Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 2, 39, 2001 TOLERANCE AVERAGING AND BOUNDARY-LAYER EQUATIONS FOR THE HEAT TRANSFER PROBLEMS IN MICRO-PERIODIC SOLIDS Czesław Woźniak Ewaryst Wierzbicki Institute of Mathematics and Computer Sciences, Technological University of Częstochowa e-mail: wozniak@matinf.pcz.czest.pl, wierzbicki@matinf.pcz.czest.pl Margaret Woźniak Department of Geotechnical and Structure Engineering, Technological University of Łódź e-mail: mwozniak@ck-sg.p.lodz.pl Themacroscopicmathematicalmodels of theheat transfer inmicro-periodic solids, obtained by the tolerance averaging approach, are represented by the partial differential equation for the averaged temperature field and the systemofordinarydifferential equations involving timederivativesof certain extra unknownfieldswhich are called internal thermal variables, cfWoźniak (2000). It follows that in the framework of the aforementioned tolerance models the boundary conditions for temperature canbe imposed exclusively on the averaged temperature field. The aim of this contribution is to show how the tolerance averaging technique can be extended in order to satisfy the boundary conditions on higher level of accuracy. Key words: composites, modelling, tolerance 1. Introduction The simplest mathematical models for the overall (macroscopic) behavior ofmicro-periodic solids can be obtained by using the results of thewell known asymptotic homogenization theory, cf Bensoussan et al. (1978), Jikov et al. (1994). These models are represented by PDEs with constant coefficients, which are referred to as homogenized equations. However, the form of the homogenized equations is independent on the microstructure size and hence 424 C.Woźniak et al. they are incapable of describing the effect ofmicrostructure size on the pheno- mena observed on the macroscopic level. To remove this drawback a number of alternative approaches to the modelling of periodic solids was proposed; the overview of these approaches can be found inWoźniak(1999). In this con- tribution we shall deal with what is called the tolerance averaging of partial differential equations with highly-oscillating micro-periodic coefficients, Woź- niak (2000). The tolerance averaging technique of heat transfer equations for nonstationary problems leads to a system of differential equations with con- stant coefficients (some of them depend on the microstructure size) for the averaged temperature field θ◦ and for certain extra unknown fields VA, A = 1, ...,N. These fields describe the disturbances of temperature caused by the periodic microheterogeneity of the solid. The characteristic feature of the tolerance averaging technique is that the equations for VA are ordinary differential equations involving only time derivatives of VA. That is why V A are called the internal thermal variables. For stationary problems VA are governed by a system of linear algebraic equations, and the tolerance model can be reduced to the homogenized one. It follows that in the framework of the tolerance averaging the boundary conditions for the temperature can be imposed only on the averaged temperature field θ◦. The problem we are going to solve in this contribution can be stated as follows: how to extend the tolerance averaging approach to the modelling of heat transfer problems in the micro-periodic solids in order to satisfy the bo- undary conditions on higher level of accuracy?This problem is strictly related to the fact that the averaged equations, describing processes inmicro-periodic solids on the macroscopic level, are deprived of the physical sense in a cer- tain near boundary layer, Woźniak (1999). Hence, the question arises how to modify these equations for describing the phenomena which take place in the boundary layer. This ”boundary layer” problem is well known if the averaging of equations is carried out by using the known asymptotic homogenization technique; we can mention here the results given by Sanchez-Palencia and Zaoui (1985) as well as more general asymptotic approach to this problem by Panasenko (1994). However, the aforementioned approaches to the ”boun- dary layer” problem cannot be directly applied to the tolerance averaging of equations, where the asymptotic method of modelling is rejected. In this paper, we propose a certain extension of the tolerance averaging of the nonstationary heat transfer equation for a micro-periodic solid, which takes into account the ”boundary layer” problem. To make the contribution self-consistentwe beginwith somemathematical notions andwe formulate the basic propositions in Section 2; theproofs of these propositions are given in the Tolerance averaging and boundary-layer equations... 425 Appendix.The general foundations of the tolerance averaging are summarized in Section 3; more detailed discussion of these results can be found in Woź- niak (1999). The extension of the tolerance averaging approach is proposed in Section 4 and an example of its application is analyzed in Section 5. 2. Mathematical preliminaries By Ω we shall denote the region in the three-dimensional reference spa- ce parametrized with the Cartesian coordinates x1,x2,x3 and occupied by the solid under consideration. Setting ∆ = (−l1/2, l1/2)× (−l2/2, l2/2)× (−l3/2, l3/2) it is assumed that the solid is ∆-periodic, i.e., lα-periodic in the direction of the xα-axis, α = 1,2,3. Moreover, the diameter l of ∆ is assumed to be very small compared with the smallest characteristic length di- mension of Ω. That is why l will be referred to as themicrostructure length. Denoting ∆(x) := x+ ∆, Ω∆ := {x ∈ Ω, ∆(x) ⊂ Ω}, for an arbitrary integrable function f defined in Ω we shall use the known averaging formula 〈f〉(x)= 1 vol∆ ∫ ∆(x) f(y) dy1dy2dy3 x∈ Ω∆ Let F(Ω) be a set of sufficiently smooth and bounded functions defined in Ω, which are the unknowns in the problem under consideration.Moreover, let ε : F(Ω)∋ ϕ → εϕ ∈ R + be a mapping which assigns to every ϕ ∈F(Ω) a positive real εϕ which will be regarded as the admissible accuracy related to the computation of the values of ϕ or to themeasurements of a physical field (such as a temperature or a temperature gradient) represented by ϕ.We shall write ϕ(x)∼= ϕ(y) iff |ϕ(x)−ϕ(y)| ¬ εϕ and say that the values of ϕ at x and y are in a tolerance. It means that the difference between them can be neglected from the computational viewpoint. The pair T = (F(Ω),ε(·)) will be referred to as the tolerance system and every εϕ = ε(ϕ) as the tolerance parameter assigned to ϕ ∈F(Ω). A sufficiently regular function F ∈ F(Ω) will be called slowly varying, F(·)∈ SV∆(T), if for every x,y from the domain of F the condition x−y∈ ∆ implies F(y)∼= F(x) and if similar conditions hold also for all derivatives of F (including time derivatives provided that F depends also on time). Hence, we deal with function F(·) which is slowly varying in the space not in time. A continuous function ψ ∈ F(Ω) will be termed periodic-like, ψ ∈ PL∆(T), if for every x ∈ Ω∆ there exist a ∆-periodic function ϕx(·) such 426 C.Woźniak et al. that the condition ψ(y) ∼= ψx(y) holds in B(x, l)∩ Ω, where B(x, l) is a ball of the radius l and the center at x. The function ψx is said to be the periodic approximation of ψ in ∆(x). If ψ ∈ PL∆(T) and 〈ρϕ〉(x) = 0 for every x ∈ Ω∆ and for some integrable positive-value function ρ defined on Ω then we shall write ψ ∈ PL ρ ∆(T) and refer ψ to as the oscillating periodic like function. The following assertion can be proved. Assertion. If F ∈ SV∆(T), ϕ ∈ PL∆(T) and ϕx is a ∆-periodic approxi- mation of ϕ in ∆(x) then for every f ∈ L∞per(∆) and h ∈ C 1 per(∆), such that max{h(y) : y ∈ ∆} ¬ l, the following propositions hold for every x∈ Ω∆: (T1) 〈fF〉(x)∼= 〈f〉F(x) for ε = 〈|f|〉εF (T2) 〈fϕ〉(x)∼= 〈fϕx〉(x) for ε = 〈|f|〉εϕ (T3) 〈f∇(hF)〉(x)∼= 〈fF∇h〉(x) for ε = 〈|f|〉(εF + lε∇F) (T4) 〈h∇(fϕ)〉(x)∼=−〈fϕ∇h〉(x) for ε = εG+ lε∇G G = 〈hfϕ〉l−1 where ε is a tolerance parameter which defines the pertinent tolerance ∼= and G,∂αG ∈F(Ω). The proofs of propositions (T1) ÷ (T4) are given in the Appendix. The tolerance averaging (in the sequel denoted by TA), which constitutes the foundation of the proposedmodelling strategy is based, roughly speaking, on replacing the tolerances in formulas (T1) ÷ (T4) by the equalities, i.e., by neglecting the terms O(εF), O(ε∇F), O(εϕ) and O(εG). This special kind of approximation is equivalent to the following assumption. Tolerance Averaging Assumption. In averaging of equations involving slowly varying and periodic-like functions the left-hand sides of formulae (T1) ÷ (T4) will be approximated respectively by their right-hand sides. It has to be emphasized that the functions F(·) and ϕ(·), occurring in (T1)÷ (T4), represent unknownfields in the problemunder consideration; all that is known about these functions (apart from the regularity conditions) is that their values have to be calculated within the tolerances determined re- spectively by certain tolerance parameters εF and εϕ. If these parameters are specified then the criteria of applicability of TA can be verified a posteriori, i.e., after finding the functions F(·) and ϕ(·). These criteria will be written in the simple form F(·) ∈ SV∆(T) and ϕ(·) ∈ PL∆(T), which involves the Tolerance averaging and boundary-layer equations... 427 pertinent tolerance parameters and constitutes the necessary condition for the physical reliability of the obtained solutions. If the tolerance parameters are not specified then they can be calculated after finding F(·) and ϕ(·) from the conditions F(·) ∈ SV∆(T), ϕ(·) ∈ PL∆(T); in this way the accuracy of the obtained solutions can be evaluated. In the tolerance averaging of equations we shall also use the following lemmas: (L0) If F ∈ SV∆(T)∩C 1(Ω) then the estimation l|∂αF| ¬ εF + lε∇F holds (L1) If g ∈ PL∆(T) and g ◦, g̃ ∈ F(Ω) then, for an arbitrary posi- tive valued integrable ∆-periodic function ρ, the decomposition g = g◦ + g̃ exists, where g◦ ∈ SV∆(T), g̃ ∈ PL ρ ∆(T) (L2) If ϕ ∈ PL∆(T), f ∈ L ∞ per(∆) and 〈fϕ〉(·) ∈ F(Ω) then 〈fϕ〉(·) ∈ SV∆(T) (L3) If F ∈ SV∆(T), f ∈ Cper(∆) and (fF)(·) ∈ F(Ω) then (fF)(·)∈ PL∆(T) (L4) If F ∈ SV∆(T), G ∈ SV∆(T) and kF +mG ∈ F(Ω) for some reals k,m, then kF +mG ∈ SV∆(T). The proofs of lemmas (L0) ÷ (L4) are given in the Appendix 3. Tolerance averaging In the linear approximation the heat conduction properties of a solid ma- terial are uniquely described by the second order heat conduction tensor A, which is symmetric and positive definite and by the specific heat scalar c, which is positive. For every periodic solid the functions A = A(·), c = c(·) are defined in Ω and are ∆-periodic (hence they can be also defined in E3) where ∆ is assumed to be the known cell cf Section 2. Let θ = θ(·, t) be the temperature field in Ω at the time t and f = f(·, t) be the known intensity of heat sources. Under the aforementioned denotations the temperature field has to satisfy the well known heat transfer equation in Ω ∇· (A ·∇θ)− cθ̇ = f (3.1) The macroscopic theory of the heat transfer phenomena in micro-periodic solids will be based on the heuristic assumption that in the problems under 428 C.Woźniak et al. consideration the temperature field conforms to the periodic structure of the solid. On the assumption that a certain tolerance system T =(F(Ω),ε(·)) is knownand that θ(·, t)∈F(Ω) for every time t, the above heuristic statement can be written in the following mathematical form. Conformability Assumption (CA). In the modelling of the heat transfer problems in microperiodic solids every temperature field θ(·, t) has to satisfy the condition θ(·, t)∈ PL∆(T) This condition may be violated only near the boundary of a solid. From (CA) and lemma (L1) it follows that there exist the decomposition θ(·, t)= θ◦(·, t)+ϑ(·, t), with θ◦(·, t)∈ SV∆(T) and ϑ(·, t)∈ PL ̺ ∆(T), where either ̺ = c or ̺ =1. It can be shown, cf Woźniak (1999), that under (CA) and (T1) the tolerance averaging of (3.1) yields ∇· [〈A〉 ·∇θ◦(x, t)+ 〈A ·∇ϑ〉(x, t)]−〈c〉θ̇◦(x, t)−〈cϑ̇〉(x, t)= 〈f〉(x) (3.2) where x ∈ Ω∆. At the same time, using (T1), (T2) and (T4), we can prove that the following periodic variational equation for the ∆-periodic function ϑx(y, t), y∈ ∆(x) holds 〈∇ϑ∗ ·A ·∇ϑx〉(x, t)+ 〈ϑ ∗ϑ̇xc〉(x, t)= (3.3) =−〈ϑ∗f〉(x, t)−〈ϑ∗c〉θ̇◦(x, t)−〈∇ϑ∗ ·A〉 ·∇θ◦(x, t) where x∈ Ω∆ and ϑ ∗(·) is a ∆-periodic test function ϑ∗ ∈ H1per(∆); here either 〈ϑ∗〉=0, 〈ϑx〉=0 or 〈cϑ ∗〉=0, 〈cϑx〉=0. Eqs (3.2), (3.3) constitute the fundamentals of the tolerance averaging approach to themodelling of heat transfer problems in micro-periodic solids on the macroscopic level. In order to obtain the model equations we shall look for the approximate solutions to the periodic problems (3.3) in the form ϑx(y, t)= h A(y)V A(x,t) (summation convention over A = 1, ...,N holds), y ∈ V (x), x ∈ Ω∆, where h A(·), A =1, ...,N, are the known V -periodicmode shape functions and VA(·, t)∈ SV∆(T) are the extra unknowns. The aforementioned mode shape functions have to satisfy the condition 〈hA〉= 0 or 〈chA〉= 0, and can be derived as solutions to a certain eigenvalue problem related to (3.3) or are resulting from a periodic discretization of the cell ∆, cfWoźniak (1999). In this way, setting ϑ∗ = hA and applying (T1), (T3), (L2), (L4), after many transformations, we Tolerance averaging and boundary-layer equations... 