1: %\documentstyle[preprint,prb,aps]{revtex} %double space
2: \documentstyle[prb,aps,floats]{revtex}
3: \input epsf.sty
4: %
5: \newcommand{\rf}[1]{(\ref{#1})}
6: \newcommand{\cao}{\c c\~ao\ }
7: \newcommand{\coes}{\c c\~oes\ }
8: \newcommand{\jpcm}{J.Phys.: Condens.\ Matter\ }
9: %
10: \begin{document}
11: %\draft
12: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname@twocolumnfalse\endcsname
13:
14: \title{Multiperiodic magnetic structures in Hubbard superlattices}
15:
16: \author{Andr\'e L.\ Malvezzi$^{\, (1)}\!$, Thereza Paiva$^{\, (2)}\!$ and
17: Raimundo R. dos Santos$^{\, (2)}\!$}
18:
19: \address{$^{(1)}$Departamento\ de F\'\i sica,
20: Faculdade de Ci\^encias,
21: Universidade Estadual Paulista,
22: Cx.P.\ 473,
23: 17015-970 Bauru SP,
24: Brazil\\
25: $^{(2)}$Instituto de F\'\i sica,
26: Universidade Federal do Rio de Janeiro,
27: Cx.P.\ 68528,
28: 21945-970 Rio de Janeiro RJ, Brazil}
29: \date{\today}
30: \maketitle
31:
32: \begin{abstract}
33: We consider fermions in one-dimensional superlattices (SL's), modeled by
34: site-dependent Hubbard-$U$ couplings arranged in a repeated pattern of
35: repulsive (i.e., $U>0$) and free ($U=0$) sites. Density Matrix
36: Renormalization Group (DMRG) diagonalization of finite systems is used to
37: calculate the local moment and the magnetic structure factor in the ground
38: state. We have found four regimes for magnetic behavior: uniform local
39: moments forming a spin-density wave (SDW), `floppy' local moments with
40: short-ranged correlations, local moments on repulsive sites forming
41: long-period SDW's superimposed with short-ranged correlations, and local
42: moments on repulsive sites solely with long-period SDW's; the boundaries
43: between these regimes depend on the range of electronic densities $\rho$
44: and on the SL aspect ratio. Above a critical electronic density,
45: $\rho_{\uparrow \downarrow}$, the SDW period oscillates both with $\rho$
46: and with the spacer thickness. The former oscillation allows one to
47: reproduce all SDW wave vectors within a small range of electronic
48: densities, unlike the homogeneous system. The latter oscillation is
49: related to the exchange oscillation observed in magnetic multilayers.
50: A crossover between regimes of `thin' to `thick' layers has also been
51: observed.
52: \end{abstract}
53: %\bigskip
54: \pacs{
55: PACS:
56: 75.75.+a, % Magnetic properties of nanostructures
57: 71.27.+a, % strongly correlated
58: 75.70.-i. % Magn. films and multilayers
59: 71.10.-w, % Theories and models of many-elctron systems}
60: }
61: \vskip2pc]
62:
63: \section{Introduction}
64: \label{Intro}
65:
66: Magnetic multilayers have been the subject of intense study over the last
67: decade. The technologically important giant magnetoresistance (GMR) is one
68: of the most interesting aspects of these compounds. Another aspect which
69: has brought attention to multilayers is the oscillation of the exchange
70: coupling between magnetic layers as the spacer layer thickness is varied.
71: While oscillations with single periods have been well understood for some
72: time, {\em multiperiodicity} has been theoretically predicted,\cite{qw,rkky}
73: and indeed observed, in trilayer materials.
74: Fe/Cr/Fe samples grown by sputtering or molecular-beam epitaxy (MBE)
75: display two periods of oscillation of the exchange coupling: a
76: so-called long-period, of about 10 to 12 monolayers thick, is
77: superimposed to a short-period component of about two monolayers
78: thick.\cite{Unguris} This superposition of short- and long-period
79: components has also been observed in other MBE-fabricated trilayer
80: materials such as Fe/Mn/Fe,\cite{FeMn} Fe/Au/Fe, \cite{FeAu}
81: Fe/Mo/Fe,\cite{FeMo} and Co/Cu/Co.\cite{CoCu}
82: Short-period oscillations, however, disappear if interface quality
83: is not carefully maintained.\cite{Unguris,Stoeffler,Wolf} Recent
84: experiments\cite{Schmidt} in Fe/Cr/Fe show that areas of constant Cr
85: thickness, with diameter larger than 3-4 nm on the interface, are
86: necessary for the development of short-period oscillations. It is
87: therefore believed that multiperiodicity has not yet been observed in
88: multilayers due to interface roughness.
89:
90: From the theoretical point of view, both the quantum well theory\cite{qw}
91: and the so-called RKKY theory \cite{rkky} can account for many features
92: related to the oscillations of the exchange coupling. For instance, a
93: direct relation between the periods of oscillation and Fermi surface extrema
94: of bulk spacers has been established.\cite{qw,rkky,mauro} However, since
95: the notion of a Fermi surface is not widely applicable to strongly
96: correlated systems, a deeper understanding of multiperiodicity is clearly
97: in order, and microscopic models should provide useful insights.
98:
99: With this in mind, here we investigate the magnetic properties of a
100: one-dimensional superlattice (SL) model\cite{tclp1,longo,mit} in which
101: electronic correlations are incorporated and treated nonperturbatively.
102: The model consists of a periodic arrangement of $L_U$ sites (``layers'')
103: in which the on-site coupling is repulsive, followed by $L_0$ free (i.e.,
104: $U=0$) sites. The role played by relative layer thicknesses on the
105: magnetic and conducting properties of these systems is then probed by
106: varying $L_0$ and $L_U$.
107:
108: The SL structure gives rise to several remarkable features,\cite{tclp1}
109: in marked contrast with the otherwise homogeneous system: Local moment
110: maxima can be transferred from repulsive to free sites, and the range of
111: parameters in which this occurs has been expressed in terms of a `phase
112: diagram'.\cite{longo} In addition, spin--density-wave (SDW) quasi-order
113: can be wiped out as a result of frustration, and the SL structure also
114: induces a shift in the density $\rho_I$ at which a Mott-Hubbard insulating
115: phase sets in.\cite{mit} Further, by examining the Luttinger liquid version
116: of the model,\cite{llsl1} one finds that these superlattices provide the
117: means to realize {\em gapless insulating phases.}\cite{llsl2}
118:
119: Previous studies of the discrete version of the
120: model\cite{tclp1,longo,mit} resorted to Lanczos diagonalization, which
121: sets limits on the system sizes used; for instance, a 24-site lattice size
122: could only be considered for the low- and high-density regimes ($\rho=1/6$
123: and $\rho=11/6$). Nonetheless, one was still able to probe the period of
124: exchange oscillations for these special densities through the analysis of
125: the magnetic structure factor: the peak position displayed oscillatory
126: behavior with the spacer thickness.\cite{longo} Here we use the Density
127: Matrix Renormalization Group (DMRG) technique\cite{DMRG} to study
128: superlattices longer than the ones available through the Lanczos method.
129: With the aid of the magnetic structure factor, we have been able to probe
130: the periodicity of the superlattice over a wider range of layer
131: thicknesses and densities. As we will see, this has led to significant
132: improvements on the phase diagram previously reported,\cite{longo} with
133: the addition of information relative to the regions in which one- and
134: two--period-oscillations are found; as it turned out, these regions are
135: closely related to the behavior of the local moment.
136: We have also been able to observe a crossover between the regimes of
137: thin and thick layers; in the latter regime, the `aspect ratio'
138: $\ell\equiv L_U/L_0$ is the only relevant geometric parameter, whereas
139: the magnetic behavior in the former regime depends on $L_U$ and $L_0$
140: separately.
141:
142: The layout of the paper is as follows: In Section \ref{model} we introduce
143: the superlattice model and comment on the calculational procedure. Section
144: \ref{localmom} focus on the local moment and how it changes with density
145: and layer thickness. The magnetic structure factor and the periodicity of
146: the superlattices are discussed in Section \ref{sq}, and Section
147: \ref{conc} summarizes our findings.
148:
149:
150: \section{Model and Calculational Procedure}
151: \label{model}
152:
153:
154: We define the Hamiltonian as
155: \begin{equation}
156: \label{Ham}
157: {\cal H}=-t\sum_{i,\ \sigma}
158: \left(c_{i\sigma}^{^{\dagger}} c_{i+1\sigma}+\text{H.c.}\right)
159: + \sum_i U_i\ n_{i\uparrow}n_{i\downarrow}
160: \label{H}
161: \end{equation}
162: where, in standard notation, $i$ runs over the sites of a one-dimensional
163: lattice, $c_{i\sigma}^{^\dagger}$ ($c_{i\sigma}$) creates (annihilates) a
164: fermion at site $i$ in the spin state $\sigma=\ \uparrow$ or $\downarrow$,
165: and $n_i=n_{i\uparrow}+n_{i\downarrow}$, with
166: $n_{i\sigma}=c_{i\sigma}^{^\dagger}c_{i\sigma}$; the on-site Coulomb
167: repulsion is taken to be site-dependent: $U_i=U>0$, for sites within the
168: repulsive layers, and $U_i=0$ otherwise.
169:
170: We consider the Hamiltonian \rf{H} on lattices with $N_s$ sites and $N_e$
171: electrons, and open boundary conditions are used. The appropriate Finite
172: Size Scaling (FSS) parameter, however, is the number of periodic cells,
173: $N_c=N_s/N_b$, for a basis with $N_b=L_U + L_0$ sites. The ground state
174: wave function and energy are obtained by numerical diagonalization using
175: the finite-system density-matrix renormalization group (DMRG)
176: method.\cite{DMRG} We used lattice sizes up to 150 sites, and
177: truncation errors in the DMRG procedure were kept around $10^{-5}$ or
178: smaller. We have performed a systematic study of the magnetic properties
179: for different values of the Coulomb repulsion $U$, different occupation
180: $\rho = N_e/N_s$ and different configurations $\{U_i\}$. Not all
181: configurations $\{U_i\}$ fit into all sizes and occupations but, since
182: DMRG allows us to study a wide range of lattice sizes, we were able to
183: establish overall trends.
184:
185:
186: \section{Local Moment Profile}
187: \label{localmom}
188:
189: The local moment at site $i$ is defined as $\langle S_i^2\rangle={3\over
190: 4} \langle (n_{i\uparrow}-n_{i\downarrow})^2 \rangle$, and is a measure of
191: both the magnetism and the degree of itinerancy of the system. Figure
192: \ref{si2xi} shows the local moment profile for the SL with $L_U=1$,
193: $L_0=2$, $U=4$, $N_s=48$, and for three different densities; effects of
194: system size on the local moment are negligible. For small densities, such
195: as for $\rho=0.25$, one identifies small-amplitude oscillations in the
196: local moment profile; their period ($2\pi/2k_F$, with $2k_F=\pi\rho$) is
197: determined not by the underlying SL structure, but by the Friedel
198: oscillations in the charge density of the otherwise homogeneous
199: system.\cite{Noack}
200:
201: As the density is increased, the SL structure dominates over the Friedel
202: oscillation as evidenced by the data: for $\rho=0.667$ the maxima lie on
203: the free sites and the modulation of the profile perfectly matches that of
204: the SL. For large enough densities the maxima migrate to the repulsive
205: sites, as shown by the data for half filling. One should also note that
206: even at the maxima $\langle S_i^2 \rangle$ is considerably reduced from
207: its value at the completely localized limit ($U = \infty$), namely
208: $\langle S_i^2\rangle=3/4$; the itinerant behavior in these cases is
209: therefore evident.
210:
211:
212: \begin{figure}
213: \epsfxsize=8.5cm
214: \begin{center}
215: %\leavevmode
216: \epsffile{si2xil12.eps}
217: \caption{Local moment as a function of the site ($i$) for the SL with
218: $L_U=1$, $L_0=2$, $U=4$, $N_s=48$, for $\rho=0.25$ (squares),
219: $\rho=0.667$ (triangles) and half-filling (circles). The local
220: moment profile changes qualitatively as the density increases.}
221: \label{si2xi}
222: \end{center}
223: \end{figure}
224:
225: The above example illustrates the existence of three regions,
226: characterized by different local moment profiles: homogeneous (or
227: Friedel-like), free-site peaked, and repulsive-site peaked. In order to
228: locate the boundaries between these regimes it is useful to determine how
229: the local moment at repulsive and free sites separately change with
230: the density. In addition, we define a bias of the local moment maxima as
231: \begin{equation}
232: \delta \equiv {\langle S_U^2 \rangle - \langle S_0^2 \rangle \over
233: \langle S_U^2 \rangle +\langle S_0^2 \rangle},
234: \label{bias}
235: \end{equation}
236: and also study its dependence with the density.
237:
238: Figure \ref{l01bias} shows the local moment [both at repulsive ($\langle
239: S_U^2\rangle$) and free sites ($\langle S_0^2\rangle$)] and the bias as
240: functions of the density, for $U=4$. In the case of Fig.\ \ref{l01bias},
241: $L_U\geq L_0$, with all SL configurations having $L_0=1$, and $L_U=1$
242: ($N_s=24$), 2 ($N_s=48$) and 3 ($N_s=64$). In order to reduce the effects
243: of open boundary conditions we have averaged over the 6 innermost cells.
244: As the density is increased from the completely empty system, we see that
245: for densities smaller than $\rho_0$, given by
246: \begin{equation}
247: \rho_0 ={1 \over {L_0+L_U}},
248: \label{rho0}
249: \end{equation}
250: the local moment increases, and is the same on both sublattices (hence
251: $\delta=0$). For the SL's with $L_0=1$ and $L_U=1$, 2 and 3 one has
252: $\rho_0=0.5$, 0.33 and 0.25, respectively, which are indicated by
253: arrows in Fig.\ \ref{l01bias}.
254: This density corresponds to having one electron on each
255: cell, so that in the case $L_U\geq L_0$ electrons have equal probability
256: of being either on a free or on a repulsive site for $\rho < \rho_0$.
257:
258: \begin{figure}
259: \epsfxsize=8.5cm
260: \begin{center}
261: %\leavevmode
262: \epsffile{l01bias.eps}
263: \caption{Bias (triangles) and local moment at repulsive (squares) and
264: free (circles) sites, as functions of density, for $U=4$
265: and for superlattices with (a) $L_U=L_0=1$, $N_s=24$; (b) $L_U=2$,
266: $L_0=1$, $N_s=48$; and (c) $L_U=3$, $L_0=1$, $N_s=64$.}
267: \label{l01bias}
268: \end{center}
269: \end{figure} % Fig 2
270:
271: From Fig.\ \ref{l01bias} one sees that there is a range of densities above
272: $\rho_0$, in which the local moment grows slower on the repulsive sites
273: than on the free ones, since added electrons will preferentially occupy
274: the free sites; hence a negative bias develops within this range. By the
275: same token, $\langle S_0^2 \rangle$ will reach its maximum value at
276: densities smaller than those at which $\langle S_U^2 \rangle$ displays its
277: maximum; for completeness, recall that the maximum value of
278: the local moment on a homogeneous free lattice is 0.375, occurring at half
279: filling.
280:
281: In strong coupling, the free sites saturate at the density
282: \begin{equation}
283: \rho_{\uparrow \downarrow} ={{2L_0} \over {L_0+L_U}},
284: \label{rhoud}
285: \end{equation}
286: which corresponds to having two electrons on each free site, while the
287: repulsive site is empty. Nonetheless, even for moderate couplings, this
288: density is special. Indeed, from Figs.\ \ref{l01bias}(b) and
289: \ref{l01bias}(c) one can see that for $L_U > L_0$ the bias reaches its
290: minimum value exactly at $\rho_{\uparrow \downarrow}$. Also, the local
291: moment at repulsive sites shows a bump at $\rho_{\uparrow \downarrow}$,
292: indicating the beginning of a steady occupation of repulsive sites.
293:
294: \begin{figure}
295: \epsfxsize=8.5cm
296: \begin{center}
297: %\leavevmode
298: \epsffile{lu1bias.eps}
299: \caption{Bias (triangles) and local moment at repulsive (squares) and
300: free (circles) sites, as functions of density, for $U=4$
301: and for superlattices with (a) $L_U=L_0=1$, $N_s=24$; (b) $L_U=1$,
302: $L_0=2$, $N_s=48$; and (c) $L_U=1$, $L_0=3$, $N_s=64$.}
303: \label{lu1bias}
304: \end{center}
305: \end{figure} % Fig 3
306:
307: Increasing the density even further, one sees that $\langle S_U^2\rangle$
308: reaches its maximum at $\rho_I$, defined as
309: \begin{equation}
310: \rho_I ={{2L_0+L_U} \over {L_0+L_U}},
311: \label{rhoi}
312: \end{equation}
313: which, in strong coupling, corresponds to having two electrons on each
314: free site and one on each repulsive site; the maximum of $\langle
315: S_U^2\rangle$, at exactly this density, is indicative of the SL being in a
316: Mott-Hubbard insulating state.\cite{mit} In the region between
317: $\rho_{\uparrow \downarrow}$ and $\rho_I$, the repulsive layer is
318: preferentially filled as the overall density is increased, causing a steep
319: rise (drop) in the local moment at the repulsive (free) sites. As Fig.\
320: \ref{l01bias} shows, the consequence is a steady increase of the bias in
321: this interval.
322:
323: For densities larger than $\rho_I$ the free sites are almost completely
324: filled, which is apparent by a considerable decrease in the magnitude of
325: the derivative of $\langle S_0^2\rangle$ with respect to the density.
326: Fermions will then start to double occupy the repulsive sites, thus
327: causing a reduction in $\langle S_U^2\rangle$.
328:
329:
330: The equivalent of Fig.\ \ref{l01bias} for the case $L_U\leq L_0$ is shown
331: in Fig.\ \ref{lu1bias}. While the overall behavior is the same,
332: a few differences are worth mentioning. The first one is the behavior of
333: the bias in the range $\rho < \rho_0$: While the bias vanishes for
334: $L_U\geq L_0$, for $L_U < L_0$ it is negative, though of small magnitude.
335: This is due to the fact that in this range of densities, and within each
336: cell, the electrons have more free sites at their disposal to resonate
337: than repulsive ones; this excess of free sites within each cell also
338: explains why the bias still decreases for densities above $\rho_0$.
339:
340: Second, for $L_U < L_0$, $\langle S_U^2 \rangle$ is boosted whenever
341: $\rho=2m\rho_0$, with $m=1,2,\ldots,L_0+L_U-1$; this can be attributed to
342: the fact that the double occupancy of the repulsive sites is least likely
343: whenever there are an even number of electrons per cell. Note also that
344: the first bump, at $2\rho_0$, coincides with the minimum of the bias.
345:
346: And, third, while for $L_U > L_0$ the bias changes sign for
347: $\rho_{\uparrow \downarrow}< \rho < \rho_I$, when $L_U \leq L_0$ this
348: occurs for $\rho_0 < \rho < \rho_{\uparrow \downarrow}$; the actual
349: location of the density at which $\delta=0$ depends on the SL
350: configuration, as well as on the Coulomb repulsion $U$.
351:
352: \begin{figure}[t]
353: \epsfxsize=8.5cm
354: \begin{center}
355: %\leavevmode
356: \epsffile{squ.eps}
357: \caption{Magnetic structure factor [Eq.\ (\ref{sofq})] for a SL with
358: $L_U=2$,
359: $L_0=1$, $N_s=60$,
360: $N_e=52$ (hence $\rho=0.87$), and for $U=4$ and $16$.}
361: \label{squ}
362: \end{center}
363: \end{figure} % Fig 4
364:
365: \section{Magnetic Structure Factor and effective densities}
366: \label{sq}
367:
368: Let us now turn to the magnetic structure factor, which is defined as
369: \begin{equation}
370: {\cal S}(q)={1\over N_c} \sum_{i,j} {\rm e}^{iq(r_i-r_j)}
371: \langle {\bf S_i}\cdot {\bf S_j}\rangle.
372: \label{sofq}
373: \end{equation}
374: As $q$ is related to
375: the repeating units, ${\cal S}(q)$ probes the relative arrangement between
376: different cells. It is important to have in mind that the homogeneous
377: system displays a single peak in the magnetic structure factor at
378: $q_{max}= 2 k_F = \pi \rho$, for $ \rho \le 1$, or $q_{max}=2 k_F= \pi (2-
379: \rho)$, for $\rho \ge 1$;\cite{Frahm90} the lattice spacing is taken to be
380: unity throughout this paper.
381:
382: \begin{figure}
383: \epsfxsize=8.5cm
384: \begin{center}
385: %\leavevmode
386: \epsffile{sqfse.eps}
387: \caption{Magnetic structure factor [Eq.\ (\ref{sofq})] for a SL with
388: $L_U=2$, $L_0=1$, $U=8$,
389: $\rho=1.2$, and for different system sizes: $N_s=30$ (squares), $N_s=60$
390: (circles), $N_s=120$ (up triangles), and $N_s=150$ (down triangles).}
391: \label{fse}
392: \end{center}
393: \end{figure} % Fig 5
394:
395: Figure \ref{squ} shows ${\cal S}(q)$ for a SL with $L_U=2$, $L_0=1$,
396: $\rho=0.87$, and for two values of $U$, namely $U=4$ and $U=16$. Two peaks
397: in the magnetic structure factor are clearly seen in this case: one at
398: $q=\pi$, and another at $q=3 \pi/ 5$. While the former is not affected by
399: an increase in $U$, the latter grows with $U$, though without changing its
400: position. Actually, for sufficiently large $U$ the peak at $q\neq \pi$
401: even becomes more pronounced than the one at $q=\pi$.
402: Further data
403: show that this happens for a range of values of $L_0$, $L_U$, and $\rho$,
404: as discussed below.
405:
406: The presence of two peaks (at, say, $q_{max}$ and $q_{max}'$, with
407: $q_{max}'<q_{max}$) in the structure factor is associated with a tendency
408: of the system to order (strictly speaking, to {\em quasi-}order, in one
409: dimension) in a magnetic arrangement dominated by the corresponding
410: periods, $\lambda=2\pi/q_{max}$ and $\lambda'=2\pi/q_{max}'$. As we will
411: see below, the long period oscillates with the spacer thickness, a
412: behavior reminiscent of the exchange oscillation observed in magnetic
413: trilayers.
414:
415: These two peaks also differ in the way they depend on the system size.
416: Figure \ref{fse} shows ${\cal S}(q)$ for the SL with $L_U=2$, $L_0=1$,
417: $U=8$, $\rho=1.2$ ($\rho_{\uparrow \downarrow} < 1.2 <\rho_I$), and for
418: four different lattice sizes, ranging from $N_s=30$ to $N_s=150$. From
419: Fig.\ \ref{fse} we see that the inflection already present for $N_s=30$ at
420: $q= 2\pi/5$ sharpens as $N_s$ increases, and that there is no change in
421: the position of the peak. We have checked that a similar slow, but steady,
422: growth of the peak height with $N_s$ occurs for the homogeneous Hubbard
423: model away from half filling. These features have been observed for other
424: SL configurations and densities, which therefore indicate that whenever a
425: peak is found at $q \ne \pi $, it is robust. On the other hand, the peak
426: at $q=\pi$ shows a much weaker size-dependence, so that it should be
427: associated with strong, although short-ranged, correlations; this point is
428: illustrated below.
429:
430: We can then turn to a systematic study of the number of peaks and their
431: positions, by analysing
432: the evolution of the structure factor as the density of electrons is
433: increased. As discussed in Sec.\ \ref{localmom}, for $\rho < \rho_0$ the
434: local moments are small, and either their maxima are on the free layers,
435: or they are evenly distributed throughout the lattice, depending on
436: whether $L_U < L_0$ or $L_U \ge L_0$, respectively. A small or zero bias
437: signals that the SL structure is not very relevant in this situation.
438: Indeed, the spatial decay of the spin-spin correlation function (not shown
439: here) in the case of a superlattice with a small bias can hardly be
440: distinguished from that of the corresponding homogeneous system; as a
441: result, the magnetic structure factor displays a single peak.
442: In addition, this single peak displays a size- and a $U$-dependence
443: similar to those for the homogeneous system.
444:
445: \begin{figure}[t]
446: \epsfxsize=8.5cm
447: \begin{center}
448: %\leavevmode
449: \epsffile{qmaxlu1.eps}
450: \caption{Maxima position, $q_{max}$, of the magnetic structure factor as
451: a function of density for $U=4$, and
452: (a) $L_U=1$ $L_0=1$ $N_s=24$,
453: (b) $L_U=1$ $L_0=2$ $N_s=48$, and (c) $L_U=1$ $L_0=3$ $N_s=64$. The dashed
454: lines indicate the presence of another peak in ${\cal S}(q)$ (see text).}
455: \label{qmaxlu1}
456: \end{center}
457: \end{figure} % Fig 6
458:
459: In order to relate the peak position with some density, one can
460: think of a free (homogeneous) lattice in which the sites are grouped
461: in cells mimicking the SL structure under consideration;
462: it then follows that a meaningful quantity is the {\em cell
463: density} of electrons,
464: \begin{equation}
465: \rho_{\rm cell}={N_e\over N_c}=\rho(L_U+L_0),
466: \label{rhocell}
467: \end{equation}
468: where $\rho$ is the overall density.
469: For the interacting SL, we have found that the peak position
470: is given by the same
471: expression as for the homogeneous case, but with $\rho_{\rm cell}$
472: replacing $\rho$; that is,
473: \begin{equation}
474: q_{max}= \pi \rho_{\rm cell},\ \ \text{for}\ \rho\leq\rho_0.
475: \label{qm1}
476: \end{equation}
477: Thus, the peak position grows linearly with $\rho$ up to $\rho=\rho_0$
478: (at which density $\rho_{\rm cell}=1$), when it reaches $q_{max}= \pi$;
479: see Figs.\ \ref{qmaxlu1} and \ref{qmaxl01}.
480:
481:
482: \begin{figure}
483: \epsfxsize=8.5cm
484: \begin{center}
485: %\leavevmode
486: \epsffile{qmaxl01.eps}
487: \caption{Maxima position, $q_{max}$, of the magnetic structure factor as a
488: function of density for $U=4$, and
489: $L_U=1$ $L_0=1$ $N_s=24$ (a),
490: $L_U=2$ $L_0=1$ $N_s=48$ (b), and
491: $L_U=3$ $L_0=1$ $N_s=64$ (c). The dashed
492: lines indicate the presence of another peak in ${\cal S}(q)$ (see text).}
493: \label{qmaxl01}
494: \end{center}
495: \end{figure} % Fig 7
496:
497: The single peak regime persists for $\rho_0 < \rho < \rho_{\uparrow
498: \downarrow}$, and, as shown in Figs.\ \ref{qmaxlu1} and \ref{qmaxl01},
499: now the
500: peak is always at $q=\pi$. The single peaks in this region show a very
501: weak dependence on the system size, which is reflected in the spatial
502: decay of the correlation functions, $\langle S_0^zS_j^z\rangle$. As
503: illustrated in Fig.\ \ref{sofr}(a) for $L_U=L_0=1$, $U=4$, and
504: $\rho=0.75$,
505: correlations with origin in either of the sublattices barely survive at
506: large distances; this should be contrasted with the case displayed in
507: Fig.\ \ref{sofr}(b), for $\rho=1.75$ (see below), in which correlations
508: in one of the superlattices are `long' ranged.
509:
510: At $\rho_{\uparrow \downarrow}$, and in strong coupling, the free
511: layers are completely filled while the repulsive layers are empty.
512: But as the density is increased beyond $\rho_{\uparrow \downarrow}$,
513: a second peak emerges,
514: %whose position, $q_{\rm max}'$, grows from $q=0$,
515: as indicated by the dotted lines in Figs.\ \ref{qmaxlu1} and \ref{qmaxl01}.
516: This second peak results from the robust moments located
517: on the repulsive sites. Indeed, if one defines an effective
518: electronic density on the {\em repulsive} layers as
519: \begin{equation}
520: \rho_{\rm eff}=\rho(L_0+L_U)-2L_0,
521: \label{rhoeff}
522: \end{equation}
523: where $\rho$ is the overall density, the long-periods are located at
524: \begin{equation}
525: q_{max}'=\pi\rho_{\rm eff}.
526: \label{qmax}
527: \end{equation}
528: With this definition, it also becomes clear that for
529: $\rho=\rho_{\uparrow \downarrow}$ there is no net moment at the repulsive
530: layers, since $\rho_{\rm eff}=0$.
531:
532:
533: \begin{figure}
534: \epsfxsize=8.5cm
535: \begin{center}
536: %\leavevmode
537: \epsffile{fig8.eps}
538: \caption{Spatial decay of correlations, for a SL with $L_U=L_0=1$
539: and $U=4$
540: for $N_s=48$ sites: in (a) $\rho =0.75$, and in (b) $\rho=1.75$.
541: Circles and squares correspond to the origin being taken on free
542: and repulsive sites, respectively.}
543: \label{sofr}
544: \end{center}
545: \end{figure} % Fig 8
546:
547:
548: This two peak structure is present until one reaches $\rho_U$, defined by
549: \begin{equation}
550: \rho_U \equiv 2-\rho_0,
551: \label{rhoU}
552: \end{equation}
553: which corresponds to a single hole per cell.
554: One should also have in mind that the overall magnetic arrangement is
555: determined by the long-period (characterized by $q_{max}'$), since,
556: as discussed above, this is the one increasing with system size.
557:
558: An interesting difference between the cases depicted in Figs.\
559: \ref{qmaxlu1} and \ref{qmaxl01} is the fact that in the
560: former $\rho_U=\rho_I$, while in
561: the latter $\rho_U>\rho_I$, and $q_{\rm max}'$ is able to go
562: through at least
563: one complete oscillation before $\rho$ reaches $\rho_U$.
564: In this case, the situation $q_{max}'=0$ does not indicate any tendency
565: towards a ferromagnetic arrangement, but is to be associated with
566: {\em frustration} of the corresponding long-period SDW.\cite{longo}
567: Indeed, when $L_U=2$ and $L_0=1$ the
568: Mott-Hubbard insulator at $\rho_I$ is frustrated, since two spins on each
569: repulsive layer form local singlets. Singlets on different repulsive
570: layers, in turn, do not couple with each other, though short ranged
571: correlations are still present; see Fig.\
572: \ref{qmaxl01}(b). The frustration at half filling for $L_U=3$ and $L_0=1$
573: can be understood by a similar strong coupling analysis: of the four
574: electrons on each cell, two occupy the free site and the remaining two
575: resonate between three sites, but always forming a singlet. Figure
576: \ref{qmaxl01}(c) shows that further addition of electrons renders these
577: singlets unfavorable, and the system again displays a SDW. At $\rho=3/2$,
578: one reenters a frustrated state, again as a result of having an even number of
579: electrons on the repulsive layer. Therefore, we can relate the
580: reentering frustrated configurations to the formation of singlets on
581: the repulsive layer, which occurs whenever there is an even number of
582: electrons per cell; that is, whenever the density goes through an even
583: multiple of $2\rho_0$.
584:
585:
586: And, finally, above $\rho_U$ all SL's return to a single peak regime:
587: ${\cal S}(q)$ has a maximum at $\pi(2-\rho_{\rm cell})$ [$=\pi(2-\rho_{\rm
588: eff})$, since $\rho_{\rm eff}=\rho_{\rm cell}-2L_0$]. The correlations
589: in this regime are quasi--long-ranged, since, as shown in Fig.\
590: \ref{sofr}(b), the correlation function with origin at a repulsive site is
591: slowly decaying.
592:
593:
594: The above analyses of the magnetic structure factor and of the local
595: moment profile can be extended to several other SL configurations, and the
596: outcome can be best summarized by a diagram in the parameter space
597: $(\rho,L_0,L_U)$, showing the presence of four different regions (or
598: phases). Cross sections of the full three-dimensional phase diagram are
599: presented in Fig.\ \ref{diag2d}(a) for $L_U = 1$ and Fig.\ \ref{diag2d}(b)
600: for $L_U=4$. In the low density region (A), located between $\rho=0$ and
601: $\rho=\rho_0$, the system behaves roughly as if it were homogeneous. The
602: local moments are small and their maxima are located preferentially on the
603: free layers. The SDW is dominated by a single density-dependent
604: wavevector, $q_{max}=\pi\rho_{cell}$.
605:
606: \begin{figure}[t]
607: \epsfxsize=8cm
608: \begin{center}
609: %\leavevmode
610: \epsffile{diag2d.eps}
611: \caption{Regions in the parameter space $(\rho,L_0)$ for (a) $L_U=1$ and
612: (b) $L_U=4$: A -- weak moments formed preferentially on free layers and
613: one single SDW period; B -- local moment maxima depend on $U$ and on
614: $L_0/L_U$ and spin correlations are predominantly antiferromagnetic, but
615: short ranged; C1 -- local moment maxima on the repulsive layers and the
616: SDW's are dominated by two periods; $C_2$ -- local moment maxima on the
617: repulsive layers and the SDW's are dominated by a single period. See text
618: for details. Full line corresponds to $\rho_0$, dotted line to
619: $\rho_{\uparrow\downarrow}$ and dashed line to $\rho_U$.}
620: \label{diag2d}
621: \end{center}
622: \end{figure} % Fig 9
623:
624:
625: At somewhat larger densities, $\rho_0 < \rho <\rho_{\uparrow \downarrow}$,
626: lies a region (B), in which the positions of the maxima in the local
627: moment profile depend on the repulsion $U$, on $L_0$, and $L_U$.
628: Presumably as a result of this `floppy' character of the local moments,
629: spin correlations in this region are strongly antiferromagnetic, but
630: short-ranged.
631:
632: As the density is further increased, one enters the double-period region
633: ($C_1$), which lies between $\rho=\rho_{\uparrow \downarrow}$ and
634: $\rho=\rho_U$. In this region the local moment on the repulsive layer
635: suffers successive boosts, and one finds SDW's with a `long' period
636: $\lambda'=2/\rho_{\rm eff}$; the latter are accompanied by strong short
637: ranged correlations, of period $\lambda=2$.
638:
639: And, finally, there is a high density region ($C_2$), with densities
640: ranging from $\rho=\rho_U$ to $\rho=2$. At $\rho_U$ the local moment bias
641: is maximum (see Figs.\ \ref{l01bias} and \ref{lu1bias}), so it decreases
642: as one increases the density. Nonetheless, one still has SDW's, now with a
643: single period given by $\lambda=2/(2-\rho_{\rm cell})$. By comparing the
644: two cases depicted in Fig.\ \ref{diag2d}, one sees that a growth of the
645: repulsive layer increases the two-peaked region, at the expense of all
646: others.
647:
648: The full three-dimensional phase diagram is shown in Fig.\ \ref{diag3d}.
649: The densities $\rho_0$, $\rho_{\uparrow \downarrow}$, and $\rho_U$ define
650: surfaces in the parameter space $(\rho,L_0,L_U)$ which act as boundaries
651: between the four regions discussed above. The $\rho_0$-surface flattens
652: considerably for thick layers, and if one imagines an $\ell=1$ line on the
653: horizontal plane of the figure, we see that the homogeneous-like region is
654: only important for moderately thin layers. The $\rho_{\uparrow
655: \downarrow}$-surface is the same whether the layers are short or long,
656: since it depends on $L_U$ and $L_0$ only through the combination
657: $L_U/L_0\equiv\ell$; to illustrate this, the intersection of the
658: $\rho_{\uparrow \downarrow}$-surface with the plane $\ell=1$, shown as a
659: dotted line in Fig.\ \ref{diag3d}, yields $\rho_{\uparrow \downarrow}=1$
660: for all $L_U=L_0$. The topmost surface ($\rho_U$) also displays a similar
661: crossover between thin and thick regimes: for thick lattices $\rho_U\to
662: 2$.
663:
664: \begin{figure}
665: \epsfxsize=8.5cm
666: \begin{center}
667: %\leavevmode
668: \epsffile{diag3d.eps}
669: \caption{Regions in the parameter space $(\rho,L_0,L_U)$: the lower
670: surface corresponds to $\rho_0$, the middle one to
671: $\rho_{\uparrow\downarrow}$, and the upper one to $\rho_U$.
672: The dotted line is the intersection of the
673: $\rho_{\uparrow \downarrow}$-surface with the plane $\ell=1$ (see text).}
674: \label{diag3d}
675: \end{center}
676: \end{figure} % Fig 10
677:
678:
679: We are now in a position to discuss the oscillation in $q_{max}$ with the
680: spacer (free layer) thickness, for a {\em fixed electron density}; as
681: mentioned before, these are related to the oscillation of the exchange
682: coupling between magnetic layers.
683: When $\rho \leq \rho_0$, one has $q_{max}\leq\pi$, so that there is no
684: oscillatory behavior in $q_{max}$. For $\rho_0 < \rho \leq
685: \rho_{\uparrow\downarrow}$, the peak is always at $q_{max}=\pi$, so that
686: again no oscillation is found. Above $\rho_{\uparrow\downarrow}$, the
687: long-period
688: maxima in the magnetic structure factor are located at
689: $q_{max}(L_0)=\pi
690: \rho_{\rm eff}(L_0)$, where we have emphasized the dependence with $L_0$
691: through $\rho_{\rm eff}$. We can then calculate the period of oscillation,
692: $\Delta L_0$, by setting $q_{max}(L_0)=q_{max}(L_0+\Delta L_0),\ {\rm
693: mod}(2\pi)$. For $\rho_{\uparrow\downarrow} < \rho \leq 1$ we get
694: \begin{equation}
695: \Delta L_0 = \left(1-{k_F \over \pi}\right)^{-1},
696: \label{period1}
697: \end{equation}
698: where $2k_F=\pi\rho$ since $\rho\leq 1$; for $1<\rho\leq\rho_U$, we
699: similarly find
700: \begin{equation}
701: \Delta L_0 = {\pi\over k_F},
702: \label{period2}
703: \end{equation}
704: where now $2k_F=\pi(2-\rho)$. Note that this result is not valid for
705: $L_U=1$, since, according to Fig.\ \ref{qmaxl01}, once $q_{max}$ vanishes,
706: it does not grow as $L_0$ increases. For $\rho>\rho_U$, on the
707: other hand, Eq.\ (\ref{period2}) is applicable.
708:
709: We have then established that (i) $\rho_{\uparrow\downarrow}$ acts as a
710: critical density for the appearance of `exchange oscillations', and that
711: (ii) our previous results\cite{longo} for $\Delta L_0$, obtained in the
712: high density region, are valid quite generally for $\rho >
713: \rho_{\uparrow\downarrow}$. Further, Eqs.\ (\ref{period1}) and
714: (\ref{period2}) reproduce previous findings, within the Hartree-Fock
715: approximation, for the periods of oscillation of the exchange coupling in
716: magnetic multilayers.\cite{qw,mauro} Also, the experimentally
717: observed short period of two
718: monolayers reported in Ref.\ \onlinecite{Unguris} corresponds, in our
719: framework, to the $\lambda = 2$ correlations. Thus, electronic
720: correlations do not modify the quantum interference effects determining
721: the periods of oscillation from the extrema of the Fermi surface of the
722: spacer material.
723:
724:
725: \section{Conclusions}
726: \label{conc}
727:
728: We have investigated one-dimensional Hubbard superlattices consisting of
729: periodic arrangements of free and repulsive layers, by means of Density
730: Matrix Renormalization Group. By considering a much wider range of lattice
731: sizes and densities than in previous studies, we have refined in several
732: aspects our earlier predictions for the magnetic behavior. There are now
733: {\em four} distinct regimes, depending on the range of electronic
734: densities. For less than one electron per periodic cell, the local moment
735: profile is approximately uniform, and spin-density waves are dominated by
736: a single density-dependent wave vector. When the density lies between
737: those corresponding to one electron per cell and to a fully occupied free
738: sublattice (with empty repulsive sites), maxima in the local moment
739: profile develop, which can be either on free sites or on the repulsive
740: sites, depending on the SL configuration, on the density, and on $U$;
741: also, spin correlations become short ranged, but dominated by a tendency
742: of neighboring cells to align antiparallel. For densities larger than two
743: electrons per free site one has a two-period magnetic structure. There is
744: a long period SDW, in which the wave vector oscillates as a function of
745: the electronic density. An immediate consequence is that SDW's with all
746: possible wave vectors are generated within an interval of densities of
747: $2\rho_0$; this should be compared with the homogeneous system, for which
748: one needs to vary between an empty lattice and a half filled one in order
749: to generate all possible wave vectors. These long-period SDW's are
750: superimposed with short ranged correlations with $q=\pi$, which disappear
751: for densities above one hole per periodic cell.
752:
753: We have also extended to a broader range of densities our earlier
754: prediction that the wave vectors for the SDW's oscillate as the free layer
755: length is varied, with a period determined solely by the electronic
756: density (through the Fermi wave vector).
757: In the context of magnetic multilayers, our results for the period of
758: oscillations exactly reproduce the relation between Fermi
759: surface extrema with exchange coupling oscillation; in addition, the
760: two-monolayer period observed experimentally corresponds in our model to
761: the short period at $q=\pi$.
762:
763: And, finally, we have been able to observe a
764: crossover between the regimes of thin and thick layers; in the latter, the
765: `aspect ratio' $\ell\equiv L_U/L_0$ is the only relevant geometric
766: parameter, whereas in the former regime the magnetic behavior depends on
767: $L_U$ and $L_0$ separately. For instance, when any of the layers are thin
768: -- less than about 6 sites long --, the SL structure is not felt at low
769: densities, and it behaves as if it were homogeneous; for thick
770: layers, this quasi-homogeneous behavior is only noticeable at very low
771: densities. Similarly, the region of singly peaked correlations at high
772: densities gets smaller as the layers get thicker.
773:
774: As a final comment, one should expect that the applicability of the
775: one-dimensional model treated here is very close to being extended beyond
776: the realm of higher dimensional superlattices. Indeed, fabrication of
777: nanowire superlattices has been recently reported.\cite{Gudiksen02}
778: Although these superlattices were made up of semiconducting materials, the
779: prospects of growing metallic and/or magnetic {\em nanosuperlattices} are
780: promising. In this case, our results indicate that a careful control of
781: the doping level leads to a wide variety of distinct magnetic behaviors in
782: the same material. Another possible realization of our model would be to a
783: (as yet hypothetical) superlattice made up of single-walled metallic
784: carbon nanotubes, since these have been successfully described in terms of
785: a Luttinger liquid; see, e.g., Ref.\ \onlinecite{llsl2} for a partial list
786: of references.
787:
788:
789: \acknowledgments
790:
791: The authors are grateful to J.~d'Albuquerque e Castro, E.~Miranda, and
792: J.~Silva-Valencia for discussions. Financial support from the Brazilian
793: Agencies CNPq (ALM and RRdS), Funda\c c\~ao de Amparo \`a Pesquisa do
794: Estado de S\~ao Paulo-FAPESP (ALM), FAPERJ (TP and RRdS), and `Millenium
795: Institute for Nanosciences/CNPq-MCT' (RRdS) is also gratefully
796: acknowledged.
797:
798: \begin{references}
799:
800: \bibitem{qw} D.\ M.\ Edwards, J.\ Mathon, R.\ B.\ Muniz, and M.\ S.\ Phan,
801: \prl{\bf 67}, 493 (1991);
802: J.\ Phys.: Cond. Matter {\bf 3}, 4941 (1991).
803:
804: \bibitem{rkky} P.\ Bruno and C.\ Chappert, \prl{\bf 67}, 1602 (1991);
805: \prb{\bf 46}, 261 (1992).
806:
807: \bibitem{Unguris} J.~Unguris, R.~Celotta, and D.~Pierce, \prl {\bf67}, 140
808: (1991).
809:
810: \bibitem{FeMn} S.~T.~Purcell, M.~T.~Johnson, N.~W.~E.~McGee, R.~Coehoorn,
811: and W.~Hoving, \prb {\bf 45}, 13064 (1992).
812:
813: \bibitem{FeAu} A.~Fuss, S.~Demokritov, P.~Gr\"unberg, and W.~Zinn,
814: J. Magn. Magn. Mater. {\bf 103}, L221 (1992).
815:
816: \bibitem{FeMo} Z.~Q.~Qiu, J.~Pearson, A.~Berger, and S.~D.~Bader, \prl
817: {\bf 68}, 1398 (1992).
818:
819:
820: \bibitem{CoCu} M.~T.~Johnson, S.~T.~Purcell, N.~W.~E.~McGee, R.~Coehoorn,
821: J.~aan de Stegge, and W.~Hoving, \prl {\bf 68}, 2688 (1992).
822:
823:
824: \bibitem{Stoeffler} D.~Stoeffler and F.~Gautier, \prb {\bf 44}, 10389
825: (1991).
826:
827:
828: \bibitem{Wolf} J.~Wolf, {\it et al.}, J.~Magn.~Magn.~Mater. {\bf121}, 253
829: (1993).
830:
831: \bibitem{Schmidt} C.~M.~Schmidt, D.~E.~B\"urgler, D.~M.~Schaller,
832: F.~Meisinger, and H.-J.~
833: G\"untherodt, \prb{\bf60}, 4158 (1999).
834:
835: \bibitem{mauro} J.\ d'Albuquerque e Castro, M.\ S.\ Ferreira, and R.\ B.\
836: Muniz, \prb {\bf 49}, 16062 (1994).
837:
838: \bibitem{tclp1} T.Paiva and R.R.dos Santos, \prl{\bf 76}, 1126 (1996).
839:
840: \bibitem{longo} T.Paiva and R.R.dos Santos, \prb{\bf 62}, 7007 (2000).
841:
842: \bibitem{mit} T.Paiva and R.R.dos Santos, \prb {\bf 58}, 9607 (1998).
843:
844: \bibitem{llsl1} J.~Silva-Valencia, E.~Miranda, and R.~R.~dos Santos, \jpcm
845: {\bf 13}, L619 (2001).
846:
847: \bibitem{llsl2} J.~Silva-Valencia, E.~Miranda, and R.~R.~dos Santos, \prb
848: {\bf 65} 115115 (2002).
849:
850: \bibitem{DMRG}
851: S. R. White, \prl {\bf 69}, 2863 (1992)
852: ;\prb {\bf 48}, 10345 (1993).
853:
854: \bibitem{Noack} G. \ Bed\"urftig, B. \ Brendel, H.\ Fr\"ahm, and R. \ M. \
855: Noack, \prb{\bf 58}, 10 225 (1998).
856:
857: \bibitem{Frahm90} H.\ Fr\"ahm and V.\ Korepin, \prb {\bf 42}, 10 553
858: (1990).
859:
860: \bibitem{Gudiksen02} M.\ S.\ Gudiksen, L.\ J.\ Lauhon, J.\ Wang, D.\ C.\
861: Smith, and C.\ M.\ Lieber, Nature (London) {\bf 415}, 617 (2002).
862:
863:
864: \end{references}
865:
866: \end{document}
867:
868: