Michigan Technological University

Digital Commons @ Michigan Tech

Dissertations, Master's Theses and Master's Reports

2021

MATHEMATICAL MODELING AND NUMERICAL SIMULATION OF

HEAT TRANSFER FROM ISOLATED OBJECTS
Esmaeil Dehdashti
Michigan Technological University, [email protected]

Copyright 2021 Esmaeil Dehdashti

Recommended Citation
Dehdashti, Esmaeil, "MATHEMATICAL MODELING AND NUMERICAL SIMULATION OF HEAT TRANSFER
FROM ISOLATED OBJECTS", Open Access Dissertation, Michigan Technological University, 2021.
https://doi.org/10.37099/mtu.dc.etdr/1231

Follow this and additional works at: https://digitalcommons.mtu.edu/etdr

MATHEMATICAL MODELING AND NUMERICAL SIMULATION OF HEAT

TRANSFER FROM ISOLATED OBJECTS

By

Esmaeil Dehdashti

A DISSERTATION

Submitted in partial fulfillment of the requirements for the degree of

DOCTOR OF PHILOSOPHY

In Mechanical Engineering–Engineering Mechanics

MICHIGAN TECHNOLOGICAL UNIVERSITY

2021

© 2021 Esmaeil Dehdashti

This dissertation has been approved in partial fulfillment of the requirements for the

Degree of DOCTOR OF PHILOSOPHY in Mechanical Engineering–Engineering

Mechanics.

Department of Mechanical Engineering–Engineering Mechanics

Dissertation Advisor: Dr. Hassan Masoud

Committee Member: Dr. Fernando L. Ponta

Committee Member: Dr. Kazuya Tajiri

Committee Member: Dr. Cécile Piret

Department Chair: Dr. William W. Predebon

Contents

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

1 Background and significance . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Conduction heat transfer from oblate spheroids and bispheres . . 6

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 Problem formulation and solutions . . . . . . . . . . . . . . . . . . . 7

2.2.1 Temperature field around an oblate spheroid . . . . . . . . . . 8

2.2.2 Temperature field around two identical spheres . . . . . . . . . 13

2.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Forced convection heat transfer from a particle at small and large

Péclet numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

iii
 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.3 Perturbation solution in the limit of conduction-dominated heat trans-

port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.4 Perturbation solution in the limit of advection-dominated heat transport 33

3.5 Specific results for spheroids in axisymmetric Stokes flow . . . . . . . 40

3.5.1 Limit of Pe  1 . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.5.2 Limit of Pe  1 . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.5.3 Comparison with direct numerical solutions . . . . . . . . . . 45

3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Heat transfer from a particle in laminar flows of a variable thermal

conductivity fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.3 Variation of Nusselt number for constant temperature boundary con-

dition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.3.1 Limit of conduction-dominated heat transport . . . . . . . . . 54

4.3.2 Limit of advection-dominated heat transport . . . . . . . . . . 59

4.3.3 Bridging results for limits of low and high Péclet numbers . . 65

4.4 Variation of Nusselt number for uniform heat flux boundary condition 66

4.4.1 Limit of conduction-dominated heat transport . . . . . . . . . 67

iv
 4.4.2 Limit of advection-dominated heat transport . . . . . . . . . . 69

4.4.3 Bridging results for limits of low and high Péclet numbers . . 70

4.5 Comparison with direct numerical solutions . . . . . . . . . . . . . . 73

4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5 Summary and future directions . . . . . . . . . . . . . . . . . . . . . . 76

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

v
List of Figures

1.1 Steady-state temperature distribution around a hot particle in uniform

Stokes flow at (a) low and (b) high Péclet numbers. The black lines

and arrows show the flow streamlines and flow direction, respectively. 2

2.1 (a) Oblate spheroidal coordinates in a meridian plane. The arrows show

the direction of the unit vectors eξ and eη . (b) Surfaces of constant ξ

and η depicted in cyan and gray, respectively. The red curve in (a) and

its corresponding surface in (b) represent a hot/cold oblate spheroid

that releases/absorbs heat at a constant uniform rate. . . . . . . . . 9

2.2 (a) Bispherical coordinates in a meridian plane. The arrows show the

direction of the unit vectors eζ and eβ . (b) Surfaces of constant ζ and

β depicted in cyan and gray, respectively. The red curves in (a) and

their corresponding surfaces in (b) represent a pair of identical hot/cold

spheres that release/absorb heat at a constant uniform rate. . . . . . 14

2.3 (a)-(b) Contour plots of the (dimensionless) temperature field θ (in

meridian planes) around an oblate spheroid of aspect ratio ε = 0.3.

The left panel illustrates the results of the two-term approximation

vi
 whereas the right panel presents those obtained by setting m = 19,

considered here as exact results. (c) The Nusselt number Nu versus the

aspect ratio ε for oblate spheroids. Red, blue, and black lines represent

the results corresponding to m = 3, linear interpolation based on the

Nu of sphere and disk, and m = 19, respectively. The inset shows the

relative difference between the approximate and exact results. . . . . 19

2.4 (a)-(b) Contour plots of the (dimensionless) temperature field θ (in

meridian planes) around two spheres whose centers are three radii

apart ( = 2/3). The left panel illustrates the results of the four-term

approximation whereas the right panel presents those obtained by set-

ting m = 9, considered here as exact results. (c) B0 as a function of

1 − , where the dashed and dotted lines represent the asymptotic ap-

proximations in the limits of  → 0 and  → 1, respectively. And, the

inset shows the relative error of the approximations. (d) The Nusselt

number Nu versus the inverse of the dimensionless separation distance

 for a pair of identical spheres. Red, blue, green, and black lines repre-

sent the results corresponding to m = 3, linear interpolation based on

the Nu of single sphere and two touching spheres, asymptotic behavior

in the limit of  → 0, and the converged solution (i.e., m  1), respec-

tively. The inset shows the relative difference between the approximate

and exact results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

vii
3.1 Schematic of a stationary particle of arbitrary shape, with surface Sp

and unit outward normal vector n, in a uniform fluid flow. The dashed

line indicates an enclosing boundary in the far field. . . . . . . . . . . 27

3.2 Schematic of uniform flow past an axisymmetric heated object at high

Péclet number. The boundary layer coordinates are represented by x

and y, and the surface heat flux is denoted by q. . . . . . . . . . . . . 34

3.3 Numerically calculated plots of the Nusselt number versus Péclet num-

ber for forced convection heat transfer from spheroids of various aspect

ratios in an axisymmetric uniform Stokes flow subject to constant heat

flux (solid lines) and isothermal (dashed lines) boundary conditions on

the surface of the spheroid. . . . . . . . . . . . . . . . . . . . . . . . . 46

3.4 Percent difference ∆ between the numerical and approximate asymp-

totic results for the Nusselt number corresponding to (a)–(b) con-

stant heat flux and (c)–(d) isothermal boundary conditions. (a)–(c)

and (b)–(d) present ∆ versus Pe curves for approximations based on

Eqs. (3.53) and (3.54), respectively. The cut-off Péclet numbers for

the aspect ratios ε = 0.2, 0.5, 1, 2, 5 in (a) and (c) are, respectively,

Pec = 0.2, 1, 1, 0.35, 0.25 and Pec = 1, 1, 1, 0.35, 0.25. . . . . . . . . . . 48

4.1 Percent difference ∆ between the results of full numerical simulations

and the predictions of Eq. (4.56) for NuT . . . . . . . . . . . . . . . . 71

4.2 Percent difference ∆ between the results of full numerical simulations

viii
and the predictions of Eq. (4.71) for NuQ . . . . . . . . . . . . . . . . 72

ix
Preface

This dissertation includes contents that were previously published as journal articles,

whose full citations are:

• Conduction heat transfer from oblate spheroids and bispheres, S. Jafari Kang,

E. Dehdashti, and H. Masoud, International Journal of Heat and Mass Transfer

139, 115–120 (2019);

• Forced convection heat transfer from a particle at small and large Peclet num-

bers, E. Dehdashti and H. Masoud, Journal of Heat Transfer 142, 061803 (2020);

• Heat transfer from a particle in laminar flows of a variable thermal conductivity

fluid, E. Dehdashti, M. Razizadeh, and H. Masoud, International Journal of

Heat and Mass Transfer 171, 121067 (2021).

As evidenced from the list of authors, the first and last publications were collaborative

works involving, respectively, Saeed Jafari Kang and Meghdad Razizadeh. In the first

collaborative study, all authors contributed to the theoretical analyses, numerical

simulations, and writing of the manuscript. And, in the second one, the primary

contributors were the author of this dissertation (Esmaeil Dehdashti) and his advisor

(Dr. Hassan Masoud), with Meghdad Razizadeh being only involved in some early

analysis.

x
Abstract

In the area of heat transfer, like other fields of science and engineering, full- and

semi-analytical solutions of elementary problems are regarded as invaluable resources

that can be used to identify relevant dimensionless parameters, to obtain basic insights

into the phenomena under consideration, to quickly quantify the effects of key factors,

and, ultimately, to pave the way for understanding more complex problems arising

in practice. These solutions can also serve as excellent benchmarks for calibrating

experimental setups and validating numerical techniques.

In this dissertation, we theoretically study three classical heat transfer problems,

with the ultimate goal of deriving analytical or approximate expressions for the Nus-

selt number (denoted by Nu), which is a key dimensionless parameter that quantifies

the transfer of heat to and from a surface. First, we consider heat transfer by conduc-

tion from oblate spheroidal and bispherical surfaces into a stationary, infinite medium.

The surfaces are presumed to maintain a constant heat flux. Assuming steady-state

condition and uniform thermal conductivity, we analytically solve the Laplace equa-

tion for the temperature distribution and discuss the challenge of dealing with the

Neumann (uniform flux) versus more convenient Dirichlet (isothermal) boundary con-

dition. The solutions are obtained in boundary-fitting coordinate systems using the

method of separation of variables and eigenfunction expansion. And, exact expressions

for the average Nusselt number are presented along with their approximations.

Next, we examine forced convection heat transfer from a single particle in uniform

xi
laminar flows. Asymptotic limits of small and large Péclet numbers (denoted by Pe)

are considered. For Pe  1 (diffusion-dominated regime) and a constant heat flux

boundary condition on the surface of the particle, we obtain a closed-form expression

for the heat transfer coefficient that is valid for arbitrary particle shapes and Reynolds

numbers, as long as the flow is incompressible. We also present a framework for

calculating the average Nusselt number of axisymmetric and two-dimensional objects

with a constant heat flux surface condition in the limits of Pe  1 and small or

moderate Reynolds numbers. Specific results are given for the heat transfer from

spheroidal particles in Stokes flow.

Finally, we revisit the problem of steady-state heat transfer from a single particle

in a uniform laminar flow with the assumption that the thermal conductivity of the

fluid changes linearly with the temperature. We use a combination of asymptotic

and scaling analyses to derive approximate expressions for the Nusselt number of

arbitrarily shaped particles. The results cover the entire range of the Péclet number.

We find that, for a constant temperature boundary condition and fixed geometry, the

Nusselt number is essentially equal to the product of two terms, one of which is only a

function of Pe while the other one is nearly independent of Pe and mainly depends on

the proportionality constant of the conductivity-temperature relation. We also show

that, in contrast, when a uniform heat flux is imposed on the surface of the particle,

the Nusselt number can be estimated as a summation of a Pe-dependent piece and

one that solely varies with the proportionality constant.

xii
Chapter 1: Background and significance

1.1 Introduction

Fundamental problems in any field of science and engineering are the foundational

blocks upon which the rest of the filed is formed. Often, complex phenomena and in-

tricate systems are interpreted and analyzed through the lens of these basic problems,

which highlights their importance in advancing the state-of-the-art and in making new

discoveries. This is certainly the case in thermal sciences, which among other topics,

deals with the transfer of thermal energy (i.e., heat) between objects and systems.

In the area of heat transfer, a subcategory is devoted to the exchange of heat

between small objects (i.e., particles) and their surrounding medium. The principal

objective of this Ph.D. research is to fundamentally contribute to this branch of

heat transfer through theoretical examination of (i) conduction heat transfer from

oblate spheroids and bispheres, (ii) forced convection heat transfer from an arbitrarily-

shaped particle, and (iii) heat transfer from a single object in laminar flows of a

variable thermal conductivity fluid.

This dissertation is organized as follows. In the rest of this introductory chapter,

we put our work in context by reviewing some of the classical theoretical investigations

1
 (a) (b)

Figure 1.1: Steady-state temperature distribution around a hot particle in

uniform Stokes flow at (a) low and (b) high Péclet numbers. The black lines
and arrows show the flow streamlines and flow direction, respectively.

on conduction and convection heat transfer from particles. In the next three chapters,

we discuss the details of our analysis for each of the above-mentioned problems. And,

in the last chapter, we summarize our study and give a few ideas for future directions.

1.2 Literature review

Convection heat transfer problems are commonly classified in terms of a dimensionless

parameter called the Péclet number (denoted by Pe), which measures the importance

of advective heat transport relative to conductive transport of the thermal energy.

According to this definition, Pe = 0 represents a class of problems known as pure

conduction. When the Péclet number is very small (Pe  1), but finite, the transfer

of heat is still dominated by conduction (see, e.g., Fig. 1.1a). However, as Pe rises,

the influence of advection grows and eventually becomes comparable with that of

conduction when Pe ∼ O(1). With further increase of the Péclet number, especially

beyond O(102 ), advection becomes the dominant mode of heat transfer and conduction

2
will be limited to a narrow region (known as the boundary layer) close to the surface

of the particle (see, e.g., Fig. 1.1b).

Theoretical study of convection heat transfer at arbitrary Péclet number is often

very challenging, if not impossible. As a result, problems treated in the vast majority

of theoretical investigations correspond to either Pe = 0 or the asymptotic limits of

Pe (i.e., Pe  1 and Pe  1) [1–16]. The latter choice makes the problem amenable

to the perturbation theory, using which very useful approximations can be derived

for the quantities of interest. One of such quantities is the dimensionless rate of total

heat transfer from a surface, which is known as the Nusselt number and is denoted

by Nu. In the following and beginning with the case of Pe = 0, we summarize several

classical studies where Nu of an object is theoretically calculated for a particular Pe

regime under steady-state condition. We note that this review is not meant to be

comprehensive.

The literature on pure conduction is vast and most of the studies on this subject

center around the solution of the Laplace equation for the temperature distribution.

Once the temperature field is known, then the integral of the local heat flux on

the surface of the object is calculated to obtain the Nusselt number. Among many

other studies, Yovanovich [17] and Hahne and Grigull [18] analytically solved for and

catalogued Nu for a number of practical geometries. The Nusselt number for addi-

tional geometries are reported in classical heat transfer textbooks (see, e.g., [19]),

with references given to the original derivations. Furthermore, Brenner and Haber

3
[20] presented a general solution for the Laplace equation, using the symbolic oper-

ator method, that is valid for a general far-filed temperature distribution and is not

restricted to the typical uniform temperature assumption at infinity. It is worth noting

that the boundary value problems associated with pure conduction appear in other

fields such as electrostatics as well. And, they are often discussed in mathematical

physics textbooks.

In the limit of Pe  1, Brenner [4] showed, in his seminal work, that the leading-

order contribution of advection to the Nusselt number for the problem of uniform flow

past an arbitrarily-shaped particle can be calculated by only knowing the conduction

Nusselt number of the same problem. This surprisingly general result was obtained by

using a singular perturbation expansion for the temperature and by the application of

the reciprocal theorem. Remarkably, he proved that the first-order correction to Nu of

pure conduction does not even depend on the details of the flow field. In the opposite

limit of Pe  1, Acrivos and Goddard [6, 7] generalized the work of Lighthill [1] in

their two-part study, where they developed a framework for calculating the Nusselt

number in cases of planar and axisymmetric laminar flows. These asymptotic results

for small and large Péclet number were later extended by [21] to problems with

variable thermal conductivity.

Overall, our survey of the literature on conduction and convection heat trans-

fer from isolated objects reveals that while there exists a great body of theoretical

analyses, there are certain aspects that have received less attention. In particular, we

4
have noticed that the boundary conditions considered on the surface of the objects

have been limited, to a large extent, to the prescribed temperature condition. And,

therefore, there is a need for more extensive examinations of other boundary condi-

tion. The theoretical calculations detailed in the subsequent chapters are performed

in response to this need.

5
Chapter 2: Conduction heat transfer from
oblate spheroids and bispheres

2.1 Introduction

In this chapter, we present analytical solutions for the problem of steady-state con-

duction heat transfer in an infinite medium due to the presence of hot/cold inclusions.

Specifically, two geometries for the heat source/sink are considered, namely an oblate

spheroid and a pair of spheres (see Figs. 2.1 and 2.2). For both cases, a uniform heat

flux is assumed to emanate from the surface of the inclusions. This boundary condi-

tion models many scenarios that appear frequently in engineering applications, such

as when a surface is covered by a thin layer of electric heater [19]. The assumption of

the Neumann boundary condition is the novel aspect of our work. In the following, we

first formulate the problem mathematically and then provide a detailed description

of the solutions for the two geometries of interest. Next, we discuss the results for

the temperature distribution and average Nusselt number and, finally, we give a brief

summary of our study.

6
2.2 Problem formulation and solutions

Consider one or more objects surrounded by an infinite medium at rest. Suppose that

heat is released/absorbed from/by the surface of the objects at a constant uniform

rate and that the temperature at infinity is maintained at a constant value. Of in-

terest, here, is the steady-state temperature distribution in the surrounding medium

assuming that the transport of heat is dominated by conduction and that the ther-

mal conductivity is constant. Given the above conditions, the boundary-value problem

governing the distribution of the temperature is

∇2 T ? = 0 with
(2.1)
? ? ?
− k n · ∇T = qs for r ∈ Sp and T → T∞ as r → ∞,

where T ? represents the temperature field, k is the thermal conductivity of the

medium, n is the unit vector outward normal to the surface of the objects denoted

?
by Sp , r is the position vector with magnitude r = |r|, and qs and T∞ are constants.

Let ` be a characteristic length scale of the problem. Then, upon the change of

variables T = k (T ? − T∞
?
) /qs `, r̃ = r/`, and r̃ = r/`, Eq. (2.1) simplifies to the

following dimensionless form:

∇2 T = 0 with
(2.2)
n · ∇T = −1 for r̃ ∈ Sp and T → 0 as r̃ → ∞.

One can define an average Nusselt number for this problem as

7
 Sp
Nu = , (2.3)
2πT

where Sp represents the dimensionless surface area of the object and T is the mean

value of T on Sp . Below, we solve Eq. (2.2) and calculate the average temperature for

cases in which Sp represents the surface of an oblate spheroid and a pair of spheres,

respectively. For each case, the solution is obtained via the method of separation

of variables and eigenfunction expansion in a coordinate system that matches the

boundaries of the problem.

The correctness of the derivations is independently verified by comparing the

results against those obtained from the numerical solution of Eq. (2.2). A second-

order finite volume method as implemented in OpenFOAM (see, e.g., [22]) is used

to perform the computations. The outer boundary at infinity is modeled as a very

large sphere, and 2D axisymmetric meshes concentrated around Sp are employed to

discretize the physical domains.

2.2.1 Temperature field around an oblate spheroid

Consider an oblate spheroid of equatorial radius ` and aspect ratio (ratio of polar to

equatorial radius) ε. Let (x, y, z) be the components of a reference Cartesian coordi-

nate system located at the center of the spheroid such that the z axis coincides with

the revolution axis of the spheroid. The coordinates are nondimentionalized by `. To

solve Eq. (2.2), we adopt an oblate spheroidal coordinate system (ξ, η, ϕ) defined as

8
 Figure 2.1: (a) Oblate spheroidal coordinates in a meridian plane. The
arrows show the direction of the unit vectors eξ and eη . (b) Surfaces of
constant ξ and η depicted in cyan and gray, respectively. The red curve in
(a) and its corresponding surface in (b) represent a hot/cold oblate spheroid
that releases/absorbs heat at a constant uniform rate.

[23, 24]
p
x / cos ϕ = y / sin ϕ = c (1 + ξ 2 ) (1 − η 2 ), z = c ξ η, (2.4)

√
where c = 1 − ε2 is the (dimensionless) radius of the focal circle and

0 ≤ ξ < ∞, −1 ≤ η ≤ 1, 0 ≤ ϕ < 2π. (2.5)

The metric coefficients associated with (ξ, η, ϕ) are

s s
ξ 2 + η2 ξ 2 + η2 p
hξ = c , hη = c , hϕ = ρ = x2 + y 2 . (2.6)
1 + ξ2 1 − η2

As shown in Fig. 2.1, the surfaces of constant ξ and η are oblate spheroids and one-
√
sheet hyperboloids of revolution, respectively. In particular, ξ = ξ0 = ε / 1 − ε2

represents Sp and ξ → ∞ corresponds to a bounding surface at large distances.

9
 In the orthogonal curvilinear coordinate system (ξ, η, ϕ), Eq. (2.2) takes the form

of [23, 24]
    
2 1 ∂
∂T
2 ∂ 2 ∂T


∇T = 2 2 1+ξ + 1−η
c (ξ + η 2 ) ∂ξ ∂ξ ∂η ∂η
1 ∂ 2T
+ = 0 with (2.7)
c2 (1 + ξ 2 ) (1 − η 2 ) ∂ϕ2
1 ∂T
= −1 and T → 0 as ξ → ∞.
hξ ∂ξ ξ=ξ0

Since the boundary conditions do not depend on ϕ, we deduce that T is azimuthally

independent, i.e., ∂T /∂ϕ = 0. Hence, Laplace’s equation for T reduces to

   
∂
∂T
2 ∂ 2 ∂T


1+ξ + 1−η = 0. (2.8)
∂ξ ∂ξ ∂η ∂η

Starting with the ansatz T (ξ, η) = X(ξ) H(η), where X and H are to-be-determined

functions, and following the steps involved in the separation of variables technique

[23, 24], it can be shown that the general solution of Eq. (2.8) is

∞
X  1
Am Pm (iξ) + A2m Qm (iξ) A3m Pm (η) + A4m Qm (η) ,
 
T = (2.9)
m=0

where m is an integer, A1m , . . . , A4m are constants, i2 = −1, and Pm and Qm are

Legendre functions of the first and second kind, respectively [25]. The latter function

is defined as
   
1 z+1 z z+1
Q0 (z) = ln , Q1 (z) = ln − 1,
2 z−1 2 z−1
(2.10)
(2n + 1) z Qn − n Qn−1
Qn+1 (z) = .
n+1
The function Pm (iξ) blows up as ξ → ∞ for m 6= 0 and so does Qm (η) at η =

10
±1 for all m. Therefore, to keep the solution finite, we need to set A1m = 0 for m 6= 0

and A4m = 0 for all m. By demanding T to vanish at infinity, we find that A10 is zero,

too, since P0 (iξ) = P0 (η) = 1 and Qm (iξ) decays to zero for large ξ. Thus, Eq. (2.9)

simplifies to
∞
X
T = Am Qm (iξ) Pm (η), (2.11)
m=0

where the constant coefficients Am are determined by applying the constant flux

boundary condition on Sp :
s
∞
1 ∂T i 1 + ξ02 X
= 2 2
Am Q0m (iξ0 ) Pm (η) = −1, (2.12)
hξ ∂ξ ξ=ξ0 c ξ0 + η m=0

with Q0m (x) = dQm (x)/dx. Note that although the gradient of T in the direction

normal to the boundary is constant, its derivative with respect to ξ (coordinate normal

to the boundary) is equal to the scale factor hξ , which varies along the boundary. Had

we been considering the constant temperature boundary condition (i.e., T = 1) at

ξ = ξ0 , we would have been dealing with a significantly simpler situation where T

would have been also independent of η and Eq. (2.8) would have further reduced to

an ordinary differential equation with the solution (see, e.g., [26])

cot−1 ξ cot−1 ξ
T = = . (2.13)
cot−1 ξ0 cos−1 ε

The eigenfunction Pm has the following orthogonality property:

(
Z 1 2 / (2m + 1) if n=m
Pm (x) Pn (x) dx = , (2.14)
−1
0 if n 6= m

11
where n is an integer. Using this feature, the unknown coefficients are obtained as

Z 1
i 2m + 1
q
Am = 0 Pm (η) ξ02 + η 2 dη. (2.15)
Qm (iξ0 ) 2(1 + ξ02 ) −1

The integral in Eq. (2.15) is zero for odd values of m and is otherwise calculated via

[27]
Z 1 q
Pm (η) ξ02 + η 2 dη
−1
m/2 (2.16)
X Γ(m/2 + n + 1/2) I2n (ξ0 )
= 2m ,
n=0
Γ(2n + 1) Γ(m − 2n + 1) Γ(n − m/2 + 1/2)

where Γ is the gamma function and

Z 1
2ξ0
q
2n −2


I2n (ξ0 ) = η ξ02 + η 2 dη = 2 F1 −1/2, n + 1/2; n + 3/2; −ξ0 . (2.17)
−1 2n + 1

Here, 2 F1 is the hypergeometric function [25]. Having determined the temperature

distribution, the average T on the surface of the spheroid can be calculated as

Z 1 q
2 π c2
Z
1
q
2
T = T dS = 1 + ξ0 T ξ02 + η 2 dη
Sp Sp Sp −1
p ∞ 2
(2.18)
2 X
4π 1 + ξ0 Am
=− iQ0m (iξ0 ) Qm (iξ0 ),
Sp m=0
2m + 1

where
!
ξ02
Z q
Sp = 1 dS = 2 π 1 + p coth−1 1 + ξ02 . (2.19)
Sp 1 + ξ02

It is worth noting that the results of this section for an oblate spheroid can be

converted to those for a prolate spheroid by allowing ε to be greater than one and by

replacing ξ with −iξ in all relations (see, e.g., [28]).

12
2.2.2 Temperature field around two identical spheres

Consider a pair of identical spheres, with radius `, that are placed distance 2`/ apart,

where 0 <  < 1. Recall the Cartesian coordinate system of §2.2.1. Only this time,

the origin is located in the middle of the line that connects the center of the spheres

and the z axis is oriented in the direction of that line. The natural coordinate system

for dealing with this geometry is a bispherical coordinate system (ζ, β, ϕ) defined as

[23, 24]
c sin β c sinh ζ
x / cos ϕ = y / sin ϕ = , z= , (2.20)
cosh ζ − cos β cosh ζ − cos β
√
where c = −2 − 1 is half the (dimensionless) focal distance and

− ∞ < ζ < ∞, 0 ≤ β ≤ π, 0 ≤ ϕ < 2π. (2.21)

The coordinates have the following scale factors:

c c sin β
hζ = hβ = , hϕ = ρ = . (2.22)
cosh ζ − cos β cosh ζ − cos β

Figure 2.2 depicts the surfaces of constant ζ and β that are non-intersecting spheres

surrounding the foci (points located at z = ±c) and intersecting tori passing through

the focal points, respectively. In this curvilinear coordinate system, the thermally

active spheres are represented by ζ = ±ζ0 = cosh−1 −1 and (ζ, β) → (0, 0) correspond

to r̃ → ∞.

The boundary-value problem introduced at the beginning of this section can be

written in terms of (ζ, β, ϕ) as [23, 24]

13
 Figure 2.2: (a) Bispherical coordinates in a meridian plane. The arrows
show the direction of the unit vectors eζ and eβ . (b) Surfaces of constant ζ
and β depicted in cyan and gray, respectively. The red curves in (a) and their
corresponding surfaces in (b) represent a pair of identical hot/cold spheres
that release/absorb heat at a constant uniform rate.

(cosh ζ − cos β)3 ∂

  
2 sin β ∂T
∇T =
c2 sin β ∂ζ cosh ζ − cos β ∂ζ
∂ 2T
  
∂ sin β ∂T 1
+ + =0 (2.23)
∂β cosh ζ − cos β ∂β sin β (cosh ζ − cos β) ∂ϕ2
1 ∂T
with = ±1 and T → 0 as (ζ, β) → (0, 0).
hζ ∂ζ ζ=±ζ0

The boundary conditions are, again, independent of ϕ, which simplifies the Laplace

equation to

   
∂ sin β ∂T ∂ sin β ∂T
+ = 0. (2.24)
∂ζ cosh ζ − cos β ∂ζ ∂β cosh ζ − cos β ∂β

Unlike Eq. (2.8), this equation is not simply separable and belongs to a class of partial

differential equations called R-separable equations [24]. Therefore, to solve Eq. (2.24),

14
 √
we start with the ansatz T (ζ, β) = cosh ζ − cos β Z(ζ) B(β), where Z and B are

the unknown functions we seek to determine. Substituting the proposed form for T

into Eq. (2.24), we arrive at a pair of ordinary differential equations for Z and B.

Upon solving those equations, the following general solution for T emerges [23, 24]:
∞
X
p  1
T = cosh ζ − cos β Bm sinh (m + 1/2)ζ
m=0 (2.25)
2
 3 4

+ Bm cosh (m + 1/2)ζ Bm Pm (cos β) + Bm Qm (cos β) ,

1 4
where Bm , . . . , Bm are constant coefficients.

4
As indicated before, the function Qm (cos β) is singular at cos β = ±1. Thus, Bm

vanish in order to retain the regularity of the solution. One can infer from the geometry

of the problem and the boundary conditions that the temperature distribution is

symmetric about the x-y plane, which means ∂T /∂ζ = 0 at ζ = 0. To satisfy this

1
condition, we must set Bm = 0. Conveniently, the infinity boundary condition is

already satisfied thanks to the term outside the summation that approaches zero as

(ζ, β) → (0, 0). Incorporating these results, the solution for T reduces to

∞
X
p
T = cosh ζ − cos β Bm cosh (m + 1/2)ζ Pm (cos β), (2.26)
m=0

where the constant coefficients Bm are determined by enforcing the normal gradient

boundary condition on the surface of the spheres, i.e.,

√
1 ∂T 1 ∂T cosh ζ0 − cos β
=− =
hζ ∂ζ ζ=ζ0 hζ ∂ζ ζ=−ζ0 c
∞ (2.27)
X
× Bm [Dm Pm (cos β) − Em cos β Pm (cos β)] = 1,
m=0

15
where

sinh ζ0 cosh (m + 1/2)ζ0

Dm =
2
+ (m + 1/2) cosh ζ0 sinh (m + 1/2)ζ0 , (2.28)

Em = (m + 1/2) sinh (m + 1/2)ζ0 .

Yet, again, we see that the application of the constant flux boundary condition results

in a more convoluted equation for the unknown coefficients compared to the situation

where the isothermal condition T = 1 is imposed at ζ = ±ζ0 (see, e.g., [29]).

In addition to the orthogonality property Eq. (2.14), the following integral rela-

tions apply to the eigenfunction Pm [27]:


 2 (m + 1)
if n=m+1


(2m + 1) (2m + 3)


Z 1 


Pm (x) Pn (x) x dx = 2m
if n=m−1 ,
−1  (2m − 1) (2m + 1)
(2.29)






 0 if n 6= m ± 1
Z 1
√
Pm (x) 2 2 −(m+1/2)ζ0
√ dx = e .
−1 cosh ζ0 + x 2m + 1

Utilizing the above relations in conjunction with Eq. (2.14), we obtain a recursive

formula for the coefficients in the form of

B1 = F0 + B0 ,
(2.30)
Bm+1 = Fm + Gm Bm + (1 − Gm ) Bm−1 ,

where
√
2 2 sinh ζ0 e−(m+1/2)ζ0
Fm = − , (2.31)
(m + 1) sinh (m + 3/2)ζ0

16
  
1 2m cosh ζ0 sinh (m + 1/2)ζ0
Gm = 1+ . (2.32)
m+1 sinh (m + 3/2)ζ0

We are now left with the task of determining B0 , accomplished by demanding the

terms in the solution that correspond to large eigenvalues to be finite. This condi-

tion is satisfied by requiring Bm → 0 as m → ∞, which, after some mathematical

manipulations, yields

B0 = − lim Hm , (2.33)
m→∞

where

H0 = F0 , H1 = F1 + G1 F0 ,
(2.34)
Hm = Fm + Gm Hm−1 + (1 − Gm ) Hm−2 .

With the coefficients known, the mean T over the surface of each sphere is calculated

via

2 π c2 π
Z Z
1 T sin β
T = T dS = 2 dβ
Sp Sp Sp 0 (cosh ζ0 − cos β)
∞ (2.35)
sinh ζ0 X
Bm 1 + e−(2m+1)ζ0 .
 
= √
2 m=0

2.3 Results and discussion

We begin with the results for oblate spheroids. First, we consider the limiting cases of

ε = 1 (sphere) and ε = 0 (disk). The results for the former have been already known

to be

T = 1/r̃, T = 1, Nu = 2. (2.36)

17
For the latter, however, we find
  2
i (2m + 1) m!
if m ≡ 0 (mod2)


(m − 1) (m + 2) 2m [(m/2)!]2

Am = , (2.37)

m ≡ 1 (mod2)

 0 if

which, consistent with [19], results in

T = 8/3π, Nu = 3π/8. (2.38)

Next, we examine other aspect ratios. Quite unexpectedly, our calculations indicate

that the series solution for the temperature (see Eq. (2.11)) is very well approximated

by its first two non-zero terms for the entire range of ε, i.e.,
q 
Sp
T ≈ 1 + ξ0 2
cot−1 ξ
4π
  
5 2

2

Sp
+ 2 + 3ξ0 + 4 + 3ξ0 1 − (2.39)
64 2π
3ξ − (3ξ 2 + 1) cot−1 ξ
  
2


× 3η − 1 .
3 (1 + ξ02 ) ξ0 cot−1 ξ0 − (2 + 3ξ02 )

In fact, the maximum percent difference between the results corresponding to m =

3 and m = 19 happens to be less than 4%. A visually convincing demonstration of

this result is provided in Fig. 2.3, where we present side-by-side contour plots of the

temperature field (in meridian planes) around an oblate spheroid of aspect ratio ε =

0.3. As you can see, the approximate results of Fig. 2.3a are barely distinguishable

from the (nearly) exact results of Fig. 2.3b.

Equally interesting, we also find that the average Nusselt number for oblate

spheroids varies almost linearly with the aspect ratio (see Fig. 2.3c). Therefore, the

18
 (a) (b)

2
(c)
Two-term approximation
Linear interpolation
1.75 Exact
Nu

1.5
3

2
δ(%)

1.25
1

1
0.2 0.4 0.6 0.8 1
ε
Figure 2.3: (a)-(b) Contour plots of the (dimensionless) temperature field
θ (in meridian planes) around an oblate spheroid of aspect ratio ε = 0.3. The
left panel illustrates the results of the two-term approximation whereas the
right panel presents those obtained by setting m = 19, considered here as
exact results. (c) The Nusselt number Nu versus the aspect ratio ε for oblate
spheroids. Red, blue, and black lines represent the results corresponding to
m = 3, linear interpolation based on the Nu of sphere and disk, and m = 19,
respectively. The inset shows the relative difference between the approximate
and exact results.

19
Nu versus ε curve can be approximated by the line that passes through its end points,

i.e.,
 
3π 3π
Nu ≈ Nudisk + (Nusphere − Nudisk ) ε = + 2− ε. (2.40)
8 8

The maximum relative error of this approximation occurs at ε ≈ 0.5 and falls below

3%, which underscores its validity for all aspect ratios (see the inset of Fig. 2.3c).

Furthermore, from Fig. 2.3c and its inset, we learn that the Nusselt number calculated

based on the two-term representation of T is extremely accurate, differing less than

0.25% from the actual values.

We now analyze the results for bispheres. In the limit  → 0, the spheres are very

far from each other and, therefore, the solution simplifies to Eq. (2.36) for a single

sphere. On the other hand, when  → 1 (i.e., when the gap between the spheres

vanish), Eq. (2.26) for the temperature distribution does not immediately reduce to

a simple form. The average surface temperature and Nusselt number in this case are

T = 1.57721 and Nu = 1.26806.

Again, we observe that setting m = 3 in the series solution for T [Eq. (2.26)]

provides excellent results, accurate to within 5% of the converged solution for 0 <

 < 4/5 (see, e.g., Figs. 2.4a and 2.4b). As the gap between the spheres narrows,

the error of the four-term approximation increases, which means that more terms are

needed to represent the temperature field accurately. For example, the error grows to

10% as the gap size reduces to `/4.

Given the recursive relation defined in Eq. (2.30), what carries the most weight in

20
(a) (b) 0
(c)
√
ln(1 − ) / 2 − 0.26085
√
−2 2
Exact

B0
−4

30

−6 20

δ(%)
10

0
−8
10−4 10−3 10−2 10−1 100
1−

6
(d)
4

1.75 δ(%) 2

0
Nu

1.5 Four-term approximation

Linear interpolation
4 / (2 + )
Exact
1.25
0.2 0.4 0.6 0.8 1


Figure 2.4: (a)-(b) Contour plots of the (dimensionless) temperature field θ

(in meridian planes) around two spheres whose centers are three radii apart
( = 2/3). The left panel illustrates the results of the four-term approx-
imation whereas the right panel presents those obtained by setting m =
9, considered here as exact results. (c) B0 as a function of 1 − , where the
dashed and dotted lines represent the asymptotic approximations in the lim-
its of  → 0 and  → 1, respectively. And, the inset shows the relative error
of the approximations. (d) The Nusselt number Nu versus the inverse of the
dimensionless separation distance  for a pair of identical spheres. Red, blue,
green, and black lines represent the results corresponding to m = 3, linear
interpolation based on the Nu of single sphere and two touching spheres,
asymptotic behavior in the limit of  → 0, and the converged solution (i.e.,
m  1), respectively. The inset shows the relative difference between the
approximate and exact results.

21
calculating Bm is determining B0 . Figure 2.4c shows the variation of this coefficient as
√
a function of 1 − . As it can be seen, B0 approaches zero as 2  when the distance
√
between the spheres is large and it asymptotes to −∞ as ln(1 − ) / 2 − 0.26085

when the spheres almost touch each other. The asymptotic formulas can be used to

estimate, with reasonable accuracy, the value of B0 over a wide range of  (see the

inset of Fig. 2.4c).

Lastly, our calculations indicate that the Nusselt number of the two-sphere system,

like that of the oblate spheroid changes almost linearly with  (see Fig. 2.4d). Hence,

again, a linear interpolation using the Nu of a single sphere and a pair of touching

spheres can be employed to effectively approximate the curve of Nu versus  (see the

inset of Fig. 2.4d). There exists another approach for approximating this curve based

on the asymptotic behavior of T in the limit  → 0. When the spheres are widely

separated, they see each other as a point source/sink. Thus, to the leading order in

, the average surface temperature and consequently the Nusselt number take the

forms of T = 1 + /2 and Nu = 4 / (2 + ). Perhaps surprisingly, this approximation

outperforms the linear interpolation for gap sizes larger than the radius of the spheres

(compare blue and green lines in the inset of Fig. 2.4d).

2.4 Summary

We derived analytical solutions for the problems of conduction heat transfer from an

isolated oblate spheroid and a pair of spheres. The derivations were carried out in

22
curvilinear coordinate systems, befitting each geometry, using the method of separa-

tion of variables and eigenfunction expansion. While the solutions are in the form of

infinite series, we showed that considering only the first few terms provides excellent

results, often accurate to within a few percent of the exact values. We also found that

the Nusselt number in both cases considered varies rather linearly with respect to the

relevant dimensionless length scale of the problem. As we shall see in the following

chapters, our results can be used to develop perturbation solutions for the problems

of forced convection heat transfer from heated spheroids and bispheres in uniform

laminar flows at small Péclet numbers.

23
Chapter 3: Forced convection heat trans-
fer from a particle at small and
large Péclet numbers

3.1 Introduction

The transport of heat from a particle via an externally driven fluid flow is a phe-

nomenon commonly observed in natural and man-made systems. The ubiquity and

importance of forced convection heat transfer have led a large number of researchers

to study various aspects of this mode of heat transport. Among the investigations

conducted to date, many have focused on incompressible laminar flows. Somewhat

surprisingly, however, the vast majority of theoretical studies in thhat area have been

limited to the case of an object with a known surface temperature distribution (see

e.g. [1, 4, 6, 7, 15, 30–36]), while little attention has been paid to the equally practical

problem of convection heat transfer from a particle with a prescribed surface heat

flux. Of course, in such a problem, the rate of heat transfer (i.e. the surface integral

of the imposed heat flux) is already known, but what is not known, and often sought

after, is the average surface temperature in response to the heat emanating from the

surface of the particle. For example, envisage a scenario where heat is dissipated from

24
an electronic element by blowing air over it. Assume that the rate at which the heat

is generated by the element is known. In this system, the goal is to set the flow speed

such that the average surface temperature stays well below a critical temperature,

e.g. the melting temperature of the element.

It is not immediately obvious as to why cases with prescribed surface heat flux

condition has been overlooked by theoreticians. However, one might surmise that the

inconvenience of applying Neumann, versus Drichlet, boundary conditions is the fac-

tor that has deterred them from considering this category of convection heat transfer

problems. To partly address this deficiency in the literature, in this chapter, we exam-

ine uniform laminar flows past a single hot/cold particle whose surface is presumed

to maintain a constant heat flux. Perturbation theory is used to derive approximate

expressions for the Nusselt number Nu (based on the average surface temperature)

in the limits of small and large Péclet numbers Pe. The accuracy of the formulas for

the specific case of a spheroidal particle in axisymmetric Stokes flow are tested via

comparison with finite-volume numerical simulations.

In the rest of this chapter, we first pose the mathematical problem and describe

the asymptotic solutions in the limits of diffusion- and advection-dominated heat

transport, respectively. Specific results are discussed next and a short summary is

given in the end.

25
3.2 Problem statement

Consider an unbounded steady laminar flow with a divergence-free velocity u past a

stationary particle of arbitrary geometry and characteristic length scale ` (see Fig.

3.1). Let the undisturbed flow field be U∞ = U∞ e, where U∞ = |U∞ | is a con-

stant and e is a unit vector. Suppose that heat is released/absorbed from the surface

of the particle at a constant uniform rate qs and that the temperature vanishes at

infinity. Neglecting viscous dissipation and assuming that the fluid properties are con-

stant, the boundary-value problem that governs the steady-state distribution of the

dimensionless temperature T outside the particle is

Pe u · ∇T = ∇2 T with
(3.1)
n · ∇T = −1 for r ∈ Sp and lim T = 0,
r→∞

where the Péclet number is defined as Pe = ρU∞ cp `/k, with ρ, cp , and k being the

density, specific heat, and thermal conductivity of the fluid, respectively. Also, n is

the unit vector outward normal to the surface of the particle denoted by Sp and r

is the position vector with magnitude r = |r| (see Fig. 3.1). Here, the temperature,

length, and fluid velocity are non-dimensionalized, respectively, by qs `/k, `, and U∞ .

We reiterate that the primary novelty of our study is the consideration of a Neumann

boundary condition on Sp in the energy transport equation (3.1).

We, again, define the average Nusselt number as

Sp
Nu = , (3.2)
2π T

26
 Figure 3.1: Schematic of a stationary particle of arbitrary shape, with
surface Sp and unit outward normal vector n, in a uniform fluid flow. The
dashed line indicates an enclosing boundary in the far field.

where Sp represents the dimensionless surface area of the particle and T is the mean

value of T on Sp . Our primary objective is to develop approximate formulas for the

variation of the Nusselt number (or equivalently T ) as a function of the Péclet number.

To this end, we use the ideas of the reciprocal theorem in conjunction with the method

of matched asymptotic expansions and boundary layer theory to derive expressions

that are valid in the limits of Pe  1 and Pe  1. Details of the calculations are

described in the following sections.

27
3.3 Perturbation solution in the limit of conduction-
dominated heat transport

Suppose that the Péclet number is small, but finite. In this limit, we seek to deter-

mine the O(Pe) contribution to the Nusselt number. It is well-known that a regular

perturbation expansion in terms of Pe is only valid in the vicinity of the particle, i.e.,

regardless of the magnitude of Pe, there exists a domain (r/` & O(Pe−1 )) where the

effect of advection outweighs that of conduction. To remedy this situation, a singu-

lar perturbation expansion is used that involves separate expansions covering regions

close to and far from the particle, i.e., the inner and outer regions, respectively (see,

e.g., [4, 32, 34]). The inner and outer expansions are matched asymptotically in an

intermediate region where both expansions are valid and, together, constitute a per-

turbation solution that is valid in the entire domain.

Specifically, the inner expansion of the temperature field takes the form of

T = T (0) + Pe T (1) + o(Pe), (3.3)

which after substitution into Eq. (3.1) leads to

∇2 T (0) = 0 with n · ∇T (0) = −1 for r ∈ Sp . (3.4)

On the other hand, the outer expansion is expressed as

T̃ = Pe T̃ (1) + o(Pe), (3.5)

28
which results in

˜ T̃ (1) = ∇
e·∇ ˜ 2 T̃ (1) with lim T̃ (1) = 0. (3.6)
r̃→∞

The ∼ overebar in Eqs. (3.5) and (3.6) denotes that the temperature field and the

∇ operator are written in terms of the stretched (rescaled) position vector r̃ = Pe r

with r̃ = |r̃|. The remaining boundary conditions of Eqs. (3.4) and (3.6) are furnished

by enforcing

lim T = lim T̃ (3.7)

r→∞ r̃→0

at every order of Pe.

We first consider the solution for T (0) . To satisfy the matching requirement at the

zeroth-order, T (0) → 0 as r → ∞, which indicates that T (0) is the conduction (Pe =

0) solution of the original problem described in Eq. (3.1). Thus, far from the particle,

the zeroth-order inner solution can be written as (see, e.g., [4])

Sp Sp
T (0) = + O(r−3 ) = Pe + O(Pe−3 ). (3.8)
4πr 4πr̃

This means that the far-field temperature distribution, to the leading order, may

be approximated as the solution of the point source of strength qs Sp `2 applied at

the “heat” center of the particle. Next, given Eq. (3.8) and applying the matching

condition again, we find

Sp
T̃ (1) → as r̃ → 0, (3.9)
4πr̃

which, together with Eq. (3.6), yields (see, e.g., [4])

29
  
(1) Sp 1
T̃ = exp − (r̃ − e · r̃) . (3.10)
4πr̃ 2

Note that T̃ (1) is also the solution of the point source qs Sp `2 .

We now implement the ideas of the reciprocal theorem (see, e.g., [37, 38]). We

multiply the Laplace equation (3.4) by T and Eq. (3.1) by T (0) and then subtract the

resulting equations to obtain, after rearranging,

∇ · T ∇T (0) = ∇ · T (0) ∇T − Pe T (0) u · ∇T.

(3.11)

Integrating this equation over the fluid domain V and using the divergence theorem,

we arrive at

Z Z Z
(0) (0)
T n · ∇T dS = T n · ∇T dS + Pe T (0) u · ∇T dV, (3.12)
Sp Sp V

where the integrands decay sufficiently fast for contributions from surfaces at infinity

to vanish. Application of the boundary conditions on Sp reduces Eq. (3.12) to

Z Z  (0) 
(0) (1)
Sp T = T dS − Pe T (0) u · ∇T dV = Sp T + Pe T + o(Pe), (3.13)
Sp V

(0)
where T is the zeroth-order contribution to the average surface temperatures and
(1)
T is the O(Pe) correction to it. Given Eqs. (3.13), (3.3), and (3.5), we deduce

Z Z  
(1)
− Sp T = T (0)
u · ∇T (0)
dV + e · T̃ (0) ˜ (1)
∇ T̃ − T̃ (0)
dr̃, (3.14)
V R3

where the second integral on the right-hand side is over the entire three-dimensional

real space R3 and T̃ (0) = Sp /4πr̃. The first integral on the right-hand side can be

30
written as
Z Z h i
(0) (0) (0) 2


2 T u · ∇T dV = ∇· T u dV
V V
Z Z (3.15)
(0) 2 (0) 2



=− T n · u dS − T n · u dS,
Sp S∞

where both surface integrals are zero because of the no penetration condition on Sp

and the incompressibility of the flow (i.e., ∇ · u = 0). Additionally,

Z
e· ˜ T̃ (0) dr̃
T̃ (0) ∇
R3
 2 Z ∞ Z π Z 2π
Sp 1 2
=− sin θ cos ϕ dϕ dθ dr̃ = 0. (3.16)
4π 0 r̃ 0 0

Hence, utilizing integration by parts, we find

 
e·r
Z
(1) Sp 1
T =− exp − (r̃ − e · r̃) dr̃. (3.17)
16π 2 R3 r̃4 2

Let ex , ey , and ez be the unit vectors of the Cartesian coordinate system (x, y, z)

located at the center of the particle, see Fig. 3.1. Then, expressing the position vector

in spherical coordinates (r̃, θ, ϕ) as

r̃ = r̃ sin θ cos ϕ ex + r̃ sin θ sin ϕ ey + r̃ cos θ ez

and setting the arbitrary unit vector e to ez (for convenience), the above relation

simplifies to
Z ∞ Z π Z 2π
(1) Sp cos θ sin θ
T =−
16π 2 0 0 0 r̃
  (3.18)
r̃ Sp
× exp − (1 − cos θ) dϕ dθ dr̃ = − ,
2 8π

31
which, in turn, yields
!
Nu(0)
 
(0) Sp (0)
T =T 1 − Pe (0)
+ o(Pe) = T 1 − Pe + o(Pe). (3.19)
8πT 4

Substitution of Eq. (3.19) into Eq. (3.2) gives the final expression for the Nusselt

number:  2
Sp Nu(0)
Nu = = Nu(0) + Pe + o(Pe). (3.20)
2πT 4

Of course, this result is consistent with the one obtained by Leal [34] for the special

case of a spherical particle in a uniform Stokes flow.

There are a couple of important points to make here. First, thanks to the reciprocal

theorem-inspired approach that was adopted, the derivation of Eqs. (3.19) and (3.20)

did not require a detailed knowledge of the velocity field. All we utilized were the

facts that the flow is divergence-free and that it does not penetrate into the particle.

Even no-slip condition was not essential and we assumed no restriction on the flow

Reynolds number defined as Re = ρU∞ `/µ, where µ is the fluid viscosity. Second, and

perhaps equally notable, Eq. (3.20) for the dependence of the Nusselt number on the

Péclet number is identical to the formula obtained by Brenner [4] for an isothermal

particle. Equation (3.20) is also the same as the expression derived by Gupalo et

al. [39] for the Sherwood number (analogue of Nu for mass transfer problems) as

a function of Pe for a particle with a first-order chemical reaction occurring on its

surface. A natural question to ask at this point is that over what range of Pe does

Eq. (3.20) produce accurate results? We will answer this question for the special case

32
of axisymmetric Stokes flow past a spheroids in §2.3.

3.4 Perturbation solution in the limit of advection-

dominated heat transport

Suppose that the Péclet number is large (i.e., Pe  1) and the Reynolds number

is small or moderate. In this limit, the temperature distribution outside the particle

is mainly restricted to a thin layer around the particle (see, e.g., Fig. 3.2), whose

thickness is proportional to Pe−1/3 (see e.g. [6]). This scaling can be deduced by

equating the order of magnitude of the advective and conductive terms in Eq. (3.1),

while assuming a linear velocity profile next to Sp . The restriction on Re is to ensure

that, unlike the temperature field, the velocity field surrounding the particle is not

confined to a narrow region. Similar to the previous section, here also we wish to

develop a two-term asymptotic approximation for the Nusselt number.

Following the standard boundary layer theory (see, e.g., [6]), we assume that the

already known velocity field and the to-be-solved-for temperature distribution are

axisymmetric or two-dimensional and that Sp is smooth. We also adopt the so-called

boundary layer coordinate system (x, y, ϕ), where the first two components measure,

respectively, the distance along and perpendicular to the surface of the particle in

the plane of flow (see Fig. 3.2) and the third component is the azimuthal angle for

axisymmetric and the distance from the x-y plane for two-dimensional problems. The

metric coefficients associated with (x, y, ϕ) are

33
 U∞
T
y
x

q = Constant
x=0

Th
erma
l bounda
ry layer

Figure 3.2: Schematic of uniform flow past an axisymmetric heated object

at high Péclet number. The boundary layer coordinates are represented by
x and y, and the surface heat flux is denoted by q.

hx = 1 + κ(x) y, hy = 1, hϕ = %(x) + α(x) y, (3.21)

where κ is the curvature of Sp . For 2D cases % = 1 and α = 0, whereas for 3D

p
axisymmetric problems % is the rotation radius of Sp and α = ± 1 − (d%/dx)2 is the

cosine of the angle between the axis of rotation and the tangent to Sp . Thus, κ =

−(d2 %/dx2 )/α for axisymmetric cases.

In the vicinity of Sp , Eq. (3.1) can be expressed in terms of the boundary layer

coordinates as
 
ux ∂T uy ∂T
Pe +
hx ∂x hy ∂y
    
1 ∂ hy hϕ ∂T ∂ hx hϕ ∂T
= + with (3.22)
hx hy hϕ ∂x hx ∂x ∂y hy ∂y
∂T
= −1 and lim T = 0,
∂y y=0
y→∞

where

34
 1 ∂ψ 1 ∂ψ
ux = and uy = − (3.23)
hϕ hy ∂y hϕ hx ∂x

are the velocity components in the x and y directions, respectively, with ψ being the

stream function. Remember that ∂T /∂ϕ and uϕ are zero. Since we are interested in

the solution near Sp , it is useful to expand hϕ /hx and ψ about y = 0 as

hϕ /hx = % + (α − %κ) y + %κ2 − ακ y2 + · · · ,

(3.24a)

ψ = ψ2 (x) y2 + ψ3 (x) y3 + · · · , (3.24b)

where
∂ux
2ψ2 = % = % τ0 (3.25)
∂y y=0

with τ0 being the dimensionless shear stress at the surface of the particle. Note that

both ψ and ∂ψ/∂y are zero at y = 0 due to the no slip condition.

Given how the thickness of the boundary layer scales with the Péclet number, we

consider

ỹ = Pe1/3 y

as the stretched coordinate and proceed with rewriting Eq. (3.22) as

h i
L (0) + Pe−1/3 L (1) + O(Pe−2/3 ) T = 0 with
(3.26)
∂T −1/3
= −Pe and lim T = 0,
∂ỹ ỹ=0
ỹ→∞

where

∂ dψ2 ∂ ∂2
L (0) = 2ỹ ψ2 − ỹ2 − % 2, (3.27a)
∂x dx ∂ỹ ∂ỹ

35
  
(1) 2 ∂ dψ3 ∂ ∂ ∂
L = 3ỹ ψ3 − ỹ3 − (α + %κ) ỹ . (3.27b)
∂x dx ∂ỹ ∂ỹ ∂ỹ

Taking the form of Eq. (4.38) into account, it is natural to expand the temperature

field, its mean value on Sp , and the Nusselt number as

 
T = Pe−1/3 T (0) + Pe−1/3 T (1) + O(Pe−1 ), (3.28a)
 (0) (1)

T = Pe−1/3 T + Pe−1/3 T + O(Pe−1 ), (3.28b)

Nu = Pe1/3 Nu(0) + Nu(1) + O(Pe−1/3 ), (3.28c)

where

Z xm −1 Z xm
T = % dx T |ỹ=0 % dx, (3.29a)
0 0

Sp
Nu(0) = (0)
, (3.29b)
2πT
 2 2π T (1)
Nu(1) = − Nu (0)
, (3.29c)
Sp

with x = 0 and xm corresponding to the forward stagnation point of Sp and the

maximum value of x, respectively. We note that, for two dimensional problems, the

quantities described by Eqs. (3.28) and (3.29) belong to the temperature boundary

layer that forms over one side of the stagnation point (see Fig. 3.2), as the energy

equation in the two layers can be treated separately, but in the same manner. Fur-

thermore, as clearly articulated by Acrivos and Goddard [6], the above expansions

are not expected to be fully applicable to all scenarios, such as where there exists

a rear stagnation point on Sp . Thus, caution needs to be exercised when employing

36
these expansions to insure the accuracy of the results.
(0) (1)
Below, we derive T and T , and, by extension, Nu(0) and Nu(1) . Substituting

for T in Eq. (3.27) and requiring the energy equation and its boundary conditions to

hold for all orders of Pe, we arrive at the following parabolic equations for T (0) and

T (1) :

L (0) T (0) = 0 with

(3.30a)
∂T (0) (0) (0)
= −1, lim T = 0, and lim T = 0,
∂ỹ ỹ=0
ỹ→∞ x→0

L (0) T (1) = −L (1) T (0) with

(3.30b)
∂T (1) (1) (1)
= 0, lim T = 0, and lim T = 0,
∂ỹ ỹ=0
ỹ→∞ x→0

We first consider Eq. (3.30a), which is indeed the standard boundary layer approxi-

mation of Eq. (3.27). One might be tempted to develop a similarity solution for this

equation, as done traditionally when Sp is considered isothermal (see, e.g., [40]). How-

ever, such an approach would fail here because of the Neumann boundary condition

at ỹ = 0. Instead, we apply a coordinate transformation of the form (see, e.g., [6])

Z x p p
t= 2ψ2 (s) %(s) ds and z = 2ψ2 (x) ỹ, (3.31)
0

which turns Eq. (3.30a) into

∂T (0) ∂ 2 T (0)
 
2ψ2 % z − = 0 with (3.32)
∂t ∂z2
∂T (0) 1
= −√ , lim T (0) = 0, and lim T (0) = 0.
∂z z=0 2ψ2 z→∞ t→0

37
According to Sutton [41], the solution of the above problem is

t
∂T (0)
Z
(0)
T =− G(t, z; t̂, 0) dt̂, (3.33)
0 ∂z z=0

where

G(t, z; t̂, ẑ) = (3.34)

 √ ! !
 z ẑ z3
+ ẑ3
2 (z ẑ)3/2


exp −
I−1/3
for t > t̂
3 t − t̂ 9 t − t̂ 9 t − t̂

(3.35)



 0 for t ≤ t̂

is the so-called Green’s function that satisfies

∂G ∂ 2 G
z − = δ(t̂, ẑ) with
∂t ∂z2
(3.36)
∂G
= lim G = lim G = 0.
∂z z=0 z→∞ t→t̂

Here, I−1/3 is the modified Bessel function of the first kind and of order −1/3 and

δ is the Dirac delta function. Replacing for G and the heat flux distribution in Eq.

(3.33), we find

Z t −1/2
1
−2/3 
e−z / 9 (t−t̂) dt̂,
3
(0)
T = 1/3 t − t̂ 2ψ̂2 (3.37)
3 Γ(2/3) 0

where Γ is the gamma function and ψ̂2 = ψ2 (x̂). Hence, the leading-order contribution

to the average surface temperature can be obtained via (see Eqs. (3.28b) and (3.29a))

Z xm Z x
(0) 1 %̂
T = 1/3 Rx
2/3 dx̂ % dx, (3.38)
3 Γ(2/3) 0 m % dx 0 0 t − t̂

where %̂ = %(x̂) and t̂ = t(x̂). Remember that all needed to calculate t(x) is the

knowledge of the shear stress distribution on Sp (see Eq. (4.33)).

38
 With the Green’s function known, the solution of Eq. (3.30b) can be written

formally as
Z t Z ∞
(1)
T = f (t̂, ẑ) G(t, z; t̂, ẑ) dẑ dt̂, (3.39)
0 0

where

L (1) T (0)
f (t, z) = −
2ψ2 %
(3.40a)
∂ 2 T (0)
∂T (0)
 
1 3
= −√ Az − Bz + C ,
2ψ2 % ∂z2 ∂z
3ψ3
A(x) = % − (α + %κ) , (3.40b)
2ψ2
!
d ψ3
B(x) = , (3.40c)
dx (2ψ2 )3/2

C(x) = α + %κ. (3.40d)

Substituting for T (0) in Eq. (3.40a), multiplying the result by Eq. (3.34), and carrying

out the ẑ integration (see, e.g., [6, 27]), we reach, after some simplifications,
 2 Z x
(1) 1
1/3
T =− 1/3
t − t̂
3 Γ(2/3) 0
!
x̂
z3
Z
1
× exp −
2  2/3
9 t − t̂

0 ˆ
t − t̂ ˆ
t̂ − t̂
n
h   i
ˆ
× t − t̂ 3A + 9 t̂ − t̂ B̂ˆ (3.41)

ˆ
!
2 t̂ − t̂ z3  
ˆ Ĉ − 2Aˆ

× 1 F1 2; ;
+ t − t̂
3 t − t̂ˆ 9 t − t̂
ˆ
!)
2 t̂ − t̂ z3
×1 F1 1; ; %̂ˆ dx̂
ˆ dx̂,
ˆ 9 t − t̂


3 t − t̂

ˆ = t(x̂),
where t̂ ˆ Aˆ = A(x̂), B̂ = B(x̂), Ĉ = C(x̂), %̂ˆ = %(x̂),
ˆ and 1 F1 is the confluent

39
hypergeometric function of the first kind [25]. The first correction to the mean surface

temperature then becomes (see Eqs. (3.28b) and (3.29a))

 2 Z xm Z x
(1) 1 1
1/3
T = − R xm 1/3
t − t̂
0
% dx 3 Γ(2/3) 0 0
Z x̂ (
1
h
ˆ
×  2  2/3 t − t̂ 3A+
0 ˆ
t − t̂ ˆ
t̂ − t̂
)
  i   
ˆ B̂ + t − t̂
9 t̂ − t̂ ˆ Ĉ − 2Aˆ %̂ˆ dx̂
ˆ dx̂ % dx. (3.42)

3.5 Specific results for spheroids in axisymmetric

Stokes flow

We choose Stokes (zero Reynolds number) flow to exemplify the general results of the

previous two sections because it is representative of many small-scale flows arising in

biology, engineering, and material science. And, we choose spheroids because of their

practical significance. In what follows, we first present the results for the limits of

small and large Péclet numbers and then show the comparison with the full numerical

solution of Eq. (3.1). We acknowledge that the choice of Stokes flow does not affect

the low Pe analytical result. For the sake of completeness, we also provide the results

for spheroids with isothermal surface condition following the works of Brenner [4] and

Acrivos and Goddard [6].

40
3.5.1 Limit of Pe  1

Consider a spheroid of equatorial radius ` and aspect ratio (ratio of polar to equatorial

radius) ε. As discussed in §3.3, when the Péclet number is small, the details of the

flow field are irrelevant for calculating the Nusselt number to the leading order in Pe.

In this limit, all needed is Nu corresponding to Pe = 0. As discussed in §2.2.1 (see

also [42]), the conduction Nusselt number of a spheroid follows

3/2
(0)
[− (1 + ξ02 )] S2p
Nu = ∞ , (3.43)
X Q2m (iξ0 ) 2
8π 2 ξ02 16m (4m + 1) 0 D
m=0
Q2m (iξ0 ) m

where

m
X Γ(m + n + 1/2)
Dm =
n=0
Γ(2n + 1) Γ(2m − 2n + 1) Γ(n − m + 1/2)
−2


2 F1 −1/2, n + 1/2; n + 3/2; −ξ0
× , (3.44)
2n + 1 !
ξ2
q
Sp = 2 π 1 + p 0 2 coth−1 1 + ξ02 , (3.45)
1 + ξ0
ε
ξ0 = √ . (3.46)
1 − ε2

Here, m and n are integers, Q2m are Legendre functions of the second kind [25] with

Q02m (x) = dQ2m (x)/dx, and 2 F1 is the hypergeometric function [25]. The summation in

the denominator of Eq. (3.43) converges very quickly, to the extent that taking only

two terms of the series produces results accurate to within 0.25% of the exact values.

Furthermore, the parameter ξ0 is real for ε ≤ 1 (oblate spheroids) and is imaginary

for ε > 1 (prolate spheroids).

41
 It is interesting to contrast Eq. (3.43) with its counterpart for the problem of

conduction heat transfer from a spheroid with a uniform surface temperature Ts ,

which is (see, e.g., [26, 43])

R √
Sp
n · ∇T dS 2 1 − ε2
(0)
Nu =− = . (3.47)
2π`Ts cos−1 ε

Allowing for imaginary values of the square root and inverse cosine functions, this

expression and also the forthcoming Eq. (3.50) are valid for the entire range of ε. Here,

the Nusselt number is denoted by a different symbol so it is easily distinguished from

its analog for the constant heat flux surface condition. Clearly, Eq. (3.47) is far less

cumbersome than Eq. (3.43). That aside, the substitution of either Eq. (3.43) or (3.47)

in Eq. (3.20) gives the Nusselt number correct to the order of Pe for incompressible

uniform laminar flows past a spheroid.

3.5.2 Limit of Pe  1

Consider the spheroid of §3.5.1 with its axis of revolution coinciding with the z axis

of the Cartesian coordinate system, and suppose that e = ez . To determine the

O(Pe1/3 ) Nusselt number and its first correction, we need to calculate the integrals in

Eqs. (3.38) and (3.42), which involve the functions %(x), t(x), A(x), B(x), and C(x).

The first and last functions depend only on the geometry of the spheroid whereas the

remaining three are additionally dependent on the flow field. It is more convenient to

express these functions in the terms of a new variable η, where

42
 s
dx ε2 + (1 − ε2 ) η 2
= −hη = − and − 1 ≤ η ≤ 1, (3.48)
dη 1 − η2

with η = 1 and η = −1 corresponding to x = 0 and x = xm , respectively. The

stream function for axisymmetric Stokes flow past a spheroid is known (see, e.g.,

[28]). Granted this, we find, following the definitions given in §3.4, that

p
%(η) = 1 − η2, (3.49a)
r
F ε  −1 p 
t(η) = cos η − η 1 − η 2 , (3.49b)
16π
5ε2
 
1 2


A(η) = 1−ε − 2 (3.49c)
2hη ε η + ε2 (1 − η 2 )
r  3/2 
η π 1 3ε2
B(η) =
3hη F (1 − η 2 ) ε [η 2 + ε2 (1 − η 2 )]2

× 1 + 3 ε2 − 1 1 − η 2 − 1 + ε2




(3.49d)
 
ε 1
C(η) = 1+ 2 , (3.49e)
hη η + ε2 (1 − η 2 )

where
3/2
8π (1 − ε2 )
F = √ (3.50)
(1 − 2ε2 ) cos−1 ε + ε 1 − ε2

is the magnitude of the Stokes drag experienced by the spheroid. This quantity is

made dimensionless by µU∞ `. The average surface temperature and its first correction

can now be calculated by replacing the foregoing relations in Eqs. (3.38) and (3.42).
(0) (1)
Substitution of T and T in Eqs. (3.29b) and (3.29c) then yields Nu(0) and Nu(1) .

Equations (3.49) can also be used to determine the O(Pe1/3 ) and O(1) contribu-

tions to the Nusselt number for high-Péclet number heat transfer from an isothermal

43
spheroid in axisymmetric Stokes flow. Availing ourselves of the general results of

Acrivos and Goddard [6], it can be shown, after much reduction, that

1
Nu(0) = (12πF ε)1/3 , (3.51a)
8 Γ(4/3)
(1) 4ε2 + 1
Nu(1) = Nusphere , (3.51b)
5ε

where
( −2/3
4 Γ(2/3) π γ − 12 sin 2γ
Z 
(1) 5
Nusphere = 0.92301 = 1−
3 [Γ(1/3)]2 0 π
1/3 )
γ − 12 sin 2γ

1 − cos γ
× 1− sin2 γ dγ (3.52)
π π

Again, simplicity-wise, the contrast between the above formulas and those for Nu(0)

and Nu(1) is quite remarkable. Equation (3.51a) was also reported by Sehlin [9], though

in a different form. However, to the best of our knowledge, Eq. (3.51b) has not been

reported elsewhere, and is, therefore, another original contribution of this work.

Lastly, we note that the existence of a rear stagnation point on the spheroid in

Stokes flow renders the perturbation expansion described by Eq. (3.28a) invalid in

the vicinity of η = −1. Fortunately, however, the contribution of this singular region

to the Nusselt number is beyond O(1), and, hence, has no effect on Nu(1) and Nu(1)

(see [6]).

44
3.5.3 Comparison with direct numerical solutions

To find out the true limits for which the perturbation calculations of §3.5.1 and §3.5.2

for the Nusselt number are valid, we compare our theoretical results with those ob-

tained from the full numerical solution of Eq. (3.1). A second-order finite volume

method as implemented in OpenFOAM (see, e.g., [22]) is used to perform the numer-

ical calculations. The Stokes equations for the velocity field u is solved first using the

SIMPLE algorithm, and the advection-diffusion equation for the temperature distri-

bution T is treated next. The outer boundary at infinity is modeled as a large cylinder,

whose center coincides with the center of the spheroid. The diameter of the cylinder

is equal to its length, which is 200 times the semi-major axis of the spheroid. 2D

axisymmetric meshes concentrated around Sp are employed to discretize the physical

domain and grid-independence tests are performed by refining the mesh in the en-

tire domain and repeating the simulations. In all cases considered, the computational

grid is chosen such that the change in the results due to the refinement is marginal.

Figure 3.3 shows the results of the numerical calculations for the spheroids of vari-

ous aspect ratios. Interestingly, the plots of Nu versus Pe for constant heat flux and

isothermal boundary conditions are very much alike, not only qualitatively but also

quantitatively (compare solid and dashed lines).

Perhaps, the simplest way to construct a global approximation for the Nusselt

number based on the two-term asymptotic expressions for Pe  1 and Pe  1 (see

Eqs. (3.53) and (3.54)) is

45
 102

ε = 0.2
ε = 0.2
ε = 0.5
ε = 0.5
ε=1
Nusselt number

ε=1
ε=2
ε=2
101
ε=5
ε=5

100
10−3 10−2 10−1 100 101 102 103 104 105
Péclet number
Figure 3.3: Numerically calculated plots of the Nusselt number versus
Péclet number for forced convection heat transfer from spheroids of various
aspect ratios in an axisymmetric uniform Stokes flow subject to constant
heat flux (solid lines) and isothermal (dashed lines) boundary conditions on
the surface of the spheroid.

  2
(0)

 Nul
 (0)
Nul + Pe if Pe < Pec

Nu = 4 , (3.53)

 Pe1/3 Nu(0) + Nu(1)

if Pe ≥ Pec

h h

where Pec is the cut-off Péclet number and the subscripts l and h indicate that

the coefficients belong to the low and high Pe limits, respectively. Alternatively, the

Nusselt number may be approximated over the entire range of the Péclet number by

(see [43, 44])

 !3 1/3
(0) (0)
Nul Nul 
(0)
3
Nu = + + Nuh Pe , (3.54)
2 2

46
which only incorporates the leading order terms in the perturbation expansions of

Nu in the asymptotic limits of Pe. Figure 3.4 presents the percent difference between

the results of Fig. 3.3 and those of Eqs. (3.53) and (3.54). Overall, we see that both

approximations are quite accurate. Specifically, Eq. (3.53) provides more precise pre-

dictions when Pe  1 and Pe  1 whereas better estimates are given by Eq. (3.54)

at intermediate values of Pe, where the approximations deviate the most from the

numerical results. Finally, we note that the kinks in Figs. 3.4a and 3.4c are associated

with transition from low- to high-Péclet-number solutions at Pec , which vary from

0.2 to 1 for the range of aspect ratios considered. In comparison with the plots for

ε = 0.2, 0.5, 1, 2, the transition in the curves corresponding to ε = 5 is smoother as,

in these cases, both asymptotic formulas overestimate the Nusselt number (i.e., ∆ is

positive) in the neighborhood of the cut-off point and no discontinuity exist at Pec .

3.6 Summary

We examined the problem of heat transfer from a stationary hot (or cold) particle im-

mersed in an unbounded fluid in the presence of a uniform background flow. We used

the perturbation theory to derive two-term approximations for the average Nusselt

number in the asymptotic limits of the Péclet number. At small Pe, Nu was approxi-

mated as the summation of the conduction Nu and the O(Pe) correction. We showed

that, for arbitrary particle shapes and flow Reynolds numbers, the correction term is

equal to the square of the zeroth-order term divided by four (see Eq. (3.20)). At high

47
 ε = 0.2 ε = 0.5 ε=1 ε=2 ε=5
15 5
(a) (b)
10
5 0
0
−5

(%)
(%)

−5

∆
∆

−10 −10
−15
−20 −15
10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe
15 5
(c) (d)
10
5 0
0
−5

(%)
(%)

−5
∆
∆

−10 −10
−15
−20 −15
10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe

Figure 3.4: Percent difference ∆ between the numerical and approximate

asymptotic results for the Nusselt number corresponding to (a)–(b) constant
heat flux and (c)–(d) isothermal boundary conditions. (a)–(c) and (b)–(d)
present ∆ versus Pe curves for approximations based on Eqs. (3.53) and
(3.54), respectively. The cut-off Péclet numbers for the aspect ratios ε =
0.2, 0.5, 1, 2, 5 in (a) and (c) are, respectively, Pec = 0.2, 1, 1, 0.35, 0.25 and
Pec = 1, 1, 1, 0.35, 0.25.

Pe, the boundary layer theory was employed to analytically solve for the temperature

distribution within the thermal boundary layer up to O(Pe−2/3 ). The solutions were

used to calculate the O(Pe1/3 ) and O(1) contributions to the Nusselt number. These

calculations were restricted to axisymmetric and two-dimensional problems with low

to moderate Reynolds numbers. It is important to note that the primary novelty of

both low and high Péclet number results are due to the assumption of constant heat

48
flux condition on the surface of the particle.

We exemplified the general perturbation calculations through the problem of ax-

isymmetric Stokes flow past a spheroid. The specific results were then compared

against those obtained from the full numerical solution of the underlying conser-

vation of thermal energy equation. The comparisons confirmed the accuracy of the

approximations for Nu over a wide range of Pe.

49
Chapter 4: Heat transfer from a particle in
laminar flows of a variable ther-
mal conductivity fluid

4.1 Introduction

Heat transfer from a hot or cold object exposed to an external fluid flow is arguably

the most basic form of forced convection heat transfer encountered in industrial pro-

cesses and technological applications. When analyzing this category of heat transfer

problems, in many cases, it is well justified to assume that the fluid properties such

as density ρ, viscosity µ, specific heat cp , and thermal conductivity k are constant. In

general, this assumption renders the energy equation (the partial differential equation

that governs the distribution of the fluid temperature) linear and decoupled from the

Navier-Stokes equations (from which the fluid velocity field is determined).

There exist, however, practical cases where at least one of the fluid properties

cannot realistically be considered constant. For instance, it has been shown that the

effective thermal conductivity of nanofluids increases considerably with the rising

temperature (see, e.g., [45]) or, for liquid metals, k has been found to vary roughly

linearly with the temperature in a wide range of operating conditions (see, e.g., [46]).

50
The energy equation in these situations is no longer linear and, therefore, becomes

more challenging to solve, which is the cost of adding realism to the mathematical

model of the underlying transport phenomenon. Perhaps for this reason, the majority

of textbook examples and classical problems in convective heat transfer from objects

are solved under the assumption that the fluid properties are constant. Hence, it is

of both academic and practical interest to revisit those problems with the goal of

extending their solutions to cases with variable fluid properties. To this end, in this

chapter, we examine the steady-state transfer of heat from a particle via an externally

driven laminar flow, with the premise that the thermal conductivity of the fluid is a

linear function of the temperature.

Building on previous theoretical efforts on the subject (see, e.g., [21, 44, 47–58]),

we derive approximate expressions for the Nusselt number of particles of arbitrary

geometry. The derivations are based on asymptotic and scaling analyses. For com-

pleteness, we consider both constant temperature and uniform heat flux boundary

conditions on the surface of the particle. The results are presented for the full range

of Péclet number Pe.

In the rest of this chapter, we first describe the problem we wish to solve. Then, we

present the solutions for the above-mentioned surface conditions. The validity ranges

of the solutions are discussed next and a brief summary is given in the end.

51
4.2 Problem statement

Consider a laminar, steady-state, incompressible flow with velocity u past a stationary

object of arbitrary geometry and characteristic length `. Suppose that the free-stream

velocity takes the form of U∞ = U∞ e, where U∞ = |U∞ | is a constant and e is a

?
unit vector, and that the temperature approaches a constant value, denoted T∞ , at

large distances from the particle. Also, let the thermal conductivity of the fluid vary

with the dimensionless temperature as k = k0 (1 + βT ), with k0 and β being the far-

field conductivity and a constant parameter, respectively. Then, neglecting viscous

dissipation, the equation that governs the steady-state distribution of T outside the

particle is

Pe u · ∇T = ∇ · [(1 + β T ) ∇T ] , (4.1)

where the Péclet number is defined as Pe = ρU∞ cp `/k0 . The boundary conditions

associated with Eq. (4.1) are

T = 1 for r ∈ Sp and lim T = 0, (4.2)

r→∞

for the case in which the surface of the particle Sp is held at a constant temperature

Ts , and

(1 + β T ) (n · ∇T ) = −1 for r ∈ Sp and lim T = 0 (4.3)

r→∞

for the one where a uniform heat flux qs is applied on Sp . As before, r is the position

vector with magnitude r = |r| and n is the unit vector outward normal to the surface

52
of the particle. Furthermore, the length and fluid velocity are non-dimensionalized,

respectively, by ` and U∞ , whereas the dimensionless temperature is defined as either

T ? − T∞
?
T ? − T∞?
or , (4.4)
Ts? − T∞
? qs `/k0

consistent with the boundary conditions in Eqs. (4.2) and (4.3). The star superscript

is used to denote the dimensional temperature. Also, the average Nusselt numberis

defined as
Z
1+β
NuT = − n · ∇T dS (4.5)
2π Sp

for the problem with a prescribed temperature on the surface of the particle (see Eq.

(4.2)) and as
Sp
NuQ = (4.6)
2πT

for the one with a prescribed heat flux on Sp (see Eq. (4.3)), where Sp represents the

dimensionless surface area of the particle and T is the mean value of T on Sp . In the

following sections, we seek to develop approximate formulas for the variations of NuT

and NuQ with Pe and β.

4.3 Variation of Nusselt number for constant tem-

perature boundary condition

We begin the derivation of NuT by first considering the asymptotic limits of small and

large Péclet numbers and then bridging the gap between the two limits by introducing

53
a smooth transition function. Our approach builds on ideas presented in the classical

works on the topic of heat and mass transfer from an isolated particle in uniform

flows (see, e.g., [4, 6, 21, 34, 44, 49]).

4.3.1 Limit of conduction-dominated heat transport

Suppose conduction is the dominant mode of heat transport, i.e., Pe  1, but finite.

In this limit, an effective approach for dealing with the nonlinearity of Eq. (4.1) is to

make a change of variable from T to θ such that

Z T Z T
1 β 2
θ= k(T) dT = (1 + βT) dT = T + T. (4.7)
k0 0 0 2

This is known as the Kirchhoff transformation (see, e.g., [21, 52]), with the reference

temperature set to T = 0 for convenience. It follows from Eq. (4.7) that

Z
1
NuT = − n · ∇θ dS, (4.8)
2π Sp

where

∇θ
Pe u · √ = ∇2 θ, with
1 + 2β θ

θ = 1 + β/2 for r ∈ Sp and lim θ = 0. (4.9)

r→∞

Our goal is to obtain an asymptotic expression for NuT , and, to that end, we

proceed with a singular perturbation expansion of θ in terms of Pe (see§3.5.1). Again,

we assume that, in the vicinity of the particle, θ can be expressed as

54
 θ = θ(0) + Pe θ(1) + o(Pe), (4.10)

which is known as the inner expansion of θ. Upon substitution of Eq. (4.10) into Eqs.

(4.8) and (4.9), we obtain

NuT = Nu(0)
T
+ Pe Nu(1)
T
+ o(Pe)
"Z Z #
1
=− n · ∇θ(0) dS + Pe n · ∇θ(1) dS + o(Pe), (4.11)
2π Sp Sp

∇2 θ(0) = 0 with θ(0) = 1 + β/2 for r ∈ Sp . (4.12)

Far from the particle, on the other hand, we consider θ to take the form of

θ̃ = Pe θ̃(1) + o(Pe), (4.13)

where the tilde overbar denotes that the transformed temperature field is written as

a function of the rescaled position vector

r̃ = Pe r with r̃ = |r̃|. (4.14)

Rewriting Eq. (4.9) in terms of r̃ and replacing Eq. (4.13) for θ̃, we find

˜ θ̃(1) = ∇
e·∇ ˜ 2 θ̃(1) with lim θ̃(1) = 0, (4.15)
r̃→∞

˜ and ∇
with ∇ ˜ 2 operators representing derivatives with respect to the stretched co-

ordinates. The inner and outer expansions are required to match asymptotically, i.e.,

lim θ = lim θ̃. (4.16)

r→∞ r̃→0

55
Enforcing the above equation at every order of Pe, furnishes the missing boundary

conditions of Eqs. (4.12) and (4.15). Following Brenner [4], the zeroth-order inner

solution away from the particle and the first-order outer solution can be written,

respectively, as

Nu(0) Nu(0)
θ(0) = T
+ O(r−3 ) = Pe T
+ O(Pe−3 ), (4.17)
2r 2r̃
(0)  
NuT 1
θ̃(1) = exp − (r̃ − e · r̃) , (4.18)
2r̃ 2

where r is measured from a proper origin located at the particle’s “heat center”.

Applying the reciprocal theorem (see §3.3), we arrive at

θ(0) ∇θ
Z Z Z
(0) (0)
θ n · ∇θ dS = θ n · ∇θ dS + Pe u· √ dV, (4.19)
Sp Sp V 1 + 2β θ

where, the contributions from surfaces at infinity (denoted by S∞ ) vanish because the

integrands decay faster than dS grows at large distances from the particle. Equation

(4.19) can be simplified to

θ(0) ∇θ
Z
(0) Pe
NuT = NuT + u· √ dV. (4.20)
2 π (1 + β/2) V 1 + 2β θ

by applying the boundary conditions on Sp .

Inspecting Eqs. (4.20), (4.13), (4.11), and (4.10), we determine that

1
Nu(1) =
T
2 π (1 + β/2)
"Z #
θ(0) ∇θ(0)
Z  
× u· p dV + e · ˜ θ̃(1) − θ̃(0) dr̃ ,
θ̃(0) ∇ (4.21)
V 1 + 2β θ (0) R3

56
where the second integral on the right-hand side is over the entire three-dimensional

real space R3 and θ̃(0) = Nu(0)

T
/2r̃. We proceed with evaluating the integral appearing

first on the right-hand side of Eq. (4.21). To that effect, we convert this volume

integral into a pair of surface integrals as

θ(0) ∇θ(0)
Z
u· p dV
V 1 + 2 β θ(0)
Z
1 hp
i
(0) βθ (0) − 1 u dV
= ∇ · 1 + 2 β θ
3 β2 V
"Z
1 p
1 + 2 β θ(0) β θ(0) − 1 n · u dS


=− 2
3β Sp

Z p 
(0)


+ 1 + 2β θ (0) β θ − 1 n · u dS . (4.22)
S∞

Due to no penetration condition, n · u = 0 on Sp . Also,

Z
n · u dS = 0 (4.23)
S∞

because the flow is incompressible (i.e, ∇ · u = 0). Therefore, the values of both

surface integrals in Eq. (4.22) amount to zero, which means that the volume integral

is zero, too. Additionally,

Z
e· ˜ θ̃(0) dr̃
θ̃(0) ∇
R3
 2
Nu(0)
T
Z ∞
1
Z π Z 2π
2
=− sin θ cos ϕ dϕ dθ dr̃ = 0. (4.24)
4 0 r̃ 0 0

Incorporating these results, we find (after algebraic manipulations) that

57
  2
Nu(0)
T
Z
e · r̃

1

(1)
NuT = exp − (r̃ − e · r̃) dr̃
8 π (1 + β/2) R3 r̃4 2
 2
(0)
NuT
= . (4.25)
4 (1 + β/2)

The evaluation of the volume integral over R3 was detailed in §3.3.

Replacing for Nu(1)

T
in Eq. (4.11), we have
 !2 
Nu(0)
T
Pe NuT (0)
NuT = (1 + β/2)  + + o(Pe)
1 + β/2 4 1 + β/2
  2 
Nu(0)
T,0
= (1 + β/2) Nu(0) + Pe + o(Pe)
 
T,0
4

= (1 + β/2) NuT,0 + o(Pe), (4.26)

where NuT,0 and Nu(0)

T,0
are the Nusselt numbers corresponding, respectively, to β =

0 and to β = 0 and Pe = 0. In other words, the former is the constant conductivity

Nusselt number and the latter is the constant conductivity pure conduction Nusselt

number. Remember that Eq. (4.12) describes a linear boundary value problem and,

therefore, changing its boundary condition on Sp from θ = 1 + β/2 to θ = 1 alters

the respective Nusselt number by a factor of (1 + β/2)−1 .

Equation (4.26) indicates that, to the leading order in Pe, NuT is the product of a

β-dependent term and the constant-conductivity Nusselt number, which depends on

Pe and the geometry of the particle. It is worth emphasizing that this relation was

derived with no restriction on the flow Reynolds number, defined as Re = ρU∞ `/µ.

58
Even no-slip condition was not necessary. The only conditions enforced were the flow

incompressibility and no flow penetration into the particle. Lastly, we note that Eq.

(4.26) recovers the classical result of Brenner [4] for β = 0 and generalizes the work

of Polyanin [21] for the special case of Stokes flow (i.e., Re = 0).

4.3.2 Limit of advection-dominated heat transport

Suppose the transport of heat is dominated by advection, i.e., Pe  1. In this limit,

the temperature outside the particle approaches T∞ at short distances from Sp . In

other words, the temperature variations are confined to a narrow layer next to the

boundary of the particle. Here, similar to §4.3.1, we aim for developing an asymptotic

approximation for the Nusselt number.

Assuming that Sp is smooth, we follow Polyanin [49] and adopt a generalized

boundary layer coordinate system (x, y, z), with the corresponding unit vectors ex ,

ey , and ez , and associated metric coefficients hx , hy , and hz . Remember that all

lengths are non-dimensionalized by `. The y component in this system measures the

distance from the surface of the particle and, thus, ey = n at y = 0 (i.e., at Sp ).

The direction of the x coordinate is chosen to be the same as the component of the

fluid velocity vector projected onto the plane normal to ey . The direction of the third

coordinate is then determined by ez = ex × ey .

Equation (4.1), written in terms of the (x, y, z) coordinates takes the form of

     
ux ∂T uy ∂T 1 ∂ hy hz ∂ β 2
Pe + = T+ T
hx ∂x hy ∂y hx hy hz ∂x hx ∂x 2

59
      
∂ hx hz ∂ β 2 ∂ hx hy ∂ β 2
+ T+ T + T+ T
∂y hy ∂y 2 ∂z hz ∂z 2

with T = 1 at y = 0 and lim T = 0. (4.27)

y→∞

In the above equation,

1 ∂Ψ 1 ∂Ψ
ux = and uy = − (4.28)
hy hz ∂y hx hz ∂x

are the velocity components in the x and y directions, respectively, where Ψ is a pseudo

stream function. Note that uz is zero per the definition of the generalized boundary

layer coordinate system. Given that the distribution of temperature is restricted to a

thin region close to Sp , it is convenient to expand the metric coefficients and Ψ about

y = 0 as

hx = hx,0 (x, z) + hx,1 (x, z) y + · · · , (4.29)

hy = 1, (4.30)

hz = hz,0 (x, z) + hz,1 (x, z) y + · · · , (4.31)

1
Ψ = Ψ2 (x, z) y2 + · · · , (4.32)
2

where
 
∂ux
Ψ2 = hz = hz,0 τ0 , (4.33)
∂y y=0

with τ0 being the dimensionless shear stress at the surface of the particle. To ensure

the validity of Eq. (4.32), the Prandtl number Pr = cp µ/k0 is assumed to be large.

60
Needless to say, both Ψ and ∂Ψ/∂y are zero at y = 0, due to the no-slip condition.

The thickness of the temperature boundary layer is known to scale with Pe−1/3 ,

which results from equating the order of magnitude of the advective and conductive

terms within the boundary layer (see, e.g., [6, 34]). Considering this scaling and the

structure of Eq. (4.27), we proceed with expanding T and NuT as

T = T (0) (x, ỹ, z) + O(Pe−1/3 ), (4.34)

NuT = Pe1/3 Nu(0)

T
+ O(1) (4.35)

and with making the following change of variables:

Y = Pe1/3 y, (4.36)
Z x
X= hx,0 hz,0 dx̂. (4.37)
x0

Next, we rewrite Eq. (4.27) to the leading order in Pe as

∂T (0) Y2 ∂Ψ2 ∂T (0) ∂2

 
(0) β (0)
2
Y Ψ2 − = 2 T + T
∂X 2 ∂X ∂Y ∂Y 2

with T (0) = 1 at Y = 0 and lim T (0) = 0. (4.38)

Y→∞

Equation (4.38) can be further simplified to its self-similar form

d2 (0)
 
(0) β (0)
2 2 dT
T + T + 3 η = 0 with
dη 2 2 dη

T (0) = 1 at η = 0 and lim T (0) = 0, (4.39)

η→∞

61
where the new independent variable η is defined, according to the von Mises trans-

formation, via
p  Z X p −1/3
η = Ψ2 9 Ψ2 dX̂ Y. (4.40)
X0

Taking everything into account, the leading-order term in the Nusselt number expan-

sion can be expressed as

∂T (0)
 Z
(0) 1+β
NuT = − dS
2π Sp ∂ỹ ỹ=0

zmax Xmax 2/3

dT (0)
  Z Z
1+β p
=− 3 Ψ2 dX dz
4π dη η=0 zmin Xmin
!−1
dT (0) dT (0)
= (1 + β) Nu(0)
dη η=0 dη η=0, β=0
T,0

dT (0)
 
4
= −Γ (1 + β) Nu(0) , (4.41)
3 dη η=0
T,0

where Nu(0)
T,0
represents the value of Nu(0)
T
corresponding to β = 0. Of course, Eq.

(4.41) reproduces the classical results of Lighthill [1] and Acrivos [40].

Below we calculate the prefactor c:

dT (0)
 
4
c = −Γ (1 + β) , (4.42)
3 dη η=0

where T (0) satisfies Eq. (4.39). We first consider the limit of β  1 and proceed with

a regular perturbation expansion of T (0) as

(0) (0)
T (0) = T0 + β T1 + O(β 2 ). (4.43)

62
Substituting Eq. (4.43) in Eq. (4.39), we find that

(0) (0)
d2 T0 dT
2
+ 3 η 2 0 = 0 with
dη dη

(0) (0)
T0 = 1 at η = 0 and lim T0 =0 (4.44)
η→∞

and

(0) (0)
d2 T1 2 dT1 1 d2  (0) 2
+ 3 η + T = 0 with
dη 2 dη 2 dη 2 0

(0) (0)
T1 = 0 at η = 0 and lim T1 = 0. (4.45)
η→∞

The solution of Eq. (4.44) is

Z η
(0) 1
exp −η̂ 3 dη̂


T0 =1− (4.46)
Γ(4/3) 0

which is derived using the integrating factor method. Applying the same approach to

Eq. (4.45), we obtain (after some mathematical manipulations)

 2
(0)
(0)
dT1   1 − T0
(0) (0)
T1 = Γ(4/3) 1 − T0 −
dη 2
η=0
Z η
1
exp −η̂ 3


−
[Γ(4/3)]2 0
 Z η̂   
(0)
× Γ(4/3) η̂ 3
T0 + η̂ exp −η̂ dη̂ˆ dη̂,
ˆ 3 ˆ 3
(4.47)
0

where the first term on the right-hand side is numerically calculated (by enforcing

the boundary condition at infinity) to be

63
 (0)
dT1
= 0.667. (4.48)
dη
η=0

Given Eqs. (4.43), (4.46), and (4.48), for small β, we can write
 
  (0)
4 dT1  β + O(β 2 )
c = 1 + 1 − Γ
3 dη
η=0

= 1 + 0.404 β + O(β 2 ). (4.49)

Now, suppose β  1, while β  Pe and β  Pr. We realize, through inspecting

Eq. (4.39), that dT (0) /dη scales with β −1/3 in this limit. Informed by this scaling, we

expand T (0) as
(0)
T (0) = T0 + O(β −1/3 ) (4.50)

and introduce

η̃ = β −1/3 η. (4.51)

Replacing for T (0) and η in Eq. (4.39), we then arrive at

(0)
d2  (0) 2
 
2 dT0
T0 + 6 η̃ = 0 with
dη̃ 2 dη̃

(0) (0)
T0 = 1 at η̃ = 0 and lim T1 = 0. (4.52)
η̃→∞

We solve the above nonlinear ordinary differential equation numerically and determine

that

  (0)
4 dT0
c = −Γ β 2/3 + O(β 1/3 )
3 dη̃
η̃=0

64
 = 0.710 β 2/3 + O(β 1/3 ). (4.53)

What is truly surprising, based on the results of Eqs. (4.49) and (4.53), is that the

prefactor c is very well approximated by a single formula (for the entire range of β)

as
 2/3
3β
c≈ 1+ . (4.54)
5

Note that the above expression captures both low- and high-β asymptotes remarkably

well. Hence, we obtain

 2/3
3β
NuT ≈ 1 + NuT,0 + O(1), (4.55)
5

with NuT,0 denoting the Nusselt number for the case of β = 0.

4.3.3 Bridging results for limits of low and high Péclet numbers

The results of the previous two subsections can be described succinctly as

  
NuT β 
≈ 1+ for Pe  1  
NuT,0 2

 NuT
⇒ ≈ (1 + a β)b , (4.56)
NuT

3β
2/3 
 NuT,0
≈ 1+ for Pe  1  
NuT,0 5


where 1/2 / a / 3/5 and 2/3 / b / 1. The functional dependency of these two

parameters on the Péclet number may be formulated as

√
(3/5) Pe + a0
a≈ √ , (4.57)
Pe + 2a0

65
 √
(2/3) Pe + b0
b≈ √ , (4.58)
Pe + b0

where

a0 = 5.78 and b0 = 5.90. (4.59)

The above relations are motivated by the full numerical solution of Eq. (4.1), subject

to the boundary conditions in Eq. (4.2), for several basic particle shapes. The accuracy

of the predictions made by Eqs. (4.56)-(4.59) and the details of the numerical solutions

will be discussed in §4.5. Note that, for some simple geometries, approximate formulas

are available for NuT,0 (see, e.g., §2).

Overall, our asymptotic analyses in this section (summarized by Eq. (4.56)) sug-

gest that the ratio between the Nusselt number and its corresponding value for the

case of constant conductivity (i.e., β = 0) is approximately equal to a term that

is mainly a function of β, while being weakly dependent on Pe. It is important to

emphasize that this result was derived for particles of arbitrary shape under a fairly

general flow condition, namely a laminar, steady-state, incompressible flow. We have

only demanded the Prandtl number to be large when the Péclet number is high.

4.4 Variation of Nusselt number for uniform heat

flux boundary condition

In §4.3, we considered Eq. (4.1) assuming that the surface of the particle is held at a

constant temperature (see Eq. (4.2)). There, the assumption of a Dirichlet boundary

66
condition on Sp allowed us to effectively employ the Kirchhoff and von Mises trans-

formations (for, respectively, Pe  1 and Pe  1) to develop a nearly analytical

formula for NuT . Unfortunately, neither techniques can be directly applied to find the

general form of the Nusselt number when a uniform heat flux is imposed on Sp . That

a Neumann boundary condition is more challenging to deal with analytically than its

Dirichlet counterpart is a known matter (see, e.g., §2 and §3). Despite this difficulty,

we derive an estimate for NuQ with the aid of the following scaling arguments.

4.4.1 Limit of conduction-dominated heat transport

Consider the limit of Pe  1, and, accordingly, an inner expansion of the form

T = T (0) + Pe T (1) + · · · (4.60)

for the temperature, where

1 + β T (0) ∇T (0) = 0 with



∇·

1 + β T (0) n · ∇T (0) = −1 for r ∈ Sp lim T (0) = 0

and (4.61)
r→∞

and

u · ∇T (0) = ∇2 T (1) + β T (0) T (1)

with

n · ∇ T (1) + β T (0) T (1) = 0 for r ∈ Sp .

(4.62)

Applying the Kirchhoff transformation (see Eq. (4.7)), it can be shown that

67
 p
(0) 1 + 2 β θ(0) − 1
T = , (4.63)
β

where

∇2 θ(0) = 0 with

n · ∇θ(0) = −1 for r ∈ Sp and lim θ(0) = 0. (4.64)

r→∞

Additionally, consistent with Eq. (4.60), the Nusselt number can be expressed as

(1)
!
Sp T
NuQ = (0)
1 − Pe (0)
+ ···
2πT T

= Nu(0)
Q
+ Pe Nu(1)
Q
+ ··· , (4.65)

where the overbar indicates average over Sp (see also Eq. (4.71)). The zeroth-order

Nusselt number is determined by the solution of Eq. (4.64), with its dependency on

β obeying Eq. (4.63). How about the dependency of Nu(1)

Q
on β? To find this out, we

examine the limits of small and large β.

Suppose β  1, in which case we can write

Nu(1)
Q
= Nu(1)
Q,0
+ β Nu(1)
Q,1
+ ··· . (4.66)

Since both Pe and β are small, then, we need to only retain Nu(1)
Q,0
in the above

expansion. This means that, to the leading order, Nu(1)

Q
is independent of β. Now,

assume β  1. Equations (4.63) and (4.62) suggest that, in this limit,

1 1
T (0) ∼ √ and T (1) ∼ . (4.67)
β β

68
Hence, we conclude that, again, to the leading order, Nu(1)
Q
is independent of β. Given

the behavior of Nu(1)

Q
in its asymptotic limits, it is reasonable to approximate NuQ

for small Péclet numbers as a summation of two terms: one that only depends on β

and another one that solely changes with Pe.

4.4.2 Limit of advection-dominated heat transport

Consider the limit of Pe  1, where the thermal boundary layer shrinks at a rate

proportional to Pe−1/3 . In this case, the temperature and Nusselt number can be

described as (see, e.g., [59])

T = Pe−1/3 T (0) + Pe−2/3 T (1) + · · · , (4.68)

(1)
!
Sp T
NuQ = (0)
Pe1/3 − (0)
+ ··· . (4.69)
2πT T

where, upon substituting for T in Eqs. (4.1) and (4.3), we have (in generalized bound-

ary layer coordinates)

∂T (0) Y2 ∂Ψ2 ∂T (0) ∂ 2 T (0)

Y Ψ2 − =
∂X 2 ∂X ∂Y ∂Y2
∂T (0)
with = −1 at Y = 0 and lim T (0) = 0. (4.70)
∂Y Y→∞

Equation (4.70) does not depend on β and, in fact, is identical to the one for β = 0.
(0)
Recognizing that T is not a function of β, we deduce from Eq. (4.69) that NuQ for

large Péclet numbers is approximately equal to the sum of a Pe-dependent term and

one that only varies with β.

69
4.4.3 Bridging results for limits of low and high Péclet numbers

The main takeaway point of our scaling analyses in §4.4.1 and §4.4.2 is that NuQ may

be estimated as the summation of two terms, one of which is a function of Pe and

the other one is a function of β. As a unified formula, we, therefore, propose

 
NuQ ≈ NuQ,0 + NucQ − NucQ,0 , (4.71)

where NuQ,0 represents the value of NuQ for β = 0 (i.e., for constant conductivity) and

the superscript c denotes the Nusselt number corresponding to Pe = 0 (i.e, conduction

Nusselt number). The above approximation possesses the required form and has zero

error for the cases of Pe = 0 and β = 0. We will test the validity of Eq. (4.71) in §4.5.

Before we conclude this section, it is worth emphasizing that, for certain geome-

tries, fairly accurate estimates of NuQ,0 exist in the literature (see, e.g., §2). Further-

more, we have discovered empirically that

Sp β p −1
c (0)
NuQ = 1 + 2β θ − 1
2π
q −1 s !−1
Sp β Sp β Sp β
≈ 1 + 2 β θ(0) − 1 = 1+ −1 . (4.72)
2π 2π π NucQ,0

Equation (4.72) is exact for a spherical particle and found to be unexpectedly precise

for other particle shapes such as ellipsoids, cylinders, cones, and cubes.

70
 β = 0.1 β=1 β = 10

Re = 0.1 Re = 1 Re = 10
(a) (b) (c)
20 20 20
∆ (%)

∆ (%)

∆ (%)
10 10 10

0 0 0

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(d) (e) (f )
20 20 20
∆ (%)

∆ (%)

∆ (%)
10 10 10

0 0 0

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(g) (h) (i)
20 20 20
∆ (%)

∆ (%)

∆ (%)
10 10 10

0 0 0

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(j) (k) (l)
20 20 20
∆ (%)

∆ (%)

∆ (%)

10 10 10

0 0 0

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(m) (n) (o)
20 20 20
∆ (%)

∆ (%)

∆ (%)

10 10 10

0 0 0

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe

Figure 4.1: Percent difference ∆ between the results of full numerical sim-
ulations and the predictions of Eq. (4.56) for NuT .

71
 β = 0.1 β=1 β = 10

Re = 0.1 Re = 1 Re = 10
(a) (b) (c)

0 0 0
∆ (%)

∆ (%)

∆ (%)
-10 -10 -10

-20 -20 -20

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(d) (e) (f )

0 0 0
∆ (%)

∆ (%)

∆ (%)
-10 -10 -10

-20 -20 -20

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(g) (h) (i)

0 0 0
∆ (%)

∆ (%)

∆ (%)
-10 -10 -10

-20 -20 -20

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(j) (k) (l)

0 0 0
∆ (%)

∆ (%)

∆ (%)

-10 -10 -10

-20 -20 -20

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe
(m) (n) (o)

0 0 0
∆ (%)

∆ (%)

∆ (%)

-10 -10 -10

-20 -20 -20

10−3 10−1 101 103 105 10−3 10−1 101 103 105 10−3 10−1 101 103 105
Pe Pe Pe

Figure 4.2: Percent difference ∆ between the results of full numerical sim-
ulations and the predictions of Eq. (4.71) for NuQ .

72
4.5 Comparison with direct numerical solutions

In §4.3 and §4.4, we presented approximate formulations for NuT and NuQ via per-

turbation analyses in the asymptotic limits of Pe and β. To give an idea about the

estimation error of the proposed formulas, here, we compare their predictions for

spherical, cubic, and ellipsoidal particles against the results obtained from the full

numerical solution of the problems described in §4.2. A finite-element approach, as

implemented in COMSOL Multiphysics [60, 61], is employed to carry out the compu-

tations. We first solve the steady-state incompressible Navier-Stokes equations for the

fluid flow, and, then, use the calculated velocity field u to compute the solution of Eq.

(4.1) for the temperature distribution T . The flow and advection-diffusion equations

are solved iteratively, while discretized by P2+P1 and quadratic Lagrange schemes,

respectively. The outer boundary at infinity is modeled as a large sphere, whose center

coincides with the center of the particle. Specifically, the diameter of the sphere is

set to 200 times the characteristic length of the particle. We use tetrahedral elements

to mesh the computational domain such that the grid density is the highest near the

particle. Grid-independence studies are performed to ensure that the results change

only marginally if the mesh is globally refined.

Figures 4.1 and 4.2 show the outcome of our calculations for NuT and NuQ , respec-

tively. The results are presented in the form of ∆ versus Pe plots for β = 0.1, 1, 10

and Re = 0.1, 1, 10, where the parameter ∆ is defined as the percent difference be-

tween the directly computed and predicted values of the Nusselt number. Each figure

73
consists of fifteen sub-figures that are organized into five rows and three columns.

The sub-figures in the left, middle, and right columns correspond to Re = 0.1, 1,

and 10, respectively. Also, those in the first and second rows belong to spherical and

cubic particles, whereas the rest are for an ellipsoidal particle of semi-axis lengths a,

b, and c, where b/a = 2/3 and c/a = 1/3. Within the bottom three rows, the plots in

the first, second, and third rows are for flows along the principal axes of the ellipsoid

associated with a, b, and c, respectively. Note that, in calculating the Nusselt and

Péclet numbers the characteristic length ` is set to the radius for the sphere, to half

the side length for the cube, and to the largest semi-axis for the ellipsoid.

Overall, we see that the predictions of Eqs. (4.56) and (4.71) are quite accurate,

with the absolute value of ∆ being less than 16.5% for all the cases considered. The

approximations are more precise for particles with more streamlined shapes. Also, as

expected, the estimations deviate the most from the numerical results when β and Re

are large and Pe is in the intermediate range. Perhaps surprisingly, however, ∆ is very

small for β . O(1), irrespective of its corresponding Reynolds and Péclet numbers.

Another observation that can be made is that ∆ is mostly positive in the plots of

Fig. 4.1 whereas it is mainly negative in those of Fig. 4.2. Note that ∆ is defined such

that it is positive when the predicted Nusselt number overestimates the numerically

calculated one. Lastly, for the same shape, Re, and β, the Péclet number at which

the approximation error is maximum is generally higher for NuQ than it is for NuT .

74
4.6 Summary

We studied the problem of forced convection heat transfer from a particle of ar-

bitrary shape immersed in an unbounded fluid whose thermal conductivity varies

linearly with the temperature. Assuming a uniform free-stream flow, we employed

asymptotic as well as scaling analyses to develop approximate relations for the varia-

tions of the Nusselt number with the Péclet number and the slope of the (normalized)

conductivity-temperature curve. We considered both constant temperature and uni-

form heat flux boundary conditions on the surface of the particle, and discovered

that, for the former, NuT can be estimated as a product of a Pe-dependent term and

one that primarily changes with β. We also found that, for the latter, NuQ may be

approximated as a sum of a Pe-dependent piece and a β-dependent one. In a nutshell,

our derivations offer a straightforward way to estimate the Nusselt number for any

β by just knowing the the Nusselt number corresponding to β = 0, i.e., the constant

conductivity Nusselt number.

We evaluated the generality and accuracy of our formulations by comparing their

predictions for NuT and NuQ with those calculated based on direct numerical solutions

of the governing equations. The comparisons confirmed that the proposed approxima-

tions are valid over a wide range of parameters. More specifically, they demonstrated

that the estimation errors are remarkably low when β . O(1).

75
Chapter 5: Summary and future directions

We theoretically examined three fundamental problems on the topic of heat trans-

fer from isolated objects, namely (i) conduction heat transfer from oblate spheroids

and bispheres, (ii) forced convection heat transfer from an arbitrarily-shaped parti-

cle, and (iii) heat transfer from a single object in laminar flows of a variable thermal

conductivity fluid. We had noticed that although there have been many studies on

mathematical modeling of heat transfer from objects of various shapes (or analogous

problems in mass transfer, electrostatics, etc.), a large number of these investigations

have focused on the isothermal (Dirichlet) boundary condition and a relatively small

number have dealt with the uniform flux (Neumann) boundary condition. In an at-

tempt to partially fill this gap in the literature, in this dissertation, we particularly

considered this former condition in our analyses, and demonstrated that mathemat-

ical derivation under this choice becomes often more challenging compared to the

situation where an isothermal surface condition is considered, which might explain

why the Neumann boundary condition has been frequently overlooked.

We employed the tools of applied mathematics to study the above-mentioned

problems. Specifically, the methods of separation of variables and eigenfuction expan-

sion were used to solve the Laplace equation in curvilinear coordinate systems for

76
the conduction problems. The convection problems, on the other hand, were treated

via a combination of the perturbation and boundary layer theories, reciprocal theo-

rem, and scaling arguments. In all cases, we developed precise, but straightforward,

approximations for the Nusselt number that can be readily used by engineers to de-

sign new thermal systems or better understand the function of the existing ones.

The accuracy of the calculated estimations were confirmed through comparison with

full analytical or numerical solution of the corresponding transport equations. It is

worth emphasizing that the formulas derived for the Nusselt number in this work

are all equally applicable for approximating the Sherwood number in equivalent mass

transfer problems.

Finally, we note that the mathematical techniques presented in §3 and §4 fur-

nish a sound basis for future theoretical investigations on heat transfer from multiple

particles, which can be considered as a natural extension of our study. We can envi-

sion using the method of reflections in addition to the approaches discussed here to

construct asymptotic expressions for the Nusselt number of multi-particle systems.

Furthermore, our work in §4 can be extended to cases where the thermal conductivity

is a non-linear function of the temperature and where other fluid properties such as

viscosity also vary with the temperature.

77
Bibliography

[1] M. J. Lighthill, “Contributions to the theory of heat transfer through a laminar

boundary layer,” Proc. Roy. Soc. (London), vol. 202, no. 1070, pp. 359–377, 1950.

[2] A. Acrivos, “Solution of the laminar boundary layer energy equation at high

prandtl numbers,” Phys. Fluids, vol. 3, no. 4, pp. 657–658, 1960.

[3] A. Acrivos and T. D. Taylor, “Heat and mass transfer from single spheres in

stokes flow,” Phys. Fluids, vol. 5, no. 4, pp. 387–394, 1962.

[4] H. Brenner, “Forced convection heat and mass transfer at small Peclet numbers

from a particle of arbitrary shape,” Chem. Eng. Sci., vol. 18, no. 2, pp. 109–122,

1963.

[5] T. D. Taylor, “Heat transfer from single spheres in a low reynolds number slip

flow,” Phys. Fluids, vol. 6, no. 7, pp. 987–992, 1963.

[6] A. Acrivos and J. D. Goddard, “Asymptotic expansions for laminar forced-

convection heat and mass transfer,” J. Fluid Mech., vol. 23, no. 2, pp. 273–291,

1965.

78
 [7] J. D. Goddard and A. Acrivos, “Asymptotic expansions for laminar forced-

convection heat and mass transfer Part 2. Boundary-layer flows,” J. Fluid Mech.,

vol. 24, no. 2, pp. 339–366, 1966.

[8] N. A. Frankel and A. Acrivos, “Heat and mass transfer from small spheres and

cylinders freely suspended in shear flow,” Phys. Fluids, vol. 11, no. 9, pp. 1913–

1918, 1968.

[9] R. C. Sehlin, “Forced-convection heat and mass transfer at large Péclet num-

bers from an axisymmetric body in laminar flow: Prolate and oblate spheroids,”

Master’s thesis, Carnegie Institute of Technology, 1969.

[10] F. Morrison Jr and S. Griffiths, “On the transient convective transport from a

body of arbitrary shape,” 1981.

[11] A. Polyanin, “An asymptotic analysis of some nonlinear boundary-value prob-

lems of convective mass and heat transfer of reacting particles with the flow,”

Int. J. Heat Mass Transfer, vol. 27, no. 2, pp. 163–189, 1984.

[12] J. Culham, M. Yovanovich, P. Teertstra, C.-S. Wang, G. Refai-Ahmed, and R.-

M. Tain, “Simplified analytical models for forced convection heat transfer from

cuboids of arbitrary shape,” J. Electron. Packag., vol. 123, no. 3, pp. 182–188,

2001.

[13] R. S. Alassar, “Forced convection past an oblate spheroid at low to moderate

reynolds numbers,” 2005.

79
[14] Z.-G. Feng, “Forced heat and mass transfer from a slightly deformed sphere at

small but finite peclet numbers in stokes flow,” J. Heat Transfer, vol. 135, no. 8,

2013.

[15] ——, “Forced heat and mass transfer from a slightly deformed sphere at small

but finite peclet numbers in stokes flow,” J. Heat Trans., vol. 135, no. 8, p.

081702, 2013.

[16] R. Alassar and B. Alminshawy, “Heat conduction from two spheres,” AlChE J.,

vol. 56, no. 9, pp. 2248–2256, 2010.

[17] M. M. Yovanovich, “A general expression for predicting conduction shape fac-

tors,” in Thermophysics and Spacecraft Thermal Control. American Institute

of Aeronautics and Astronautics, 1974, pp. 265–291.

[18] E. Hahne and U. Grigull, “Shape factor and shape resistance for steady multi-

dimensional heat conduction,” Int. J. Heat Mass Transf., vol. 18, pp. 751–767,

1975.

[19] F. P. Incropera, A. S. Lavine, T. L. Bergman, and D. P. DeWitt, Fundamentals

of Heat and Mass Transfer. New York: Wiley, 2011.

[20] H. Brenner and S. Haber, “Symbolic operator solutions of Laplace’s and Stokes’

equations: Part I Laplace’s equation,” Chem. Eng. Commun., vol. 27, no. 5-6,

pp. 283–295, 1984.

80
[21] A. D. Polyanin, “An asymptotic analysis of some nonlinear boundary-value prob-

lems of convective mass and heat transfer of reacting particles with the flow,”

vol. 27, no. 2, pp. 163–189, 1984.

[22] F. Moukalled, L. Mangani, and M. Darwish, The Finite Volume Method in Com-

putational Fluid Dynamics: An Advanced Introduction with OpenFOAM® and

Matlab® . Springer, 2015, vol. 113.

[23] P. M. Morse and H. Feshbach, Methods of Theoretical Physics. New York:

McGraw-Hill, 1953.

[24] P. Moon and D. E. Spencer, Field Theory Handbook: Including Coordinate Sys-

tems, Differential Equations and Their Solutions. New York: Springer-Verlag,

1988.

[25] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions: With

Formulas, Graphs, and Mathematical Tables. New York: Dover, 1972.

[26] R. S. Alassar, “Heat conduction from spheroids,” J. Heat Trans., vol. 121, no. 2,

pp. 497–499, 1999.

[27] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products.

Boston: Academic Press, 2007.

[28] J. Happel and H. Brenner, Low Reynolds Number Hydrodynamics, with Special

81
 Applications to Particulate Media. The Hague, The Netherlands: Martinus

Nijhoff, 1983.

[29] R. S. Alassar and B. J. Alminshawy, “Heat conduction from two spheres,” AIChE

J., vol. 56, no. 9, pp. 2248–2256, 2010.

[30] G. W. Morgan and W. H. Warner, “On heat transfer in laminar boundary layers

at high Prandtl number,” J. Aero. Sci., vol. 23, no. 10, pp. 937–948, 1956.

[31] G. W. Morgan, A. C. Pipkin, and H. W. Warner, “On heat transfer in laminar

boundary-layer flows of liquids having a very small Prandtl number,” J. Aero.

Sci., vol. 25, no. 3, pp. 173–180, 1958.

[32] A. Acrivos and T. D. Taylor, “Heat and mass transfer from single spheres in

Stokes flow,” Phys. Fluids, vol. 5, no. 4, pp. 387–394, 1962.

[33] C. R. Robertson and A. Acrivos, “Low Reynolds number shear flow past a rotat-

ing circular cylinder. Part 2. Heat transfer,” J. Fluid Mech., vol. 40, no. 4, pp.

705–718, 1970.

[34] L. G. Leal, Advanced Transport Phenomena: Fluid Mechanics and Convective

Transport Processes. Cambridge: Cambridge University Press, 2007.

[35] L. M. Relyea and A. S. Khair, “Forced convection heat and mass transfer from

a slender particle,” Chem. Eng. Sci., vol. 174, pp. 285–289, 2017.

82
[36] R. Shah and T. Li, “The thermal and laminar boundary layer flow over prolate

and oblate spheroids,” Int. J. Heat Mass Transf., vol. 121, pp. 607–619, 2018.

[37] V. Vandadi, S. Jafari Kang, and H. Masoud, “Reciprocal theorem for convective

heat and mass transfer from a particle in Stokes and potential flows,” Phys. Rev.

Fluids, vol. 1, no. 2, p. 022001, 2016.

[38] H. Masoud and H. A. Stone, “The reciprocal theorem in fluid dynamics and

transport phenomena,” J. Fluid Mech., vol. 879, p. P1, 2019.

[39] Y. P. Gupalo, Y. S. Ryazantsev, and Y. N. Syskov, “Diffusion to a reacting

particle of arbitrary shape having a liquid flowing around it,” Fluid Dyn., vol. 10,

no. 2, pp. 274–281, 1975.

[40] A. Acrivos, “Solution of the laminar boundary layer energy equation at high

Prandtl numbers,” Phys. Fluids, vol. 3, no. 4, pp. 657–658, 1960.

[41] W. G. L. Sutton, “On the equation of diffusion in a turbulent medium,” Proc.

R. Soc. London A, vol. 182, no. 988, pp. 48–75, 1943.

[42] L. A. Romero, “Low or high Péclet number flow past a prolate spheroid in a

saturated porous medium,” SIAM J. Appl. Math., vol. 55, no. 4, pp. 952–974,

1995.

[43] R. Clift, J. R. Grace, and M. E. Weber, Bubbles, Drops, and Particles. New

York: Academic Press, 1978.

83
[44] A. D. Polyanin and V. V. Dil’man, “The method of asymptotic analogies in the

mass and heat transfer theory and chemical engineering science,” Int. J. Heat

Mass Transfer, vol. 33, no. 6, pp. 1057–1072, 1990.

[45] S. K. Das, N. Putra, P. Thiesen, and W. Roetzel, “Temperature dependence of

thermal conductivity enhancement for nanofluids,” vol. 125, no. 4, pp. 567–574,

2003.

[46] W. M. Kays and M. E. Crawford, Convective Heat and Mass Transfer. New

York: McGraw-Hill, 1980.

[47] M. Arunachalam and N. R. Rajappa, “Forced convection in liquid metals with

variable thermal conductivity and capacity,” vol. 31, no. 1-2, pp. 25–31, 1978.

[48] ——, “Thermal boundary layer in liquid metals with variable thermal conduc-

tivity,” vol. 34, no. 2-3, pp. 179–187, 1978.

[49] A. D. Polyanin, “Method for solution of some non-linear boundary value prob-

lems of a non-stationary diffusion-controlled (thermal) boundary layer,” vol. 25,

no. 4, pp. 471–485, 1982.

[50] T. C. Chiam, “Heat transfer in a fluid with variable thermal conductivity over a

linearly stretching sheet,” vol. 129, no. 1-2, pp. 63–72, 1998.

[51] W. M. Rohsenow, J. P. Hartnett, and Y. I. Cho, Handbook of Heat Transfer.

New York: McGraw-Hill, 1998.

84
[52] S. Kakaç, Y. Yener, and C. P. Naveira-Cotta, Heat Conduction. Boca Raton,

Florida: CRC Press, 2018.

[53] M. A. Ezzat, “State space approach to solids and fluids,” vol. 86, no. 11, pp.

1241–1250, 2008.

[54] M. A. Ezzat, A. S. El-Karamany, and A. A. El-Bary, “On thermo-viscoelasticity

with variable thermal conductivity and fractional-order heat transfer,” vol. 36,

no. 7, pp. 1684–1697, 2015.

[55] M. A. Ezzat and A. A. El-Bary, “Effects of variable thermal conductivity and

fractional order of heat transfer on a perfect conducting infinitely long hollow

cylinder,” vol. 108, pp. 62–69, 2016.

[56] ——, “Effects of variable thermal conductivity on Stokes’ flow of a thermoelectric

fluid with fractional order of heat transfer,” vol. 100, pp. 305–315, 2016.

[57] M. A. Ezzat, “Fractional thermo-viscoelastic response of biological tissue with

variable thermal material properties,” pp. 1–18, 2020.

[58] ——, “The effects of thermal and mechanical material properties on tumorous

tissue during hyperthermia treatment,” vol. 92, p. 102649, 2020.

[59] E. Dehdashti and H. Masoud, “Forced convection heat transfer from a particle

at small and large Peclet numbers,” vol. 142, no. 6, 2020.

85
[60] W. B. J. Zimmerman, Multiphysics Modeling with Finite Element Methods. Sin-

gapore: World Scientific Publishing Company, 2006.

[61] D. W. Pepper and J. C. Heinrich, The Finite Element Method: Basic Concepts

and Applications with MATLAB, MAPLE, and COMSOL. New York: CRC

Press, 2017.

86