ and $\mathbf{S}\mathbf{S}$
198: interactions. The estimated errorbar in numerical data is
199: $10^{-3}$ or less.
200:
201: Although the problem of DOS-calculations from $G(\tau)$ data is
202: ill-defined, quite reliable estimations can be made at our level
203: of the numerical accuracy. For example, DOS of the above-mentioned
204: Hamiltonian models is at most contributed by the four
205: $\delta$-peaks located at $\pm U/2, \pm (U/2+J)$ for Ising and
206: $\pm U/2, \pm (U/2+2 J)$ for Heisenberg interaction. Fit of the
207: numerical data with several $\delta$-peaks indeed resolves their
208: positions (with a $5\%$ errorbar) and relative heights (with a
209: $20\%$ errorbar). Further, we study models with a continuous DOS,
210: and standard maximum-entropy analytical continuation method
211: \cite{maxent} is used to recover DOS. Normally it resolves the DOS
212: features from imaginary time QMC data with a similar accuracy.
213:
214:
215: \begin{figure}
216: \includegraphics[width=\columnwidth]{f1.EPS}
217: \caption{\label{fig:epsart} Imaginary part of the Green function
218: at Matsubara frequencies for the model (\ref{S}) with
219: ${\cal{G}}^{-1}_{i}(i\omega_n)=\mu + i \omega_n$ for
220: $\mathbf{S}\mathbf{S}$ and $S_zS_z$ interactions. Symbols are
221: CT-QMC data, lines are exact diagonalization (ED) results.
222: Parameters of the model: $U=2, J=0.2, t=0, \beta=8, \mu=U/2$. }
223: \end{figure}
224:
225: Let us discuss correlated adatom trimer in the metallic bath
226: depending on type of the effective exchange interaction ($S_zS_z$
227: or $\mathbf{S}\mathbf{S}$) for AFM and FM cases. First we show
228: that $\mathbf{S}\mathbf{S}$ type of interaction suppresses the
229: resonance in AFM case. We study the equilateral triangle at
230: half-filling. The Green functions at the Matsubara frequencies
231: obtained by the CT-QMC technique and corresponding DOS are
232: presented in Fig. 2. The case of $J=0$ corresponds to the single
233: Kondo impurity in the bath. One can see that there is no essential
234: difference between $S_zS_z$ and $\mathbf{S}\mathbf{S}$ types of
235: interaction in FM but for the AFM one this difference is very
236: important. The AFM $\mathbf{S}\mathbf{S}$ interaction leads to
237: pronounced suppression of the Kondo resonance at Fermi level for
238: physically relevant values of $J$ (Fig. 2). On the other hand, in
239: the FM case we do not observe any essential difference between the
240: $S_zS_z$ and $\mathbf{S}\mathbf{S}$ types of the interaction for a
241: wide range of the model parameters.
242:
243: \begin{figure}
244: \includegraphics[width=\columnwidth]{f2.EPS}
245: \caption{\label{fig:epsart} Imaginary part of the Green functions
246: at Matsubara frequencies for the correlated adatom equilateral
247: triangle in the metallic bath for AFM (upper figure) and FM (lower
248: figure) types of effective exchange interaction. Parameters: $U=2,
249: J=\pm 0.2, t=0, \beta=16, \mu=U/2$. There are three dependencies
250: on each picture for $\mathbf{S}\mathbf{S}$, $S_zS_z$ and $J=0$
251: (which corresponds to single atom in the metallic bath) types of
252: interaction. The insets show DOS.}
253: \end{figure}
254:
255: To explain these results we need to calculate the spectral density
256: $D(\omega)$ of the on-site spin-flip operators $S_{i}^{\pm}$ for
257: an AFM equilateral triangle
258: $D_{\mathbf{S}\mathbf{S}}(\omega)=\frac{2}{3}\delta(\omega)+
259: \frac{1}{3}\delta(\omega-3J)$, $
260: D_{S_zS_z}(\omega)=\frac{1}{3}\delta(\omega)+
261: \frac{2}{3}\delta(\omega-2J)$. In both cases there is a part of
262: the spectral density with zero frequency due to degeneracy of the
263: ground state and spin-flip transitions between its components, but
264: for the $\mathbf{S}\mathbf{S}$ case this part is twice as large.
