cond-mat0203297/pap.tex
1: \def\a{1}  %%%%%%%%%%%%%%%%% 1 = twocolumn, 0 = preprint
2: 
3: \ifodd\a
4: \documentclass[a4paper,10pt,twocolumn]{article}
5: \usepackage{epsf}
6: \usepackage{amsbsy}
7: \usepackage{cite}
8: \setlength{\textwidth}{16.0cm}
9: \setlength{\textheight}{20.2cm}
10: \setlength{\columnsep}{0.8cm}
11: \setlength{\oddsidemargin}{0pc}
12: \setlength{\columnwidth}{7.6cm}
13: \else
14: \documentclass{article}
15: \usepackage{epsf}
16: \usepackage{amsbsy}
17: \usepackage{cite}
18: \fi
19: 
20: \newcommand{\kB}{k_{\mathrm{B}}}
21: \newcommand{\ul}{\underline}
22: 
23: \begin{document} 
24: 
25: \date{\today}
26: 
27: \title{Modeling exchange bias microscopically}
28: 
29: \author{U.\ Nowak, A.\ Misra and K.\ D.\ Usadel\\
30:   Theoretische Physik, Gerhard-Mercator-Universit\"{a}t
31:   Duisburg,\\
32:  47048 Duisburg, Germany\\ e-mail: uli@thp.uni-duisburg.de}
33: \date{\today} 
34: \maketitle
35: 
36: \begin{abstract}
37:   Exchange bias is a horizontal shift of the hysteresis loop observed
38:   for a ferromagnetic layer in contact with an antiferromagnetic
39:   layer. Since exchange bias is related to the spin structure of the
40:   antiferromagnet, for its fundamental understanding a detailed
41:   knowledge of the physics of the antiferromagnetic layer is
42:   inevitable.  A model is investigated where domains are formed in the
43:   volume of the AFM stabilized by dilution.  These domains become
44:   frozen during the initial cooling procedure carrying a remanent net
45:   magnetization which causes and controls exchange bias. Varying the
46:   anisotropy of the antiferromagnet we find a nontrivial dependence of
47:   the exchange bias on the anisotropy of the antiferromagnet.
48: \end{abstract}
49: 
50: {\bf Keywords:} Exchange biasing, magnetic multilayers, Heisenberg
51: model, numerical simulations 
52: 
53: 
54: %{\bf Contact author:} U.~Nowak, Theoretische Tieftemperaturphysik,
55: %Gerhard-Mercator-Universit\"{a}t Duisburg, 47048 Duisburg/ Germany\\ 
56: %phone : +49 203 379 2969, fax: +49 203 379 3665, e-mail:
57: %uli@thp.uni-duisburg.de\\ 
58: 
59: 
60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
61: \section{Introduction}%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
62: \label{s:intro}
63: 
64: When a ferromagnet (FM) is in contact with an antiferromagnet (AFM) a
65: shift of the hysteresis loop along the magnetic field axis can occur
66: which is called exchange bias (EB).  Usually, this shift is observed
67: after cooling the entire system in an external magnetic field below
68: the N\'eel temperature $T_{\mathrm{N}}$ of the AFM.  Although this
69: effect is well known since many years\cite{meiklejohnPR56} and is
70: already intensively exploited in magnetic devices its microscopic
71: origin is still under debate. For a review of the experimental work
72: see the recent article by Nogu\'es and Schuller \cite{noguesJMMM99}.
73: 
74: In the approach of Malozemoff \cite{malozemoffPRB87} exchange bias is
75: attributed to the formation of domain walls in the AFM perpendicular
76: to the FM/AFM interface due to interface roughness. However, the
77: formation of domains in the AFM only due to interface roughness is
78: unlikely to occur because the creation of the domain walls is
79: energetically unfavorable.
80: 
81: Koon considered a spin-flop coupling between FM and the compensated
82: AFM as responsible for EB \cite{koonPRL97}, but recently, Schulthess
83: and Butler \cite{schulthessPRL98,schulthessJAP99} showed that
84: spin-flop coupling alone cannot account for this effect. Instead, in
85: their model EB is only obtained if uncompensated AFM spins are assumed
86: at the interface --- their occurrence is not explained
87: microscopically.
88: 
89: In a previous Letter Milt\'enyi et al.\ \cite{miltenyiPRL00} showed
90: that it is possible to strongly influence EB in Co/CoO bilayers by
91: diluting the antiferromagnetic CoO layer, i.\ e. by inserting
92: nonmagnetic substitutions (Co$_{1-x}$Mg$_x$O) or defects (Co\( _{1-y}
93: \)O) not at the FM/AFM interface, but rather throughout the volume
94: part of the AFM.  In these systems the observed EB is primarily not
95: due to disorder or defects at the interface. Rather, the full
96: antiferromagnetic layer must be involved and it was argued that in
97: these systems EB has its origin in a domain state in the volume part
98: of the AFM which triggers the spin arrangement and the FM/AFM exchange
99: interaction at the interface. This domain state develops due to the
100: dilution of the AFM: the domain walls pass preferentially through
101: non-magnetic sites thus reducing considerably the energy necessary to
102: create a wall.  The domain state strongly depends on the dilution of
103: the AFM resulting in a strong dependence of EB on dilution. Since
104: dilution favors the formation of domains it leads to an increase of
105: the magnetization in the AF and thus to a strong increase of the EB
106: upon dilution (see also \cite{mouginPRB01}, where it was shown that it
107: is possible to influence (to increase or even to revers) EB by a
108: subsequent ion irradiation of the sample).
109: 
110: In the same letter this picture was further supported by Monte Carlo
111: simulations. Later it was shown \cite{nowakJAP01,nowakPRB01} that a
112: variety of experimental facts associated with exchange bias can be
113: explained within our model, like positive exchange bias after cooling
114: in strong magnetic fields, the temperature dependence of exchange
115: bias, especially the relation between the so-called blocking
116: temperature and the N\'eel temperature, and the training effect, among
117: others. In these studies the AFM CoO investigated experimentally was
118: due to its rather strong uniaxial anisotropy modeled as Ising system
119: which is from a numerical point of view an ideal candidate to study
120: basic properties of EB. However, since the occurrence of EB is not
121: restricted to systems with a strong anisotropy in the AFM, in the
122: present paper we will extend the previous model
123: \cite{miltenyiPRL00,nowakJAP01,nowakPRB01} to the non-Ising case, i.\ 
124: e.\ we will vary the strength of the anisotropy of the AFM.
125: 
126: In the next section we give a brief review of the physics of domains
127: in diluted Ising antiferromagnets in an external field. These systems
128: have been studied in great detail in the past and the physics which
129: emerge from these studies are important for understanding EB.  In Sec.
130: \ref{s:model} our model is described and in Sec.  \ref{s:results} our
131: results from Monte Carlo simulations are discussed. Finally, we
132: summarize in the last section.
133: 
134: \section{Domains in disordered antiferromagnets} %%%%%%%%%%%%%%%%%%%%%%%%
135: \label{s:domains}  
136: 
137: Considerable interest has been focused in recent years on the
138: understanding of {\bf d}iluted Ising {\bf a}nti{\bf f}erromagnets in
139: an external magnetic {\bf f}ield (DAFF) as they are ideal candidates
140: for the study of disordered systems. A typical material for
141: experimental studies is $\mathrm{Fe}_{1-p}\mathrm{Zn}_p \mathrm{F_2}$
142: where $\mathrm{Fe} \mathrm{F_2}$ is the AFM which is randomly diluted
143: with probability $p$ by non-magnetic Zn ions. Theoretically, due to
144: the very strong uniaxial anisotropy this system is usually treated as
145: Ising model. Properties which have been extensively exploited are the
146: critical behavior, domain structures, metastability and slow dynamics
147: (for reviews on DAFF see \cite{kleemannIJMP93,belangerBOOK98}).
148: Additionally, many of the findings of the DAFF are also relevant for
149: the {\bf r}andom {\bf f}ield {\bf I}sing {\bf m}odel (RFIM) which has
150: been shown to be in the same universality class
151: \cite{fishmanJPC79,cardyPRB84}.
152: 
153: In zero field the system undergoes a phase transition from the
154: paramagnetic phase to the long-range ordered, antiferromagnetic phase
155: at the disorder dependent N\'eel temperature $T_N$ as long as the
156: dilution $p$ is small enough so that the lattice of occupied sites is
157: above the percolation threshold. In the low temperature region, for
158: small magnetic fields $B$ the long-range ordered phase remains stable
159: in three dimensions \cite{imbriePRL84,bricmontPRL87}, while for higher
160: fields the DAFF develops a disordered domain state
161: \cite{montenegroPRB91,nowakPRB91} with a spin-glass-like behavior. The
162: reason for the domain formation was originally investigated by Imry
163: and Ma for the RFIM \cite{imryPRL75}.  The driving force for the
164: domain formation is a statistical imbalance of the number of
165: impurities of the two antiferromagnetic sublattices within a finite
166: region of the DAFF. This leads to a net magnetization of this region
167: which couples to the external field. A spin reversal of this region,
168: i. e. the creation of a domain can hence lower the energy of the
169: system. The necessary energy increase due to the formation of a domain
170: wall can be minimized if the domain wall passes preferentially through
171: non-magnetic defects at a minimum cost of exchange energy.  Hence,
172: these domains have non-trivial shapes following from an energy
173: optimization.  They have been shown to have a fractal structure with a
174: broad distribution of domain sizes and with scaling laws
175: quantitatively deviating from the original Imry-Ma assumptions
176: \cite{nowakPRB92,esserPRB97}.
177: 
178: In small fields the equilibrium phase of the three-dimensional DAFF is
179: long-range ordered. However, if cooled in a field $B$ below a certain
180: temperature $T_i(B)$, the system usually also develops metastable
181: domains \cite{birgeneauJSP84,belangerJAP85}.  The reason for this
182: metastability is a strong pinning which hinders domain wall motion.
183: These pinning effects are due to the dilution (random-bond pinning) as
184: well as due to the fact that a rough domain wall also carries
185: magnetization in a DAFF (following again the Imry-Ma argument) which
186: couples to the external field and hinders domain wall motion
187: (random-field pinning) \cite{villainPRL84}.  Consequently, after
188: cooling the system from the paramagnetic phase within an external
189: field, a DAFF freezes in a metastable domain state which survives even
190: after switching off the field, then leading to a remanent
191: magnetization which decays extremely slow \cite{hanPRB92}.
192: 
193: In the following we will argue that these well established properties
194: of the DAFF are the key for understanding exchange bias. Indeed,
195: during preparation of the system, the AFM is cooled in an external
196: magnetic field and additionally under the influence of an effective
197: interface exchange field stemming from the magnetized FM.  Hence, the
198: AFM --- as far as it is diluted in any sense --- must develop a domain
199: state with a surplus magnetization similar to that of a DAFF after
200: field cooling.
201: 
202: \section{Model for exchange bias} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
203: \label{s:model}  
204: 
205: The model which we consider in the following consists of one FM
206: monolayer exchange coupled to a diluted AFM film consisting of $t$
207: monolayers. In Fig.\ \ref{f:skizze} a sketch of our model is shown for
208: $t=3$.
209: 
210: \ifodd\a
211: \begin{figure}[h]
212:   \epsfxsize=55mm
213:   \hspace*{10mm} \epsffile{skizze.eps}
214:   \caption{Sketch of the model with one FM layer and three diluted AFM 
215:     layers. The dots mark defects (non-magnetic ions).}
216:   \label{f:skizze}
217: \end{figure}
218: \fi
219: 
220: The system is described by a classical Heisenberg model,
221: 
222: \begin{eqnarray}
223:   {\cal H} = & \! \! \!- J_{\mathrm{FM}} \sum\limits_{\langle i, j \rangle}
224:                  {\ul S}_i \! \cdot \! {\ul S}_j - \sum\limits_i
225:                  \left( d_z S_{iz}^2 \! + \! d_x S_{ix}^2
226:                  \! + \! {\ul S}_i \! \cdot \! {\ul B} \right)  \nonumber \\
227:             & \! \! \! - J_{\mathrm{AFM}} \sum\limits_{\langle i, j \rangle}
228:                 \epsilon_i \epsilon_j {\ul \sigma}_i \! \cdot \! {\ul \sigma}_j
229:                  -\sum\limits_i
230:                  \left( k_z \epsilon_i \sigma_{iz}^2 + \epsilon_i {\ul
231:                  \sigma}_i \! \cdot \! \ul B \right) \nonumber \\ 
232:             & - J_{\mathrm{INT}} \sum\limits_{\langle i, j \rangle}
233:                           \epsilon_i {\ul \sigma}_i \cdot {\ul S}_j, \nonumber
234:      \label{e:ham}                                     
235: \end{eqnarray} 
236: where the first line contains the energy contributions of the FM. Here,
237: the first term is the ferromagnetic nearest neighbor interaction with
238: exchange constant $J_{\mathrm{FM}}$.  The second term introduces an
239: easy axis in the FM ($z$-axis, anisotropy constant $d_z =
240: 0.1J_{\mathrm{FM}}$) which sets the Stoner-Wohlfarth limit of the
241: coercive field, i. e. the zero temperature limit for magnetization
242: reversal by coherent rotation ($B_c = 2 d_z$, in our units, for a
243: field parallel to the easy axis).  The shape anisotropy is
244: approximated by the next term (anisotropy constant $d_x = -
245: 0.1J_{\mathrm{FM}}$) leading to a magnetization which is
246: preferentially in the $y-z$-plane.  We checked, however, that its
247: value does not influence our results. The last term of the first line
248: is the Zeemann energy.
249: 
250: The second line describes the diluted AFM ($\epsilon_i = 0,1$;
251: dilution $p$) correspondingly except of the shape anisotropy.  For the
252: exchange constant of the AFM which mainly determines its N\'eel
253: temperature (also depending on the dilution and the uniaxial
254: anisotropy $k_z$) we set $J_{\mathrm{AFM}} = - J_{\mathrm{FM}}/2$.
255: Finally, the third line includes the interface coupling between FM and
256: AFM and for simplicity we assume $J_{\mathrm{INT}} = -
257: J_{\mathrm{AFM}}$).
258: 
259: Our magnetic field $B$ will always be along the $z$ axis.  In earlier
260: publications \cite{miltenyiPRL00,nowakJAP01,nowakPRB01} the AFM was
261: described by an Ising model. In the present work, we relax this
262: restriction on the AFM. In order to investigate a broader class of
263: systems for the AFM we vary the uniaxial anisotropy $k_z$ of the AFM.
264: 
265: 
266: \section{Results from Simulations} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
267: \label{s:results}
268: 
269: We use Monte Carlo methods with a heat-bath algorithm and single-spin
270: flip methods for the simulation of the model explained above. Each
271: spin is subject to a trial step consisting of a small deviation from
272: the original direction followed by a second trial step in form of a
273: total flip.  This two-fold trial step can take care of a broad range
274: of anisotropies starting from very soft spins up to the Ising limit
275: \cite{nowakARCP00}. We perform typically 25000 Monte Carlo steps per
276: spin for a complete hysteresis loop.
277: 
278: Since we are not interested in the critical behavior of the model
279: above, we do not perform a systematic finite-size analysis.  However,
280: in order to observe the domain structure of the AFM we have to
281: guarantee that typical length scales of the domain structure fit into
282: our system.  Therefore, we show here only results for rather large
283: systems of lateral extension $128 \times 128$. Nevertheless, we also
284: varied the system size and checked that our results are not influenced
285: by the system size as long as the system is not much smaller.
286: 
287: In our simulations the system is cooled from above to below the
288: ordering temperature of the AFM. During cooling the FM is long-range
289: ordered along the easy $z$ axis and its magnetization is practically
290: constant, resulting in a nearly constant exchange field for the AFM
291: monolayer at the interface. In addition to this exchange field the
292: external, magnetic field acts also on the whole AFM.