429 obtain the following system of equations for θ◦, VA, A =1, ...,N ∇· [〈A〉 ·∇θ◦(x, t)+ 〈A ·∇hA〉V A(x, t)]−〈c〉θ̇◦(x, t)= 〈f〉(x, t) (3.4) 〈chAhB〉V̇ A(x, t)+ 〈∇hA ·A ·∇hB〉V B(x, t)+ +〈∇hA ·A〉 ·∇θ◦(x, t)=−〈hAf〉(x, t) At the same time, using (L3) we can prove that the temperature field can be approximated bymeans of the formula θ(x, t)≃ θ◦(x, t)+hA(x)V A(x, t) x∈ Ω∆ (3.5) where theapproximation ≃dependson thenumber N of terms in the formula ϑx(y, t) = h A(y)VA(x, t), y ∈ ∆(x). The solutions to problems described by Eqs (3.4) are physically reliable only if θ◦(·, t)∈ SV∆(T), V A(·, t)∈ SV∆(T), A =1, ...,N, for every time t. Eqs (3.4), (3.5) together with the conditions θ◦(·, t) ∈ SV∆(T) and VA(·, t)∈ SV∆(T), represent the tolerance model of nonstationary heat trans- fer problems in a periodic microheterogeneous solid. Themain features of this model are: 1◦ governing Eqs (3.4) have constant coefficients (which can be calculated after obtaining themode shape functions hA, A =1, ...,N), 2◦ the coefficients 〈chAhB〉 depend on themicrostructure length l, 3◦ the unknowns VA are governed by the system of ordinary differential equations involving only time derivatives of VA. The aforementioned equations have been derived only for every x∈ Ω∆, but from the formal point of view they can be assu- med to hold for every x∈ Ω being deprived of the physical interpretation in Ω\Ω∆. The basic unknowns in (3.4) are: the averaged temperature θ ◦(·) and the functions VA(·). Moreover, there are no boundary conditions for V A(·); that is why the unknowns VA(·) are called the internal thermal variables. 4. Boundary layer equations In order to solve the ”boundary layer” problem formulated in the Intro- duction we shall extend the results of the tolerance averaging, outlined in Section 3, by incorporating an extra term of the boundary layer type into (3.5). To this end denote by n= n(x) the inner unit normal to ∂Ω at the points x belonging to the smooth parts of ∂Ω. Let us introduce in every subregion of Ω, situated near a certain smooth part Σ of ∂Ω, the system 430 C.Woźniak et al. of orthogonal coordinates ξα, α = 1,2,3, such that ξ1 = ξ1(y) is the di- stance between the point y belonging to this subregion and Σ. Let us also assume that the smallest curvature radius of Σ is sufficiently large compared with the microstructure length parameter l. The orthogonal coordinates ξα, α =1,2,3, where ξ1 ­ 0, assigned to an arbitrary but fixed smooth part Σ of ∂Ω will be called the near boundary normal coordinates. They parametrize uniquely points belonging to a certain region of the boundary layer. If the bo- undary ∂Ω is smooth then the whole boundary layer can be parametrized by one near boundary normal coordinate system; otherwise we have to introduce more than one system. In this case it may happen that some from the points belonging to the boundary layer will not be parametrized at all or parame- trized independently by two ormore normal coordinate systems generated by the adjacent smooth parts Σ1, Σ2 of ∂Ω. The ”thickness” of the boundary layer is not specified (cf Sanchez-Palencia and Zaoui, 1985) but has to include the region Ω \Ω∆. Let {g1,g2,n} be the vector basis assigned to an arbitrary point in a re- gion of theboundary layer and related to an arbitrarybutfixednear boundary normal coordinate system. We shall assume that the vector functions g1(·), g2(·), n(·) defined in the region of the boundary layer as well as their first derivatives are slowly varying. This requirement imposes certain conditions on the shape of the boundary ∂Ω; and as we have stated above, the minimum curvature radius of every smooth part Σ of ∂Ω has to be sufficiently large compared with the microstructure length l. Throughout the considerations related to the boundary layer problems we shall use the following denotations ∂ =(n ·∇, 0, 0) ∇=(0, g1·∇, g2·∇) which are assigned to an arbitrary normal coordinate system. Itmeans that if F = F(ξ1,ξ2,ξ3) then ∂F = (∂F ∂ξ1 ,0,0 ) ∇F = ( 0, ∂F ∂ξ2 , ∂F ∂ξ3 ) We shall also denote ∂nF = ∂F/∂ξ 1 and ∂2nF = ∂ 2F/(∂ξ1)2. In the subsequent part of this section the considerations will be restricted to an arbitrarybutfixed region of theboundary layerwhich isparametrized by the known normal coordinate system. Let the temperature field in this region be approximated by θ(x, t)≃ θ◦(x, t)+hA(x)VA(x, t)+ψ(x, t) (4.1) Tolerance averaging and boundary-layer equations... 431 where the right-hand side of formula (3.5) was supplemented by a certain boundary layer term ψ(x, t)= hA(x)Y A(x, t) (4.2) with Y A(x, t) as the extra unknowns which are assumed to decay in the direction of n(x), x∈ ∂Ω, i.e. they decay while passing from the boundary ∂Ω to the bulk region of Ω. Hence the functions Y A(·, t) and their derivatives ∂Y A(·, t) are not slowly varying functions. At the same time the gradients ∇Y A(·, t) are assumed to be slowly varying. We shall also assume that Y A depend on the microstructure length parameter l such that ∂2nY A ∈ O(l−2), ∂nY A ∈ O(l−1), Y A ∈ O(1) and hence the functions Y A, l∂nY A, l2∂2nY A are of O(1) order. Using (4.1) and (4.2) we have to replace the sum hAVA in (3.5) by hAVA+hAY A. Denoting ĥA = l−1hA and setting I =∇· (〈A ·∇hAY A〉+ l〈ĥAA ·∇Y A〉)− l〈cĥAẎ A〉 (4.3) IA = l〈ĥA∇· (A ·∇hBY B + lĥBA ·∇YB)〉− l2〈cĥAĥBẎ B〉 after applying the procedure described by Woźniak (1999) and leading from Eqs (3.2), (3.3) to Eqs (3.4), we obtain ∇· (〈A〉 ·∇θ◦+ 〈A ·∇hA〉V A)−〈c〉θ̇◦+ I = 〈f〉 (4.4) 〈∇hA ·A ·∇hB〉V B + 〈∇hA ·A〉 ·∇θ◦+ 〈chAhB〉V̇ B + 〈fhA〉+ IA =0 Eqs (4.4) have the form similar to that of Eqs (3.4) but incorporate the extra boundary-layer type terms I, IA. These terms can be neglected in the bulk region of Ω situated outside the boundary layer. To satisfy this requirement we shall introduce into considerations the formal asymptotic assumption that IA → 0 together with l → 0. Bearing inmind that the functions Y A, l∂nY A, l2∂2nY A are of O(1) order and under the limit passage l → 0 they behave like slowly varying functions, from the above asymptotic assumption we obtain l2〈ĥAĥBn ·A ·n〉∂2nY B − l〈(ĥB∇hA− ĥA∇hB) ·A ·n〉∂nY B − (4.5) −〈∇hA ·A ·∇hB〉Y B =0 The bounded solutions Y A(·, t) to Eqs (4.5) decay across the boundary layer and will be treated, together with their derivatives, as negligibly small in the bulk region of Ω. That is why Eqs (4.5) will be referred to as the boundary layer equations. For the same reason also the terms I, IA in Eqs (4.4) can be neglected in the bulk region. Let us also observe that if l tends to zero 432 C.Woźniak et al. then the boundary layer ”thickness” also tends to zero; i.e. the bulk region can be ”approximated” by the region Ω. Hence the approximate description of the boundary-layer problem proposed in this contribution is based on the formal asymptotic assumption that l → 0 implies IA → 0, leading to the boundary layer Eqs (4.5). The above assumption can be also supplemented by the heuristic statement that the terms I, IA are neglected in the bulk region and the boundary layer ”thickness” is negligibly small when compared with the length dimensions of Ω. Hence the proposed statement means, roughly speaking, that Eqs (4.4) can be approximated by Eqs (3.4). This simplified approach to theboundary layer phenomena is describedbyEqs (3.4) and (4.5). However, more general analysis of the boundary layer phenomena can be also carried out on the basis of Eqs (4.4), (4.5). Summarizing the obtained results we conclude that the tolerancemodel of the nonstationary heat transfer problems under consideration is described by Eqs (3.4) (or bymore general Eqs (4.4)), by the boundary layer equation (4.5) and by formulae (4.1), (4.2) for the temperature field. This model makes it possible to satisfy the boundary conditions for the temperature field, given by θ(x, t)= θ̃(x, t), x∈ ∂Ω, (as well as other kinds of the boundary conditions) provided that the boundary temperature θ̃(x, t) can be approximated by θ◦(x, t)≃ θ̃◦(x, t)+hA(x)θ̃A(x, t) x∈ ∂Ω (4.6) where the functions θ̃◦(·, t), θ̃A(·, t) are known.Hence theboundaryconditions for the temperature are θ◦(x, t)= θ̃◦(x, t) (4.7) VA(x, t)+Y A(x, t)= θ̃A(x, t) x∈ ∂Ω where Y A(·, t), A =1, ...,N, is a bounded solution to the boundary layer Eqs (4.5). It should be remembered that usingEqs (3.4) and formulae (3.5) we are able to satisfy the boundary conditions only for the averaged part θ◦ of the temperature field by assuming that θ◦(x, t)= θ̃◦(x, t), x∈ ∂Ω. It has to be emphasized that the approach to the boundary layer pheno- mena outlined in this section can be treated only as a certain first approxi- mation of the boundary layer theory. The main advantage of the proposed simplified approach is that the system of the boundary layer Eqs (4.5) is not coupled with the governing Eqs (3.4) of the tolerance model of heat trans- fer in micro-periodic composites. A more general approach to the boundary layer problems can be based on the asymptotic analysis of this problem, Pa- nasenko (1994). Tolerance averaging and boundary-layer equations... 433 5. Example problem In order to illustrate themodelEqs (3.4), (4.5) and formulae (4.1), (4.2) let us take into account the nonstationary problem of heat transfer in the infinite microheterogeneous half-space x1 ­ 0 subjected on its boundary x1 = 0 to the time dependent temperature oscillations Acosωt, t ∈ R, with the known amplitude A and frequency ω. For the sake of simplicity let us confine ourselves to the simplest case of the tolerance model in which N =1, taking intoaccount onlyonemode shape function h = h1(x),x=(x1,x2,x3), related to the smallest eigenvalue λ of the periodic eigenvalue problem related to Eqs (3.3), cfWoźniak (1999). Denoting by n the versor of the x1-axis and setting h = hl−1 we shall introduce the coefficients a0 = 〈n ·A ·n〉= 〈A11〉 a1 = 〈n ·A ·∇h〉= 〈A1αh,α 〉 a2 = 〈∇h ·A ·∇h〉= 〈Aαβh,α h,β 〉 c0 = 〈c〉 c2 = 〈c(h) 2〉 α = 〈n ·A ·n(h)2〉= 〈A11(h) 2〉 k = a2 α where the subscripts α,β runover 1,2,3. Let us also define V = V 1,Y = Y 1. Under the aforementioned denotations Eqs (3.4) reduce to the form a0θ ◦,11+a1V,1−c0θ̇ ◦ =0 (5.1) l2c2V̇ +a1θ ◦,1+a2V =0 the boundary layer equation (4.5) is given by l2αY,11−a2Y =0 (5.2) and formulae (4.1), (4.2) yield θ(x, t)≃ θ◦(x1, t)+h(x)[V (x1, t)+Y (x1, t)] (5.3) where x=(x1,x2,x3), x1 ­ 0, (x2,x3)∈ R 2, t ∈ R. Thus, the problemunder consideration consists its finding the functions θ◦,V ,Y on the right-hand side of Eq (5.3) satisfying Eqs (5.1), (5.2) for x1 > 0 and the following boundary conditions for x1 =0 θ◦(0, t)= Acosωt V (0, t)+Y (0, t)= 0 (5.4) 434 C.Woźniak et al. To solve this problem we look for a solution to Eqs (5.1) in the complex form θ◦(x1, t)= Ae γx1+iωt V (x1, t)= Be γx1+iωt (5.5) with the real constants A, B, γ, ω. Substituting this solution into (5.1) we arrive at the system of homogeneous linear algebraic equations for A and B. Thedeterminant of this systemhas to be equal to zerowhichmakes it possible to determine the constant γ. Under the extra denotation a◦ = a0− (a1) 2 a2 we obtain γ2 = ωc0(−l 2c2ω+ia2) a2a ◦+il2a0c2ω (5.6) It can be proved that the constant a◦ is positive and can be interpreted as the first approximation (because in (4.1), (4.2) we have assumed N = 1) of the effective heat transfermodulus in the direction of the x1-axis. The square roots of (5.6) will be written down in the general form γ(ω)=±[γRe(ω)+ iγIm(ω)] (5.7) where γRe and γIm are assumed to be negative and determined by rather complicated formulae which are not given here. However, bearing in mind that l2 in formula (5.6) can be interpreted as a small parameter, we can obtain the following asymptotic expansion for γRe and γIm γRe(ω)=− √ c0ω 2a◦ [ 1+ l2 c2 2a0 (a1 a2 )2 ω ] +O(l4) (5.8) γIm(ω)=− √ c0ω 2a◦ [ 1− l2 c2 2a0 (a1 a2 )2 ω ] +O(l4) At the same time we obtain the interrelation B = ϕ(ω)A between the real constants A and B, where ϕ(ω) =− a1a2γRe(ω) (a2)2+ l4(c2)2ω2 (5.9) The asymptotic expansion for ϕ(ω) is given by ϕ(ω)= a1 a2 √ c0ω 2a◦ [ 1+ l2 c2 2a0 (a1 a2 )2 ω ] +O(l4) (5.10) Tolerance averaging and boundary-layer equations... 435 Now, we shall pass to the boundary layer equation (5.2).We shall look for solutions to Eq (5.2) in the form Y (x1, t) = Y (x1)e iωt, where Y satisfies the equation Y ,11− (k l )2 Y =0 Hence, the real part of Y will be given by Y (x1, t)= Bcos(ωt)exp ( − k l x1 ) (5.11) where B is an arbitrary real constant. It can be seen that the real part of θ◦(x1, t) derived from formula (5.5) satisfies the first from the boundary conditions (5.4). Taking into account formulae (5.5), (5.11) it can be seen that the real part of V (x1, t) together with Y (x1, t) satisfy the second from the boundary conditions (5.4) provided that B = −B. It follows that B = −ϕ(ω)A. Summarizing all the derived results, bymeans of (5.3) we obtain the solution to the problem under consideration in the form θ(x, t)≃ A { eγRe(ω)x1 cos(γIm(ω)x1+ωt)+ (5.12) +h(x)ϕ(ω) [ eγRe(ω)x1 cos(γIm(ω)x1+ωt)− e −k l x1 cos(ωt) ]} where x = (x1,x2,x3), x1 ­ 0, (x2,x3) ∈ R 2 and t ∈ R. In the first approximation the values of γRe(ω), γIm(ω) and ϕ(ω) can be calculated from (5.8) and (5.10), respectively, after neglecting the terms O(l4). This approximation can be applied only if the second term in the square brackets of formulae (5.8) and (5.10) is much smaller than 1. Now, let us compare the obtained result with that which can be derived usingthehomogenizedmodel of theproblemunderconsideration.Thesimplest form of this model (in which N =1, i.e. if we deal with only onemode shape function h) can be directly derived from the governing Eqs (5.1) and (5.2) by neglecting in the second equation from (5.1) the term l2c2V̇ involving the microstructure length parameter l. In this case, instead of (5.1) we obtain a◦θ◦,11−c0θ̇ ◦ =0 V =− a1 a2 θ◦,1 Using the general homogenization procedurewe have to calculate a◦. In both cases we shall look for the solutions of the form given only by the first from Eqs (5.5). Taking into account boundary layer equation (5.2), after simple 436 C.Woźniak et al. calculations we arrive at formula (5.12), in which γRe(ω)= γIm(ω)=− √ c0ω 2a◦ (5.13) ϕ(ω) = a1 a2 √ c0ω 2a◦ It can be seen that this result is also implied by the formal limit passage l → 0 in formulae (5.8) and (5.10). Hence the important conclusion that the homogenized model of the problem we deal with can be applied only if the second term in the square brackets in formulae (5.8) and (5.10) is negligibly small compared with 1. This situation takes place if (a1 a2 )2 ω ≪ 2a0 l2c2 Roughly speaking, the homogenizedmodel of the problemunder consideration can be applied only if the frequencies ω of the temperature oscillations are not too big; otherwise we have to use the tolerance averaging model detailed in this paper. For a homogeneous solid a1 = 0, a ◦ = a0 = A11, c0 = c; hence V = 0, Y =0, ϕ =0 and (5.12) reduces to the known formula for the temperature θ. Appendix In this Appendix the proofs of propositions (L0)÷ (L4) and (T1)÷ (T4), which constitute themathematical background of the tolerance averaging ap- proach are given. Lemma (L0). If F ∈ SV∆(T)∩C 1(Ω) then the estimation l|∂αF| ¬ εF + lε∇F holds. Proof. Recall that the smooth function F(·) is called slowly varying, F ∈ SV∆(T), if ∀x,y∈ Ω [(‖x−y‖¬ l =⇒|DF(x)−DF(y)| ¬ εDF ] for every DF ∈F(Ω) where DF stands for F and for an arbitrary partial derivative ∂αF which belongs to F(Ω). Hence ∀x,y∈ Ω [(‖x−y‖¬ l =⇒|F(x)−F(y)| ¬ εF ] Tolerance averaging and boundary-layer equations... 437 and ∀x,y∈ Ω [(‖x−y‖ < l =⇒|∂αF(x)−∂αF(y)| ¬ ε∇F ] If F ∈ C1(Ω) then for every vector h such that |h| ¬ l and for y+νh∈ Ω, ν ∈ (0,1], and for some η ∈ [0,1] we obtain ∇F(y+ηh) ·h= F(y+h)−F(y) y∈ Ω Moreover, if F ∈ SV∆(T) then |∇F(y) ·h|= |∇F(y+ηh) ·h+[∇F(y)−∇F(y+ηh)] ·h| ¬ ¬ |∇F(y+ηh) ·h|+ |[∇F(y)−∇F(y+ηh)] ·h| ¬ ¬ |F(y+h)−F(y)|+ |[∇F(y)−∇F(y+ηh)] ·h| ¬ εF +ε∇F |h| Let h = el, where e is an arbitrary unit vector, ‖e‖ = 1. Then the above estimate yields |∇F(y) ·e|l ¬ εF + lε∇F For e = eα, where eα is the versor of the xα-axis, α = 1,2,3, we obtain ∇F(y) ·eα = ∂αF(y) and finally l|∂αF| ¬ εF + lε∇F which was to be proved. Lemma (L1). If g ∈ PL∆(T) and g ◦, g̃ ∈ F(Ω) then, for an arbitra- ry positive valued integrable ∆-periodic function ρ, the decomposition g = g◦ + g̃ exists, where g◦ ∈ SV∆(T), g̃ ∈ PL ρ ∆(T). Proof. Setting g◦(·)= 〈̺〉−1〈̺g〉(·) bymeans of (̺g)(·)∈ PL∆(T) and after using (L1) we obtain 〈̺g〉(·)∈ SV∆(T) Hence g◦(·) is a slowly varying function and the condition g̃(y)= g(y)−g◦(y)∼= g(y)−g ◦(x) y∈ B(x, l)∩Ω holds true for an arbitrary but fixed x ∈ Ω. It follows that g̃(·) is also a periodic-like function.Moreover, bymeans of proposition (T1) (see below) we obtain 〈̺g̃〉= 〈̺g〉−〈̺g◦〉∼= 〈̺g〉−〈̺〉g ◦ 438 C.Woźniak et al. by the definition of g◦ the term 〈̺g〉 − 〈̺〉g◦ is equal to zero and hence 〈̺g̃〉=0.Thuswe jump to the conclusion that g̃ is an oscillating periodic like function (with the weight ̺) g̃ ∈ PL ρ ∆ (T), which ends the proof of (L1). Lemma (L2). If ϕ ∈ PL∆(T), f ∈ L ∞ per(∆) and 〈fϕ〉(·) ∈ F(Ω) then 〈fϕ〉(·)∈ SV∆(T). Proof. If ϕ ∈ PL∆(T) then for every y1, y2 such that ‖y1 −y2‖ ¬ l and for x=(y1+y2)/2 we obtain |〈ϕf〉(y1)−〈ϕxf〉(y1)| ¬ 〈|f|〉εϕ |〈ϕf〉(y2)−〈ϕxf〉(y2)| ¬ 〈|f|〉εϕ where ϕx(·) is a periodic approximation of ϕ(·) on ∆(x). Because of 〈ϕxf〉(y1)= 〈ϕxf〉(y2)= const we conclude that the condition |〈ϕf〉(y1)−〈ϕf〉(y2)| ¬ 2〈|f|〉εϕ holds for every y1,y2 ∈ Ω such that ‖y1−y2‖¬ l. It follows that 〈ϕf〉(·)∈ SV∆(T) with ε〈ϕf〉 =2〈|f|〉εϕ, which ends the proof of (L2). Lemma (L3). If F ∈ SV∆(T), f ∈ Cper(∆) and (fF)(·) ∈ F(Ω) then (fF)(·)∈ PL∆(T). Proof. If F(·) is a slowly varying function defined in Ω then for every y∈ B(x, l)∩Ω and every x∈ Ω |f(y)F(y)−f(y)F(x)|= |f(y)||F(y)−F(x)| ¬ |f(y)|εF Bearing in mind that f ∈ Cper(∆) and setting εfF = εF max{|f(y)| : y∈ ∆} we obtain |f(y)F(y)−f(y)F(x)| ¬ εfF y∈ B(x, l)∩Ω It follows that f(y)F(x),y ∈ B(x, l)∩Ω can be treated as a certain periodic approximation of the function (fF)(·) on ∆(x), i.e., f(y)F(x)= (fF)x(y), y∈ B(x, l)∩Ω; hence (fF)(·) is a periodic-like function. This ends the proof of (L3). Tolerance averaging and boundary-layer equations... 439 Lemma (L4). If F ∈ SV∆(T), G ∈ SV∆(T) and kF + mG ∈ F(Ω) for some reals k,m, then kF +mG ∈ SV∆(T). Proof. It is easy to see that |(kF +mG)(x)− (kF +mG)(y)|= |k[F(x)−F(y)]+m[G(x)−G(y)]| ¬ ¬ |k||F(x)−F(y)|+ |m||G(x)−G(y)| holds for every x,y ∈ Ω. If ‖x−y‖ ¬ l then by means of F,G ∈ SV∆(T) we obtain |(kF +mG)(x)− (kF +mG)(y)| ¬ |k|εF + |m|εG If kF(·)+mG(·)∈F(Ω) then setting εkF+mG = |k|εF + |m|εG we conclude that kF(·)+mG(·) is a slowly varying function. This ends the proof of (L4). Assertion. If F ∈ SV∆(T), ϕ ∈ PL∆(T) and ϕx is a ∆-periodic approxi- mation of ϕ on ∆(x) then for every f ∈ L∞per(∆) and h ∈ C 1 per(∆), such that max{|h(y)| : y ∈ ∆}¬ l, the following propositions hold for every x∈ Ω∆: (T1) 〈fF〉(x)∼= 〈f〉F(x) for ε = 〈|f|〉εF (T2) 〈fϕ〉(x)∼= 〈fϕx〉(x) for ε = 〈|f|〉εϕ (T3) 〈f∇(hF)〉(x)∼= 〈fF∇h〉(x) for ε = 〈|f|〉(εF + lε∇F) (T4) 〈h∇(fϕ)〉(x)∼=−〈fϕ∇h〉(x) for ε = εG+ lε∇G G = 〈hfϕ〉l−1 where ε is the tolerance parameter which defines the pertinent tolerance ∼= and G,∂αG ∈F(Ω). Proof (T1) If F ∈ SV∆(T) and f ∈ L ∞ per(∆) then for every x∈ Ω∆ we obtain |〈fF〉(x)−〈f〉F(x)|= = 1 |∆| ∣∣∣ ∫ ∆ [f(x+y)F(x+y)−f(x+y)F(x)] d∆(y) ∣∣∣¬ ¬ 1 |∆| ∫ ∆ |f(x+y)[F(x+y)−F(x)]| d∆(y) 440 C.Woźniak et al. Bymeans of F ∈ SV∆(T), for every y∈ ∆ and for every x∈ Ω∆ |F(x+y)−F(x)| ¬ εF Hence |〈fF〉(x)−〈f〉F(x)| ¬ 〈|f|〉εF for every x∈ Ω∆. This ends the proof of (T1) (T2) If f ∈ L∞per(∆), ϕ ∈ PL∆(T) and ϕx is a certain ∆-periodic approximation of ϕ on ∆(x), x∈ Ω∆, then we obtain |〈fϕ〉(x)−〈fϕx〉(x)|= = 1 |∆| ∣∣∣ ∫ ∆ [f(x+y)ϕ(x+y)−f(x+y)ϕx(x+y)] d∆(y) ∣∣∣¬ ¬ 1 |∆| ∫ ∆ |f(x+y)[ϕ(x+y)−ϕx(x+y)]| d∆(y) By means of ϕ ∈ PL∆(T) we obtain |ϕ(x+y)−ϕx(x+y)| ¬ εϕ for every x∈ Ω∆ and every y∈ ∆. Hence |〈fϕ〉(x)−〈fϕx〉(x)| ¬ 〈|f|〉εϕ which ends the proof of (T2). (T3) If F ∈ SV∆(T), f ∈ L ∞ per(∆) and h ∈ C 1 per(∆), such that max{|h(y)| : y ∈ ∆} ¬ l then, bearing in mind that l|∂αF| ¬ εF + lε∇F (see (L0)), for every x∈ Ω∆ we obtain |〈f∂α(hF)〉(x)−〈fF∂αh〉(x)|= = 1 |∆| ∣∣∣ ∫ ∆ [f(x+y)∂α(hF)(x+y)−f(x+y)(F∂αh)(x+y)] d∆(y) ∣∣∣¬ ¬ 1 |∆| ∫ ∆ |f(x+y)[∂α(hF)(x+y)− (F∂αh)(x+y)]| d∆(y)= = 1 |∆| ∫ ∆ |(fh∂αF(x+y)| d∆(y)¬〈|f|〉(εF + lε∇F) This shows that |〈f∂α(hF)〉(x)−〈fF∂αh〉(x)| ¬ 〈|f|〉(εF + lε∇F) α =1,2,3 Tolerance averaging and boundary-layer equations... 441 which ends the proof of (T3). (T4) The difference between both sides of (T4) is equal to |〈h∂α(fϕ)〉(x)+ 〈fϕ∂αh〉(x)|= |〈∂α(fϕh)〉(x)| Setting G = 〈hfϕ〉l−1 andusing (L2)we conclude that G ∈ SV∆(T). Bearing in mind that l|∂αG| ¬ εG+ lε∇G (see (L0)) we obtain |〈h∂α(fϕ)〉(x)+ 〈fϕ∂αh〉(x)|= l|∂αG(x)| ¬ εG+ lε∇G α =1,2,3 which ends the proof of (T4). References 1. Bensoussan A., Lions J.L., Papanicolau G., 1978, Asymptotic Analysis for Periodic Structures, North-Holland, Amsterdam 2. Jikov V.V., Kozlov C.M., Oleinik O.A., 1994, Homogenization of Diffe- rential Operators and Integral Functionals, Springer, Berlin-Heidelberg 3. Panasenko G.P., 1994, Asymptotic Analysis of Bar Systems,Russian Journ. of Math. Phys., 2, 325-331 4. Sanchez-Palencia E., Zaoui A., 1985, Homogenenization Techniques for Composite Media, Lecture Notes in Physics, 272, Springer Verlag 5. Woźniak C., 1999, AModel for Analysis ofMicro-Heterogeneous Solids,Me- chanik Berichte, Nr.1, IAM, RWTH, Aachen 6. Woźniak C., 2000, ComputationalModels of Periodic Composites. Tolerance Averaging Versus Homogenization, J. Theor. Appl. Mech., 38, 447-459 Uśrednianie tolerancyjne i równania warstwy brzegowej dla przewodnictwa ciepła w ośrodku mikroperiodycznym Streszczenie Modelemakroskopoweprzewodnictwaciepławośrodkumikroperiodycznym, sfor- mułowanenadrodzeuśredniania tolerancyjnego (modele tolerancyjne), są reprezento- wane równaniem różniczkowym cząstkowymdla uśrednionego pola temperatury oraz równaniami rózniczkowymi zwyczajnymi zawierającymi pochodne czasowe pewnych 442 C.Woźniak et al. dodatkowych niewiadomych, zwanych termicznymi zmiennymi wewnętrznymi, Woź- niak (2000). Niewiadome te opisują zaburzenia pola temperatury spowodowane pe- riodycznąmikroniejednorodnością ośrodka.Tym samymwramachmodeli tolerancyj- nych, warunki brzegowemogą być określone tylko dla uśrednionego pola temperatu- ry. Celemopracowania jest pokazanie, jakmetoda tolerancyjnego uśredniania równań może zostać rozszerzona w celu zapewnienia dokładniejszego spełnienia warunków brzegowych. Manuscript received October 23, 2000; accepted for print January 4, 2001