265: For the case $J\gg T_K$ ($T_K$ is one-site Kondo temperature) only
266: this ``soft'' component of the spectral density will lead to Kondo
267: screening which means that the suppression of the Kondo effect is
268: twice more efficient for the $\mathbf{S}\mathbf{S}$ case than for
269: the $S_zS_z$ one. Since the trimer as a whole has a degenerate
270: ground state there is still a strong-coupling regime and an
271: effective Kondo temperature $T_{K}^{*}$ which however is much
272: smaller than $T_K$. Scaling considerations similar to one proposed
273: in Ref.\onlinecite{irkhin} gives an estimation of $T_{K}^{*}\simeq
274: T_{K}^{3/2}/J^{1/2}$ and $T_{K}^{*}\simeq T_{K}^{3}/J^2$ for the
275: $\mathbf{S}\mathbf{S}$ and $S_zS_z$ case respectively. We assume
276: that this quantity is too small to be visible in our simulations
277: (as well as in the experimental data \cite{CrAu}) so what is
278: observed corresponds to the one-site Kondo resonance at the
279: condition $T_{K}\geq J$. Similar estimation for the case of FM
280: interactions shows that there is no difference between the
281: $S_zS_z$ and $\mathbf{S}\mathbf{S}$ model there and
282: $T_{K}^{*}\simeq T_{K}^{3}/J^2$ as in the AFM $S_zS_z$ case.
283:
284: In order to describe the experimental situation we changed the
285: geometry of the adatom trimer. An observation of the Kondo
286: resonance reconstruction was reported for one isosceles geometry
287: of three Cr atoms on a gold surface \cite{CrAu}. Thus we study
288: the isosceles triangles for AFM and FM types of effective exchange
289: interaction. We have chosen the following parameters of $J_{ij}$
290: to imitate the experimental system: $J_{23}=J, J_{12}=J/3,
291: J_{13}=J/3$. The computational results are presented in Fig. 3,
292: where one can see the reconstruction of resonance in AFM and FM
293: cases in accordance with experimental data. Note that the Kondo
294: resonance appears only for the more weakly bonded adatom in AFM case.
295:
296: \begin{figure}
297: \includegraphics[width=\columnwidth]{f3.EPS}
298: \caption{\label{fig:epsart} DOS for equilateral triangle (ET) and
299: isosceles triangle (IT) geometries with AFM (upper figure) and FM
300: (lower figure) types of effective exchange interaction. Parameters
301: are the same as in Fig.2. Values of the effective exchange
302: integrals for IT are as follows: $J_{23}=J, J_{12}=J/3,
303: J_{13}=J/3$. There are two dependencies in case of IT: one for
304: adatom $1$ and another for equivalent adatoms $2$ and $3$. All
305: adatoms are equivalent in the case of ET (one dependence).}
306: \end{figure}
307:
308: The observed picture can be drastically changed by introducing a
309: non-zero value of the hopping parameter $t_{ij}$. As noted above
310: the trimer ground state is degenerate at $t_{ij}=0$, however this
311: degeneracy is lifted for $t_{ij}\neq0$. If one of the obtained
312: states lies below the Fermi level, DOS can be changed drastically
313: and at certain parameters it leads to the appearance of a
314: resonance on the Fermi level (see Fig.4). In Fig.4, DOS is shown
315: for various values of filling in the system and at nonzero
316: $t_{ij}$. It is necessary to point out that the introduction of
317: the parameter $t_{ij}$, the variation of the filling in the system
318: ($\mu$), and different geometries ($J_{ij}$) can lead to various
319: results in DOS. This can be used as one of the possible
320: explanations of the experimental data regarding the Cr trimer on a
321: Au surface which look initially appear counter-intuitive: the
322: Kondo temperature for the trimer is much larger than for the
323: single site, despite the suppression of the Kondo effect by
324: $J_{ij}$. One can assume that this is a consequence of the change
325: of the number of $d$-electrons in ground-state configuration for
326: the Cr atom in the trimer in comparison with the isolated one. In
327: order to describe specific experimental results quantitatively the
328: method developed here should be combined with first-principle
329: calculations in the spirit of recently developed ``local density
330: approximation plus dynamical mean-field theory'' (LDA + DMFT)
331: scheme \cite{ldadmft}.
332:
333: \begin{figure}
334: \includegraphics[width=\columnwidth]{f4.EPS}
335: \caption{\label{fig:epsart} DOS for equilateral triangle with an
336: AFM type of effective exchange interaction at various values of
337: filling in the system. Parameters: $U=2, J=0.2, t=0.3, \beta=16$.
