PHYSICAL REVIEW E 86, 046601 (2012) Boundary effects on effective conductivity of random heterogeneous media with spherical inclusions A. Rabinovich, G. Dagan, and T. Miloh School of Mechanical Engineering, Tel Aviv University, Tel Aviv, Israel (Received 1 May 2012; revised manuscript received 11 September 2012; published 3 October 2012) It is common to determine the effective conductivity of heterogeneous media by assuming stationarity of the random local properties. This assumption is not obeyed in a boundary layer of a body of finite size. The effect of different types of boundaries is examined for a two-phase medium with spherical inclusions of given conductivity distributed randomly in a matrix of a different conductivity. Exact solutions are derived for the apparent conductivity and the boundary layer thickness. The interaction between the spheres and the boundaries is fully incorporated in the solutions using a spherical harmonics expansion and the method of images. As applications, the corrections for the effective conductivity are given for two cases of finite bodies: the Maxwell sphere and a cylinder of flow parallel to the axis. DOI: 10.1103/PhysRevE.86.046601 PACS number(s): 41.20.Cv, 44.35.+c, 47.56.+r I. INTRODUCTION The classical problem of determining the effective con- ductivity of a medium composed of spherical inclusions of one substance embedded in a matrix of a different material has been investigated extensively for more than a century. Many approximate relationships have been proposed since and comprehensive and up-to-date compendia have been published recently [1,2]. We consider a process that is described mathematically by q =−K T , · q = 0, where q is the flux vector, x = (x,y,z) is a Cartesian coordinate, K (x) is the spatially variable conductivity, and T drives the flow. There are at least eight different physical processes that obey such a law [3], e.g., electrical conduction, magnetism, flow in porous media, dielectrics, etc., but our primary interest is in steady heat flow through solids for which q is the heat flux, K is the thermal conductivity, and T is the temperature. Without loss of generality we shall maintain this nomenclature. We regard K as a random isotropic space function and pur- sue determining qand T , the mean flux and temperature, respectively. Most of the literature dealt with stationary K fields and mean uniform flow, i.e., qand T are constant. With the effective conductivity defined by q〉=−K eff T , one of the central problems of physics is to derive its dependence on the statistical properties of K [1,2]. Strictly speaking, stationarity of K and mean flow uniformity imply an unbounded spatial domain. However, in applications, as well as in numerical or experimental determination of K eff , the domain is bounded. It is generally assumed that the impact of the boundary is felt only in a thin layer, which has a negligible effect upon K eff of the bulk. Our aim is to quantify this effect and to provide estimates of the error incurred by the presence of boundaries onto K eff . Such estimates may prove to be of importance in applications in which the heterogeneity scale is not much smaller than other scales, e.g., for natural formations or thin composite layers. The few past attempts to investigate this problem in the specialized literature dealt with particular configurations and under limiting assumptions. Thus, the works of [46] derive analytical expressions for the mean fields and effective conductivity near a boundary. However, the conductivity is regarded as a continuous random function and weak heterogeneity is assumed. Other relevant studies such as [79] deal with heterogeneous media of periodic nature using the various periodic homogenization techniques. To the best of our knowledge, the only work addressing the classical problem of randomly distributed spherical inclusions near a boundary is [10]. However their approach is quite different from ours for a number of reasons: (i) they use a Green’s function formulation to arrive at the averaged properties (see [11]), which is not straightforward and involves a tedious mathe- matical derivation; (ii) while accounting for the presence of the boundary by using images of the inclusions, the nonlinear interaction between a sphere and its image is determined in an approximate manner, making the solution imprecise for spheres close to the boundary; (iii) only a flow normal to the boundary of a given temperature is considered. In contrast, in this work we use a simple formulation, based on the exact solution of an isolated sphere near a boundary, to solve with great precision, a few types of flow and boundary conditions. II. FORMULATION The structure we consider here is a matrix of conductivity K 0 with nonoverlapping spherical inclusions of radius R and of conductivity K . The domain of the heterogeneous medium is the half space z 0; the influence of the boundary at z = 0 is similar to the one of a finite body, provided that the radius of curvature and other length scales are much larger than R. Furthermore, we consider the dilute limit, for which the volume density n (average volume of inclusions per volume of media) is much smaller than unity. This substantially simplifies the problem and yet has been shown to be applicable for a significant range of values of n ([12] and [2], Sec. 19.1.3). The randomness of K (x) =K 0 + j (K 0 K )I (xx j ), where the unit function I = 1 inside the sphere of center coordinate x j and I = 0 outside, stems from the randomness of x j . An ideal random distribution is defined as the one for which centers are scattered uniformly and independently in space. Such a setup is not possible for relatively dense systems due to the nonoverlap requirement. The random generation of centers for nonoverlapping inclusions is treated in a considerable body of literature (see, for instance, recent discussions in [13] and [14]). In any case the marginal probability density function (PDF) f ( x) is uniform and the nonoverlapping condition 046601-1 1539-3755/2012/86(4)/046601(6) ©2012 American Physical Society