293: 
294: \ifodd\a
295: \begin{figure}[h]
296:   \epsfxsize=78mm \epsffile{domains.ps} 
297:   \caption{Frozen domain states of a 40\% diluted AFM consisting of 6
298:   monolayers for different values of the AFM anisotropy, $k_{z} =0.1
299:   J_{\mathrm{FM}}, 1.0J_{\mathrm{FM}}, 30J_{\mathrm{FM}}$ (from
300:   top). The shading codes the $z$-component of the staggered
301:   magnetization. } \label{f:domains}
302: \end{figure}
303: \fi
304: 
305: As already argued in the section before, during the cooling procedure
306: the AFM becomes frozen in a domain state, the structure of which
307: depends on the system parameters. The influence on dilution
308: \cite{nowakPRB01} and the influence of the AFM film thickness
309: \cite{nowakJAP01} was already discussed before for the case of an
310: Ising AFM. In the present case, typical staggered domain
311: configurations of the bulk AFM are shown for three different values of
312: the AFM anisotropy (Fig.\ \ref{f:domains}). For low anisotropies $k_z
313: < J_{\mathrm{AFM}}$ domain walls have a width of the order of
314: $\sqrt{J_{\mathrm{AFM}} / k_z}$.  Even for the lowest anisotropy shown
315: in Fig.\ \ref{f:domains} the width is only of the order of a few
316: lattice constants which can hardly be detected in our figure. Also,
317: due to the dilution walls tend to follow the holes so that the wall
318: width is further reduced at those places.  Interestingly, the domain
319: structure itself depends also on $k_z$.  The system has the smallest
320: domains for an intermediate value of $k_z = J_{\mathrm{FM}}$ and not
321: for the Ising case corresponding to the high anisotropy limit $k_z =
322: 30 J_{\mathrm{FM}}$ as one might expect. We will discuss the results
323: following from this behavior later in connection with the anisotropy
324: dependence of the EB.
325: 
326: Typical hysteresis loops taken after cooling in a field of $B =
327: 0.25J_{\mathrm{FM}}$ are depicted in Fig.\ \ref{f:hysteresis}.  Shown
328: are results for the magnetization of the FM (upper figure) as well as
329: that of the AFM interface layer (lower figure). An exchange bias is
330: observed clearly and we determine the corresponding exchange bias
331: field as $B_{\mathrm{EB}} = (B^+ + B^-)/2$ where $B^+$ and $B^-$ are
332: those fields of the hysteresis loop branches for increasing and
333: decreasing field, where the easy axis component of the magnetization
334: of the FM becomes zero.
335: 
336: \ifodd\a
337: \begin{figure}[h]
338:   \epsfxsize=70mm \epsffile{hys-fm.eps} \epsfxsize=70mm
339:   \epsffile{hys-afm.eps}
340:   \caption{Typical hysteresis loop along $z$ of a) the FM and b) the
341:   interface layer of the AFM, for a dilution $p=0.4$ and an AFM
342:   thickness $t=2$. The magnetization is in units of the saturation
343:   value and the field in units of $J_{\mathrm{FM}}$.}
344:   \label{f:hysteresis}
345: \end{figure}
346: \fi
347: 
348: The interface magnetization of the AFM also shows a hysteresis,
349: following the coupling to the FM. Additionally, its curve is shifted
350: upwards due to the fact that after field cooling the AFM is in a
351: domain state with a surplus magnetization. The upward shift of the
352: hysteresis loop for the interface AFM proves the existence of remanent
353: magnetization in the AFM domains. This shifted interface magnetization
354: of the AFM acts as an additional effective field on the FM, resulting
355: in EB.  The magnitude of the EB field strongly depends on the amount
356: of this upward shift. Note that the shift is of the order of a few
357: percent of the saturation magnetization of the AFM while approximately
358: 10\% of the spins of the AFM contribute to the AFM hysteresis. The
359: saturation field for the AFM is much larger than that of the FM so
360: that the AFM is never saturated during the simulation.
361: 
362: In absence of any anisotropy in the FM and at very low dilution of the
363: AFM we observe a perpendicular coupling between FM and AFM.  The
364: magnetization reversal in the FM is here by coherent rotation. The
365: picture changes with increasing uniaxial anisotropy in the FM and upon
366: further dilution of the AFM. The magnetization reversal in the FM is
367: now by domain wall motion and the perpendicular coupling becomes less
368: significant.  This is because uniaxial anisotropies in both the FM and
369: the AFM having the same axis no longer lead to an energy minimum for a
370: perpendicular coupling across the interface. Moreover AFM spins with
371: missing AFM neighbors can lower their energy by rotating parallel to
372: their FM neighbors. Therefore, in the framework of our calculations a
373: spin-flop coupling is not an essential mechanism for EB.
374: 
375: \ifodd\a
376: \begin{figure}[h]
377:   \epsfxsize=60mm
378:   \epsffile{eb_d.eps}
379:   \caption{EB field versus anisotropy of the AFM for
380:     different AFM thicknesses (numbers of AFM layers). Field and
381:     anisotropy are in units of $J_{\mathrm{FM}}$.}
382:  \label{f:anisotropy}
383: \end{figure}
384: \fi
385: 
386: We have calculated the EB field for a wide range of values of $k_{z}$,
387: starting from very soft spins to rigid, Ising-like spins. Fig.
388: \ref{f:anisotropy} shows result for different thicknesses of the AFM
389: and for a dilution of $p=0.4$. Interestingly, we find a peak in the EB
390: field at an intermediate value of $k_{z}$ for a sufficiently thick AFM
391: while at lower thicknesses the EB field increases with the anisotropy
392: and saturates in the Ising limit.
393: 
394: The key for the understanding of EB is the understanding of AFM domain
395: configurations and domain walls. AFM domains are required to carry a
396: surplus magnetization at the interface which must be stable along the
397: $z$ direction during hysteresis in order to produce any EB.  In
398: general, one might expect that the most stable domain configurations
399: are obtained for the Ising limit. But Fig.\ \ref{f:anisotropy} and
400: also Fig.\ \ref{f:domains} suggest that the behavior of domains is
401: more complex. Let us start considering the Ising limit where some
402: domain wall is formed upon field cooling. This domain wall
403: preferentially passes through defects thereby minimizing the exchange
404: energy and at the same time it gathers magnetization thereby lowering
405: the Zeeman energy.  When the anisotropy $k_{z}$ is decreased the
406: energy to create a domain wall will decrease. Thus the system will
407: respond by roughening the domain boundaries (see Fig.\ 
408: \ref{f:domains}, where the domain configuration in the middle is more
409: complex than the lower one which represents the Ising limit). This
410: roughening enhances the possibility for the domains to carry any
411: surplus magnetization and hence the EB will increase.  However, there
412: exists a counter effect.  While further decreasing $k_{z}$ the width
413: of the domain wall increases, so that less energy can be saved through
414: the dilution.  Hence, for still lower anisotropy the domain walls will
415: smoothen thereby lowering again the exchange energy (see once again
416: Fig.\ \ref{f:domains}, now comparing the domain configuration in the
417: middle with the upper one for still lower anisotropy which has much
418: smoother domain walls). Since flat walls carry less remanent
419: magnetization the EB will decrease now with decreasing anisotropy.
420: The compromise between these two opposite effects is achieved at some
421: intermediate value of $k_{z}$ where the bias shows a peak.
422: 
423: However, the peak disappears at lower values of $t$. This happens
424: since for only one monolayer of AFM we are close to the percolation
425: threshold where the domain walls pass nearly exclusively through the
426: defects costing very little or no energy.  Therefore the first
427: mechanism discussed above is less important and the EB increases with
428: $k_{z}$ till it saturates in the Ising limit.
429: 
430: \section{Conclusions}
431: In conclusion, we find that the domain state model for EB proposed
432: originally for the Ising AFM is not restricted to this limit. Rather
433: under certain combination of thickness and dilution of the AFM, the
434: softness of the AFM spins can lead to an even stronger bias field.
435: Since disorder in the AFM of an exchange bias system is rather common,
436: our model yields a general understanding of the microscopic origin of
437: exchange bias.  Within our model there are several properties which
438: influence the bias field, such as dilution, thickness and anisotropy
439: of the AFM. Although a qualitative understanding regarding the
440: dependence of EB on these parameters has been achieved, a quantitative
441: study of the domain structure both at the interface and in the bulk of
442: the AFM would provide a deeper understanding to the problem.
443: 
444: \section*{Acknowledgments}
445: This work has been supported by the Deutsche Forschungsgemeinschaft
446: through SFB 491 and Graduiertenkolleg 277.
447: 
448: \begin{thebibliography}{10}
449: 
450: \bibitem{meiklejohnPR56}
451: W.~H. Meiklejohn and C.~P. Bean, Phys.\ Rev. {\bf 102},  1413  (1956).
452: 
453: \bibitem{noguesJMMM99}
454: J. Nogu\'es and I.~K. Schuller, J.\ Magn.\ Magn.\ Mat. {\bf 192},  203  (1999).
455: 
456: \bibitem{malozemoffPRB87}
457: A.~P. Malozemoff, Phys.\ Rev.\ B {\bf 35},  3679  (1987).
458: 
459: \bibitem{koonPRL97}
460: N.~C. Koon, Phys.\ Rev.\ Lett. {\bf 78},  4516  (1998).
461: 
462: \bibitem{schulthessPRL98}
463: T.~C. Schulthess and W.~H. Butler, Phys.\ Rev.\ Lett. {\bf 81},  4516  (1998).
464: 
465: \bibitem{schulthessJAP99}
466: T.~C. Schulthess and W.~H. Butler, J.\ Appl.\ Phys. {\bf 85},  5510  (1999).
467: 
468: \bibitem{miltenyiPRL00}
469: P. Milt\'enyi, M. Gierlings, J. Keller, B. Beschoten, G. G\"untherodt, U.
470:   Nowak, and K.~D. Usadel, Phys.\ Rev.\ Lett. {\bf 84},  4224  (2000).
471: 
472: \bibitem{mouginPRB01}
473: A. Mougin, T. Mewes, M. Jung, D. Engel, A. Ehresmann, H. Schmoranzer, J.
474:   Fassbender, and B. Hillebrands, Phys.\ Rev.\ B {\bf 63},  60409  (2001).
475: 
476: \bibitem{nowakJAP01}
477: U. Nowak, A. Misra, and K.~D. Usadel, J.\ Appl.\ Phys. {\bf 89},  7269  (2001).
478: 
479: \bibitem{nowakPRB01}
480: U. Nowak and K.~D. Usadel, Phys.\ Rev.\ B  (2001), in preparation.
481: 
482: \bibitem{kleemannIJMP93}
483: W. Kleemann, Int.\ J.\ Mod.\ Phys.\ B {\bf 7},  2469  (1993).
484: 
485: \bibitem{belangerBOOK98}
486: D.~P. Belanger,  in {\em Spin Glasses and Random Fields}, edited by A.~P. Young
487:   (World Scientific, Singapore, 1998).
488: 
489: \bibitem{fishmanJPC79}
490: S. Fishman and A. Aharony, J.\ Phys.\ C {\bf 12},  L729  (1979).
491: 
492: \bibitem{cardyPRB84}
493: J.~L. Cardy, Phys.\ Rev.\ B {\bf 29},  505  (1984).
494: 
495: \bibitem{imbriePRL84}
496: J.~Z. Imbrie, Phys.\ Rev.\ Lett. {\bf 53},  1747  (1984).
497: 
498: \bibitem{bricmontPRL87}
499: J. Bricmont and A. Kupiainen, Phys.\ Rev.\ Lett. {\bf 59},  1829  (1987).
500: 
501: \bibitem{montenegroPRB91}
502: F.~C. Montenegro, A.~R. King, V. Jaccarino, S.-J. Han, and D.~P. Belanger,
503:   Phys.\ Rev.\ B {\bf 44},  2255  (1991).
504: 
505: \bibitem{nowakPRB91}
506: U. Nowak and K.~D. Usadel, Phys.\ Rev.\ B {\bf 44},  7426  (1991).
507: 
508: \bibitem{imryPRL75}
509: Y. Imry and S. Ma, Phys.\ Rev.\ Lett. {\bf 35},  1399  (1975).
510: 
511: \bibitem{nowakPRB92}
512: U. Nowak and K.~D. Usadel, Phys.\ Rev.\ B {\bf 46},  8329  (1992).
513: 
514: \bibitem{esserPRB97}
515: J. Esser, U. Nowak, and K.~D. Usadel, Phys.\ Rev.\ B {\bf 55},  5866  (1997).
516: 
517: \bibitem{birgeneauJSP84}
518: R.~J. Birgeneau, R.~A. Cowley, G. Shirane, and H. Yoshizawa, J.\ Stat.\ Phys.
519:   {\bf 34},  817  (1984).
520: 
521: \bibitem{belangerJAP85}
522: D.~P. Belanger, M. Rezende, A.~R. King, and V. Jaccarino, J.\ Appl.\ Phys. {\bf
523:   57},  3294  (1985).
524: 
525: \bibitem{villainPRL84}
526: J. Villain, Phys.\ Rev.\ Lett. {\bf 52},  1543  (1984).
527: 
528: \bibitem{hanPRB92}
529: S.-J. Han, D.~P. Belanger, W. Kleemann, and U. Nowak, Phys.\ Rev.\ B {\bf 45},
530:   9728  (1992).
531: 
532: \bibitem{nowakARCP00}
533: U. Nowak,  in {\em Annual Reviews of Computational Physics IX}, edited by D.
534:   Stauffer (World Scientific, Singapore, 2000), p.\ 105.
535: 
536: \end{thebibliography}
537: 
538: \ifodd\a
539: \end{document}
540: \fi
541: 
542: \begin{figure}[p]
543: %  \epsfxsize=60mm
544: %  \epsffile{skizze.eps}
545:   \caption{Sketch of our model with one FM layer and three AFM
546:   layers.}
547:   \label{f:skizze}
548: \end{figure}
549: 
550: \begin{figure}[p]
551: %  \epsfxsize=70mm
552: %  \epsffile{domains.ps}
553:   \caption{Frozen domain states of a 40\% diluted AFM consisting of 6
554:     monolayers for different values of the AFM anisotropy, $k_{z}
555:     =0.1 J_{\mathrm{FM}}, 1.0J_{\mathrm{FM}},
556:       30J_{\mathrm{FM}}$ (from top).}
557:  \label{f:domains}
558: \end{figure}
559: 
560: \begin{figure}[p]
561: %  \epsfxsize=70mm  \epsffile{hys-fm.eps}
562: %  \epsfxsize=70mm  \epsffile{hys-afm.eps}
563:   \caption{Typical hysteresis loop along $z$ of (a) the FM and (b)
564:     the interface monolayer of AFM, for $p=0.4$ $t=2$. The net
565:     magnetization is shown in units of the saturation magnetization  
566:     and the field in units of $J_{\mathrm{FM}}$.} 
567:   \label{f:hysteresis}
568: \end{figure}
569: 
570: \begin{figure}[p]
571: %  \epsfxsize=70mm
572: %  \epsffile{eb_d.eps}
573:   \caption{Exchange bias field versus anisotropy of the AFM for
574:   different AFM thicknesses (numbers of AFM layers).}
575:  \label{f:anisotropy}
576: \end{figure}
577: 
578: \clearpage
579: \setcounter{figure}{0}
580: \begin{figure}[h]
581:   \epsfxsize=60mm
582:   \epsffile{skizze.eps}
583:   \vspace{3cm}
584:   \caption{Nowak et al.}
585: \end{figure}
586: 
587: \newpage
588: \begin{figure}[h]
589:   \epsfxsize=70mm
590:   \epsffile{domains.ps}
591:   \vspace{3cm}
592:   \caption{Nowak et al.}
593: \end{figure}
594: 
595: \newpage
596: \begin{figure}[h]
597:   \epsfxsize=70mm  \epsffile{hys-fm.eps}
598:   \epsfxsize=70mm  \epsffile{hys-afm.eps}
599:   \vspace{3cm}
600:   \caption{Nowak et al.}
601: \end{figure}
602: 
603: \newpage
604: \begin{figure}[h]
605:   \epsfxsize=70mm
606:   \epsffile{eb_d.eps}
607:   \vspace{3cm}
608:   \caption{Nowak et al.}
609: \end{figure}
610: 
611: \end{document}
612: 
613: