1: \documentstyle[aps,prl,epsf,twocolumn]{revtex}
2: \def\o{\omega}
3: \def\a{\alpha}
4: \def\b{\beta}
5: \def\g{\gamma}
6: \def\d{\delta}
7: \def\e{\varepsilon}
8: \def\s{\sigma}
9: \def\chib{\overline{\chi}}
10: \def\@cite#1{{\footnotesize $^{#1}$}}
11: \begin{document}
12: \draft
13: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname @twocolumnfalse\endcsname
14: %
15: \title{
16: %Abstract designation: 1E-04 (April 30, 2000)\\
17: %Kondo insulator droplets
18: Griffiths phase of the Kondo insulator fixed point
19: %A novel Griffiths phase near the localization transition of a disordered Anderson lattice
20: }
21: %
22: \author{E. Miranda$^{\, (1)}$ and V. Dobrosavljevi\'c$^{\, (2)}$}
23: %
24: \address{$^{(1)}$Instituto de F\'{\i}sica ``Gleb Wataghin'', Unicamp,
25: C.P. 6165, Campinas, SP, CEP 13083-970, Brazil.\\
26: $^{(2)}$Department of Physics and
27: National High Magnetic Field Laboratory,
28: Florida State University,
29: Tallahassee, FL 32306.}
30: %
31: \maketitle
32:
33: \begin{abstract}
34: %
35: Heavy fermion compounds have long been identified as systems
36: which are extremely sensitive to the presence of impurities
37: and other imperfections. In recent years, both experimental and theoretical
38: work has demonstrated that such disorder can lead to unusual,
39: non-Fermi liquid behavior for most physical quantities.
40: In this paper, we show that this anomalous sensitivity to
41: disorder, as well as the resulting Griffiths phase behavior,
42: directly follow from the proximity of metallic heavy fermion systems
43: to the Kondo insulator fixed point.
44: %
45: \end{abstract}
46:
47: %\vspace*{0.5cm}
48:
49: %Keywords: Heavy-fermion systems, Non-Fermi liquid, Disordered systems,
50: %Anderson localization.
51:
52: %\vspace*{0.5cm}
53:
54: %$^*$ Corresponding author: Unicamp-IFGW-DFMC, C. P. 6165, CEP
55: %13083-970 Campinas, SP, Brazil. Fax: +55-19-3289-3137, E-mail:
56: %emiranda@ifi.unicamp.br
57:
58: %\vspace*{0.5cm}
59:
60: \pacs{PACS Numbers: 71.10.Hf, 71.27.+a, 72.15.Rn,75.20.Hr}
61: %\pacs{PACS Numbers: 71.10.Hf, 71.27.+a, 72.15.Rn,75.20.Hr}
62: %71.10.Hf - Non-Fermi-liquid ground states, electron phase diagrams
63: %and phase transitions in model systems
64: %71.27.+a - Strongly correlated electron systems; heavy fermions
65: %72.15.Rn - Localization effects (Anderson or weak localization)
66: %75.20.Hr Local moment in compounds and alloys; Kondo effect, valence
67: %fluctuations, heavy fermions
68: %71.20.Eh - Rare earth metals and alloys
69: %71.23.-k - Electronic structure of disordered solids
70: %75.30.Mb Valence fluctuation, Kondo lattice, and heavy-fermion
71: %phenomena
72: ]
73: \narrowtext
74:
75: Much of the recent interest in strongly correlated electronic systems,
76: such as heavy fermion compounds, has been sparked by a gamut of
77: unusual properties found in experiments. Most notably, many such
78: materials display unusual, non-Fermi liquid (NFL) behavior which is
79: often consistently observed in both the thermodynamic and transport
80: measurements. While the precise origin of these anomalies remains a
81: highly controversial issue, both the experimental and the theoretical
82: advances have provided convincing evidence\cite{general} that disorder
83: may be at the origin of such behavior, at least for certain classes of
84: materials. The purpose of this paper is to clarify the {\em physical
85: content} behind one of the proposed mechanisms for such
86: disorder-driven NFL behavior.
87:
88: Based on an initial success in explaining the anomalous behavior of
89: $\rm UCu_{5-x}Pd_x$ by means of the so-called Kondo disorder model
90: (KDM)\cite{nmr-msr,ourprev}, we have recently extended the analysis to
91: describe Anderson localization effects in a strongly correlated
92: environment\cite{sces,grifprl}. It is our intention to show here how
93: the emergence of NFL properties can be described in terms of a quantum
94: Griffiths phase induced by Anderson localization effects.
95: Furthermore, the onset of anomalous behavior occurs already at
96: relatively weak disorder, an effect we ascribe to the {\em proximity
97: to the Kondo insulator fixed point}. In a clean compound, unitary
98: Kondo scatterers act coherently to create a hybridization gap and the
99: state is the familiar Kondo insulator. A small deviation from
100: unitarity leads to the formation of heavy fermionic quasi-particles.
101: In a disordered system, however, large spatial fluctuations induce the
102: appearance of random unitary scatterers or ``Kondo insulator
103: droplets'' which are responsible for a strong {\em renormalized
104: effective disorder}. These droplets, in turn, are regions of depleted
105: density of states (DOS) which fail to quench nearby localized
106: moments. The latter ultimately lead to the NFL behavior.
107:
108:
109: Heavy fermion non-Fermi liquids are characterized by logarithmic or
110: weak power law divergence of the magnetic susceptibility $\chi(T)$ and
111: the specific heat coefficient $\gamma(T)=C_V(T)/T$ and a resistivity
112: that behaves as $\rho(T)=\rho_o+AT^\alpha$, with
113: $\alpha<2$. Deviations from normal Fermi-liquid behavior have also
114: been observed in optical conductivity\cite{optics},
115: magneto-resistance\cite{magnet}, dynamic magnetic
116: susceptibility\cite{neutron}, NMR and $\mu$SR\cite{nmr-msr}
117: measurements.
118:
119:
120: Several mechanisms have recently been proposed to account for such
121: anomalous behavior. One of them is provided by the proximity to an
122: ordering transition at zero temperature, a quantum critical point
123: (QCP)\cite{qcp-th}, where critical modes mediate unscreened long-range
124: interactions between carriers. Indeed, magnetic ordering is fairly
125: common in heavy-fermion physics and in many of the cleaner compounds a
126: strong case can be made in favor of a QCP
127: interpretation\cite{qcp-ex}. However, our current treatment of QCP's
128: has not produced a unified picture able to account for all the
129: observed behavior of these systems\cite{andysces}.
130:
131: Another possible route relies on the local dynamical frustration of
132: exotic impurity models, where the local moment cannot decide with
133: which conduction electron channel it will Kondo
134: bind\cite{dan}. Although a complete understanding of the single
135: impurity case is available, there remains the question of the
136: relevance of inter-site correlations\cite{jarrell}. This is important
137: because most of the studied systems are {\em not} in the dilute
138: limit\cite{amitsuka}.
139:
140: More recently, the idea that NFL behavior can occur due to the
141: presence of disorder in a strongly correlated environment has been
142: proposed. The pioneering work relied heavily on the Cu NMR line-widths
143: of $\rm UCu_{5-x}Pd_x$ and its temperature
144: dependence\cite{nmr-msr}. The KDM, a phenomenological model of a
145: system with a broad distribution of Kondo temperatures $T_K$, was then
146: proposed to account for the data. The presence of a wide range of
147: energy scales led to a picture where different spins were quenched at
148: widely different temperatures. Thus, the singular behavior could be
149: attributed to a few rare spins which remained unquenched (and
150: therefore, highly polarizable) at the lowest temperatures. These ideas
151: were put on a firmer foundation by means of a dynamical mean field
152: theory (DMFT)\cite{dmft} treatment of strong correlations and
153: disorder, where the KDM found its natural setting\cite{ourprev}. The
154: KDM had considerable success in describing a whole series of
155: experiments, including thermodynamic responses\cite{nmr-msr,ourprev},
156: neutron scattering\cite{neutron,ourprev}, optical
157: conductivity\cite{optics} and magneto-resistance\cite{magnet}. More
158: recently, we have introduced a microscopic model which incorporates
159: Anderson localization effects\cite{sces,grifprl}, a feature absent
160: from the KDM. We will discuss some of these recent results below.
161:
162: Alternatively, it has been proposed that the physics of these
163: compounds is intimately tied to disorder effects in the proximity to
164: magnetic ordering\cite{antonio}. The effect of spatial inhomogeneities
165: together with the tendency of the Kondo effect to destroy magnetic
166: order would then lead to the formation of large clusters of
167: magnetically ordered material close to the phase boundary but still
168: within the disordered phase. Similarly, these large ordered clusters
169: would be responsible for the NFL behavior.
170:
171:
172: In all proposed disorder-based mechanisms, the low temperature
173: anomalies result from a broad distribution of local energy scales in
174: the system. In addition, these energy scales are viewed as the
175: appropriate Kondo temperatures describing either individual spins or
176: clusters of spins embedded in a metal. In all the theories, these
177: Kondo temperatures assume a strong, exponential dependence on the
178: parameters describing the fluctuator in question, hence their broad
179: distribution even for moderate disorder. In the simplest KDM, the
180: emergence of low $T_K$ sites is a result of the randomness in the
181: immediate environment of a given spin. In the ``spin cluster''
182: picture, the ``cluster Kondo temperature'' is exponentially dependent
183: on the cluster size, a quantity which can be expected to be large
184: close to any magnetic quantum critical point. Finally, in very recent
185: work, the depression of $T_K$ was proposed to be a result of Anderson
186: localization effects, which lead to a reduction of the local density
187: of conduction electron states required for Kondo screening.
188:
189: From a general point of view, all three proposed scenarios may be
190: relevant and can be expected to contribute. A more immediate question
191: is the relative practical significance of these processes and their
192: general robustness with respect to different material characteristics.
193: In this respect, recent experiments have provided evidence that the
194: local disorder inherent to the KDM picture may not be sufficient to
195: account for the observed anomalies\cite{booth}. Similarly, the
196: emergence of large magnetically ordered clusters can be anticipated
197: only in the very close vicinity of magnetic transitions. More
198: importantly, it is difficult to imagine how the cluster picture could
199: even come close to providing enough residual entropy to account for
200: the observed specific heat anomalies in the physical temperature
201: range.
202:
203: In contrast, we will show that the localization-based scenario
204: provides a very robust and quantitatively relevant mechanism for the
205: NFL behavior. Since the corresponding fluctuations must be present in
206: any moderately disordered system, this route should be of direct
207: relevance to most disordered heavy fermion compounds, irrespective of
208: the proximity to any magnetic ordering. Our key point is that the
209: corresponding Griffiths phase is a direct consequence of the proximity
210: not to any magnetic, but rather to the Kondo insulator fixed
211: point. The deviation from the Kondo insulator provides an energy scale
212: which is universally small for any heavy fermion metal, since it is
213: defined by the underlying Kondo energy. This observation also provides
214: a simple explanation for the notorious sensitivity of HF systems to
215: disorder, a feature which we believe is at the origin of all the
216: observed anomalies.
217:
218: We start from a disordered infinite-U Anderson lattice Hamiltonian
219:
220: \begin{eqnarray}
221: H &=& \sum\limits_{ij\sigma} \left(-t_{ij} + \e_i\d_{ij}\right)
222: c^{\dagger}_{i\sigma} c^{\phantom{{\dagger}}}_{j\sigma}
223: + \sum\limits_{j\sigma} E_{fj} f^{\dagger}_{j\sigma}
224: f^{\phantom{{\dagger}}}_{j\sigma} \nonumber \\
225: &+& \sum\limits_{j\sigma} V_j (c^{\dagger}_{j\sigma}
226: f^{\phantom{{\dagger}}}_{j\sigma} + {\rm H. c.} ),
227: \label{hammy}
228: \end{eqnarray}
229: where, $c_{i\sigma}$ destroys a conduction electron at site $i$ and
230: spin $\sigma$ from a broad uncorrelated band with hopping $t_{ij}$ and
231: $f_{j\sigma}$ destroys an f-electron at site $j$ with spin
232: $\sigma$. Since $U\to\infty$, the constraint $n^f_j
233: =\sum_{\sigma}f^{\dagger}_{j\sigma} f^{\phantom{{\dagger}}}_{j\sigma}
234: \le 1$ is assumed.
235:
236: Typically, alloying introduces substitutions either in the f-shell
237: sub-lattice (``Kondo holes''), e. g. $\rm U_{1-x}Y_xPd_3$, or the
238: non-f-shell subsystem (``Ligand disorder''), e. g. $\rm
239: UCu_{5-x}Pd_x$. It is reasonable to assume that the former case should
240: be modeled by a distribution of $E_{fj}$, whereas the latter is
241: expected to introduce randomness both in $V_j$ and $\e_j$. Most of
242: our results remain unchanged irrespective of the kind of disorder.
243:
244: We work within the framework of the recently introduced statistical
245: dynamical mean field theory (SDMFT)\cite{sdmft}. It is a natural
246: generalization of the DMFT, which retains the latter's treatment of
247: local correlations while going beyond it by incorporating Anderson
248: localization effects. It is most easily implemented on a Bethe lattice
249: of coordination $z$ (we have used $z=3$ in our simulations). Each
250: lattice site $j$ defines an effective local action for the f-orbital
251: which should be thought of as resulting from integrating out all the
252: other electronic degrees of freedom. It is written as
253: ($U\to\infty$)\cite{sces}
254: \begin{eqnarray}
255: S_{{\rm eff}}^{(j)} &=&\int_{0}^{\beta }d\tau
256: \sum_{\sigma }f_{j,\sigma }^{\dagger }(\tau )\left( \partial _{\tau
257: }+E_{fj}\right)f_{j,\sigma
258: }(\tau ) +\nonumber \\
259: &&\int_{0}^{\beta }d\tau \int_0^{\beta}d\tau ^{\prime }
260: \sum_{\sigma }f_{j,\sigma }^{\dagger }(\tau )
261: \Delta _{j}(\tau -\tau ^{\prime })f_{j,\sigma
262: }(\tau ^{\prime });
263: \label{seff}\\
264: &&\Delta _{j}(\omega ) = \frac{V_{j}^{2}}{\omega -\e
265: _{j}-\sum_{k=1}^{z-1}t_{jk}^{2}G_{ck}^{(j)}(\omega )}. \label{hybrid}
266: \end{eqnarray}
267: Here, $G_{ck}^{(j)}(\omega )$ is the local c-electron Green's function
268: on the nearest neighbor site $k$ with site $j$ removed, in other
269: words, a ``cavity'' has been created where there once was site
270: $j$. Note how, as in the DMFT\cite{dmft}, we neglect higher order
271: Green's functions in the process. $G_{ck}^{(j)}(\omega )$, on the
272: other hand, is determined recursively by relating it to a similar
273: quantity on the next nearest neighbor site $l$
274: \begin{eqnarray}
275: G_{ck}^{(j)(-1)}(\omega ) &=&\omega -\e
276: _{k}-\sum_{l=1}^{z-1}t_{kl}^{2}G_{cl}^{(k)}(\omega )
277: - \Phi_k (\omega ), \label{recur}\\
278: \Phi_k (\omega )
279: &=&\frac{V_{k}^{2}}{\omega -E_{fk}-\Sigma _{fk}(\omega
280: )}. \label{phi}
281: \end{eqnarray}
282: Finally, the self-consistency loop is closed by requiring that the
283: {\em local} self-energy $\Sigma _{fk}{(\omega )} $ be obtained from
284: the solution of the effective action $S_{{\rm eff}}^{(k)}$ , see
285: Eq.~(\ref {seff})\cite{sdmft,sces}. We note that the self-consistent
286: set of stochastic Eqs.~(\ref{seff}-\ref{phi}) reduces to the DMFT when
287: the limit $z\to\infty$ is taken (with the appropriate rescaling
288: $t_{ij}\sim t/\sqrt{z}$), in which case the disorder is treated on the
289: CPA level and, therefore, shows no localization
290: effects\cite{dinfdis}. This was a severe limitation of the previous
291: treatment\cite{ourprev} which is here remedied. On the other hand, the
292: non-interacting limit ($U=0$) reduces to the self-consistent theory of
293: localization of Abou-Chacra, Anderson and Thouless\cite{abou}, which
294: is known to exhibit an Anderson metal-insulator transition (MIT) for
295: $z\ge3$.
296:
297: \begin{figure}
298: \epsfxsize=3.2in \epsfbox{fig1.eps}
299: \caption{The conduction electron wave function $\Psi_c(\vec{r})$ has
300: strong amplitude fluctuations due to disorder and correlate several
301: local moments within a correlation length's distance.
302: \label{fig1}}
303: \end{figure}
304:
305:
306: We would like to stress that the hybridization function
307: Eq.~(\ref{hybrid}) ``seen'' by each f-orbital has strong spatial
308: fluctuations, reflecting the disorder inside a correlation volume
309: enclosing several lattice sites in its neighborhood. The spectral
310: information is carried by the extended conduction electron wave
311: function, which acts to correlate the different Anderson impurity
312: problems defined by Eq.~(\ref{seff}) (see Fig.~\ref{fig1}). This leads
313: to a distribution of Kondo temperatures. On the other hand, the effect
314: of this {\em ensemble} of single-impurity problems is to create a
315: renormalized effective disorder ``seen'' by the conduction electrons,
316: {\it cf.} Eqs.~(\ref{recur},\ref{phi}). The net effect on the
317: relevant distribution functions of this {\em highly non-local}
318: self-consistency turns out to be robust and universal.
319:
320:
321:
322:
323: The stochastic equations~({\ref{seff}-\ref{phi}) were solved by
324: standard sampling techniques\cite{abou} and provided us with the
325: statistical distributions of the most important physical
326: quantities. The single-impurity problem of~(\ref{seff},\ref{hybrid})
327: was solved in the large-N mean-field approximation at
328: $T=0$\cite{largeN}, which has the desirable feature of correctly
329: reproducing the exponential form of the Kondo temperature. We have
330: made sure the statistics and numerical procedures were accurate enough
331: to obtain Kondo temperatures spanning many orders of magnitude ($\sim
332: 15$) in order to probe the long tails of the distribution functions.
333:
334: \begin{figure}
335: \epsfxsize=3.2in \epsfbox{fig2.eps}
336: \caption{Distribution of $T_K$ showing the emergence of NFL
337: behavior. Here, $\e_i$'s are distributed uniformly with width $W$ and
338: we have used $z=3$, $E_f=-1$, $V=0.5$ and $\mu=-0.1$. Inset: the
339: exponent $\alpha$ of Eq.~(\ref{power-law}). The dashed line indicates
340: the marginal case $\alpha=1$.
341: \label{fig2}}
342: \end{figure}
343:
344:
345: One of our main results is the identification of a NFL region at
346: relatively weak disorder\cite{grifprl}. This is triggered by the
347: appearance of a fraction of localized moments with very low $T_K$'s,
348: as can be seen in Fig.~\ref{fig2}. Indeed, the $T_K$ distribution
349: exhibits {\em power law} behavior as $T_K\to 0$
350: \begin{equation}
351: P(T_K)\sim T_K^{(\alpha-1)},
352: \label{power-law}
353: \end{equation}
354: where the exponent $\alpha$ varies continuously with the amount of
355: disorder (see the inset of Fig.~\ref{fig2}). This can be shown to lead
356: to singular thermodynamic responses
357: \begin{equation}
358: \chi(T) \sim \gamma(T) \sim \frac{1}{T^{(1-\alpha)}},
359: \end{equation}
360: as has been observed in several heavy fermion alloys\cite{marcio}.
361: The ``marginal'' case $\alpha=1$ leads to a logarithmic divergence of
362: the same quantities. The occurrence of such NFL behavior in a system
363: with a wide distribution of $T_K$'s had been proposed in the context
364: of the KDM\cite{ourprev}. However, in the KDM the presence of these
365: spins could only be obtained through a finely tuned choice of the bare
366: disorder distribution. By contrast, in our current treatment this
367: behavior is an unavoidable consequence of the spatial fluctuations of
368: the conduction electron wave function amplitude. Due to the extended
369: nature of this wave function and the consequent correlation between
370: $T_K$ values on different sites (Fig.~\ref{fig1}) we should expect a
371: high degree of robustness and universality in these
372: distributions. This is indeed what we have found for different types
373: of disorder distributions. We also note that the case depicted in
374: Fig.~\ref{fig2} corresponds to conduction band disorder {\em only},
375: for which the KDM would predict {\em no} $T_K$ fluctuations.
376:
377:
378:
379:
380:
381:
382: Diverging thermodynamic responses with disorder-dependent exponents
383: due to rare regions with very low $T_K$'s are characteristic of
384: Griffiths phases\cite{griffiths}. Since Anderson localization effects
385: are the driving mechanism here, we associate this Griffiths phase to
386: the proximity to a disorder-driven metal-insulator transition (MIT).
387: This can be checked through an examination of the typical density of
388: states (DOS) of the conduction electrons $\rho_{typ} =
389: \exp\{ <\ln \rho_j >\}$; $\rho_j = (1/\pi )Im G_{cj} (\omega =0)$, a
390: quantity known to vanish at the MIT. We show our results in
391: Fig.~\ref{fig3}. The Griffiths phase is observed for relatively small
392: amounts of disorder ($W/t\approx1$), whereas the MIT occurs at much
393: higher values ($W/t\approx 12$). Surprisingly, however, the typical
394: DOS is a {\em non-monotonic} function of the disorder strength (for
395: $-0.2 \alt \mu \alt 0.3$), in sharp contrast to the non-interacting case
396: (Fig.~\ref{fig3}).
397:
398:
399: \begin{figure}
400: \epsfxsize=3.2in \epsfbox{fig3.eps}
401: \caption{Localization properties of the conduction electrons as
402: monitored by their typical DOS as a function of disorder, for several
403: values of the chemical potential $\mu$. The other parameters are the
404: same as in Fig.~\ref{fig2}. $\mu=0.1$ corresponds to a Kondo insulator
405: in the clean case. While the non-interacting case is monotonically
406: decreasing, the interaction-induced renormalized disorder leads to a
407: dip for small values of $W/t$. We attribute this dip to the proximity
408: to the Kondo insulator fixed point.
409: \label{fig3}}
410: \end{figure}
411:
412: In order to gain more insight into this rather non-intuitive behavior,
413: we look at the effective renormalized disorder ``seen'' by the
414: conduction electrons. It is clear from Eq.~(\ref{recur}) that, in
415: addition to the bare disorder in the $\e_k$'s, scattering from the
416: strongly correlated f-sites adds an {\em effective} disorder described
417: by the quantity $\Phi_k(\omega)$ of Eq.~(\ref{phi}). In particular,
418: unitary scattering ($\delta=\pi/2$) corresponds to
419: $\Phi(0)\to\infty$. A diverging $\Phi(0)$ in the clean case leads to
420: the formation of a hybridization gap and Kondo insulating behavior. It
421: is precisely the appearance of the first unitary scatterers (USC's)
422: once disorder is introduced which is responsible for the sharp drop in
423: $\rho_{typ}$ seen in Fig.~\ref{fig3}. This can be clearly seen from
424: the distribution of $\Phi_k^{-1}(0)$ as a function of disorder in
425: Fig.~\ref{fig4}.
426:
427: \begin{figure}
428: \epsfxsize=3.2in \epsfbox{fig4.eps}
429: \caption{Distribution of $\Phi_k^{-1}(0)$ as a function of
430: disorder. The inset shows the concentration of unitary scatterers
431: ($\Phi_k^{-1}(0)=0$). Same parameters as in Fig.~\ref{fig2}, but with
432: $\mu=0$.
433: \label{fig4}}
434: \end{figure}
435:
436: Therefore, the following picture emerges from our results. The clean
437: system has a small energy scale, the Kondo temperature $T_K$, which
438: sets the distance from the Kondo resonance to the Fermi energy. When
439: disorder is introduced, the positions of the Kondo resonances start to
440: fluctuate and a distribution of $\Phi_k(0)$ ensues, giving rise to a
441: renormalized effective disorder. The latter initially broadens
442: monotonically and USC's are quickly formed, having a dramatic effect
443: on the conducting properties. The extreme sensitivity of the system to
444: even weak disorder is therefore a result of the existence of the small
445: energy scale $T_K$, which sets the proximity to the Kondo
446: insulator. Thus, the deeper in the Kondo limit, the more sensitive the
447: system will be to disorder. Note that the value of $\Phi_k(0)$ is {\em
448: not} set by $T_K$ but is rather a local measure of {\em particle-hole
449: asymmetry} (see the first ref. of~\cite{ourprev}). As the disorder is
450: further increased, the DOS depletion of the Kondo (pseudo-)gap is
451: washed way and the concentration of USC's saturates. Finally, once the
452: nearby Kondo fixed point is securely destroyed, conventional
453: localization effects set in and the system proceeds to an
454: Anderson-like MIT. That the actual MIT is not very much affected by
455: the renormalized disorder is evidenced by the fact that we observe it
456: to occur at the same point as the non-interacting system. It is the
457: competition between renormalized and bare disorder which, in a
458: three-stage process, leads to the non-monotonic behavior of
459: Fig.~\ref{fig3}.
460:
461: It is worth stressing that the emergence of the Griffiths phase
462: behavior already at very weak disorder directly follows from the
463: described disorder renormalization as induced by the proximity of an
464: incipient Kondo insulator. Indeed, this mechanism leads to a very
465: large {\em effective} disorder as seen by conduction electrons, which
466: in turn produces the localization-induced local density of states
467: fluctuations, and the resulting broad distributions of Kondo
468: temperatures. In other correlated systems, such as the disordered
469: Hubbard model\cite{sdmft}, the Kondo gap is not present, and the
470: relevant Griffiths phase proves to be restricted to the immediate
471: vicinity of the Mott-Anderson transition, in dramatic contrast to what
472: we find here.
473:
474:
475:
476: In conclusion, we have elucidated the mechanism of the Griffiths phase
477: observed in our studies of disordered Anderson lattices. Within our
478: theory, the NFL behavior proves to be intimately related to the
479: physics of disordered metals close to the Kondo insulator fixed point.
480:
481: We would like to acknowledge useful discussions with M. C. Aronson,
482: D. L. Cox, G. Kotliar, D. E. MacLaughlin, A. J. Millis, P. Schlottmann
483: and G. Zarand. This work was supported by FAPESP and CNPq (EM), NSF
484: grant DMR-9974311 and the Alfred P. Sloan Foundation (VD).
485:
486:
487:
488: \begin{references}
489:
490:
491: \bibitem{general} For an overview, see the papers presented at the
492: {\it Conference on Non-Fermi Liquid Behavior in Metals}, J. Phys.:
493: Cond. Matter {\bf 8} (48) 1996.
494:
495: \bibitem{nmr-msr} O. O. Bernal {\it et al.}, Phys. Rev. Lett.
496: 75 (1995) 2023; D. E. MacLaughlin {\it et al.} J. Phys.: Cond. Matter
497: 8 (1996) 9855; D. E. MacLaughlin {\it et al.}, Phys. Rev. B 58 (1998)
498: 11849; Chia-Ying Liu {\it et al.}, cond-mat/9909215.
499:
500:
501: \bibitem{ourprev} E. Miranda, V. Dobrosavljevi\'c, and G. Kotliar,
502: J. Phys.: Cond. Matter {8} (1996) 9871; Phys. Rev. Lett. {78} (1997)
503: 290; Physica B {230} (1997) 569.
504:
505: \bibitem{sces} E. Miranda and V. Dobrosavljevi\'{c}, Physica B 259-261
506: (1999) 359.
507:
508: \bibitem{grifprl} E. Miranda and V. Dobrosavljevi\'c,
509: cond-mat/0023209.
510:
511: \bibitem{optics} L. Degiorgi L and H. R. Ott, J. Phys.:
512: Condens. Matter 8 (1996) 9901; A. Chattopadhyay and M. Jarrell,
513: Phys. Rev. 56 (1997) 2920.
514:
515: \bibitem{magnet} A. Chattopadhyay {\it et al.}, cond-mat/9805127.
516:
517: \bibitem{neutron} M. C. Aronson {\it et al.}, Phys. Rev. Lett. 75
518: (1995) 725.
519:
520:
521: \bibitem{qcp-th} M. A. Continentino, G. Japiassu, and A. Troper,
522: Phys. Rev. B 39 (1989) 9734; A. J. Millis, Phys. Rev. B 48 (1993)
523: 7183; A. Rosch {\it et al.}, Phys. Rev. Lett. 79 (1997) 159.
524:
525:
526: \bibitem{qcp-ex} H. v. L\"{o}hneysen {\it et al.},
527: Phys. Rev. Lett. {\bf 72}, 3262 (1994); F. M. Grosche {\it et al.},
528: Physica B {\bf 223+224}, 50 (1996); N. D. Mathur {\it et al.}, Nature
529: {\bf 394}, 39 (1998); A. Schr\"{o}der {\it et al.},
530: Phys. Rev. Lett. 80 (1998) 5623; O. Stockert {\it et al.},
531: Phys. Rev. Lett. 80 (1998) 5627.
532:
533: \bibitem{andysces} A. J. Millis, Physica B 259-261 (1999) 1169.
534:
535:
536: \bibitem{dan} P. Nozi\`{e}res and A. Blandin, J. Physique 41 (1980)
537: 193; D. L. Cox, Phys. Rev. Lett. 59 (1987) 1240.
538:
539: \bibitem{jarrell} M. Jarrell, H. B. Pang, D. L. Cox, and K. H. Luk,
540: Phys. Rev. Lett. 77 (1996) 1612; M. Jarrell, H. B. Pang, D. L. Cox,
541: and F. Anders, Physica B 230 (1997) 557.
542:
543: \bibitem{amitsuka} See, however, H. Amitsuka and T. Sakakibara
544: J. Phys. Soc. Jpn. 63 (1994) 736.
545:
546: \bibitem{dmft} A. Georges, G. Kotliar, W. Krauth, and M. J. Rozenberg;
547: Rev. Mod. Phys. 68 (1996) 13.
548:
549: \bibitem{antonio} A. H. Castro Neto, G. E. Castilla, and B. A. Jones,
550: Phys. Rev. Lett 81 (1997) 3531; A. H. Castro Neto, cond-mat/9911092;
551: A. H. Castro Neto and B. A. Jones, cond-mat/0003085.
552:
553: \bibitem{booth} C. H. Booth, D. E. MacLaughlin, R. H. Heffner,
554: R. Chau, M. B. Maple, and G. H. Kwei, Phys. Rev. Lett. 81 (1998) 3960;
555: E. Miranda and V. Dobrosavljevi\'{c}, unpublished.
556:
557:
558: \bibitem{sdmft} V. Dobrosavljevi\'{c} and G. Kotliar,
559: Phys. Rev. Lett. 78 (1997) 3943; V. Dobrosavljevi\'{c} and
560: G. Kotliar, Phil. Trans. R. Soc. Lond. A356 (1998) 1.
561:
562: \bibitem{dinfdis} V. Janis and D. Vollhardt, Phys. Rev.
563: B 46 (1992) 15712; V. Dobrosavljevi\'{c} and G. Kotliar, Phys. Rev.
564: Lett. 71 (1993) 3218 and Phys. Rev. B 50 (1994) 1430.
565:
566: \bibitem{abou} R. Abou-Chacra, P. W. Anderson, and D. J. Thouless,
567: J. Phys. C 6 (1973) 1734.
568:
569:
570: \bibitem{largeN} S. Barnes, J. Phys. F 7 (1977) 2637; N. Read and
571: D. M. Newns, J. Phys. C 16 (1983) L1055; P. Coleman, Phys. Rev. B 35
572: (1987) 5072.
573:
574: \bibitem{marcio} M. C. de Andrade {\it et al.}, Phys. Rev. Lett. {\bf
575: 81}, 5620 (1998).
576:
577: \bibitem{griffiths} R. B. Griffiths, Phys. Rev. Lett. 23 (1969) 17.
578:
579:
580:
581: \end{references}
582: \end{document}