Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 41, 1, pp. 75-88, Warsaw 2003 ON THE ASYMPTOTIC BEHAVIOR OF THERMAL TAYLOR DISPERSION PROCESSES IN PERIODIC MEDIA Claudia Timofte Department of Mathematics, Faculty of Physics, University of Bucharest, Romania e-mail: claudiatimofte@hotmail.com Using thegeneralizedmethodofmomentsandacentral limit theorem,we shall describe a large class of thermal dispersion phenomena occurring in some macrohomogeneous systems. We shall be interested in computing the macroscale coefficients in terms of the microscale coefficients and the system geometry. Also, the functional dependence of the effective coefficients on the velocity and the spatial scale parameters is analyzed. Key words: macrotransport equation, macroscale coefficients, scale parameters 1. Introduction The field of macrotransport processes constitutes a natural extension of classical Taylor dispersion theory for unidirectional, rectilinear flows (see Tay- lor, 1953) to a large class of flow and dispersion problems. It is well-known by now that G.I. Taylor used rather intuitive semi-analytical arguments to prove the Fickian nature of the mean axial dispersion of a diffusing solute injected into aviscous fluidflowing througha circular cylindrical tube.He showed that, asymptotically, in the long-time limit, suchadispersionprocess is describedby a one-dimensional convective-diffusive equation and the dispersion coefficient characterizing this axialmacrotransport equation governing thecross-sectional mean solute concentration is D∗ =D+ a2U 2 48D (1.1) 76 C.Timofte where a is the radius of the tube, U is the mean velocity of flow and D is the coefficient of molecular diffusion. This new coefficient combines the mi- croscopically distinct effects of radial molecular diffusion and axial convective solute flow. Dispersion is caused by the radial inhomogeneity of the Poiseuil- le velocity field which interacts with the lateral diffusion of solute molecules. In fact, all macrotransport processes combine such a Brownian (stochastic) diffusive transportmechanismwith an inhomogeneous, convective (determini- stic) transport mechanism. The stochastic dispersion is assumed to act over a period of time large enough to allow the sampling of all such velocities by molecular diffusion across the streamlines of the flow. Taylor‘s technique, later extended to the case of solute dispersion in turbu- lent flows, provided thefirst framework for the generic phenomenon referred to in the literature as ”Taylor dispersion”. In 1956, Aris (see Aris, 1956) extends these results for cylinders of noncircular cross-section and develops a rigorous theory, based on a method of moments scheme. Also, he analyzes the effects of time-periodic convection on dispersion (see Aris, 1960). In 1971 Horn (see Horn, 1971) made another important step to put the foundations of the so- called generalized Taylor dispersion theory. He extended the classical theory to multidimensional phase spaces. A decisive step was done then by Brenner (see Brenner, 1980) who developed a paradigmatic dispersion theory for very general and complex systems. In 1993, recognizing the analogy existing be- tween the mass transport and other modes of transport phenomena, Brenner and Edwards extended this theory to non-material transport processes (see Brenner and Edwards, 1993). Hence, based upon a rigorous description of microtransport processes oc- curring in heterogeneous systems,macrotransport theory, alternatively known in the literature as thegeneralizedTaylor dispersion theory, allowsus todescri- be a large class of material and nonmaterial dispersive phenomena occurring in macrohomogeneous systems. Applications of themacrotransport theory are presently recognized inma- ny fields of scientific and engineering research. Various other methods have been developed for obtaining the macroscale behavior and properties of so- me heterogeneous complex systems. These include homogenization techniques (see Bensoussan et al., 1978, Sanchez-Palencia, 1980), statistical and volume- averaging methods (see Koch and Brady, 1985) and probabilistic methods based on central limit theorems (see Bhattacharya et al., 1989). In this paper, we shall be especially interested in getting a macrotran- sport paradigm for a class of thermal transport phenomena occurring in some complex multidimensional adiabatic systems. We shall deduce simple macro- On the asymptotic behavior of thermal... 77 transport equations (and effective coefficients appearing therein) which apply, for long times, at a coarse-grained level of description of our systems. Our analysis is based on two alternative methods: the above mentioned generalizedmethod ofmoments and a probabilisticmethod based on a central limit theorem forMarkov processes. We are also interested in getting the asymptotic behavior of the macrodi- spersion coefficients as functions of the velocity and the spatial scale parame- ters which characterize our transport processes. Specific examples are given in Section 3 to illustrate the computation of these macrotransport coefficients as functions of the prescribed microscale data. 2. A macrotransport paradigm for thermal dispersion phenomena in adiabatic systems Let us introduce two distinctly different classes of independent coordinate variables which characterize a generic microtransport process. These will be designated as global and local variables and denoted by Q and q Q= [Q1,Q2, ...,Qr] q= [q1,q2, ...,qs] (2.1) Together, the vectors (Q,q) define amultidimensional phase space Q ∞ ⊕ q0 within which our transport processes occur. The global subspace Q∞, representing the domain of the values taken by Q, will be unbounded, while the subspace q0 (q∈ q0)will, generally, bebounded.Theglobal coordinate Q properly corresponds to a long-time scale, while the local variable corresponds to a short-time scale. They are also called slow and, respectively, fast variables (see Sanchez-Palencia, 1980). Thegenericmicrotransport equation governing the evolution of the tempe- raturefield T =T(Q,q, t) in continuous adiabatic systemsmaybe represented as ρCp ∂T ∂t +∇Q ·J+∇q ·j =0 (2.2) where the constitutive equations for the global and the local internal energy flux-density vectors are J = ρ(q)Cp(q)U(q)T −KT(q) ·∇Q(T) (2.3) j = ρ(q)Cp(q)u(q)T −kT(q) ·∇q(T) 78 C.Timofte Here, (KT ,kT) denote the global and the local-space thermal conductivities and (U,u) the comparable velocity vectors. Here, ρ and Cp are positive functions and KT and kT are positive definite tensors. This system of equations is subjected to the following conditions n ·j =0 n ·KT ·∇qT =0 on ∂q0 |Q|m{T,J}→{0,0} as |Q|→∞ ∀(q, t­ 0) m=0,1, ... (2.4) T(Q,q,0)=T0(Q,q) with the right-hand side a prescribed function. Note that the energy dissipation and kinetic energy contribution are ne- glected in the microtransport equation. We shall limit ourselves to the class of problems for which n ·u = 0 on ∂q0 and u ·∇qCp = 0. Also, it will be supposed that the thermal properties are everywhere nonnegative definite. It proves useful to reformulate linear microscale problem (2.2) in terms of a Green’s function. In this context, let us define a quantity P = P(Q,q, t |Q′,q′, t′) such that ρCpP to be interpreted as the conditio- nal probability density of having the temperature T(Q,q, t) at the position (Q,q) at the moment t if we had the temperature T(Q′,q′, t′) at (Q′,q′) at the earlier moment t′. P will generally depend only on the differences Q−Q′ and t− t′ and so, choosing Q′ = 0 and t′ =0, we can consider, without loss of generality, that P =P(Q,q, t |q′). Since we are modeling the transport of conserved entities, we have ∫ Q ∞ ∫ q0 ρCpP dqdQ=1 t­ 0 and P =0 t< 0 Also, the relationship between the temperaturefield T and theGreen function P is T(Q,q, t)= ∫ Q′ ∞ ∫ q′ 0 ρ(q′)Cp(q ′)P(Q,q, t |q′)T(Q′,q′,0) dq′dQ′ (2.5) and the microtransport equation of energy dispersion in continuous systems may be represented as ρCp ∂P ∂t +∇Q ·JP +∇q ·jP = δ(Q)δ(q−q ′)δ(t) On the asymptotic behavior of thermal... 79 where JP = ρ(q)Cp(q)U(q)P −KT(q) ·∇Q(P) jP = ρ(q)Cp(q)u(q)P −kT(q) ·∇q(P) Defining the macroscale Green’s function P(Q, t |q′)= 1 ρCp ∗ ∫ q0 ρ(q)Cp(q)P(Q,q, t |q ′) dq (2.6) this will become asymptotically independent of q′ P(Q, t |q′)∼=P(Q, t) and, hence, following a moment analysis (see Brenner and Edwards, 1993; Timofte, 1996), we are led to the following macrotransport equation ρCp ∗ (∂P ∂t +U ∗ ·∇QP ) = kT ∗ :∇Q∇QP + δ(Q)δ(t) (2.7) subjected to the conditions P =0 t< 0 P → 0 when |Q|→∞ (2.8) Here, the macroscale coefficients ρCp ∗ and U ∗ are given by ρCp ∗ = 1 τ0 ∫ q0 ρCp dq (2.9) where τ0 = ∫ q0 dq (2.10) and by U ∗ = ∫ q0 ρCpP ∞ 0 U dq (2.11) The effective thermal conductivity dyadic has the expression: kT ∗ = k M +k C (2.12) 80 C.Timofte where k M = 1 τ0 sim ∫ q0 KT dq (2.13) is the ”molecular” contribution and k C = sim ∫ q0 [( P∞0 − 1 τ0ρCp ∗ ) KT +ρCpP ∞ 0 B(U −U ∗ ) ] dq (2.14) is the convective contribution. These phenomenological coefficients are to be obtained after solving the associated local problems for P∞0 (q) and B(q) ∇q ·j ∞ 0 =0 j∞0 = ρCpuP ∞ 0 −kT ·∇qP ∞ 0 (2.15) n ·KT ·∇qP ∞ 0 =0 on ∂q0 ∫ q0 ρCpP ∞ 0 dq=1 and j∞0 ·∇qB−∇q · (P ∞ 0 kT ·∇qB)= ρCpP ∞ 0 (U −U ∗ ) (2.16) ρCpP ∞ 0 n ·kT ·∇qB=0 on ∂q0 More, we shall require that P∞0 is nonnegative for all q ∈ q0. Also, P ∞ 0 and B must be single-valued for all q∈ q0. So, we can express the macrotransport coefficients ρCp ∗ , U ∗ and kT ∗ in terms of the prescribed microscale data and the system geometry. It is worthwhile to notice that, in fact, ρCp, which is inhomogeneous in q, acts like a biasing potential. This causes a redistribution of the internal energy in the local space, which is finally reflected in the magnitude of the macrotransport coefficients U ∗ and kT ∗ . The coarse-grained macroscale temperature field T(Q, t)= 1 τ0 ∫ Q′ ∞ ∫ q ′ 0 ρ(q′)Cp(q ′)P(Q−Q′, t |q′)T(Q′,q′,0) dq′dQ′ (2.17) On the asymptotic behavior of thermal... 81 will satisfy themacrotransport equation ρCp ∗ (∂T ∂t +U ∗ ·∇QT ) = kT ∗ :∇Q∇QT (2.18) subjected to appropriate initial and boundary conditions (see Brenner and Edwards, 1993; Timofte, 1996). A similar analysis can be done for the problem of thermal dispersion in discontinuous adiabatic systems. Thegeometrical structureof thediscontinuousmediumis idealizedasbeing spatially periodic (porousmedia, compositematerials, laminatedmedia). The periodic medium is represented as a spatially periodic array in R3, composed of topologically indistinguishable unit cells of periodicity, having the same shape, orientation, volume and ”content”. If we denote by τ0 the volume of such an elementary cell and by τp the volume of the solid part, we have τp = τ0− τf where τf is the interstitial fluid volume within such a cell. Arbitrarily designating one of the elementary cells as being the zeroth cell, it is convenient tomeasure the position vector R of any point in space relative to the centroid of this cell. Denoting by Rn the position vector of the centroid of the n-th cell relative to the centroid of the zeroth cell and by r ∈ τ0{n} the local position vector for anypointwithin the n-th cell relative to an origin at its center, we have R=Rn+r (2.19) If we suppose that at the particle-fluid interface Sp we have a local equ- ilibrium described by a linear partitioning relationship, using similar nota- tions as in the continuous case and introducing microscale Green’s function P = P(Rn,r, t |r ′), this will obey the following system of cellular-level equ- ations (see Brenner and Edwards, 1993; Timofte, 1996) ρ(r)Cp(r) ∂P ∂t +∇·J = δnn′δ(r−r ′)δ(t) J = ρ(r)Cp(r)U(r)P −KT(r) ·∇P (2.20) ν ·∆SpJ =0 on Sp |Rn−Rn′| mP → 0 |Rn−Rn′| mJ → 0 } as { {n−n′}→∞ m=0,1, ... 82 C.Timofte In (2.20)3, for an arbitrary tensor field f, ∆Sp defines the ”jump” of f, across the discontinuous phase surface Sp and ν is a unit vector which is normal to this surface. The thermophysical properties ρ, Cp, KT and the local fluid velocity are regarded as being spatially periodic. Following amoment-matching scheme and consideringmacroscale Green’s function P P(Rn, t |r ′)= 1 ρCp ∗ ∫ τ0 ρ(r)Cp(r)P(Rn,r,t |r ′) d3r∼=P(R, t) (2.21) we get the following macrotransport equation ρCp ∗ (∂P ∂t +U ∗ ·∇P ) = kT ∗ :∇∇P +δ(R)δ(t) (2.22) with P → 0 as |R|→∞ (2.23) Here, R is the macroscale (Darcy) position vector of a lattice point relative to an origin O arbitrarily chosen in the unit periodic cell. Themacroscale coefficients ρCp ∗ , U ∗ and KT ∗ are given by the following formulas ρCp ∗ = 1 τ0 ∫ τ0 ρ(r)Cp(r) d 3 r U ∗ = ∫ τ0 J ∞ 0 (r) d 3 r (2.24) α∗ = KT ∗ ρCp ∗ = ∫ τ0 P∞0 (r)(∇B) ⊤(r) ·simKT(r) ·∇B(r) d 3r The fields P∞0 (r) and B(r) satisfy, for r ∈ τ0, the following boundary- value problems ∇·J∞0 =0 J ∞ 0 = ρCpUP ∞ 0 −KT ·∇P ∞ 0 ν ·∆SpJ ∞ 0 =0 ∆SpP ∞ 0 =0 on Sp (2.25) ‖P∞0 ‖=0 ‖∇P ∞ 0 ‖=0 on ∂τ0 ∫ τ0 ρ(r)Cp(r)P ∞ 0 (r) d 3r=1 On the asymptotic behavior of thermal... 83 and ∇· (P∞0 KT ·∇B)−J ∞ 0 ·∇B= ρCpP ∞ 0 U ∗ B is continuous across Sp (2.26) ν ·∆Sp(KT ·∇B)= 0 on Sp ‖B‖=−‖r‖ ‖∇B‖=0 on ∂τ0 Here, for any tensor-valued field F, ‖F‖ defines the ”jump” in the value of F between the equivalent points lying on opposite pairs of the cell faces. In this manner, we can obtain a macrotransport paradigm for a class of thermal dispersion phenomena occurring in periodic media. Moreover, for this case, introducing two positive scalars U0 and a, we can express the fluid velocity U in the form U(r)=U0V (r/a) (2.27) U0 and a will be interpreted as being the velocity and the spatial scale para- meters which characterize our transport processes. We are interested in getting the functional dependence of the asymptotic dispersion coefficients α∗ in terms of these two parameters. Using a central limit theorem for Markov processes, it can be proved (see Timofte, 1999; Bhattacharya et al., 1989) that for a special case of thermal dispersion phenomena in periodic media, the macroscale coefficients α∗ij de- pend only on the product aU0, the result being in accordance with all the experimental studies that have been done. In fact α∗(a,U0)=α ∗(U0,a)=α ∗(aU0,1) (2.28) This interchangeability of the velocity and spatial scale parameters in the large-scale dispersionmatrix enables us to consider, if needed, that the spatial scale parameter a is held fixed at a= 1, while the velocity parameter U0 is allowed to vary. Amore precise analysis of the asymptotic behavior of the dispersion coeffi- cients α∗ij canbedone if the thermophysical properties ρ,Cp,KT are supposed to be constant and if wemakemore restrictive assumptions about the velocity field (see Timofte, 1999). 84 C.Timofte 3. Applications As a first example, we shall consider the problem of internal energy di- spersion in an incompressible viscous fluid moving under laminar flow condi- tions between two parallel, insulated porous plates separated by a distance h. The upper plate moves at a velocity U0 parallel to it in the x-direction. Si- multaneously, there exists a uniform flow across the channel (in the negative y-direction) at a constant velocity v0. In this case, the fluid velocity field U is given by U = U0y h i−v0j (3.1) At t=0, an amount of heat is instantaneously added into our system over some region of the infinite domain between the plates in the form of some initial temperature distribution T0(x). Assuming that the thermophysical properties ρ,Cp and KT are constant, the evolution of the temperature T(t,x) will be governed by the following equation ∂T ∂t + U0y h ∂T ∂x −v0 ∂T ∂y =α∆T (3.2) with the initial condition T(0,x)=T0(x) and with α=KT/(ρCp). Introducing the dimensionless parameter β= v0h α (3.3) and considering the incomplete gamma function γ(n+1,β)= β ∫ 0 ξnexp(−ξ) dξ n=0,1,2, ... (3.4) the macroscale thermal velocity U is given by U ∗ =U ∗ i (3.5) where U ∗ = U0γ(2,β) βγ(1,β) (3.6) If we consider the mean axial fluid velocity V = U0 2 (3.7) On the asymptotic behavior of thermal... 85 we get U ∗ V = 2γ(2,β) βγ(1,β) (3.8) So, the thermal velocity U ∗ is different from themean axial fluid velocity V . As the cross flow velocity v0 → 0, corresponding to β → 0, it is easy to see that lim β→0 U ∗ V =1 If v0 →∞, then β→∞ and lim β→∞ U ∗ V =0 Using the general formulas given by the abovemethod ofmoments, we see that the only component of the effective thermal dispersivity dyadic α∗ which is different from zero is α∗11 =α+k(β) h2V 2 α (3.9) with k(β)= 4 β4 [γ(2,β) γ(1,β) ]2[ 2 γ(2,β) γ(1,β) +3 γ(3,β) γ(2,β) ] (3.10) We notice that if v0 =0we get a formula which is similar to classical formula (1.1) for the case of Taylor’s solute dispersion. As a second example, let us consider the problemof internal energy disper- sion in a layered periodic porousmedium, saturatedwith a viscous incompres- sible fluid.We shall choose as a periodic cell the parallelepiped τ0 having the sides lx, ly and lz. Let us suppose that the thermophysical properties ρ, Cp, KT are constant and the velocity field U is periodic, with the period lz U = [ U0 ( 1+sin 2πz lz ) ,U0 sin 2πz lz ,U0ω ] (3.11) Here, U0 and ω are given real parameters (see Timofte, 1996). Initially, the medium has an uniform temperature T0 (we can choose T0 = 0). At t = 0, an amount of heat Q is instantaneously introduced in- to the system as the initial distribution of temperature T(0,x)=T0(x). With α = KT/(ρCp), the evolution of the temperature T(t,x) will be governed by the following equation ∂T ∂t =α∆T −U0 ( 1+sin 2πz lz )∂T ∂x −U0 sin 2πz lz ∂T ∂y −U0ω ∂T ∂z (3.12) 86 C.Timofte subjected to the initial condition T(0,x)=T0(x). Obviously U ∗ =(U0,0,U0ω) (3.13) Following the general scheme offered by the abovemethod ofmoments, we can compute the macroscale coefficients α∗ij α∗11 =α ∗ 22 =α+ αl2zU 2 0 2[(2πα)2 +(U0lzω)2] α∗33 =α (3.14) α∗12 =α ∗ 21 = αl2zU 2 0 2[(2πα)2 +(U0lzω)2] α∗13 =α ∗ 31 =α ∗ 23 =α ∗ 32 =0 It is simple to see that for small values of lzU0, α ∗ ij depend quadratically on lzU0. However, as lzU0 →∞, each α ∗ ij becomes asymptotically constant. As the final example, we shall consider the problem of internal energy dispersion in a periodic porous medium, saturated with an incompressible viscous fluid having the velocity field U(x)=U0V (x) given by V = [0,2+sin2πx,2+cos2πx cos2πy] (3.15) We assume that the spatial scale parameter a is fixed at a = 1 and the phenomenological coefficients ρ, ,Cp and KT are strictly positive constants. Obviously U ∗ = [0,2U0,2U0] (3.16) It is easy to see that in this case α∗11 =α (3.17) α∗12 =α ∗ 21 =α ∗ 13 =α ∗ 31 =α ∗ 23 =α ∗ 32 =0 For this example, closed-form solutions of themacrotransport coefficients α∗22 and α∗33 cannot be obtained. However, the analytical theory developed by Timofte (1999) and Bhattacharya et al. (1989) shows that, as U0 → ∞, α∗22 =α+O(U 2 0) and α ∗ 33 =α+O(1). This example reflects the influence of the geometry of the flow curves on the asymptotic behavior of the macrotransport coefficients. On the asymptotic behavior of thermal... 87 References 1. Aris R., 1956, On the dispersion of a solute in a fluid flowing through a tube, Proc. Roy. Soc., Ser. A, 235, 67-78 2. Aris R., 1960, On the dispersion of a solute in pulsating flow through a tube, Proc. Roy. Soc., Ser. A, 259, 370-376 3. Bensoussan A., Lions J.L., Papanicolaou G., 1978,Asymptotic Analysis for Periodic Structures, Amsterdam, North-Holland 4. BhattacharyaR.N.,GuptaV.K.,WalkerH., 1989,Asymptotics of solute dispersion in periodic porous media, SIAM J. Appl. Math., 49, 86-98 5. Brenner H., 1980, Dispersion resulting from flow through spatially periodic porous media,Phil. Trans. Roy. Soc. London, A, 297, 81-133 6. Brenner H., Edwards D.A., 1993,Macrotransport Processes, Butterworth- Heinemann, Boston 7. Horn F.J.M., 1971, Calculation of dispersion coefficients by means of mo- ments,AIChE J., 613-620 8. KochD.L., Brady J.F., 1985,Dispersion in fixed beds, J. FluidMech., 154, 399-427 9. Sanchez-PalenciaE., 1980,Non-HomogeneousMedia andVibration Theory, Lect. Notes Phys., 127, Springer-Verlag, Berlin 10. Taylor G.I., 1956, Dispersion of soluble matter in a solvent flowing slowly through a tube, Proc. Roy. Soc., Ser. A, 219, 186-203 11. Timofte C., 1996, Stochastic Processes and their Applications to Fluid Me- chanics, Ph.D. Thesis, Institute of Mathematics of the Romanian Academy, Bucharest (in Romanian) 12. Timofte C., 1999, Asymptotics of thermal dispersion in periodic media, Journal of Theoretical and Applied Mechanics, 37, 1, 95-108 O asymptotycznej naturze procesu termicznej dyspersji Taylora w ośrodkach periodycznych Streszczenie Wpracy dokonano przeglądu szerokiej klasy zjawisk związanych z termiczną dys- persją w wybranych układach jednorodnych w skali makro. W opisie wykorzystano uogólnioną metodę momentów i twierdzenie o granicy centralnej. Jako szczególnie 88 C.Timofte interesujący przedstawiono problem obliczania współczynnikówmakroskali w funkcji współczynników mikroskali i geometrii badanego układu. Ponadto przeanalizowano funkcjonalną zależność współczynników efektywnych od pola prędkości i przestrzen- nych parametrów skali. Manuscript received October 16, 2001; accepted for print October 22, 2002