338: Corresponding numbers of particles are $n=0.8$, $n=1$ (half-filled
339: case) and $n=1.3$.}
340: \end{figure}
341:
342: In conclusion, we have shown that the electronic structure of a
343: correlated adatom trimer on a metallic surface drastically depends
344: on the symmetry of magnetic interactions. The effective exchange
345: interaction of $\mathbf{S}\mathbf{S}$ type leads to more efficient
346: suppression of the Kondo resonance in the AFM case than in the
347: case of $S_zS_z$ interactions. The experimental STM data
348: \cite{CrAu} can be reproduced qualitatively well by variation of
349: the geometry of the problem, hopping integrals and electronic
350: filling for magnetic nanosystems.
351:
352: The work was supported by FOM project N0703M, NWO project
353: 047.016.005 and "Dynasty" foundation.
354:
355: \begin{thebibliography}{99}
356:
357: \bibitem{plummer} For review, see special issue of Surf. Sci. {\bf
358: 500}, 1-1053 (2002), ed. by C. B. Duke and E. W. Plummer;
359: especially E. W. Plummer et al, Surf. Sci. {\bf 500}, 1 (2002); J.
360: Schen and J. Kirschner, Surf. Sci. {\bf 500}, 300 (2002).
361:
362: \bibitem{STM} V. Madhavan et al, Science \textbf{280}, 567 (1998);
363: J. Li et al, Phys. Rev. Lett. \textbf{80},
364: 2893 (1998).
365:
366: \bibitem{coral} H. C. Manoharan et al, Nature
367: \textbf{403}, 512 (2000).
368:
369: \bibitem{kol} O. Yu. Kolesnichenko et al, Nature \textbf{415}, 507 (2002).
370:
371: \bibitem{gruner} G. Gr\"uner and A. Zawadowski, Rep. Prog. Phys.
372: {\bf 37}, 1497 (1974).
373:
374: \bibitem{hewson} A. C. Hewson, \textit{The Kondo Problem to Heavy
375: Fermions}, Cambridge University Press, Cambridge, 1993.
376:
377: \bibitem{CoAu}W. Chen et al,
378: Phys. Rev. B {\bf 60}, R8529 (1999).
379:
380: \bibitem{CrAu} T. Jamneala et al, Phys. Rev.
381: Lett. {\bf 87}, 256804 (2001).
382:
383: \bibitem{nanotube} T. W. Odom et al, Science {\bf 290}, 1549 (2000).
384:
385: \bibitem{doniach} S. Doniach, Physica B {\bf 91}, 231 (1977).
386:
387: \bibitem{irkhin} V. Yu. Irkhin and M. I. Katsnelson, Phys. Rev. B
388: {\bf 56}, 8109 (1997); ibid. {\bf 59}, 9348 (1999); ibid. {\bf
389: 61}, 14640 (2000).
390:
391: \bibitem{si} Q. Si et al,
392: Nature \textbf{413}, 804 (2001).
393:
394: \bibitem{castro} E. Novais et al, Phys. Rev. B {\bf 66}, 174409 (2002).
395:
396: \bibitem{zhu} J.-X. Zhu et al, Phys. Rev. Lett.
397: {\bf 91}, 156404 (2003).
398:
399: \bibitem{stewart} G.R. Stewart, Rev. Mod. Phys. {\bf 56}, 755
400: (1984); ibid. {\bf 73}, 797 (2001).
401:
402: \bibitem{amato} A. Amato, Rev. Mod. Phys. {\bf 69}, 1119 (1997).
403:
404: \bibitem{uzdin} Yu.B. Kudasov and V.M. Uzdin, Phys. Rev. Lett.
405: {\bf 89}, 276802 (2002).
406:
407: \bibitem{rubtsov} A. N. Rubtsov and A. I. Lichtenstein, JETP Lett.
408: {\bf 80}, 61 (2004); A. N. Rubtsov, cond-mat/0302228.
409:
410: \bibitem{georges} A. Georges et al,
411: Rev. Mod. Phys. {\bf 68}, 13 (1996).
412:
413: \bibitem{hirsch} J. E. Hirsch and R. M. Fye, Phys. Rev. Lett. {\bf 56},
414: 2521 (1986).
415:
416: \bibitem{maxent} M. Jarrell and J. E. Gubernatis, Phys. Rep. {\bf
417: 269}, 133 (1996).
418:
419: \bibitem{ldadmft} V. I. Anisimov {\it et~al.}, J. Phys.: Condens.
420: Matter {\bf 9}, 7359 (1997); A. I. Lichtenstein and M. I.
421: Katsnelson, Phys. Rev. B {\bf 57}, 6884 (1998).
422:
423: \end{thebibliography}
424:
425:
426: \end{document}
427: