cond-mat0609058/p.tex
1: \documentclass[aps, pre, eqsecnum, amsmath,amssymb,showpacs]{revtex4}
2:  \usepackage[final]{graphicx}  % to include figures
3: 
4: \newcommand{\dd}{\text{d}}
5: \newcommand{\oD}{\overline{D}}
6: \newcommand{\om}{\overline{m}}
7: \newcommand{\oalpha}{\overline{\alpha}}
8: \newcommand{\oepsilon}{\overline{\epsilon}}
9: \newcommand{\ophi}{\overline{\phi}}
10: 
11: \pagestyle{myheadings}
12: \markboth{\today}{\today}
13: \begin{document}
14: \renewcommand{\thefootnote}{\fnsymbol{footnote}}
15: 
16: \title{Magnetization of concentrated polydisperse ferrofluids: Cluster
17: expansion}
18: \author{B.~Huke and M.~L\"{u}cke \\}
19: \affiliation{Institut f\"{u}r Theoretische Physik,Universit\"{a}t des Saarlandes,
20: D-66041 Saarbr\"{u}cken, Germany \\}
21: 
22: \renewcommand{\thefootnote}{\arabic{footnote}}
23: \setcounter{footnote}{0}
24: \date{\today}
25: 
26: 
27: % ABSTRACT
28: 
29: \begin{abstract}
30: 
31: The equilibrium magnetization of concentrated ferrofluids
32: described by a system of {\it polydisperse} dipolar hard spheres is calculated 
33: as a function of the internal magnetic field using the
34: Born--Mayer or cluster expansion technique. This paper extends the results
35: of Phys. Rev. E {\bf 62}, 6875 (2000) obtained for {\it monodisperse\/} ferrofluids.
36: The magnetization is given as a power 
37: series expansion in two parameters related to the volume 
38: fraction and the coupling strength of the dipolar interaction, respectively.
39: 
40: \end{abstract}
41: 
42: \pacs{PACS: 75.50.Mm, 05.70.Ce, 05.20.Jj} 
43: \maketitle
44: 
45: % BEGIN OF THE TEXT
46: 
47: \section{Introduction}
48: \label{intro}
49: 
50: Ferrofluids \cite{R85} are suspensions of ferromagnetic particles of about 
51: 10 nm diameter in a carrier fluid. The particles are stabilized against 
52: aggregation by coating with polymers or by electrostatic repulsion of charges 
53: brought on their surface. As long as the concentration of the particles is
54: low, the equilibrium magnetization of a ferrofluid is that of an ideal 
55: paramagnetic gas. In highly concentrated ferrofluids on the other hand, the
56: magnetization is influenced by effects of particle--particle interactions.
57: 
58: We studied these effects for ferrofluids that are described by a system of 
59: identical
60: dipolar hard spheres in \cite{HL}, from now on referred to as paper~I.
61: In that paper we used the technique of the Born--Mayer or cluster expansion
62: technique to evaluate the equilibrium magnetization as
63: a series expansion in terms of the volume fraction $\phi = N \pi D^3/6V$, and
64: a dipolar coupling parameter $\epsilon = m^2/4 \pi \mu_0 kT$, with $N/V$
65: being the particle density, and $D$ and $m$ being the common hard sphere
66: diameter and magnetic moment of the particles, respectively.
67: 
68: However, real ferrofluids are polydisperse, i.~e.\ the particles vary in size and 
69: magnetic moment. This property has a strong influence on the equilibrium 
70: magnetization, for concentrated as well as for dilute fluids. The goal of
71: this paper is to generalize the findings of paper~I to include the effects
72: of polydispersity. 
73: 
74: The linear response problem of determining the static initial susceptibility
75:  $\chi$ of a mixture of dipolar hard spheres was investigated already for the 
76: equivalent electric case in the framework of integral theories: 
77: The mean spherical model \cite{W71} was extended to binary or multicomponent 
78: mixtures \cite{AD73,IB74,HS78,FHI79,RH81,CB86}. The  
79: reference hypernetted chain method \cite{FP84} was also applied to bidisperse 
80: systems \cite{LL87,LL89}. Recently \cite{KS02}, the mean spherical model was 
81: used within the algebraic perturbation theory \cite{K99}, however without
82: leading to new results for the initial susceptibility. The mean spherical model
83: was also extended to polydisperse ferrofluids in arbitrary high fields
84: \cite{SPMS90,ML90}. Another theory dealing with arbitrary fields is the
85: high temperature approximation \cite{BI92}. A variant of this theory was
86: proposed in \cite{PML96} and extended in \cite{IK01}.
87: 
88: Our calculation follows closely that of paper~I. Therein the application of 
89: the cluster expansion technique to a monodisperse system of dipolar hard 
90: spheres resulted in an expression for the magnetization $M$ that can
91: be put into the form 
92: \begin{equation}
93: M = M_{\text{sat}} \sum_{m=0}^\infty \sum_{n=0}^\infty \phi^m \epsilon^n
94: L_{m,n} (\alpha) \;\; , 
95: \end{equation}
96: where $M_{\text{sat}}$ is the saturation magnetization of the fluid. 
97: The functions $L_{m,n} (\alpha)$ were given explicitly in terms of analytic 
98: expressions in the dimensionless 
99: magnetic field $\alpha$. We calculated $L_{2,2}(\alpha)$ and some 
100: of the $L_{1,n}(\alpha)$. Lower orders vanish, except for the Langevin function 
101: $L_{0,0}(\alpha)$. 
102: 
103: In the polydisperse case discussed here the parameters $\phi$, $\epsilon$ and 
104: $\alpha$ are replaced by more
105: generally defined quantities $\ophi$, $\oepsilon$ and $\oalpha$ 
106: (cf. Sec.~\ref{SEC3}). The 
107: calculated $L_{m,n}$ transform into one--, two-- or threefold sums over all 
108: particles, where the individual addends are analytical functions of the 
109: magnetic moments and diameters of the involved particles, and the reduced 
110: magnetic field $\oalpha$.
111: 
112: The paper is organized as follows. In Sec.~\ref{SEC2} we explain the
113: principles of the cluster expansion technique. The main part of the paper is 
114: Sec.~\ref{SEC3}, where we generalize the results of paper~I for the 
115: equilibrium magnetization in the monodisperse case to polydisperse 
116: ferrofluids. The findings are discussed in Sec.~\ref{SEC4} using example
117: distributions. 
118: In Sec.~\ref{SEC4a} the results are compared to experimental data.
119: We conclude in Sec.~\ref{SEC5}.
120: 
121: \section{Cluster expansion: Application to the system of dipolar hard
122: spheres}
123: \label{SEC2}
124: 
125: Here we recapitulate briefly the principle of the Born--Mayer or cluster 
126: expansion technique: Consider a system of particles $i = 1, \, ...\, N$ 
127: interacting with an 
128: external potential $V_i$ and with each other via a potential $V_{ij}$. 
129: To calculate thermodynamic properties of the system one has to find the
130: canonical partition function
131: \begin{equation}
132: Z = \int e^{ -\sum_k v_k - \sum_{i<j} v_{ij}} \;
133: d \Gamma \;\; . \label{candistrigen}
134: \end{equation} 
135: Here $v_i = V_i/kT$, $v_{ij} = V_{ij}/kT$, and $d \Gamma$ means integration 
136: over the configuration space. The kinetic energy of the particles, if important,
137: can be thought to be included in the terms $V_i$. One now writes
138: \begin{equation}
139: Z = \int \prod_k e^{-v_k}
140: \prod_{i<j} (1 + f_{ij} ) \; d \Gamma \;\; ,
141: \label{Zinf}
142: \end{equation} 
143: where
144: \begin{equation}
145: f_{ij} = e^{- v_{ij}} -1 \;\; .
146: \end{equation} 
147: If the typical interaction energy is small compared to $kT$, the $f_{ij}$ can 
148: be considered as small parameters for the expansion of the integrand in 
149: Eq.~(\ref{Zinf}).
150: The leading terms factorize into low dimensional integrals that can be 
151: calculated at least numerically.
152: 
153: In the system of dipolar hard spheres (monodisperse or polydisperse) the 
154: interaction potential $V_{ij}$ consists of a dipole--dipole (DD) interaction 
155: and a hard core (HC) repulsion part,
156: $V_{ij} = V^{DD}_{ij} + V^{HC}_{ij}$, where the first part is given by
157: \begin{equation}
158: V^{DD}_{ij} = 
159: - \frac{ 3 ({\bf m}_i \cdot {\bf \hat{r}}_{ij} )({\bf m}_j \cdot 
160: {\bf \hat{r}}_{ij} ) - {\bf m}_i \cdot {\bf m}_j }
161: {4 \pi \mu_0 r_{ij}^3} \;\; , \label{VDDijdef}
162: \end{equation}
163: for two particles with magnetic moments ${\bf m}_i$ and ${\bf m}_j$ at a
164: distance ${\bf r}_{ij} = {\bf x_i} - {\bf x_j}$,  with 
165: $r_{ij} = |{\bf r}_{ij}|$, and ${\bf \hat{r}}_{ij} = {\bf r}_{ij}/r_{ij}$.
166: For particles with diameters $D_i$ and $D_j$ one has $V_{ij}^{HC}(r_{ij}) = \infty$,
167: if $r_{ij} < D_{ij} = (D_i+D_j)/2$, and $V_{ij}^{HC}(r_{ij}) = 0$ otherwise.
168: 
169: Taking the thermodynamic limit in a system of dipolar particles requires some
170: care because of the long range character of the forces \cite{BGW98}. 
171: We circumvented this problem by decomposing the dipolar potentials into a short
172: range and a long range part, and replacing the latter by an effective mean
173: field. Within this approach a particle experiences the local magnetic field
174: \begin{equation}
175: {\bf H}_{local} = {\bf H}_s + {\bf H}_{dipole,near} = 
176: {\bf H} + \frac{{\bf M}}{3} + {\bf H}_{dipole,near} \;\; .
177: \end{equation}
178: It consists of the dipolar near field ${\bf H}_{dipole,near}$ that is produced 
179: by the other particles within a sphere of radius $R_s$ and of an effective 
180: "external" field
181: \begin{equation} 
182:  {\bf H}_s = {\bf H} + \frac{\bf M}{3} \;\; , \label{Hsdef}
183: \end{equation}
184: seen by the particle in question at the center of the sphere. Here ${\bf H}$ is
185: the macroscopic internal magnetic field and ${\bf M}$ the sought after 
186: equilibrium magnetization. Thus, when
187: evaluating the partition function one has to take 
188: \begin{equation}
189: V_i = - {\bf m}_i \cdot {\bf H}_s 
190: \end{equation} 
191: as the external potential. The radius $R_s$ of the sphere has to be taken to be
192: sufficiently large to allow the far-field dipolar contributions to be replaced by
193: those of a continuum -- cf. paper~I for details. Neither the kinetic energy
194: of the magnetic particles, nor the carrier fluid has to be taken into account in 
195: the partition function, since these terms do not contribute
196: to the equilibrium magnetization. The configuration space is thus given by
197: the positions ${\bf x_i}$ of all particles and the orientations $\Omega_i$
198: of their magnetic moments: $d \Gamma = d^n {\bf x}_i d^n \Omega_i$. 
199: 
200: In paper~I we used in addition also an expansion in the dipolar interaction: 
201: The $f$--terms were expanded as
202: \begin{equation}
203: f_{ij} = e^{-v^{HC}_{ij}} e^{-v^{DD}_{ij}} -1 
204: = f_{ij}^{(0)} + f_{ij}^{(1)} + f_{ij}^{(2)} + ... ,
205: \end{equation}
206: with
207: \begin{eqnarray}  
208: f_{ij}^{(0)} &=& e^{-v^{HC}_{ij}} - 1\\
209: f_{ij}^{(n)} &=& \frac{\left(-v^{DD}_{ij}\right)^n}{n!} 
210: e^{-v^{HC}_{ij}} \label{fijndef}
211: ... \;\; ,  n \geq 1 \;\;.
212: \end{eqnarray}
213: The two expansions concerning the $f$--terms and $v_{ij}^{DD}$ together 
214: translate in the monodisperse case into a double power expansion of $Z$ in the 
215: the volume fraction $\phi$ of the particles and the dipolar coupling constant 
216: $\epsilon$. We calculated the terms in 
217: $O(\phi \epsilon^n)$ and in $O(\phi^2 \epsilon^2)$ of $Z$ and from that the 
218: equilibrium magnetization in the same order.
219: 
220: \section{Calculating the equilibrium magnetization}
221: \label{SEC3}
222: 
223: \subsection{Notation}
224: \label{SEC31}
225: 
226: Consider a system of $N$ spherical hard particles with diameters 
227: $D_1$, ..., $D_N$ carrying permanent magnetic moments $m_1$, ..., $m_N$
228: contained in a volume $V$ and subjected to a magnetic field ${\bf H}$. 
229: Let $\oD$ and $\om$ be some "typical" values for diameters and magnetic 
230: moments that are discussed further below. We then define the parameter 
231: $\ophi$ related to the volume fraction
232: of the hard spheres and the dipolar coupling parameter $\oepsilon$ as 
233: \begin{equation}
234: \ophi = \frac{N\pi \oD^3}{6 V} \;\; ,\qquad
235: \oepsilon = \frac{\om^2}{4 \pi \mu_0 kT \oD^3} \;\; .
236: \end{equation}
237: The equilibrium magnetization is calculated as a power expansion in these 
238: two parameters. The dimensionless magnetic fields are defined as
239: \begin{equation}
240: \oalpha = \frac{\om H}{k T} \;\; , \;\; \oalpha_s = \frac{\om H_s}{k T} \;\; .
241: \end{equation}
242: Diameters and magnetic moments will be expressed in units of the typical
243: values via
244: \begin{equation}
245: \Delta_i = \frac{D_i}{\oD} \;\; , \qquad \mu_i = \frac{m_i}{\om} \;\; .
246: \end{equation}
247: We will also use the minimal possible distance between two hard spheres 
248: $i$ and $j$ given by
249: \begin{equation}
250: \Delta_{ij} = \frac{1}{2} \left( \Delta_i + \Delta_j \right) = 
251: \frac{D_{ij}}{\oD} \;\; .
252: \end{equation}
253: Furthermore we introduce the reduced magnetic fields for each particle $i$ by
254: \begin{equation}
255: \alpha_i = \frac{m_i H}{k T} = \mu_i \oalpha \;\; ,\;\;
256: \alpha_{si} = \frac{m_i H_s}{k T} = \mu_i \oalpha_{s} \;\; .
257: \end{equation}
258: 
259: Our cluster expansion does not depend on how the characteristic 
260: values of $\oD$ and $\om$ are defined in detail. For example they could be taken
261: as as some weighted mean of the $D_i$ and $m_i$, respectively, or their most
262: probable values. To preserve this 
263: freedom of choice in our expansion offers some advantages for the comparison 
264: with magneto--granulometric analyses where the distribution of the diameters
265: and magnetic moments is not known a priori but on the contrary the goal of the 
266: calculations. 
267: 
268: Note, however, that $\ophi$ coincides with the actual 
269: volume fraction $\phi$ of the hard spheres only if one defines $\oD$ via the 
270: mean volume of the particles
271: \begin{equation}
272: \oD^3 = \frac{1}{N} \sum_i D_i^3
273: = \int D^3 P(D) \, dD \equiv \left< D^3 \right>_P \;\; . 
274: \label{defDbar}
275: \end{equation}
276: Here $P(D)$ is the normalized distribution function of the hard sphere
277: diameters. 
278: 
279: Similarly $\om$ is related to the saturation magnetization 
280: $M_{\text{sat}}$ of the ferrofluid via $M_{\text{sat}} = N \om/\mu_0 V$ only if
281: $\om$ is defined by 
282: \begin{equation}
283: \om = \frac{1}{N} \sum_i m_i = \int m(D) P(D) \, dD \;\; . 
284: \label{defmbar}
285: \end{equation}
286: The second equality of Eq.~(\ref{defmbar}) holds when the magnetization of each
287: particle is given by a function of its volume. We assume that this is the case
288: and thus describe in this paper polydispersity effects of the ferrofluid by a 
289: distribution function $P(D)$ depending only on the hard sphere diameter $D$.
290: The generalization to a distribution 
291: function $P(D,m)$ of independently varying diameters and moments is 
292: straightforward. Averages weighted with the distribution function $P(D)$ of the
293: diameters will mostly appear in the reduced version as integrals over the
294: reduced diameter $\Delta = D/\oD$ with the appropriate weight function 
295: $P(\Delta)$.
296: 
297: The thermodynamic mean with respect to the noninteracting system will be
298: denoted by $\left< ... \right>_0$ and the corresponding canonical
299: partition function by $Z_0$. With this notation integrals over the $f$--terms
300: appearing in $Z$ (\ref{Zinf}) can be written in the form
301: \begin{equation}
302: \int \prod_k e^{-v_k} f ...  \; d \Gamma = Z_0 \left< f ... \right>_0 \;\; .
303: \end{equation}
304: But in contrast to paper~I we derive here an approximation directly for the 
305: free energy $F = - kT \ln Z$. If the particle--particle
306: interaction would depend only on interparticle distance then $F$ would be given
307: by 
308: \begin{equation}
309: F = F_0 - 
310: k T \left( \frac{1}{2} {\sum}^\prime \left< f_{ij} \right>_0 + 
311: \frac{1}{6} {\sum}^\prime \left< f_{ij} f_{jk} f_{ki} \right>_0 \right) \;\; ,
312: \label{Fseries}
313: \end{equation}
314: including orders up to $O(\ophi^2)$, or, more generally speaking, up to terms 
315: of second order in the number density. The primed sums are taken over all
316: particle pairs $i$,$j$, resp.\ all triples $i$,$j$,$k$.
317: While Eq.~(\ref{Fseries}) does not hold for a system of dipolar particles in a 
318: magnetic field in arbitrary order of $\oepsilon$ it still is correct in the 
319: orders we want to calculate.
320:  
321: The polydisperse generalization affects the calculation of the integrals
322: in Eq.~(\ref{Fseries}) in two ways: ({\it i}) The fact that the individual dimensionless
323: magnetic fields $\alpha_i$ are different leads to more complicated 
324: expressions for some resulting functions compared to the monodisperse case
325: -- see the definitions of $G^P_n$ and $K^P$ below. ({\it ii}) 
326: The dispersion in the hard sphere diameters requires more difficult geometrical
327: considerations concerning the $v_{ij}^{HC}$--terms, especially in the three
328: particle integral.
329: 
330: \subsection{The leading term: polydisperse Weiss model}
331: \label{SEC32}
332: 
333: The leading term in Z is the partition function of the (formally) 
334: noninteracting paramagnetic gas in the magnetic field $H_s$
335: \begin{equation}
336: Z_{0} = \int \prod_k e^{-v_k} \; d \Gamma = \prod_k z_k \;\; ,
337: \end{equation}  
338: \begin{equation}
339: z_k = \int e^{-v_k} d {\bf x}_k d \Omega_k = 
340: 4 \pi V \frac{\sinh \alpha_{sk}}{\alpha_{sk}} \;\; .
341: \end{equation}  
342: The equilibrium magnetization $M(\oalpha_s)$ obtained from
343: \begin{equation}
344: M(\oalpha_s) = -\frac{1}{\mu_0 V} \frac{\partial F}{\partial H_s} = 
345: \frac{1}{\mu_0 V} \frac{\partial (k T \ln Z)}{\partial H_s} = 
346: \frac{\om}{\mu_0 V} \frac{\partial \ln Z}{\partial \oalpha_s} 
347: \end{equation} 
348: reads in leading order 
349: \begin{equation}
350: M(\oalpha_s)= \frac{N \om}{\mu_0 V} {\cal L}_{\text{poly}}(\oalpha_s) \;\; . 
351: \label{Mofalphas}
352: \end{equation}
353: Here 
354: \begin{equation}
355: {\cal L}_{\text{poly}}(\oalpha_s) = \frac{1}{N} \sum_i \mu_i 
356: {\cal L} \left( \mu_i \oalpha_s \right)  
357: = \int \mu(\Delta)
358: {\cal L} \left[ \mu(\Delta)  \oalpha_s \right] P(\Delta) d \Delta \;\; .
359: \label{Lpoly}
360: \end{equation}
361: is given by the sum of the Langevin paramagnetic contributions coming from
362: each (reduced) magnetic moment $\mu_i=m_i/\om$ with ${\cal L}$ being the 
363: Langevin function. The second equality in Eq.~(\ref{Lpoly}) is the continuous
364: analog of the sum with $\mu(\Delta)=m(\Delta)/\om$ and $\Delta=D/\oD$.
365: If one defines $\om$ via $N \om = \sum_i m_i$ so that 
366: $N \om/\mu_0 V = M_{\text{sat}}$ then 
367: ${\cal L}_{\text{poly}}(\oalpha_s \rightarrow \infty) = 1$.
368: 
369: The result (\ref{Mofalphas}) reduces to the well known expression for the
370: magnetization of a polydisperse ideal paramagnetic gas as a superposition of
371: Langevin functions, if one replaces $\oalpha_s$ by $\oalpha$ 
372: (see e.~g.~\cite{ML90}). However, the dipolar far-field contributions enter 
373: via (\ref{Hsdef}) as a mean field into 
374: \begin{equation}
375: \oalpha_s = \oalpha + \om M/3kT \;\; .
376: \label{oalphas}
377: \end{equation} 
378: Thus, the lowest order result (\ref{Mofalphas}) for $M(\oalpha)$  
379: \begin{equation}
380: M = \frac{N \om}{\mu_0 V} {\cal L}_{\text{poly}} \left( \oalpha + 
381: \frac{\om M}{3kT} \right) 
382: \label{Mofalpha+M}
383: \end{equation}
384: contains already corrections from the particle--particle interaction
385: in the mean field approximation and (\ref{Mofalpha+M}) is the polydisperse 
386: generalization of the Weiss model \cite{Ce82}. 
387: 
388: By replacing M on the right
389: hand side of Eq.~(\ref{Mofalpha+M}) by the expression for the ideal paramagnetic 
390: gas, one arrives at the equation proposed in \cite{PML96}.
391: 
392: \subsection{The magnetization in $O(\ophi)$ }
393: \label{SEC33}
394: 
395: To calculate the canonical partition function in linear order of $\ophi$
396: we follow the lines of paper~I, Sec.~IV.
397: One needs to include only the linear $f$--terms in the expansion 
398: (\ref{Fseries}). Thus we write
399: \begin{equation}
400: F = F_0 - k T \frac{1}{2} {\sum}^\prime \left< f_{ij} \right>_0 \;\; .
401: \label{FexpinOphi}
402: \end{equation} 
403: 
404: For the second term in (\ref{FexpinOphi}), the trivial integrations over the 
405: degrees of freedom of all particles except $i$ and $j$ are performed 
406: first. This gives
407: \begin{eqnarray}
408: && \frac{1}{2} {\sum}^\prime \left< f_{ij} \right>_0 = 
409: \frac{1}{2 Z_0} {\sum}^\prime \int \prod_k e^{-v_k} f_{ij}  \; d \Gamma 
410: \\ && \qquad
411: ={\sum}^\prime \frac{1}{2 z_i z_j} \int e^{-v_i - v_j} f_{ij} 
412: \; d{\bf x}_i d{\bf x}_j d\Omega_i d\Omega_j \;\; . \nonumber
413: \end{eqnarray} 
414: Now we expand $f_{ij}$. Let 
415: \begin{equation}
416: A_{n,ij} = \int e^{-v_i - v_j} f^{(n)}_{ij} 
417: \; d{\bf x}_i d{\bf x}_j d\Omega_i d\Omega_j \;\; ,
418: \end{equation} 
419: such that 
420: \begin{equation}
421: F = F_0 - {\sum}^\prime \frac{kT}{2 z_i z_j} \sum_{n=0}^\infty A_{n,ij} \;\; .
422: \label{FexpinOphiwithA}
423: \end{equation} 
424: 
425: We need not calculate $A_0$, since this term does not contribute to the equilibrium
426: magnetization. Furthermore $A_1 = 0$, because a dipolar magnetic field vanishes when 
427: averaged over a spherical surface. This is explained in more detail in paper~I. 
428: Using the definition (\ref{fijndef}) of $f^{(n)}_{ij}$, and the dipolar 
429: potential (\ref{VDDijdef}), we can write
430: \begin{eqnarray}
431: A_n &=& \frac{V}{n!} \int 
432: e^{\alpha_{si} \cos \vartheta_i + \alpha_{sj} \cos \vartheta_j} \nonumber \\
433: && \times \left( \frac{\om^2 \mu_i \mu_j}{4 \pi \mu_0 k T r^3_{ij}} \right)^n 
434: P^n(\varphi_i,\vartheta_i,\varphi_j,\vartheta_j,\varphi,\vartheta)
435: \label{Anresult1}\\
436: && \times e^{-v^{HC}_{12}}
437: r^2_{ij} dr_{ij}  d\omega_{ij} d\Omega_i  d\Omega_j \;\; . \nonumber
438: \end{eqnarray}
439: Here we have integrated over ${\bf x_i}$ and decomposed ${\bf r}_{ij}$ into
440: the distance $r_{ij}$ and a spherical angle $\omega_{ij}$. The angles
441: $(\varphi_i,\vartheta_i)$, $(\varphi_j,\vartheta_j)$, and 
442: $(\varphi,\vartheta)$ represent the spherical angles $\Omega_i$, $\Omega_j$,
443: and $\omega_{ij}$, respectively. The function
444: \begin{equation}
445: P(\varphi_i,\vartheta_i,\varphi_j,\vartheta_j,\varphi,\vartheta) = 
446: 3 (\hat{\bf m}_i \cdot {\bf \hat{r}}_{ij} )(\hat{\bf m}_j \cdot 
447: {\bf \hat{r}}_{ij} ) - \hat{\bf m}_i \cdot \hat{\bf m}_j 
448: \end{equation}
449: comes from the dipolar interaction.
450: The integration over the directions of ${\bf r}_{ij}$, ${\bf m}_i$, and
451: ${\bf m}_j$ can still be done analytically. But in contrast to the monodisperse
452: calculation, the result is now a function of two parameters $\alpha_{si}$
453: and $\alpha_{sj}$. We define
454: \begin{eqnarray} \label{Gndef}
455: &&G^P_n(\alpha_{si},\alpha_{sj}) = \frac{V^2}{n! (n-1) \pi z_i z_j} 
456: \nonumber \\ 
457: && \;\;\; \times 
458: \int e^{\alpha_{si} \cos \vartheta_i + \alpha_{sj} \cos \vartheta_j} \\
459: && \qquad \times  
460: P^n(\varphi_i,\vartheta_j,\varphi_i,\vartheta_j,\varphi,\vartheta) 
461: \; d\Omega_i d\Omega_j d\omega_{ij} \;\; . \nonumber
462: \end{eqnarray}
463: $G^P_n(\alpha_{si},\alpha_{sj})$ is symmetric in its two arguments and a
464: polydisperse counterpart to the function $G_n(\alpha_{s})$ defined in 
465: paper~I. It is $G^P_n(\alpha_{s},\alpha_{s}) = G_n(\alpha_{s})$.
466: Some of the $G_n^P$ are given in the appendix.
467: 
468: Inserting Eq.~(\ref{Gndef}) into Eq.~(\ref{Anresult1}), integrating over 
469: $r_{ij}$ between the minimal distance $D_{ij}$ and $\infty$, and 
470: introducing $\ophi$ and $\oepsilon$ yields
471: \begin{eqnarray}
472: A_n &=& \frac{2}{N} z_i z_j\ophi \oepsilon^n (\mu_i \mu_j)^n 
473: \Delta_{ij}^{3-3n} G^P_n(\alpha_{si},\alpha_{sj}) \;\; , 
474: \end{eqnarray}
475: and together with (\ref{FexpinOphiwithA}) the free energy
476: \begin{eqnarray}
477: F &=& F_0 + ... \\ 
478: &&  - \frac{k T}{N} \sum_{n=2}^\infty \ophi \oepsilon^n 
479: {\sum}^\prime  (\mu_i \mu_j)^n 
480: \Delta_{ij}^{3-3n} G^P_n(\alpha_{si},\alpha_{sj})  \;\; . \nonumber 
481: \end{eqnarray}
482: Here the dots 
483: represent the contribution from $A_0$ that was not calculated. It can easily 
484: be shown that $F$ does not depend on a particular definition of $\om$ or
485: $\oD$.
486: 
487: Now, the equilibrium magnetization $M(\oalpha_s)$ is given in $O(\ophi)$ by
488: \begin{equation}
489: M(\oalpha_s)= 
490: \frac{N \om}{\mu_0 V}  \left[ {\cal L}_{\text{poly}}(\oalpha_s) + 
491: \sum_{n=2}^\infty  
492: \ophi \oepsilon^n G_{\text{poly},n}^{\prime}(\oalpha_s) \right] \;\; . 
493: \label{MofalphasOphi}
494: \end{equation}
495: The function $G_{\text{poly},n}^{\prime}$  is the derivative of 
496: \begin{eqnarray}\label{Gpolydef}
497: G_{\text{poly},n}(\oalpha_s) &=& \frac{1}{N^2} {\sum}^\prime 
498: (\mu_i \mu_j)^n \Delta_{ij}^{3-3n} 
499: G^P_n(\alpha_{si},\alpha_{sj}) \nonumber \\ 
500: &=& \int [\mu(\Delta_i) \mu(\Delta_j)]^n 
501: \Delta_{ij}^{3-3n} \nonumber \\
502: &\times& G^P_n(\alpha_{si},\alpha_{sj}) P(\Delta_i) P(\Delta_j) d \Delta_i d \Delta_j 
503: \end{eqnarray}
504: which is is a generalization of $G_n$ \cite{HL} and reduces to the latter in 
505: the monodisperse case $D_i = \oD = D$ and $m_i = \om = m$. 
506: 
507: In a last step, we convert the expression for $M$ as a 
508: function of $\oalpha_s$ into a function of $\oalpha$ using the definition 
509: of $H_s$ in Eq.~(\ref{Hsdef}). By expanding and iterating in a way that is
510: analogous to the procedure in paper~I, Sec.~IV~C we obtain the 
511: final result for $M$ up to order $\ophi$ 
512: \begin{eqnarray}
513: M(\oalpha) &=& \frac{N \om}{\mu_0 V } \Big[ {\cal L}_{\text{poly}}(\oalpha) +
514: 8 \ophi \oepsilon {\cal L}_{\text{poly}}(\oalpha)
515: {\cal L}^\prime_{\text{poly}}(\oalpha)  \nonumber \\
516: && \qquad \qquad+ 
517: \sum_{n=2}^\infty  \ophi \oepsilon^n G_{\text{poly},n}^\prime(\oalpha)
518: \Big] \;\; . \label{Mofalpha}
519: \end{eqnarray}
520: The two leading terms can be seen as the polydisperse extension of the high 
521: temperature approximation derived in \cite{BI92} for monodisperse systems.
522: 
523: \subsection{The contribution in $O(\ophi^2 \oepsilon^2)$}
524: \label{SEC34}
525: 
526: For the monodisperse system, the magnetization contribution in 
527: $O(\ophi^2 \oepsilon^2)$ was calculated in Sec.~V and appendix~B 
528: of paper~I. The cluster
529: integrals needed in that order are shown in Fig.~3 of paper~I. Some of them
530: vanish for the same reason as the contribution $A_1$ in $O(\ophi \oepsilon)$:
531: They involve the averaging of a dipolar field over a spherical surface.
532: Most of the remaining integrals cancel when the free energy $F = - kT \ln Z$
533: is calculated. Up to $O(\oepsilon^2)$ we can write the remaining term in
534: Eq.~(\ref{Fseries}) as
535: \begin{eqnarray}
536: \left< f_{ij} f_{ik} f_{jk} \right>_0 &=& 
537: \left< f^{(0)}_{ij} f^{(0)}_{ik} f^{(0)}_{jk} \right>_0 +
538: 3 \left< f^{(2)}_{ij} f^{(0)}_{ik} f^{(0)}_{jk} \right>_0 \nonumber \\
539: && +
540: 3 \left< f^{(1)}_{ij} f^{(1)}_{ik} f^{(0)}_{jk} \right>_0 \;\; .
541: \label{FtermsinOphi2}
542: \end{eqnarray}
543: These three terms correspond to the graphs E, G and H in paper~I. The first one 
544: is an $O(\ophi^2 \oepsilon^0)$--term that does not contribute to the
545: magnetization. 
546: 
547: \subsubsection{Graph G}
548: 
549: Here we give an outline of the polydisperse generalization of the
550: calculation pertaining to graph G (appendix~B.7 of paper~I). The quantity 
551: $Z_G$ calculated in paper~I, is given by the sum
552: \begin{equation}
553: Z_G = Z_0 \frac{1}{2} {\sum}^\prime 
554: \left< f^{(2)}_{ij} f^{(0)}_{jk} f^{(0)}_{ki} \right>_0 
555: \end{equation}
556: over distinct particles $i, j, k$.
557: For calculating $\left< f^{(2)}_{ij} f^{(0)}_{jk} f^{(0)}_{ki} \right>_0$ one
558: starts with the trivial integrations: the degrees of freedom of all particles
559: except $i$, $j$, and $k$, the position of the center of mass of the three
560: remaining particles, and the orientation of ${\bf m}_k$, since particle $k$
561: is not involved in dipolar interactions in this cluster. Then, for a fixed
562: distance $r_{ij}$ the integrations over $\omega_{ij}$, $\Omega_i$, and
563: $\Omega_j$, defined as in Sec.~\ref{SEC33} are carried out. This introduces
564: the function $G_2^P$ into the result:
565: \begin{eqnarray}
566: && \left< f_{ij}^{(2)} f_{ik}^{(0)} f_{jk}^{(0)} \right>_0 = 
567: \frac{36}{\pi N^2} (\mu_i \mu_j)^2 \ophi^2 \oepsilon^2 
568: G^P_2(\alpha_{si},\alpha_{sj}) \nonumber \\
569: && \qquad \times \int
570: r_{ij}^{-4} e^{-v_{ij}^{HC}}  
571:  f_{ik}^{(0)} f_{jk}^{(0)} \; 
572: d r_{ij} d {\bf r}_{ik} \;\; .\label{GGfirst} 
573: \end{eqnarray}
574: The integral over ${\bf r}_{ik}$ can be described by the following geometrical 
575: considerations:
576: The volume of possible positions of particle $k$ has to be found, such that
577: this particle overlaps with both, particle $i$ (i.~e.\ $r_{ik} < D_{ik}$)
578: and $j$ ($r_{jk} < D_{jk}$). Otherwise the integral would vanish
579: because of the factor $f_{ik}^{(0)} f_{jk}^{(0)}$. This is only possible, if 
580: $r_{ij} < D_{ik} + D_{jk}$. 
581: 
582: In a final step the integration over $r_{ij}$
583: between $D_{ij}$ and  $D_{ik} + D_{jk}$ is carried out. The final result is
584: \begin{eqnarray}
585: \left< f_{ij}^{(2)} f_{ik}^{(0)} f_{jk}^{(0)} \right>_0 &=&
586: \frac{3}{N^2} (\mu_i \mu_j)^2 \ophi^2 \oepsilon^2 
587: G^P_n(\alpha_{si},\alpha_{sj})\nonumber \\
588: && \times f^G( \Delta_{ij},\Delta_{ik}, \Delta_{jk}) \;\; .
589: \end{eqnarray} 
590: The function $f^G( \Delta_{ij},\Delta_{ik}, \Delta_{jk})$ is given in the
591: appendix. The contribution to the free energy is according to 
592: Eq.~(\ref{Fseries}) and Eq.~(\ref{FtermsinOphi2}) given by
593: \begin{eqnarray}
594: && - k T \frac{1}{6} {\sum}^\prime 
595: 3 \left< f^{(2)}_{ij} f^{(0)}_{jk} f^{(0)}_{ki} \right>_0 \nonumber \\ 
596: &&\qquad = - N k T \ophi^2 \oepsilon^2 \frac{1 + 6 \ln 2}{4} 
597: \widetilde{G}_{\text{poly},2} (\oalpha_s) 
598: \label{FcontriG} \;\; .
599: \end{eqnarray}
600: The function
601: \begin{eqnarray}\label{Gpoly2tilde}
602: && \widetilde{G}_{\text{poly},2} (\oalpha_s) = \frac{6}{1 + 6 \ln 2}\\ 
603: && \qquad \times \frac{1}{N^3} {\sum}^\prime
604: (\mu_i \mu_j)^2 G^P_n(\alpha_{si},\alpha_{sj}) 
605: f^G( \Delta_{ij},\Delta_{ik}, \Delta_{jk}) \;\; , \nonumber
606: \end{eqnarray}
607: was defined in such a way, that it reduces to $G_2(\alpha_s)$ in the
608: monodisperse case. Finally introducing the diameter distribution function 
609: $P(\Delta)$ requires the replacement
610: \begin{equation}
611: \frac{1}{N^3} {\sum}^\prime \rightarrow 
612: \int P(\Delta_i) \,d \Delta_i P(\Delta_j) \,d \Delta_j 
613: P(\Delta_k) \,d \Delta_k \label{Pdintro} 
614: \end{equation}
615: in eqs.~(\ref{FcontriG},\ref{Gpoly2tilde}).
616: 
617: \subsubsection{Graph H}
618: 
619: The integrations to calculate 
620: $\left< f^{(1)}_{ij} f^{(1)}_{jk} f^{(0)}_{ki} \right>_0$ are performed as
621: follows (compare to appendix B.8 in paper~I): After doing the trivial
622: integrations (concerning the possible configurations of the particles except 
623: $i$, $j$, and $k$, and the center of mass of the cluster), the possible 
624: orientations of the cluster are integrated out. Then, the integrations over the
625: orientations of ${\bf m}_i$, ${\bf m}_j$, and ${\bf m}_k$ are performed.
626: One arrives at
627: \begin{eqnarray}
628: && \left< f_{ij}^{(1)} f_{ik}^{(1)} f_{jk}^{(0)} \right>_0 = 
629: \mu_i^2 \mu_j \mu_k \oepsilon^2 \frac{4 \pi^2}{15 V^2} 
630: K^P(\alpha_{si}, \alpha_{sj}, \alpha_{sk}) 
631: \nonumber \\
632: && \qquad \times \int 
633: \frac{\oD^6}{r_{ij} r_{ik}} \left( 3 \cos^2 \vartheta_{jk} - 1 \right)
634: e^{-v_{ij}^{HC}} e^{-v_{ik}^{HC}} f_{jk}^{(0)} 
635: \nonumber \\
636: && \qquad \qquad \times \dd r_{ij} 
637: \dd r_{ik} \sin \vartheta_{jk} \dd \vartheta_{jk} \;\; . \qquad\qquad
638: \label{Hresult1}
639: \end{eqnarray}
640: This is up to a factor $N^2 Z_0/2$ a polydisperse generalization of 
641: Eq.~(B21) of paper~I. The remaining integrations concern the distances $r_{ij}$
642: and $r_{ik}$, and the angle $\vartheta_{jk}$ between ${\bf r}_{ij}$ and
643: ${\bf r}_{ik}$. 
644: The function $K^P$ is defined by
645: \begin{eqnarray}
646: && K^P(\alpha_{si}, \alpha_{sj}, \alpha_{sk}) 
647: = \frac{3}{8} \frac{(4 \pi V)^3}{z_i z_j z_k}
648: \nonumber \\ && \qquad \times
649: \int_{-1}^1 \int_{-1}^1 \int_{-1}^1
650: e^{\left( \alpha_{si} u_i + \alpha_{sj} u_j +\alpha_{sk} u_k \right)} 
651: \\ && \qquad \qquad \times
652: \left( 3 + u^2_i \right) u_j u_k \, \dd u_i \dd u_j \dd u_k 
653: \nonumber
654: \label{Kdef}
655: \end{eqnarray}
656: [compare to Eq.~(B22) of paper~I]. An expression for the monodisperse
657: counterpart $K(\alpha_s)$ was given in appendix A of paper~I. Here we express $K^P$
658: via Langevin functions: 
659: \begin{eqnarray}
660: && K^P(\alpha_{si}, \alpha_{sj}, \alpha_{sk}) =  \nonumber \\ 
661: && \qquad 3 {\cal L}(\alpha_{sj}) {\cal L}(\alpha_{sk}) 
662: \left[ 3 + {\cal L}^\prime(\alpha_{si}) + {\cal L}(\alpha_{si})^2 \right]\;\; .
663: \label{Kresult}
664: \end{eqnarray}
665: Note that $r_{ij} > D_{ij}$ and $r_{ik} > D_{ik}$ is required, otherwise the 
666: integral (\ref{Hresult1}) vanishes. The requirement that particles $j$ and $k$ 
667: have to overlap imposes the additional restrictions 
668: $|r_{ij} - r_{ik}| < D_{jk}$ and  
669: \begin{equation}
670: \vartheta_{jk} < \vartheta^{\text{max}}_{jk} = \arccos 
671: \frac{r_{ij}^2 + r_{ik}^2 -D_{jk}^2}{2 r_{ij} r_{ik}} \;\; .
672: \end{equation}
673: After performing the remaining integrations within these limits the result is 
674: \begin{eqnarray}
675: &&\left< f_{ij}^{(1)} f_{ik}^{(1)} f_{jk}^{(0)} \right>_0 = 
676: \mu_i^2 \mu_j \mu_k \ophi^2 \oepsilon^2 \frac{48}{5 N^2} \nonumber \\
677: && \qquad \times 
678: K^P(\alpha_{si}, \alpha_{sj}, \alpha_{sk}) 
679: f^K(\Delta_{ij}, \Delta_{ik}, \Delta_{jk}) \;\; . \label{Hresult2}
680: \end{eqnarray}
681: The function $f^K(\Delta_{ij}, \Delta_{ik}, \Delta_{jk})$ is given in the 
682: appendix. 
683: The contribution to the free energy is
684: \begin{equation}
685: - k T \frac{1}{6} {\sum}^\prime 
686: 3 \left< f^{(1)}_{ij} f^{(1)}_{jk} f^{(0)}_{ki} \right>_0 = 
687: N k T \ophi^2 \oepsilon^2 K_{\text{poly}} (\oalpha_s) \;\; ,
688: \label{FcontriH}
689: \end{equation}
690: with
691: \begin{eqnarray}
692: && K_{\text{poly}} (\oalpha_s) = -\frac{24}{5} \label{Kpolydef}  
693: \frac{1}{N^3} {\sum}^\prime \mu_i^2 \mu_j \mu_k 
694: \\ && \qquad \times
695: K^P(\alpha_{si},\alpha_{sj},\alpha_{sk}) 
696: f^K( \Delta_{ij},\Delta_{ik}, \Delta_{jk}) 
697: \;\; . \nonumber
698: \end{eqnarray}
699: 
700: \subsubsection{The magnetization contribution}
701: 
702: Inserting Eq.~(\ref{FcontriG}) and Eq.~(\ref{FcontriH}) in Eq.~(\ref{Fseries})
703: and calculating the equilibrium magnetization results in an additional
704: $O(\ophi^2 \oepsilon^2)$--term in Eq.~(\ref{MofalphasOphi}):
705: \begin{eqnarray}
706: \frac{N \om}{\mu_0 V} \ophi^2 \oepsilon^2 
707: \left[ \frac{1 + 6 \ln 2}{4} 
708: \widetilde{G}_{\text{poly},2}^\prime (\oalpha_s) - 
709: K_{\text{poly}}^\prime (\oalpha_s) \right] \;\; .
710: \end{eqnarray}
711: The expansion and iteration procedure to switch from $\oalpha_s$ to $\oalpha$
712: is identical to the monodisperse case in paper~I. The full expression for the 
713: magnetization containing all calculated terms reads
714: \begin{eqnarray}\label{Mfinal}
715: M &=& 
716: \frac{N \om}{\mu_0 V}  \Biggl[ {\cal L}_{\text{poly}}(\oalpha) + 
717: 8 \ophi \oepsilon {\cal L}_{\text{poly}} (\oalpha) 
718: {\cal L}_{\text{poly}}^\prime (\oalpha) \nonumber \\
719: && \qquad + \sum_{n=2}^\infty
720: \ophi \oepsilon^n G_{\text{poly},n}^\prime (\oalpha) \nonumber \\
721: && \qquad + 64 \ophi^2 \oepsilon^2 {\cal L}_{\text{poly}} (\oalpha) 
722: {\cal L}_{\text{poly}}^\prime (\oalpha)^2\\
723: && \qquad + 32 \ophi^2 \oepsilon^2 {\cal L}_{\text{poly}} (\oalpha)^2 
724: {\cal L}^{\prime\prime}_{\text{poly}} (\oalpha) \nonumber \\
725: && \qquad + \ophi^2 \oepsilon^2 \frac{1 + 6 \ln 2}{4} 
726: \widetilde{G}_{\text{poly},2}^\prime (\oalpha)  - 
727: \ophi^2 \oepsilon^2 K_{\text{poly}}^\prime (\oalpha) \Biggr]
728: \;\; .\nonumber
729: \end{eqnarray}
730: 
731: \section{Selected results for the magnetization}
732: \label{SEC4}
733: 
734: The figures \ref{L11L22fig} and \ref{L13fig} showing 
735: separate contributions to $M(\oalpha)$ in polydisperse systems were obtained 
736: by using lognormal distributions
737: \begin{equation} \label{logndist}
738: P(D) = \frac{1}{\sqrt{2 \pi} \sigma D_0 } e^{-\sigma^2/2}
739: e^{-\ln^2(D/D_0)/(2 \sigma^2)}
740: \end{equation}
741: for the particle diameters as a representative and often used example for size 
742: distibutions of model polydisperse systems. Using similar distributions like, 
743: e.~g., \ Gamma distributions, however, would not qualitatively modify the results. 
744: In Sec.~\ref{SEC4a} we use also an experimentally determined size distribution.
745: Here, the quantity $\oD$ was taken to 
746: be defined via the mean volume  
747: (\ref{defDbar}), i.e., $\oD^3 = \left< D^3 \right>_P =
748: D_0^3 e^{15 \sigma^2/2}$ so that $\ophi$ is 
749: the volume fraction 
750: $\phi$. The magnetic moments of the particles were taken to scale with their
751: volumes, $m \sim D^3$, allowing to set $\mu(\Delta) = \Delta^3$ such that
752: $\om = \left< m \right>_P$.
753: 
754: Figures \ref{L11L22fig} and \ref{L13fig} show the contributions
755: \begin{mathletters}
756: \begin{eqnarray}
757: L_{1,1} (\oalpha) &=& 8 {\cal L}_{\text{poly}} (\oalpha) 
758: {\cal L}_{\text{poly}}^\prime (\oalpha) \;\; ,\\
759: L_{2,2} (\oalpha) &=&  64 {\cal L}_{\text{poly}} (\oalpha) 
760: {\cal L}_{\text{poly}}^\prime (\oalpha)^2 \nonumber \\
761: && + 32 {\cal L}_{\text{poly}} (\oalpha)^2 
762: {\cal L}^{\prime\prime}_{\text{poly}} (\oalpha)  \\
763: && + \frac{1 + 6 \ln 2}{4} 
764: \widetilde{G}_{\text{poly},2}^\prime (\oalpha)  - 
765: K_{\text{poly}}^\prime (\oalpha) \;\; , \nonumber \\
766: L_{1,2} (\oalpha) &=& G_{\text{poly},2}^\prime (\oalpha)\;\; ,\\
767: L_{1,3} (\oalpha) &=& G_{\text{poly},3}^\prime (\oalpha)
768: \end{eqnarray}
769: \end{mathletters}
770: to $M(\oalpha)$ (\ref{Mfinal})
771: for different values of the width $\sigma$ of the distribution (\ref{logndist}).
772: The contributions of higher-order terms increase with growing $\sigma$. 
773: This is so because they depend
774: on higher moments of the distribution $P(\Delta)$ that grow with 
775: the width of the distribution, even if the third moment 
776: $\left< \Delta^3 \right>_P$ is kept fixed. The shift of the maxima
777: of the curves to the right has a similar reason: Bigger particles, that react to smaller
778: fields get more and more important when the width of the distribution
779: grows.
780: 
781: The assumption $\mu(\Delta) = \Delta^3$ that the magnetic moments in our 
782: ferrofluid model scale $\sim D^3$ with the total volume 
783: of its hard sphere constituents is somewhat too simple for particles in real
784: ferrofluids for two reasons: First, in the common case of steric stabilization 
785: by polymers surfactants providing the repulsion the surfactant layer of about 
786: 1 -- 3 nm does not contribute to the magnetic moment. Second, an outer layer of
787: the magnetic material might be magnetically dead so that it does not contribute
788: to the magnetic moment either. For magnetite particles a dead layer depth 
789: of $\approx$ 0.8 nm has been reported \cite{SISI87}. To account for the sum of 
790: these
791: two effects we have introduced in our calculation for $M$ an effective magnetic 
792: diameter $D_{\text{mag}}$ via the relation $D = D_{\text{mag}} + 5.6 \text{ nm}$.
793: It ascribes to every particle that has a magnetically effective core of diameter 
794: $D_{\text{mag}}$ a hard sphere with a magnetically inert layer of depth 2.8 nm.
795: The magnetic moment of each particle was then taken to scale with is magnetically
796: effective volume, i.e., $\mu = (D_{\text{mag}}/\oD)^3$.
797: 
798: The full line in Fig.~\ref{magcomp} shows the reduced equilibrium magnetization
799: of an interacting polydisperse ferrofluid
800: as function of $\oalpha$. Here each particle has a total magnetic inactive layer of 
801: 2.8 nm and the diameters $D_{\text{mag}}$ of the magnetic 
802: cores are lognormally distributed with $\sigma = 0.25$
803: and $\left< D_{\text{mag}}^3 \right>_P^{1/3}$ = 10 nm. The latter is 
804: taken as the reference diameter $\oD$ in the calculation.
805: The particle density is chosen in 
806: such a way that $\ophi = 0.05$. Note that here $\ophi = \phi_{\text{mag}}$,
807: the volume fraction of the magnetically active material. The magnetic
808: moment $\om$ for $D_{\text{mag}} =$ 10 nm is chosen such that $\oepsilon = 2$.
809: The magnetization of the core is then about 550 kA/m at room temperature which
810: is a little bit larger than that of magnetite. 
811: 
812: Expansion terms of order 
813: $\ophi \oepsilon^n$ are taken into account up to order $n$=5 for the 
814: magnetization
815: curves in Fig.~\ref{magcomp}. Higher $O(\ophi \oepsilon^n)$--terms have only a 
816: small effect on the magnetization. Here $\oepsilon$ is by
817: definition of $\om$ and $\oD$ a typical interaction energy divided by $kT$ 
818: for particles at a distance of $\oD$ = 10 nm. But with the two additional dead 
819: layers
820: of total size 5.6 nm in between the particles of our model ferrofluid the real 
821: typical dipolar energies at contact are smaller. 
822: 
823: The magnetization is compared to that of a
824: polydisperse fluid without particle--particle interaction (long dashed line), 
825: and to a 
826: monodisperse ferrofluid, both with and without taking into account the 
827: particle--particle interaction (short dashed and dotted line, respectively). 
828: The monodisperse system consists of particles with $D_{\text{mag}} =$ 10 nm
829: with the same nonmagnetic layer thickness, bulk magnetization, and 
830: $\phi_{\text{mag}}$ as before. 
831: 
832: One sees that 
833: taking into account polydispersity or particle interaction alone strengthens
834: the magnetization and especially the initial susceptibility. Both effects are 
835: comparable for the given parameters. Together, they result in an even higher
836: equilibrium magnetization. 
837: 
838: Figure \ref{magofsigma} shows magnetization curves for magnetite--based
839: ferrofluids with a bulk magnetization of 480 kA/m for distributions of different
840: widths. $D_{\text{mag}}$ is taken 
841: to be lognormally distributed 
842: with $\left< D_{\text{mag}}^3 \right>_P = (8 \text{ nm})^3$ and $\sigma =$ 0.2,
843: 0.3, and 0.4. The volume fraction
844: of the magnetic material is $\phi_{\text{mag}} = 0.1$. The particles are again 
845: assumed to carry a nonmagnetic layer of 2.8 nm thickness. 
846: 
847: The increase in $\sigma$ causes an increase of the initial
848: susceptibility already in the noninteracting case (long dashed lines). 
849: Including the  $O(\ophi \oepsilon)$--terms (short dashed lines) has a positive 
850: effect on the magnetization. The relative increase is maximal for small 
851: $\oalpha$. The magnetization decreases again at higher $\oalpha$, if higher 
852: order terms are taken into account (solid lines). For the considered 
853: ferrofluids the $O(\ophi^2 \oepsilon^2)$--term that is negative for higher
854: $\oalpha$ (see Fig.~\ref{L11L22fig}) is almost solely responsible for this 
855: decrease. The positive contributions from the higher 
856: $O(\ophi \oepsilon^n)$--terms ($n \geq 2$) are again negligible, except for 
857: $\sigma = 0.4$ and small $\oalpha$, 
858: where they cause a further increase of the initial susceptibility. 
859: For small $\oalpha$ the $O(\ophi^2 \oepsilon^2)$--term also has a positive 
860: effect, but this effect is too small to be visible. The plots show again,
861: that the influence of higher order terms is larger for broad distributions. 
862: 
863: \section{Comparison to experiments}
864: \label{SEC4a}
865: 
866: We compared our theorerical predictions for the 
867: magnetization curves with experimental results of
868: two different magnetite--based ferrofluids. S. Odenbach (ZARM, Bremen) provided 
869: data on the equilibrium magnetization of
870: the ferrofluid EMC~905 produced by FerroTec. We fitted our theoretical result 
871: (\ref{Mfinal}) taking into account the terms in $O(\phi \epsilon^n)$ 
872: up to $n = 5$ to the data assuming lognormally distributed $D_{\text{mag}}$
873: and a nonmagnetic layer of depth 2.8 nm. The bulk magnetization of magnetite
874: was taken to be 480 kA/m. The result is shown in Fig.~\ref{magEMC}.
875: According to the fit, the saturation magnetization of the ferrofluid is
876: $M_{\text{sat}} = 37.4$ kA/m. The parameters defining the distribution turn out
877: to be $D_0 = 8.3$ nm and $\sigma = 0.28$. There are small differences between
878: the data and the fit curve that are in our opinion due to deviations of the real
879: diameter distribution curve from the idealized lognormal form. 
880: 
881: J. Embs (Universit\"at des Saarlandes, Saarbr\"ucken) measured the equilibrium
882: magnetization curve of the ferrofluid APG~933 of FerroTec. In addition he
883: determined the diameter distribution of its particles by transmission electron
884: microscopy (TEM) which was then used in our theoretical analysis. Diameters found 
885: in TEM measurements are those of the magnetite particles. We assumed the 
886: magnetically effective 
887: diameters $D_{\text{mag}}$ to be $2 \times 0.8$ nm $= 1.6$ nm smaller and to be 
888: zero for particles smaller than 1.6 nm.  The hard core diameters $D$ were taken
889: to be $2 \times 2$ nm $= 4$ nm larger than the diameters obtained from the TEM
890: measurements. As above, we took into account terms 
891: up to $O(\phi \epsilon^5)$ and set the bulk magnetization of magnetite to 
892: 480 kA/m. Fig.~\ref{magAPG} shows the TEM data and 
893: the experimental magnetization curve together with the results of our theory. 
894: Both agree very well.
895: 
896: 
897: \section{Conclusion}
898: \label{SEC5}
899: 
900: In this paper we used the technique of cluster expansion to derive 
901: an approximation to the equilibrium magnetization for the system of dipolar
902: hard spheres in a magnetic field with diameter and/or magnetic moment 
903: dispersion as a model system for a polydisperse ferrofluid. The calculation
904: results in an expression for the magnetization $M$ in form of a twofold 
905: series expansion in the parameters $\ophi$, closely related to the volume 
906: fraction $\phi$, and $\oepsilon$, a coupling
907: parameter measuring the strength of the dipolar interaction:
908: \begin{equation}
909: M = \frac{N \om}{\mu_0 V}  
910: \sum_{m=0}^\infty \sum_{n=0}^\infty \ophi^m \oepsilon^n
911: L_{m,n} (\oalpha) \;\; .
912: \end{equation}
913: $\ophi$, $\oepsilon$, and the dimensionless magnetic field $\oalpha$ are 
914: defined for some typical values $\oD$ and $\om$ for the hard sphere diameters
915: and magnetic moments respectively. $\om$ can be chosen in such a way that
916: the prefactor $N \om/\mu_0 V$ reduces to the saturation magnetization of the
917: system. We gave expressions for $L_{1,n}$ ($n \leq 5$) and $L_{2,2}$. Lower
918: orders vanish, except for $L_{0,0}$ reducing to the Langevin function in the
919: monodisperse case. The calculated $L_{m,n}$ can be written as multiple sums
920: over all particles whose addends are analytical expressions.
921: 
922: The influence of particle--particle interaction grows with increasing width
923: of the considered diameter distribution.
924: Taking into account only the $L_{1,1}$--term results in an increase of the 
925: magnetization relative to the non interacting system, 
926: whereas the $L_{2,2}$--term leads again to somewhat smaller values at higher
927: $\oalpha$. Only at very small $\oalpha$ its contribution is positive. 
928: The $L_{1,n}$--terms have little effect for realistic, 
929: magnetite--based ferrofluids, except for broad distributions, where they 
930: increase the initial magnetization.
931: 
932: \acknowledgments
933: 
934: We would like to thank Jan Embs and Stefan Odenbach for providing
935: the experimental data discussed in Sec.~\ref{SEC4a}.
936: This work was supported by the Deutsche Forschungsgemeinschaft (SFB 277).
937: 
938: \appendix
939: 
940: \section{The functions $G_n^P, f^G, f^K$}
941: \label{appa}
942: 
943: %\section*{}
944: 
945: The functions $G_n^P(x_1,x_2)$ are symmetric in their
946: arguments and have the form 
947: \begin{eqnarray}
948: && G_n^P (x_1,x_2 ) = 
949: G^{P(0)}_{n}( 1/x_1 , 1/x_2) \nonumber \\ 
950: && \qquad + \coth (x_1) G^{P(1)}_{n}(1/x_1, 1/x_2) 
951: \\ 
952: && \qquad + \coth (x_2 ) G^{P(1)}_{n}(1/x_2, 1/x_1) 
953: \qquad \nonumber \\ && \qquad +
954: \coth (x_1 ) \coth (x_2 )  
955: G^{P(2)}_{n}(1/x_1 , 1/x_2 ) \;\; .\nonumber 
956: \end{eqnarray}
957: The $G^{P(i)}_{n}$ are polynomials and read for $n \leq 5$
958: \begin{mathletters}
959: \begin{eqnarray}
960: G_2^{P(0)}(y_1,y_2) &=&
961: \frac{8}{5}  
962: +\frac{4}{5} y_1^{2} 
963: +\frac{4}{5}  y_2^{2}
964: +\frac{12}{5} y_1^{2} y_2^{2} \;\; ,
965: \nonumber \\
966: G_2^{P(1)}(y_1,y_2) &=&
967: -\frac{4}{5} y_1 
968: -\frac{12}{5} y_1 y_2^{2} \;\; ,
969: \\
970: G_2^{P(2)}(y_1,y_2) &=&
971: \frac{12}{5}  \;\; .
972: \nonumber
973: \end{eqnarray}
974: \begin{eqnarray}
975: G_3^{P(0)}(y_1,y_2) &=&
976: -\frac{4}{35} y_1 y_2
977: -\frac{24}{35} y_1^{3} y_2 \nonumber\\ &&
978: -\frac{24}{35} y_1 y_2^{3}
979: -\frac{12}{7} y_1^{3} y_2^{3} \;\; ,
980: \nonumber \\
981: G_3^{P(1)}(y_1,y_2) &=&
982: -\frac{4}{35}  y_2
983: +\frac{24}{35} y_1^{2} y_2 \\ && \nonumber
984: +\frac{4}{35}  y_2^{3}
985: +\frac{12}{7} y_1^{2} y_2^{3} \;\; ,
986: \\
987: G_3^{P(2)}(y_1,y_2) &=&
988: \frac{16}{105}  
989: -\frac{4}{35} y_1^{2} 
990: -\frac{4}{35}  y_2^{2}
991: -\frac{12}{7} y_1^{2} y_2^{2} \;\; .
992: \nonumber
993: \end{eqnarray}
994: \begin{eqnarray}
995: G_4^{P(0)}(y_1,y_2) &=&
996: \frac{8}{105}  
997: +\frac{4}{35} y_1^{2} 
998: +\frac{4}{35} y_1^{4} 
999: +\frac{4}{35}  y_2^{2}\nonumber \\ &&
1000: +\frac{12}{5} y_1^{2} y_2^{2} 
1001: +\frac{36}{7} y_1^{4} y_2^{2}
1002: +\frac{4}{35}  y_2^{4} \nonumber \\ &&
1003: +\frac{36}{7} y_1^{2} y_2^{4}
1004: +12 y_1^{4} y_2^{4}\;\; ,
1005: \nonumber \\
1006: G_4^{P(1)}(y_1,y_2) &=&
1007: -\frac{8}{105} y_1 
1008: -\frac{4}{35} y_1^{3} 
1009: -\frac{24}{35} y_1 y_2^{2}\\ &&
1010: -\frac{36}{7} y_1^{3} y_2^{2}
1011: -\frac{8}{7} y_1 y_2^{4}
1012: -12 y_1^{3} y_2^{4} \;\; ,\nonumber
1013: \\
1014: G_4^{P(2)}(y_1,y_2) &=&
1015: \frac{32}{105} y_1 y_2
1016: +\frac{8}{7} y_1^{3} y_2
1017: +\frac{8}{7} y_1 y_2^{3}
1018: +12 y_1^{3} y_2^{3} \;\; .
1019: \nonumber
1020: \end{eqnarray}
1021: \begin{eqnarray}
1022: G_5^{P(0)}(y_1,y_2) &=&
1023: \frac{12}{385} y_1 y_2
1024: -\frac{104}{385} y_1^{3} y_2
1025: -\frac{60}{77} y_1^{5} y_2\nonumber \\ &&
1026: -\frac{104}{385} y_1 y_2^{3}
1027: -\frac{732}{77} y_1^{3} y_2^{3}
1028: -\frac{240}{11} y_1^{5} y_2^{3}\nonumber \\ &&
1029: -\frac{60}{77} y_1 y_2^{5}
1030: -\frac{240}{11} y_1^{3} y_2^{5}
1031: -\frac{540}{11} y_1^{5} y_2^{5} \;\; ,
1032: \nonumber \\
1033: G_5^{P(1)}(y_1,y_2) &=&
1034: -\frac{4}{231}  y_2
1035: +\frac{4}{385} y_1^{2} y_2
1036: +\frac{60}{77} y_1^{4} y_2 \nonumber \\ &&
1037: +\frac{4}{385}  y_2^{3}
1038: +\frac{172}{77} y_1^{2} y_2^{3}
1039: +\frac{240}{11} y_1^{4} y_2^{3}  \\ &&
1040: +\frac{4}{77}  y_2^{5}
1041: +\frac{60}{11} y_1^{2} y_2^{5}
1042: +\frac{540}{11} y_1^{4} y_2^{5} \;\; ,\nonumber 
1043: \\
1044: G_5^{P(2)}(y_1,y_2) &=&
1045: \frac{16}{1155}  
1046: +\frac{8}{1155} y_1^{2} 
1047: -\frac{4}{77} y_1^{4} \nonumber \\ &&
1048: +\frac{8}{1155}  y_2^{2}
1049: -\frac{32}{77} y_1^{2} y_2^{2}
1050: -\frac{60}{11} y_1^{4} y_2^{2}\nonumber \\ &&
1051: -\frac{4}{77}  y_2^{4} 
1052: -\frac{60}{11} y_1^{2} y_2^{4}
1053: -\frac{540}{11} y_1^{4} y_2^{4} \;\; .
1054: \nonumber
1055: \end{eqnarray}
1056: \end{mathletters}
1057: 
1058: It is
1059: \begin{eqnarray}
1060: &&f^G( \Delta_{ij},\Delta_{ik}, \Delta_{jk}) = 
1061: \ln \left( \frac{\Delta_{ik} + \Delta_{jk}}{\Delta_{ij}} \right)
1062: - \frac{ 3( \Delta_{ik}^2 - \Delta_{jk}^2)^2}
1063: {4 \Delta_{ij}^4} 
1064: \nonumber \\ && \qquad 
1065: + \frac{ 8 ( \Delta_{ik}^3 + \Delta_{jk}^3)}
1066: {3 \Delta^3_{ij} }
1067: -\frac{3( \Delta_{ik}^2 + \Delta_{jk}^2)}{\Delta_{ij}^2} 
1068: \\ && \qquad 
1069: + \frac{ 13 \Delta_{ik}^2 + 14 \Delta_{ik} \Delta_{jk} + 13 \Delta_{jk}^2}
1070: {12 ( \Delta_{ik} + \Delta_{jk})^2} \;\; , \nonumber
1071: \end{eqnarray}
1072: and
1073: \begin{eqnarray}
1074: && f^K(\Delta_{ij}, \Delta_{ik}, \Delta_{jk}) = -\frac{2}{9} 
1075: - \frac{5}{36} \frac{\Delta_{ij}^6 + \Delta_{ik}^6 + 
1076: \Delta_{jk}^6}{\Delta_{ij}^3 \Delta_{ik}^3} 
1077: \nonumber \\ && \qquad
1078: -\frac{\Delta_{jk}^2}{4 \Delta_{ij} \Delta_{ik}} - 
1079: \frac{2 \Delta_{jk}^3}{9 \Delta_{ij}^3} 
1080: - \frac{2 \Delta_{jk}^3}{9 \Delta_{ik}^3} \\
1081: &&\qquad + \frac{(\Delta_{ij}^4 + \Delta_{ik}^4 + \Delta_{jk}^4)
1082: (\Delta_{ij}^2 + \Delta_{ik}^2 + \Delta_{jk}^2)}
1083: {8 \Delta_{ij}^3 \Delta_{ik}^3} \;\; . \nonumber
1084: \end{eqnarray}
1085: When all diameters are equal, these functions reduce to
1086: \begin{eqnarray}
1087: f^G(\Delta,\Delta,\Delta) &=& \frac{1}{6} + \ln 2 \;\; ,\\
1088: f^K(\Delta,\Delta,\Delta) &=& -\frac{5}{24} \;\; .
1089: \end{eqnarray}
1090: 
1091: \clearpage
1092: \begin{thebibliography}{\addcontentsline{toc}{section}{RERFERENCES}}
1093: 
1094: \bibitem{HL}
1095: B.~Huke and M.~L\"ucke,
1096: Phys.~Rev.~E~{\bf 62}, 6875 (2000).
1097: 
1098: \bibitem{R85}
1099: R.~E.~Rosensweig,
1100: {\it Ferrohydrodynamics},
1101: (Cambridge University Press, Cambrigde, U.K., 1985).
1102: 
1103: \bibitem{W71} 
1104: M.~S.~Wertheim,
1105: J.~Chem.~Phys.~{\bf 55}, 4291 (1971).
1106: 
1107: \bibitem{AD73}
1108: S.~A.~Adelman and J.~M.~Deutch,
1109: J.~Chem.~Phys.~{\bf 59}, 3971 (1973).
1110: 
1111: \bibitem{IB74}
1112: D.~Isbister and R.~J.~Bearman,
1113: Molec.~Phys.~{\bf 28}, 1297 (1974).
1114: 
1115: \bibitem{HS78}
1116: J.~S.~Hoye and G.~Stell, 
1117: J.~Chem.~Phys.~{\bf 70}, 2894 (1978).
1118: 
1119: \bibitem{FHI79}
1120: B.~Freasier, N.~Hamer, and D.~Isbister,
1121: Molec.~Phys.~{\bf 38}, 1661 (1979).
1122: 
1123: \bibitem{RH81}
1124: J.~D.~Ramshaw and N.~D.~Hamer,
1125: J.~Chem.~Phys.~{\bf 75}, 3511 (1981).
1126: 
1127: \bibitem{CB86}
1128: P.~T.~Cummings and L.~Blum,
1129: J.~Chem.~Phys.~{\bf 85}, 6158 (1986).
1130: 
1131: \bibitem{FP84}
1132: P.~H.~Fries and G.~N.~Patey,
1133: J.~Chem.~Phys.~{\bf 82}, 429 (1984).
1134: 
1135: \bibitem{LL87}
1136: P.~H.~Lee and B.~M.~Ladanyi,
1137: J.~Chem.~Phys.~{\bf 87}, 4093 (1987).
1138: 
1139: \bibitem{LL89}
1140: P.~H.~Lee and B.~M.~Ladanyi,
1141: J.~Chem.~Phys.~{\bf 91}, 7063 (1989). 
1142: 
1143: \bibitem{KS02}
1144: G.~Kronome, I.~Szalai, and J.~Liszi,
1145: J.~Chem.~Phys.~{\bf 116}, 2067 (2002). 
1146: 
1147: \bibitem{K99}
1148: V.~I.~Kalikmanov, 
1149: Phys.~Rev.~E {\bf 59}, 4085 (1999).
1150: 
1151: \bibitem{SPMS90}
1152: M.~I.~Shliomis, A.~F.~Pshenichnikov, K.~I.~Morozov, and I.~Yu.~Shurubor,
1153: J.~Magn.~Magn.~Mater.~{\bf 85}, 40 (1990).
1154: 
1155: \bibitem{ML90}
1156: K.~I.~Morozov and A.~V.~Lebedev,
1157: J.~Magn.~Magn.~Mater.~{\bf 85}, 51 (1990).
1158: 
1159: \bibitem{BI92}
1160: Y.~A.~Buyevich and A.~O.~Ivanov,
1161: Physica A, {\bf 190}, 276 (1992).
1162: 
1163: \bibitem{PML96}
1164: A.~F.~Pshenichnikov,  V.~V.~Mekhonoshin, and A.~V.~Lebedev,
1165: J.~Magn.~Magn.~Mater.~{\bf 161}, 94 (1996).
1166: 
1167: \bibitem{IK01}
1168: A.~O.~Ivanov and O.~B.~Kuznetsova,
1169: Phys.~Rev.~E~{\bf 64}, 041405 (2001).
1170: 
1171: \bibitem{Ce82}
1172: A.~O.~Cebers,
1173: Magn.~Gidrodinamika {\bf 2}, 42 (1982).
1174: 
1175: \bibitem{BGW98}   
1176: S.~Banerjee, R.~B.~Griffiths, and M.~Widom,
1177: J.~Stat.~Phys.~{\bf 93}, 109 (1998).
1178: 
1179: \bibitem{SISI87}
1180: T.~Sato, T.~Iijima, M.~Seki, and N.~Inagaki:
1181: J.~Magn.~Magn.~Mater.~{\bf 65}, 252 (1987).
1182: 
1183: \end{thebibliography}
1184: 
1185: 
1186: \clearpage
1187: 
1188: \begin{figure}
1189: \includegraphics[width=16cm,angle=0]{fig1.eps}
1190: \caption[]
1191: {The contributions $L_{1,1}$ and $L_{2,2}$ as functions of $\oalpha$ for
1192: lognormal distributions with different widths $\sigma$. Here
1193: $\left< \Delta^3 \right>_P = 1$ and $\mu(\Delta) = \Delta^3$.}
1194: \label{L11L22fig} 
1195: \end{figure}
1196: 
1197: \begin{figure}
1198: \includegraphics[width=16cm,angle=0]{fig2.eps}
1199: \caption[]
1200: {The contributions $L_{1,2}$ and $L_{1,3}$ as a function of $\oalpha$ for
1201: lognormal distributions with different widths $\sigma$. $\Delta$ and 
1202: $\mu(\Delta)$ as in Fig.~\ref{L11L22fig}.}
1203: \label{L13fig}
1204: \end{figure}
1205: 
1206: \begin{figure}
1207: \includegraphics[width=12cm,angle=270]{fig3.eps}
1208: \caption[]
1209: {Equilibrium magnetization of a polydisperse ferrofluid 
1210: ($\left< D_{\text{mag}}^3 \right>_P^{1/3}$ = 10 nm, $\sigma = 0.25$, 
1211: $\mu = (D_{\text{mag}}/\oD)^3$)
1212: and a comparable monodisperse ferrofluid, both with and without 
1213: taking into account the particle interaction. See text for further details.}
1214: \label{magcomp}
1215: \end{figure}
1216: 
1217: \begin{figure}
1218: \includegraphics[width=13cm,angle=0]{fig4.eps}
1219: \caption[] 
1220: {Equilibrium magnetization as a function of $\oalpha$ for lognormal 
1221: distributions with $\left< D_{\text{mag}}^3 \right>_P^{1/3}$ = 8 nm and different 
1222: widths $\sigma$. Long dashed: without particle interaction, dashed: 
1223: particle interaction only in $O(\ophi \oepsilon)$, solid lines: all calculated
1224: terms.}
1225: \label{magofsigma}
1226: \end{figure}
1227: 
1228: \begin{figure}
1229: \includegraphics[width=15cm,angle=270]{fig5.eps}
1230: \caption[] 
1231: {Equilibrium magnetization of the ferrofluid EMC 905. Dots: experiment,
1232: solid line: the theoretical magnetization curve assuming lognormally distributed
1233: $D_{\text{mag}}$, c.f. text.}
1234: \label{magEMC}
1235: \end{figure}
1236: 
1237: \begin{figure}
1238: \includegraphics[width=12.5cm,angle=0]{fig6corr.eps}
1239: \caption[] 
1240: {Top: equilibrium magnetization of the ferrofluid APG 933. 
1241: Circles: experiment, solid line: the theoretical magnetization curve. Bottom: 
1242: diameter distribution according to TEM measurements.}
1243: \label{magAPG}
1244: \end{figure}
1245: 
1246: 
1247: \end{document}
1248: