Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

Contents lists available at ScienceDirect

Journal of Non-Newtonian Fluid Mechanics

journal homepage: www.elsevier.com/locate/jnnfm

A particle distribution function approach to the equations of continuum T

mechanics in Cartesian, cylindrical and spherical coordinates: Newtonian
and non-Newtonian fluids
⁎
R.R. Huilgol ,a, G.H.R. Kefayatib
a
College of Science and Engineering, Flinders University of South Australia, GPO Box 2100, Adelaide, SA 5001, Australia
b
Department of Mechanical Engineering, The Hong Kong Polytechnic University, Kowloon, Hong Kong SAR, China

A R T I C L E I N F O A B S T R A C T

Keywords: The evolution equations for the particle distribution functions are written in a divergence form applicable in
Lattice Boltzmann equation three dimensions. From this set, it is shown that the continuity equation and the equations of motion are satisfied
BGK approximation in Cartesian, cylindrical and spherical coordinates for all fluids when additional source terms are added to the
Particle distribution function equations of evolution in the latter two coordinate systems. If the body forces are present, a new set of source
Continuum mechanics
functions is required in each coordinate system and these are described as well. Next, the energy equation is
derived by using a separate set of particle distribution functions. Modifications of the relevant equations to be
applicable to incompressible fluids is described. The incorporation of boundary conditions and the description of
the numerical scheme for the simulation of the flows employing the new approach is given. Validation results
obtained through the modelling of a mixed convection flow of a Bingham fluid in a lid-driven square cavity, and
the steady flow of a Bingham fluid in a pipe of square cross-section are presented. Next, using the cylindrical
coordinate version of the evolution equations, numerical modelling of the steady flow of a Bingham fluid and the
Herschel–Bulkley fluid in a pipe of circular cross-section have been performed and compared with the simulation
results using the augmented Lagrangian method as well as the analytical solutions for the velocity field and the
flow rate. Finally, some comments on the theoretical differences between the present approach and the existing
formulations regarding Lattice Boltzmann Equations are offered.

1. Introduction properties of all the particles which lie within the range of the kernel.
For example, using Monaghan’s cubic spline kernel [3], the temperature
In the numerical modelling of the flows of Newtonian and non- at position x depends on the temperatures of all the particles within a
Newtonian fluids, the finite element method (FEM) is the preferred radial distance 2h of x.
option. However, it has become increasingly clear that FEM requires a The contributions of each particle to a property are weighted ac-
lot of CPU and is not fast enough. This outcome has led to the devel- cording to their distance from the particle of interest, and their density.
opment of two different numerical schemes: the first is based on the Mathematically, this is governed by the kernel function, W. The kernel
smoothed-particle hydrodynamics (SPH) method which has been ap- functions commonly used include the Gaussian function and the cubic
plied to non-Newtonian moulding flow by Fan et al. [1], and to the spline. The latter function is exactly zero for particles further away than
simulation of solid bodies suspended in a shear flow of an Oldroyd-B two smoothing lengths, unlike the Gaussian, where there is a small
fluid by Hashemi et al. [2]; and the second scheme is the Lattice contribution at any finite distance away. That is, the cubic spline has
Boltzmann equation (LBE) and its variant, namely the particle dis- the advantage of saving computational effort by not including the re-
tribution function method. latively minor contributions from distant particles. While the ad-
Briefly, SPH works by dividing the fluid domain into a set of discrete vantages of SPH are many, one drawback over grid-based techniques is
elements, also referred to as particles. These particles have a spatial the need for large numbers of particles to produce simulations of
distance, known as the “smoothing length” h, over which their prop- equivalent resolution.
erties are “smoothed” by a kernel function. This means that the physical In contrast with SPH, the particle distribution function method
quantity of any particle can be obtained by summing the relevant employed here depends on fifteen (resp. twenty two) particles only to

⁎
Corresponding author.
E-mail addresses: Raj.Huilgol@flinders.edu.au (R.R. Huilgol), [email protected] (G.H.R. Kefayati).

https://doi.org/10.1016/j.jnnfm.2017.10.004
Received 26 June 2017; Received in revised form 12 October 2017; Accepted 16 October 2017
Available online 11 December 2017
0377-0257/ © 2017 Elsevier B.V. All rights reserved.
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

mimic the continuity equation and the equations of motion (resp. the augmented Lagrangian method as well as the analytical solutions for
full set of continuity equation, the equations of motion and the energy the velocity field and the flow rate. Finally, in the Concluding Remarks,
equation) applicable to the motion of a fluid. No averaging process as in some comments on the theoretical differences between the present
SPH is needed. As summarised by Huilgol and Kefayati [4], it is possible approach and the existing formulations regarding Lattice Boltzmann
to derive the continuity equation and Cauchy’s equations of motion for Equations are offered.
a compressible Newtonian fluid, when one uses the Bhatnagar–- It has to be noted that additional source functions have been em-
Gross–Krook (BGK) approximation. In LBE, the derivations are based on ployed in modelling non-swirling, axisymmetric flows of in-
expanding the particle distribution functions as a Taylor series in u, compressible Newtonian fluids in cylindrical coordinates; for example,
retaining terms up to order |u|2, where u is the macroscopic velocity, see the review by Huang and Lu [11], which has been extended to every
with coefficients depending on the relaxation time and the grid spacing axisymmetric flow in the review by Zhang et al. [12]. In these two
and the time step. Hence, when one considers incompressible New- reviews, the derivations are based on expanding the particle distribu-
tonian fluids, it is not surprising that the kinematic viscosity ν is re- tion functions as a Taylor series in u, retaining terms up to order |u|2,
laxation time and grid-dependent, for it is given by [5] where u is the macroscopic velocity. In the derivation employed here,
the expansion occurs as a Taylor series in the microscopic particle ve-
(2τ − 1) (▵x )2
ν= · , locity ξα, retaining terms up to |ξα|2. As shown in Appendix A, it is not
6 ▵t (1.1)
easy to compare the two methods of derivation. Leaving this aside, in
where τ is the relaxation time, △x is the grid size and △t is the time the reviews [11,12], one finds that two time partial derivative time
step. Clearly, these restrictions on the viscosity make it difficult to scales may be needed in some models. Or, two different sets of source
model the flows of non-Newtonian, incompressible fluids. For a com- terms are needed in recovering the Navier–Stokes equations. In the
plete description of these matters, see Huilgol and Kefayati [4]. In ad- present work, only one relaxation time is needed and only one set of
dition to the problems arising from Eq. (1.1), Fu et al. [6–8], So et al. source functions is needed in cylindrical and spherical coordinates and
[9] and Kam et al. [10] have reiterated the difficulties in employing the none in Cartesian coordinates. Additionally, there are other differences
LBE formulation in solving Navier–Stokes equations. between the source functions in [11,12] and those needed here to
In order to overcome these inherent problems, new models for the model the flows in cylindrical coordinates; see Section 3.2.
particle distribution functions are needed. In the Finite Difference Finally, it has to be emphasised here that the evolution equations
Lattice Boltzmann Method (FDLBM) due to Fu and So [6], the particle deal with all fluids, compressible of incompressible, and apply to every
distribution function leads to the conservation of mass and the equa- coordinate system employed regularly in fluid dynamics. Some of the
tions of motion applicable to incompressible fluids, when the flows are dyadic products and related material used in the body of the paper,
assumed to occur in a two dimensional setting underpinned by a D2Q9 along with direct verifications that the equations of motion are cor-
lattice, using Cartesian coordinates only. These results were refined by rectly derived in cylindrical and spherical coordinates, are provided in
Huilgol and Kefayati [4] through the use of vector analysis and linear Appendix B.
algebra.
Next, in the Thermal Finite Difference Discrete Flux Method
2. Particle distribution function
(TFDDFM) proposed by Fu et al. [8], their approach was extended to
three dimensional problems using a D3Q15 lattice. The resulting
First of all, the evolution equation is usually written as follows:
equations are capable of incorporating body forces; moreover, a new set
of particle distribution functions was employed to obtain the balance of ∂fα 1
energy equation. These results were reformulated in [4] using simple + ξα ·∇fα = − (f − f αeq ),
∂t ɛτ α (2.1)
results from vector analysis and linear algebra, once again. The im-
portant point to note is that the previous restrictions on the pressure where ξα are the lattice vectors, modelled after the D3Q15 lattice and
and the viscosity are eliminated in these derivations [4,6,8], meaning defined in Table 1 below, and ε is a small parameter to be prescribed
that one is free to choose a constitutive equation. That is, one can model when numerical simulations are considered, τ is the collision relaxation
a Newtonian fluid, or power law fluids, or viscoelastic and viscoplastic time, and f αeq is the equilibrium distribution function. One notes that in
fluids. However, the derivations in [4,6,8] are suitable for Cartesian numerical modelling, the product ετ is replaced by a suitably chosen
coordinates only. non-dimensionalised time step.
In the present work, we unify and extend the derivation of the In order to consider three dimensional flows, whether they be in
conservation of mass equation and Cauchy’s equations of motion to Cartesian or cylindrical or spherical coordinates, the evolution Eq. (2.1)
Cartesian, cylindrical and spherical coordinates using linear algebra, has to be modified. That is, the evolution equation for the particle
vector and dyadic analysis, in a 3D format. While the methodology
follows some of the features of the earlier work [4], the evolution Table 1
Subscripts and microscopic velocities.
equation for the particle distribution function is written in a divergence
form with the addition of a new set of source functions; see Section 2. In Value of α ξα/σ
Section 3, it is shown that this new set of particle distribution functions
delivers the relevant equations applicable to flows in Cartesian, cy- 0 0
1 e1
lindrical and spherical coordinates. In Section 4, additional source
2 e2
functions to incorporate body forces are described and, in Section 5, the 3 − e1
energy equation is derived employing a new set of particle distribution 4 − e2
functions. In Section 6, the simple modifications necessary for the 5 e3
6 − e3
equations to be applicable to incompressible fluids are listed. Next, in
7 e1 + e2 + e3
Section 7 some comments on the incorporation of Dirichlet and stress 8 − e1 + e2 + e3
boundary conditions, and validation results, based on the works which 9 − e1 − e2 + e3
have appeared, are presented. Further, in Section 8, using the cylind- 10 e1 − e2 + e3
rical coordinate version of the evolution equations, numerical model- 11 e1 + e2 − e3
12 − e1 + e2 − e3
ling of the steady flow of a Bingham fluid and the Herschel–Bulkley
13 − e1 − e2 − e3
fluid in a pipe of circular cross-section have been performed have been 14 e1 − e2 − e3
performed and compared with the simulation results using the

120
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

distribution function fα(x, ξα, t) is expressed in a divergence form, dif- 14 14

ferent from that in Eq. (2.1) above, with additional source functions: ∑ fα ξα ⊗ ξα = ∑ f αeq ξα ⊗ ξα + O (ɛ) = M + O (ɛ),
α=0 α=0 (2.13)
3
∂fα 1
+ ∇ ·(fα ξα ) + ∑ Sα ξα ·ej = − (fα − f αeq ), α = 0, 1, ⋯, 14. 14 14
∂t j=1
ɛτ (2.2) ∑ (fα ξα )·∇ξα = ∑ (f αeq ξα )·∇ξα + O (ɛ).
α=0 α=0 (2.14)
The source functions Sα have to be defined and depend on the chosen
The results in (2.11)–(2.14) play a crucial role later on when we turn to
coordinate system. This matter will be discussed later on in Section 3
the derivations of equations of continuum mechanics in Section 3.
where it is shown that at most six, Sα , α = 1, …, 6, are required.
Returning to (2.4), and noting that α varies from 0 to 14, there are three
To proceed, one assumes that fα has the following Chapman–Enskog
sets of separate coefficients: Aα, Bα, and Cα. However, when α = 0, only the
expansion:
coefficient A0 is required. Regarding the others, one makes the assumption
fα = f αeq + ɛf α(1) + ɛ2f α(2) + O (ɛ3). that the coefficients having the same energy shell of the lattice velocities are
(2.3)
equal. Thus, Aα = A1 , α = 1, …, 6; Aα = A2 , α = 7, …, 14.
Next, we expand the equilibrium lattice function f αeq as a quadratic Since B0 and C0 do not affect the value of f 0eq in (2.4), they can be put
[4,6,8] in the particle velocity ξα: equal to zero. By the just mentioned assumption about certain coefficients
being equal, we see that Bα = B1, α = 1, …, 6; Bα = B2 , α = 7, …, 14.
f αeq = Aα + Bα ·ξα + Cα : (ξα ⊗ ξα ), (2.4) Similarly, Cα = C1, α = 1, …, 6; Cα = C2 , α = 7, …, 14. Keeping in mind
where Aα is a scalar, Bα is a three-dimensional vector and Cα is a 3 × 3 that the matrices are symmetric, there are twenty one independent quan-
symmetric matrix. These coefficients depend on the density ρ, the ve- tities: three scalars in A0, A1, A2; six components of the vectors B1, B2; six
locity field u, and the physical components of the stress tensor Tij, along components each in C1, C2. Looking at Eqs. (2.5)–(2.7), it is obvious that the
with the lattice speed σ; see Eqs. (2.15)–(2.22) below. number of available constraints is ten only. Thus, there is a lot of latitude in
The constraints on the functions f αeq , f α(1) , f α(2) , are given by choosing the variables, as remarked by Fu et al. [6,8], and reiterated by
Huilgol and Kefayati [4].
14
In Table 1, σ has the dimension of speed; it is usually called the
∑ f αeq = ρ,
lattice speed in the literature. Here, it is a parameter which determines
α=0 (2.5)
the stability of the numerical solution and has to be varied at each step.
14 Thus, σ ≠ △x/△t, where △x and △t are the lattice displacement and
∑ f αeq ξα = ρu, u = u e1 + v e 2 + w e 3 , time step, respectively. This matter will be discussed further in
α=0 (2.6) Section 7.
Returning to Eq. (2.4), we shall assume, in analogy with Fu and So
and
[8], and Huilgol and Kefayati [4], that
14
∑ f αeq ξα ⊗ ξα = M, ρu2 T + T22 + T33
A0 = ρ − + 11 , A1 = A2 = 0,
α=0 (2.7) σ2 σ2 (2.15)

14
and
∑ f α(n) = 0, n ≥ 1, ρu
(2.8) B1 = , B2 = 0.
α=0 2σ 2 (2.16)

14 Next,
∑ f α(n) ξα = 0, n ≥ 1.
α=0 (2.9) ⎡C11 0 0 ⎤
C1 = ⎢ 0 C22 0 ⎥.
Note that the set of base vectors {ei}, i = 1, 2, 3, stands for {i, j, k} in ⎢0 0 C33 ⎥
⎣ ⎦ (2.17)
Cartesian coordinates; {er, eθ, ez} in cylindrical coordinates; in sphe-
rical coordinates, it stands for {er, eθ, eϕ}, where θ is the polar angle Here,
and ϕ is the azimuthal angle. In terms of the base vectors, we can de- 1 1
C11 = (ρu2 − T11), C22 = (ρv 2 − T22),
scribe the microscopic velocities and these are listed in Table 1. 2σ 4 2σ 4 (2.18)
In physical components, M has the matrix form
1
2 C33 = (ρw 2 − T33).
⎡ ρ u − T11 ρ uv − T12 ρ uw − T13 ⎤ 2σ 4 (2.19)
M = ⎢ ρ uv − T12 ρ v 2 − T22 ρ vw − T23 ⎥.
⎢ ⎥ Finally,
2
⎣ ρ uw − T13 ρ vw − T23 ρ w − T33 ⎦ (2.10)
⎡ 0 C12 C13 ⎤
In the above set, ρ is the density, and C2 = ⎢C21 0 C23 ⎥,
T11, T22, T33, T12 = T21, T13 = T31, T23 = T32 are the stresses in a fluid which ⎢C C 0 ⎥
⎣ 31 32 ⎦ (2.20)
can be defined through any relevant constitutive relation. The tensor M
contains terms similar to Reynolds’ stresses, viz., ρu2, and the compo- where
nents of the stress tensor T. 1 1
C12 = C21 = (ρuv − T12), C13 = C31 = (ρuw − T13),
At this juncture, it is important to note that while fα = f αeq + O (ɛ), 16σ 4 16σ 4
for each α = 0, …, 14, it follows from (2.3), (2.8)and (2.9) that:
(2.21)
14 14
1
∑ fα = ∑ f αeq + O (ɛ3) = ρ + O (ɛ3), C23 = (ρvw − T23).
α=0 α=0 (2.11) 16σ 4 (2.22)
Since B1 = Bα , α = 1, …, 6, we obtain
14 14
∑ fα ξα = ∑ f αeq ξα + O (ɛ3) = ρu + O (ɛ3), ρ
B1·ξ1 = − B3·ξ3·= u,
α=0 α=0 (2.12) 2σ (2.23)

121
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

ρ 14
B2 ·ξ2 = − B4 ·ξ4 = v,
2σ (2.24) ∑ f αeq ξα ⊗ ξα = M,
α=0 (2.45)
ρ
B5·ξ5 = − B6·ξ6 = w. or, Eqs. (2.7) and (2.10) have been verified.
2σ (2.25)
Finally, one observes that:
Moreover, B0 = B2 = Bα = 0, α = 7, …, 14, means that
f 0eq ξ0 = 0, (2.46)
Bα ·ξα = 0, α = 0, 7, …, 14. (2.26)
ρ
Noting that C1 = Cα , α = 1, …, 6 and using Eqs. (B.1)–(B.3) in f1eq ξ1 = σ ⎡ u + σ 2C11⎤ e1,
Appendix B, we obtain ⎣ 2σ ⎦ (2.47)

C1: ξ1 ⊗ ξ1 = C3: ξ3 ⊗ ξ3 = σ 2C11, (2.27) ρ

f3eq ξ3 = − σ ⎡− u + σ 2C11⎤ e1,
2
⎣ σ ⎦ (2.48)
C2 : ξ2 ⊗ ξ2 = C4 : ξ4 ⊗ ξ4 = σ 2C22, (2.28)
ρ
f 2eq ξ2 = σ ⎡ v + σ 2C22 ⎤ e 2,
C5: ξ5 ⊗ ξ5 = C6: ξ6 ⊗ ξ6 = σ 2C33. (2.29) ⎣ 2σ ⎦ (2.49)

From the assumption that C2 = Cα , α = 7, …, 14, we find from ρ

(B.4)–(B.7) that f 4eq ξ4 = − σ ⎡− v + σ 2C22 ⎤ e 2,
⎣ 2σ ⎦ (2.50)
C7 : ξ7 ⊗ ξ7 = C13: ξ13 ⊗ ξ13
ρ
= 2σ 2 (C12 + C13 + C23), (2.30) f5eq ξ5 = σ ⎡ w + σ 2C33 ⎤ e3,
⎣ 2σ ⎦ (2.51)
C8: ξ8 ⊗ ξ8 = C14 : ξ14 ⊗ ξ14 ρ
= 2σ 2 (−C12 − C13 + C23), f 6eq ξ6 = − σ ⎡− w + σ 2C33 ⎤ e3,
(2.31) ⎣ 2σ ⎦ (2.52)

C9: ξ9 ⊗ ξ9 = C11: ξ11 ⊗ ξ11 f 7eq ξ7 = − f13eq ξ13, (2.53)

= 2σ 2 (C12 − C13 − C23), (2.32)
f 8eq ξ8 = − f14eq ξ14, (2.54)
C10 : ξ10 ⊗ ξ10 = C12 : ξ12 ⊗ ξ12
= 2σ 2 (−C12 + C13 − C23). (2.33) f 9eq ξ9 = − f11eq ξ11, (2.55)
Finally, we can see that f10eq ξ10 = − f12eq ξ12. (2.56)
f 0eq = A0 , (2.34)
Hence,
ρ ρ 14
f1eq = u + σ 2C11, f3eq = − u + σ 2C11,
2σ 2σ (2.35) ∑ f αeq ξα = ρu,
α=0 (2.57)
ρ ρ
f 2eq = v + σ 2C22, f 4eq = − v + σ 2C22, which shows that Eq. (2.6) has been verified.
2σ 2σ (2.36)
At this juncture, it is worth emphasising that while each one of the
ρ ρ functions, f αeq , depends on the parameter σ, the sums in Eqs. (2.42) and
f5eq = w + σ 2C33, f 6eq = − w + σ 2C33,
2σ 2σ (2.37) (2.57) are independent of this lattice speed. These comments apply to
the sums of fα as well.
f 7eq = f13eq = 2σ 2 (C12 + C13 + C23), (2.38)

f 8eq = f14 = 2σ 2 (−C12 − C13 + C23), 3. Equations of continuum mechanics

(2.39)

f 9eq = f11eq = 2σ 2 (C12 − C13 − C23), Let us return to the evolution Eq. (2.2) and sum it over
(2.40)
α = 0, 1, ⋯, 14, using the expansion (2.3). We obtain
f10eq = f12eq = 2σ 2 (−C12 + C13 − C23). (2.41) 14 14 14 3 14
∂⎛
∑ f ⎞ + ∇·⎛⎜ ∑ fα ξα ⎞⎟ + ∑ ∑ Sα ξα·ej = − 1 ⎛⎜ ∑ f α(1) ⎞
+ ɛf α(2) ⎟
Using Eqs. (2.34)–(2.42), one finds that ∂t ⎜ α = 0 α ⎟ τ
⎝ ⎠ ⎝ α=0 ⎠ α=0 j=1 ⎝ α=0 ⎠
14
+ O (ɛ2).
∑ f αeq = ρ,
α=0 (2.42) (3.1)
which means that Eq. (2.5) is satisfied. Now, at this juncture, it is sufficient to choose the source functions Sα
Next, using Eqs. (2.35)–(2.38) and (B.1)–(B.3), it is easy to obtain such that except for S1, S2, S3, S4, S5, S6, the remaining functions are all
the following matrix: zero. Moreover, we demand that
S1 = S3, S2 = S4, S5 = S6. (3.2)
6 ⎡C11 0 0 ⎤
∑ f αeq ξα ⊗ ξα = 2σ 4 ⎢ 0 C22 0 ⎥. On noting that the vectors ξ1 = −ξ3 = σe1, ξ2
α=0 ⎢0 0 C33 ⎥
⎣ ⎦ (2.43) = −ξ4 = σ e 2, ξ5 = −ξ6 = σ e3, and that in the set {e1, e2, e3}, each
vector is orthogonal to the other two, one finds that
From (B.8)–(B.11) in Appendix B, it follows that:
14 3
14 ⎡ 0 C12 C13 ⎤ ∑ ∑ Sα ξα·ej = 0.
∑ f αeq ξα ⊗ ξα = 16σ 4 ⎢C12 0 C23 ⎥. α=0 j=1 (3.3)
α=7 ⎢C C 0 ⎥
⎣ 13 23 ⎦ (2.44)
In a fluid, we see that the constraints in Eqs. (2.5), (2.6), (2.8), (2.11),
In sum, one finds that (2.12) and Eqs. (3.1)–(3.3) lead to the following result:

122
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

∂ρ 14 3
+ ∇ ·(ρ u) = O (ɛ2), ⎡ ∂fα ⎤
∂t (3.4) ∑ ⎢ ∂t + ∇ ·(fα ξα ) + ∑ Sα ξα·ej⎥ ξα = ρa − ∇·T = O (ɛ).
α=0 ⎣ j=1 ⎦ (3.14)
or the continuity equation for a fluid is satisfied to within O(ε2). Since
this equation deals with a scalar, it holds true in all coordinate systems. Note that Eq. (3.14) is a vector equation and is valid in all coordinate
However, the divergence operator varies from one system to another systems. However, one has to determine the source terms in the Car-
and this has to be taken into account in writing down the equation in tesian, cylindrical and spherical coordinates to satisfy Eq. (3.13).
terms of the physical components of the velocity field.
The derivation of the equations of motion requires a different choice 3.1. Cartesian coordinates
of the source functions Sα , α = 1, …, 6, for the Cartesian, cylindrical and
spherical coordinates. Thus, it is convenient to discuss the three co- The derivation of the equations of motion in Cartesian coordinates
ordinate systems separately. has been given by Fu et al. [6] and Huilgol and Kefayati [4]. It is in-
As a preamble, in order to obtain the equations of motion, one has to cluded here for the sake of completeness and also because source
multiply the evolution Eq. (2.2) by each lattice vector ξα and take its functions were not considered by them. First of all, the base vectors
sum. This results in the following: {e1, e 2, e3} = {i, j , k} in Cartesian coordinates and, consequently, each
lattice vector ξα is constant in magnitude and direction and has zero
14 3 14
⎡ ∂fα ⎤ 1 divergence. Hence, ∇ξα = 0. Thus
∑ ⎢ ∂t + ∇ ·(fα ξα ) + ∑ Sα ξα·ej⎥ ξα = − ∑ (f − f αeq ) ξα .
ɛτ α
α=0 ⎣ j=1 ⎦ α=0 (3.5) ∇ ·(fα ξα ) = ∇fα ·ξα . (3.15)
Since each lattice vector is independent of time t, it follows from Eq. Consequently, in Cartesian coordinates, one does not need any
(2.12) that: source terms. That is, we put
14 14 Sα = 0, α = 1, …, 6, (3.16)
∂fα ∂⎛ ∂ (ρ u) ∂ρ ∂u
∑ ∂t
ξα = ∑ f ξ ⎞ = ∂t + O (ɛ3) = ∂t u + ρ ∂t + O (ɛ3).
∂t ⎜ α = 0 α α ⎟
α=0 ⎝ ⎠ as well. Consequently, Eq. (3.14) reduces to the following:
(3.6) ρa − ∇ ·T = O (ɛ). (3.17)
Next, one notes that if A and B are vectors, the product A⊗B is a
That is, we have obtained Cauchy’s equations of motion in Cartesian
dyad. From dyadic analysis, it is known that [13]
coordinates to within O(ε).
∇ ·(A ⊗ B) = (∇ ·A) B + A ·(∇B). (3.7)
3.2. Cylindrical coordinates
Here, we can let A = fα ξα , B = ξα . Thus, from Eqs. (2.13)–(2.14), it
follows that:
Here, the base vectors {e1, e2, e3} correspond to {er, eθ, ez}, re-
14 14 14 spectively. From dyadic analysis [13],
⎡ ⎤
∑ [∇ ·(fα ξα )] ξα = ∇ ·⎢ ∑ fα ξα ⊗ ξα⎥ − ∑ (fα ξα )·∇ξα
1 1
α=0 ⎣ α=0 ⎦ α=0 ∇e r = eθ eθ , e θ · ∇e r = eθ ,
14 r r (3.18)
= ∇·M − ∑ (fα ξα )·∇ξα + O (ɛ),
α=0 (3.8) 1 1
∇e θ = − eθ e r , e θ · ∇e θ = − e r .
r r (3.19)
where the matrix M is defined in Eq. (2.10). We note that M can be
rewritten as From Table 3 in Appendix B, we find that

M = ρ u ⊗ u − T. (3.9) 2σ 4
f 2eq ξ2·∇ξ2 + f 4eq ξ4·∇ξ4 = − C22 er ,
r (3.20)
Hence,
4σ 4
∇ ·M = [∇ ·(ρ u)] u + ρ u ·(∇u) − ∇ ·T. (3.10) f 7eq ξ7·∇ξ7 + f13eq ξ13·∇ξ13 = (C12 + C13 + C23)(eθ − er ),
r (3.21)
Combining Eqs. (3.5), (3.6), (3.8) and (3.10) and using the continuity
Eq. (3.4), it follows that: 4σ 4
f 8eq ξ8·∇ξ8 + f14eq ξ14·∇ξ14 = − (−C12 − C13 + C23)(eθ + er ),
r (3.22)
14 3 14 3
⎡ ∂fα
∑ ⎢ ∂t + ∇ ·(fα ξα ) + ∑ Sα ξα·ej⎤⎥ ξα = ρa − ∇·T + ∑ ∑ (Sα ξα·ej) ξα 4σ 4
α=0 j=1 α == 0 j = 1
f 9eq ξ9·∇ξ9 + f11eq ξ11·∇ξ11 = (C12 − C13 − C23)(eθ − er ),
⎣ ⎦ r (3.23)
14
− ∑ (fα ξα )·∇ξα = O (ɛ), 4σ 4
α=0 f10eq ξ10·∇ξ10 + f12eq ξ12·∇ξ12 = − (−C12 + C13 − C23)(eθ + er ).
r
(3.11) (3.24)
where a is the acceleration vector: Hence, on appealing to Eq. (2.14), it follows that:
∂u 14 14
a= + u · ∇u .
∂t (3.12) ∑ fα ξα·∇ξα = ∑ f αeq ξα·∇ξα + O (ɛ)
α=0 α=0
It is now obvious that the source terms, Sα, have to be chosen so that 2σ 4 16σ 4
= − C22 er + C12 eθ + O (ɛ)
14 3 14 r r
∑ ∑ (Sα ξα·ej) ξα − ∑ (fα ξα )·∇ξα = O (ɛ).
=
1 1
(Tθθ − ρv 2) er + (ρuv − Trθ ) eθ + O (ɛ).
α == 0 j = 1 α=0 (3.13) r r (3.25)

Hence, Eqs. (3.5), (3.11) and (3.13) deliver Cauchy’s equation of mo- It is now easy to see that an appropriate choice of the scalars,
tion to within O(ε): Sα , α = 1, …, 6, to satisfy Eq. (3.13) is given by

123
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

Table 2 Table 4
Comparisons between the present results, ALM [32] and analytical solutions. Subscripts, microscopic velocities and dyads.

Comparisons Velocity of plug Flow rate Sp Computing Value of α ξα σ eα · (∇eα)

time (s)
0 0 0
Bingham Fluid 1 er 0
Present study (50 × 50) 9.07 × 10−2 0.18846 0.4008 37 2 eθ − (1/ r ) er
Present study (70 × 70) 9.05 × 10−2 0.18818 0.4006 1102 3 − er 0
Present study (90 × 90) 9.03 × 10−2 0.18782 0.4004 2458 4 − eθ − (1/ r ) er
Present study 0.18761 0.4002 3485 5 eϕ − (1/ r ) er − (cot θ/ r ) eθ
9.02 × 10−2
(100 × 100) 6 − eϕ − (1/ r ) er − (cot θ/ r ) eθ
Present study 9.02 × 10−2 0.18761 0.4002 4958 7 er + eθ + e ϕ eθ·∇ (er + eθ) + eϕ·∇ (er + eθ + eϕ)
(110 × 110) 8 − er + eθ + e ϕ eθ·∇ (−er + eθ) + eϕ·∇ (−er + eθ + eϕ)
ALM [32] 9.01 × 10−2 0.186 0.4002 – 9 − er − eθ + e ϕ − eθ·∇ (−er − eθ) + eϕ·∇ (−er − eθ + eϕ)
Analytical solutions 9 × 10−2 0.187 0.4 – 10 er − eθ + e ϕ eθ·∇ (er − eθ) + eϕ·∇ (er − eθ + eϕ)
[32]
11 er + eθ − e ϕ eθ·∇ (er + eθ) − eϕ·∇ (er + eθ − eϕ)
Herschel–Bulkley Fluid,
n = 0.75 12 − er + eθ − e ϕ eθ·∇ (−er + eθ) − eϕ·∇ (−er + eθ − eϕ)
Present study (50 × 50) 5.21 × 10−2 0.11447 0.4061 37 13 − er − eθ − e ϕ − eθ·∇ (−er − eθ) − eϕ·∇ (−er − eθ − eϕ)
Present study (70 × 70) 5.20 × 10−2 0.11374 0.4052 1102 14 er − eθ − e ϕ − eθ·∇ (er − eθ) − eϕ·∇ (er − eθ − eϕ)
Present study (90 × 90) 5.19 × 10−2 0.11312 0.4038 2458
Present study 5.18 × 10−2 0.11282 0.4021 3485
(100 × 100) sets of source functions are needed; see Eqs. (8) and (9). In the Li et al.
Present study 5.18 × 10−2 0.11282 0.4021 4958 model [14] quoted by Zhang et al. [12], a term proportional to feq, LBEu/
(110 × 110)
ALM [32] 0.112 0.4050 –
r appears; see Eqs. (6) and (7) therein. As seen above, this term does not
5.16 × 10−2
Analytical solutions 0.1119 0.4 – appear. Secondly, in the Zhou model [15] which appears in [12], two
5.166 × 10−2
[32] source functions are needed; see Eqs. (27a) - (27b). Such complexities
are avoided and shown to be unnecessary in the approach employed
here.
1
S1 = S3 = (Tθθ − ρv 2),
2σ 2r
1 3.3. Spherical coordinates
S2 = S4 = (ρuv − Trθ ), S5 = S6 = 0.
2σ 2r (3.26)
Here, the base vectors {e1, e2, e3} correspond to {er, eθ, eϕ}, re-
In sum, Cauchy’s equations of motion (3.14) are satisfied in cylindrical spectively, where r is the radial distance from the centre of the co-
coordinates. For a direct verification, see Eqs. (B.13)–(B.19) and ordinate system, θ is the polar angle and ϕ is the azimuthal angle. From
Table 3. dyadic analysis [13],
We shall now offer some comments regarding the differences be-
1 1 1
tween the source functions in Eq. (3.26) and those in [11,12]. First of ∇e r = (eθ eθ + eϕ eϕ), e θ · ∇e r = eθ , e ϕ · ∇e r = eϕ ,
r r r (3.28)
all, Huang and Lu [11] examine non-swirling axisymmetric flows in
Newtonian fluids only, while Zhang et al. [12] extend the derivations to
1
axisymmetric flows of such fluids. In contrast, in the work presented ∇e θ = (−eθ er + cot θ eϕ eϕ),
r (3.29)
here, the derivations apply to all fluids, whether be compressible or
incompressible, Newtonian or non-Newtonian. Restricting the attention 1 cot θ
to Newtonian fluids only, it is clear from Eq. (3.26) that two products, eθ ·∇eθ = − er , e ϕ · ∇e θ = eϕ ,
r r (3.30)
viz., ρuv and ρv2, and two stresses
1
u 1 ∂v ⎤ 1 ∂u ∂v v ∇ eϕ = − (eϕ er + cot θ eϕ eθ),
Tθθ = 2η ⎡ + , Trθ = η ⎡ + − ⎤, r (3.31)
⎣ r r ∂θ ⎦ ⎣ r ∂θ ∂r r ⎦ (3.27)

where η is the viscosity, are required. In contrast, in [11] two separate 1 cot θ
e ϕ · ∇ eϕ = − er − eθ .
r r (3.32)
Table 3
From Table 4 in Appendix B, one finds that
Subscripts, microscopic velocities and dyads.
2σ 4
Value of α ξα/σ eα · (∇eα) f 2eq ξ2·∇ξ2 + f 4eq ξ4·∇ξ4 = − C22 er ,
r (3.33)
0 0 0
1 er 0 2σ 4 2σ 4
2 eθ − (1/ r ) er f5eq ξ5·∇ξ5 + f 6eq ξ6·∇ξ6 = − C33 er − C33 cot θeθ ,
r r (3.34)
3 − er 0
4 − eθ − (1/ r ) er
4σ 4
5 ez 0 f 7eq ξ7·∇ξ7 + f13eq ξ13·∇ξ13 = (C12 + C13 + C23)(−2er + (1 − cot θ) eθ
6 − ez 0 r
7 er + eθ + e z eθ·∇ (er + eθ) = (1/ r ) eθ − (1/ r ) er + (1 + cot θ) eϕ),
8 − er + eθ + e z eθ·∇ (−er + eθ) = −(1/ r ) eθ − (1/ r ) er
9 − er − eθ + e z − eθ·∇ (−er − eθ) = (1/ r ) eθ − (1/ r ) er (3.35)
10 er − eθ + e z − eθ·∇ (er − eθ) = −(1/ r ) eθ − (1/ r ) er
4σ 4
11 er + eθ − e z eθ·∇ (er + eθ) = (1/ r ) eθ − (1/ r ) er f 8eq ξ8·∇ξ8 + f14eq ξ14·∇ξ14 = (−C12 − C13 + C23)(−2er − (1 + cot θ) eθ
12 − er + eθ − e z eθ·∇ (−er + eθ) = −(1/ r ) eθ − (1/ r ) er r
13 − er − eθ − e z − eθ·∇ (−er − eθ) = (1/ r ) eθ − (1/ r ) er − (1 − cot θ) eϕ),
14 er − eθ − e z − eθ·∇ (er − eθ) = −(1/ r ) eθ − (1/ r ) er
(3.36)

124
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

4σ 4 1 1
f 9eq ξ9·∇ξ9 + f11eq ξ11·∇ξ11 = (C12 − C13 − C23)(−2er + (1 − cot θ) eθ F2 = − F4 = ρ b·ξ2 = ρb2 ,
r 2σ 2 2σ (4.6)
− (1 + cot θ) eϕ),
1 1
F5 = − F6 = ρ b·ξ5 = ρb3,
(3.37) 2σ 2 2σ (4.7)

4σ 4 Fα = 0, α = 0, 7, ⋯, 14. (4.8)
f10eq ξ10·∇ξ10 + f12eq ξ12·∇ξ12 = (−C12 + C13 − C23)(−2er − (1 + cot θ) eθ
r
+ (1 − cot θ) eϕ).
5. The energy equation
(3.38)
Hence, In order to obtain the energy equation, an internal energy dis-
14 14
tribution function gα is introduced and it is assumed to satisfy an evo-
∑ fα ξα·∇ξα = ∑ f αeq ξα·∇ξα + O (ɛ) lution equation similar to that for fα. Thus,
α=0 α=0 ∂gα 1
2σ 4 2σ 4 + ∇ ·(gα ξα ) − Gα = − (g − gαeq), α = 0, ⋯, 6,
= − (C22 + C33) er + (16C12 − C33) eθ ∂t ɛτ α (5.1)
r r
16σ 4 where the source functions Gα will be specified below. Note that only
+ (1 + cot θ) C23 eϕ + O (ɛ) seven distribution functions are required.
r
1 2 In Eq. (5.1), each equilibrium distribution function gαeq is linear in
= [Tθθ + Tϕϕ − ρ (v 2 + w 2)] er + [ρuv − Trθ + Tϕϕ the lattice velocity vector as follows:
r r
1
2
− ρw ] eθ + (1 + cot θ)[ρvw − Tθϕ] eϕ + O (ɛ). gαeq = Dα + ξα ·Eα . (5.2)
r
(3.39) And,
Now, in order to satisfy Eq. (3.13), one can select the source func- gα = gαeq + ɛgα(1) + ɛ2gα(2) + O (ɛ3), (5.3)
tions Sα , α = 1, …, 6, as follows:
with the requirement that
1
S1 = S3 = [Tθθ + Tϕϕ − ρ (v 2 + w 2)], 6
2σ 2r (3.40)
∑ gα(n) = 0, n ≥ 1.
α=0 (5.4)
1
S2 = S4 = 2 [ρuv − Trθ + Tϕϕ − ρw 2],
σr (3.41) The energy equation applicable to a continuous media is given by

1 de 1
S5 = S6 = (1 + cot θ)[ρvw − Tθϕ]. ρ = T: A1 − ∇ ·q + ρr ,
2σ 2r (3.42) dt 2 (5.5)

In conclusion, Cauchy’s equations of motion have now been obtained where e is the internal energy, A1 is the first Rivlin–Ericksen tensor
from Eq. (3.14) in spherical coordinates. For a direct verification, see [16], q is the heat efflux vector and r is the external supply. The deri-
Eqs. (B.20)–(B.26) and Table 4. vation of this equation can be found in standard books on continuum
mechanics and rheology; for example, see Tanner [17].
In order to derive the above equation from the internal energy
4. Body forces
distribution function, it is assumed that et is the total energy given by
the sum of the internal and kinetic energies [4,8], i.e.,
We shall now turn to the problem of incorporating body force terms
into the equations of motion requires that the evolution Eq. (2.2) has to 1 2
et = e + u .
be modified through the addition of a new set of source functions, Fα, as 2 (5.6)
follows [4,8]:
Next, the following consistency relations must hold:
3
∂fα 1 6
+ ∇ ·(fα ξα ) + ∑ Sα ξα ·ej − Fα = − (fα − f αeq ), α = 0, 1, …, 14.
∂t j=1
ɛτ ∑ gαeq = ρet ,
α=0 (5.7)
(4.1)
6
This new set of source functions Fα must be such that ∑ gαeq ξα = ρet u − Tu + q,
14 α=0 (5.8)
∑ Fα = 0, 6
α=0 (4.2)
∑ Gα = ρ b·u − ρr .
which guarantees that the mass conservation Eq. (3.4) is unchanged. To α=0 (5.9)
derive the equations of motion in the form
One way of satisfying the above is to assume, as before, that the scalars
ρa − ∇ ·T − ρ b = O (ɛ), (4.3) Dα are such that Dα = D1, α = 1, …, 6 and set

where b is the body force per unit mass, one requires that D0 = ρet , D1 = 0. (5.10)
14 In addition, it is assumed that the vectors Eα are defined through
∑ Fα ξα = ρb1 e1 + ρb2 e2 + ρb3 e3. E0 = 0, Eα = E1, α = 1, …, 6, where
α=0 (4.4)
1
A simple choice for the set Fα is [4,8]: E1 = (ρet u − Tu + q).
2σ 2 (5.11)
1 1 Next, the scalars Gα describe the power produced by the body forces
F1 = − F3 = ρ b·ξ1 = ρb1,
2σ 2 2σ (4.5) and the external supply. They can be chosen such that G0 = 0, and

125
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

1 1 To emphasise further the advantages of the methodology adopted

Gα = ρ b·(ξα ⊗ ξα ) u − ρr (ξα ⊗ ξα ): 1, α = 1, …, 6.
2σ 2 6σ 2 (5.12) here, one notes that there is no need to appeal to special equations, such
showing that Eq. (5.9) holds. as bounce-back or the Zou–He linear equations [18] to determine the
Finally, summing Eq. (5.1) over all α, and using Eqs. (5.7)–(5.9), one boundary values of fα or gα. Recently, Kam et al. [10] have discussed
finds that this matter at length. They have pointed out that the bounce back rule
used to model solid-fluid boundary conditions leads to mass leakage at
6
∂ (ρet ) ⎛ ⎞ a solid boundary; consequently, mass conservation is not satisfied. In
+ ∇ ·⎜ ∑ gαeq ξα ⎟ − ρ b·u + ρr = O (ɛ).
∂t (5.13) the approach adopted by Kam et al. [10], Huilgol and Kefayati [4] and
⎝ α=0 ⎠
here, physical boundary conditions are specified and mass leakage does
Next, not exist. Additionally, the bounce back rule fails for flat as well as
∇ ·(ρet u) = ((∇ρ) et + ρ (∇et ))·u + ρet ∇ ·u . (5.14) curved surfaces; in the case of curved surface, the error is compounded
by the fact that the lattice grid points cannot be made to coincide with
Finally, using indicial notation for convenience, noting the symmetry of the solid boundary [10].
the stress tensor, and that of the Rivlin–Ericksen tensor(A1 )ij = ui, j + uj, i , Even though σ is called “the lattice speed” in the literature, it is not
one finds that related to the lattice spacing and the time step. Instead, its value affects
1 numerical stability and has to be chosen to suit the problem at hand.
(Tij uj ), i = Tij, i uj + Tij ui, j = Tji, j ui + Tij (A1 )ij .
2 (5.15) For instance, in their solution to the Stokes second problem, Fu and So
[6] recommend varying it at every iteration step and the selected value
Thus,
is determined through the Courant–Friedrichs–Lewy (CFL) condition
1 for stability [19,20]. Using the procedure in [6], the CFL condition for
∇ ·(Tu) = (∇ ·T)·u + T: A1.
2 (5.16) stability in the steady flow of a Bingham fluid in a pipe of square cross-
Combining all of the above and using the continuity Eq. (3.4), we ob- section has been derived by Kefayati and Huilgol [23]. Thus, the value
tain of σ is altered and changes in each iteration. For an application of this
procedure to the mixed convection problem of a Bingham fluid in a lid
de 1 driven cavity, see Kefayati and Huilgol [22]. More recently, Kam et al.
ρ + (ρa − ∇ ·T − ρ b)·u − T: A1 + ∇ ·q − ρr = O (ɛ),
dt 2 (5.17) [10] have imposed upper and lower limits on the choice of σ. That is,
which reduces to Eq. (5.5) within O(ε), when one appeals to the there is no unique choice of σ and it has to be selected with care.
equations of motion (3.14). Since Eq. (5.5) is a scalar, it is valid in all In numerical modelling of the flows of incompressible fluids
coordinate systems although the terms in the stress power, (T: A1)/2, [6,8,22,23], the iteration and recovery of the pressure field is similar to
and the divergence (∇ · q) vary from one to the other. the SIMPLE method of Patankar and Spalding [24,25]. As is well
known, the SIMPLE method is a guess-and-correct procedure for the
6. Incompressible fluids calculation of the pressure field. In each iteration, the velocity field is
obtained from the first guessed pressure field. Next, using the corrected
If the fluid is incompressible, the density ρ is constant and the velocity field, it is possible to find the corrected pressure and this
conservation of mass Eq. (3.4) becomes process continues till a very small or zero mass residual is obtained,
since the zero mass residual demonstrates that the divergence of the
∇ ·u = O (ɛ). (6.1) velocity vector field is zero. In the numerical scheme used in
The total stress tensor T = −p1 + τ , where p is the pressure and τ is the [6,8,22,23], the mass residual is the difference between the sum of the
extra stress tensor. Hence, Eq. (3.14) has the form distribution functions and the fixed density; see Eq. (2.5). Thus, the
correction of the pressure field and the subsequent correction of the
ρa + ∇p − ∇ ·τ − ρ b = O (ɛ). (6.2)
velocity field continues till a small or zero difference exists between this
The energy equation, Eq. (5.5), is now satisfied to O(ε) sum and the density.
At this juncture, it is essential to mention the numerical procedure
de 1
ρ − τ : A1 + ∇ ·q − ρr = O (ɛ), adopted by Fu et al. [6–8], So et al. [9], Kam et al. [10], and Kefayati
dt 2 (6.3)
and Huilgol [22,23] in the simulation of problems expressed in Carte-
since the incompressibility condition is equivalent to A1: 1 = 0. sian coordinates. Typically, the main equation of the discrete particle
In order to model incompressible fluids, Eq. (2.5) imposes a strict distribution function is solved by the splitting method of Toro [26].
condition on the sum of the particle distribution functions. This can That is, the set of Eqs. (2.2), which are the same as (2.1), are separated
been used in numerical modelling to ensure that the sum is constant to into two parts. The first one is the streaming section which is written for
within an acceptable error term. This matter will be discussed in the use in Cartesian coordinates as
next section.
∂fα
+ ξα ·∇fα = 0, α = 0, 1, …, 14,
∂t (7.1)
7. Numerical modelling: Boundary conditions and validation
results when body forces are ignored. Eqs. (7.1) are solved with the method of
Lax and Wendroff [27] with second order differences in space and time
The incorporation of boundary conditions in solving the evolution variables. The second step is the collision stage
equations is quite easy, for the coefficients Aα, Bα, Cα in determining the
∂fα 1
particle distribution functions fα, and the coefficients Dα, Eα in the so- =− (f (x, t ) − f αeq (x, t )), α = 0, 1, …, 14.
∂t ɛτ α (7.2)
lution of gα, are defined in terms of the macroscopic variables which
include the velocity field and the stress tensor. That is, whether the This can be solved by the Euler method, with ɛτ = ▵t , the time step.
boundary conditions are of the Dirichlet type or depend upon surface Note that we need to non-dimensionalise the equations before applying
tractions, the velocity field u and the total stress tensor T appear in Eq. the numerical method. For a full description of the methodology, see
(2.10). Hence, whenever a node appears on the surface, the relevant [23] for example.
boundary values appear in the evolution equations; it does not matter We shall now turn to some validation results from two flows to
whether the fluid is compressible or incompressible. Thus, the in- demonstrate that evolution equations for the particle distribution
corporation of the boundary conditions is no different from that in FEM. functions deliver results which are equivalent to those in the existing

126
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

literature. The first one is concerned with the benchmark problem of based on the assumption that the fluid remains at rest or moves as a
modelling the mixed convection flow of a Bingham fluid in a lid-driven rigid body if the second invariant of the extra stress tensor τ is less than
cavity [22]. Both the Papanastasiou model and the Bingham fluid were or equal to the yield stress τy. If the second invariant exceeds the yield
used to simulate this flow at different Reynolds numbers, using the stress, the material flows like a fluid. This second invariant is defined
particle distribution function method. An extensive mesh testing pro- through
cedure was conducted to guarantee a grid independent solution. Seven
II (τ ) = (1/ 2 ) τ : τ . (8.1)
different mesh combinations were explored for the case of Re = 1000
and Bn = 10. The present code was tested for grid independence by Hence, using the first Rivlin–Ericksen tensor A1 [16], the rigidity con-
calculating the u and v velocities in the middle of the cavity. It was dition is given by
confirmed that the grid size (250–250) ensures a grid independent so-
lution as shown in Table 1 [22]. The accuracy of the applied code in a A1 = 0, II (τ ) ≤ τy. (8.2)
lid-driven cavity using the Papanastasiou model was validated through When the magnitude of the extra stress tensor exceeds the yield
a comparison with the work of Neofytou [28], who chose a smaller grid stress, one defines τ as a function of the tensor A1 leading to the fol-
size (200–200). The results are shown in Fig. 4 [22], where the u and v lowing relation:
velocities profiles demonstrate the accuracy of the present simulation
τy
for Bn = 1 and Re = 100. For the Bingham fluid, the results were τ = η A1 + A1, II (τ ) > τy, II (A1) = (1/ 2 ) A1: A1 .
shown to be similar to those of Dean and Glowinski [29], who used the II (A1) (8.3)
operator-splitting method. The above results demonstrate that the Under Dirichlet boundary conditions for the velocity field which
particle distribution function method can be employed to simulate the applies to the flow in a pipe of circular cross-section, a new constitutive
lid-driven flow of Bingham fluids in a cavity using the Papanstasiou equation for a Bingham fluid fully equivalent to the original form can
model as well as the Bingham model. be used. This idea is due to Duvaut and Lions [35] and Glowinski [36]
Next, the problem of natural convection of a Bingham fluid in a and the constitutive equation takes the form
square cavity was also examined by Kefayati and Huilgol [22]. The
results were compared with those of Huilgol and Kefayati [21], who τ = η A1 + 2 τy Λ, 1: Λ = 0, (8.4)
obtained their results with the operator-splitting method; it was found
where one may call the second order, symmetric, tensor Λ the visco-
that the agreement between the two sets of results was excellent; see
plasticity constraint tensor ([32]). Note that the traceless condition
Figs. 5 and 6 in [22]. The two sets of agreements described above led to
1: Λ = 0 has been imposed on this tensor so that the stress tensor τ
the investigation of the mixed convection flow of a Bingham fluid in a
satisfies the condition tr τ = 0, leading to a precise definition of pres-
lid-driven cavity [22]. Both the Papanastasiou model and the Bingham
sure p [33,34]. In order to demarcate the flow field into unyielded/
fluid were included in this study and, as expected, differences arise
yielded zones, one requires that the tensor Λ meet the following con-
between the predictions of the Papanstasiou and the Bingham fluid as
ditions:
the Bingham number increases.
The second problem that has been solved is the steady flow of a < 1, A1 = 0,
Bingham fluid in a pipe of square cross-section [23]. The results are in Λ: Λ = ⎧
⎨ 1, A1 ≠ 0. (8.5)
good agreement with those available in the literature [30–32]; see ⎩
Tables 3 and 4 in [23]. The advantage of the new method lies in the its These conditions satisfy those imposed on the stress tensor, viz.,
ability to model the flow with a sharp reduction in the computing time II (τ ) ≤ τy when A1 = 0, and τy < II (τ ) when A1 ≠ 0. The problem of
without any loss of accuracy. The mesh size of 80-80 was found to be determining where the flow is rigid and where it is liquid-like has been
sufficient for the determination of the plug velocity, with an accuracy shifted to finding the tensor Λ in the flow field such that is satisfies Eq.
up to five decimal places; and the running time for the 2D simulation (8.5). Moreover, just as the magnitude of the shear stress σ satisfies
was just 1025 s on a PC. Regardless of the Bingham number, each si- 0 ≤ σ < τy in the rigid core in a shearing flow, it is found that
mulation required the same running time on the PC. 0 ≤ Λ < 1 in the rigid core regions. What has been proposed is im-
In sum, the particle distribution function approach has shown itself portant for the following reasons:
to be quite successful in modelling the above flows of Bingham and
Papanastaiou models. Without loss of accuracy, the results are obtained 1. The constitutive equations Eqs. (8.4)–(8.5) are defined over the
with a distinct reduction in computing time. It has to be emphasised entire flow domain, not just where the fluid has yielded.
that the evolution equations used in these investigations employ the 2. One searches for the solution velocity field u and the viscoplasticity
version suitable for Cartesian coordinates. constraint tensor Λ to determine the yielded/unyielded regions.
There are no singularities because one is not trying to find the lo-
8. Numerical modelling: Steady flow of Bingham and cation of the yield surface(s) through the limit of A1/II(A1) as A1 →
Herschel–Bulkley fluids in a pipe of circular cross-section 0.
3. However, the equations of motion now involve two unknown fields:
We shall now turn to the numerical modelling of the steady flow of a a vector field u, and a symmetric tensor field Λ. The latter requires
Bingham and Herschel–Bulkley fluids in a pipe of circular cross-section that there should exist a connection between the velocity field u and
using the evolution equations in their cylindrical coordinate formula- Λ. Under Dirichlet boundary conditions, it is possible to prove such
tion. That is, one uses Eq. (2.2) along with Eq. (3.26) for the source a relation; see [32,33,35,36]. Here, we provide a summary of the
functions. Since the velocity field depends on the radial coordinate only results. First, we define a set
and is in the axial direction, it is quite straight forward to model the
flow, in a manner similar to that for the flow of a Bingham fluid in a M = {μ μ=μT , μ = (μij )1 ≤ i, j ≤ 2 ∈ (L2 (Ω))2 , μ ≤ 1 a. e. on Ω}
pipe of square cross-section [23]; see Section 4.3 therein for a de- (8.6)
scription of the algorithm. Comparisons of our results with the simu-
and a projection operator PM through
lations based on the augmented Lagrangian method (ALM) [32] as well
as the analytical solutions are presented below. q
PM (q) = , a. e. in Ω, ∀ q ∈ (L2 (Ω))2 .
The constitutive equation for an incompressible Bingham fluid is max(1, q ) (8.7)

127
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

Thus, let Λ0 be given, say it is 0. If Λn is known, use the constitutive π (1 − r0)1 + β [1 − [2(1 − r0)/(2 + β )] + [2(1 − r0)2 /(2 + β )(3 + β )]]
Q=
relation Eq. (8.4) to solve for the velocity field un, and find Λn + 1 2 β (1 + β )
through the projection (8.17)
.
Λn + 1 = PM (Λn + rτy A1n), (8.8) It can be shown quite easily that Eqs. (8.16)–(8.17) reduce to Eqs.
(8.10) and (8.13), respectively if β = 1/ n = 1.
Having assembled all of the relevant equations, we shall now pre-
sent the results of the simulations for the two fluids considered above
where r > 0 is a real number to be specified. Successive iterations are
when Od = 0.2 and r0 = 0.4. The numerical values for the Bingham and
performed till convergence is achieved to the desired level of accuracy.
Herschel–Bulkley fluids, obtained by using the ALM, are taken from
Note that the yield surface is the boundary between Λ < 1 and
Tables 1, 3 and 6 in [32]. The analytical solutions for the velocity of the
Λ = 1. Hence, the solution of the boundary value problem delivers in
plug and the flow rates have been calculated anew (Table 2).
the limit both the velocity field as well as the shape and location of the
First of all, it is easy to see that simulation results are accurate up to
yield surface. For a recent application, see the numerical solution to the
five decimal figures, when one uses a grid size of 100 × 100, or larger.
natural convection problem [21] and the mixed convection problem of
The present study overestimates the velocity of the plug and the flow
a Bingham fluid [22].
rate in both fluids slightly, when compared with those obtained by
We shall now turn to the steady flow of a viscoplastic fluid in a pipe
using the ALM [32] or the analytical solutions; however, its prediction
of circular cross-section. Let G > 0 be the constant pressure drop per
of Sp, which is the radius of the plug, is closer to the analytical solution
unit length and the radius of the pipe, R, be the length scale. The
than that based on the ALM. While it has not been possible to ascertain
constitutive equation for the Bingham fluid (8.4), when the stresses are
the running time using the ALM, the running time in the present study
scaled with GR2 and the velocity with U = GR2 / η , becomes
is quite small. Interestingly, the computing times do not differ from one
τ = A1 + 2 Od Λ. (8.9) fluid to another.

In this fluid, the analytical solution for the axial velocity field,
9. Concluding remarks
w = w (r ), has the following non-dimensional form:

1 To conclude, it is worth emphasising that the evolution equations

⎧ (1 − r0)2 , 0 ≤ r ≤ r0,
⎪ 4 for the particle distribution functions with source functions, selected as
w (r ) =
⎨1 2 2 required, deliver the equations of continuum mechanics, applicable to
⎪ 4 [(1 − r0) − (r − r0) ], r0 ≤ r ≤ 1,
(8.10) both compressible and incompressible fluids. The motivation for the
⎩
derivation arises from the format of the D3Q15 lattice, although sig-
where r = r0 is the location of the yield surface; see Eq. (5.3) in [32]. It nificant changes have to be made. Secondly, there are no restrictions
has been shown by Huilgol and You [32] that placed on the constitutive equations for the fluids. That is, one can
r0 = 2 Od, (8.11) model Newtonian, power law, viscoelastic and viscoplastic fluids.
Moreover, no additional source terms are needed in Eq. (3.14) to obtain
where the Oldroyd number is given by the equations of motion in Cartesian coordinates; four are needed in
τy cylindrical coordinates and six in spherical coordinates only.
Od = . In the present work, in order to derive the equations of continuum
GR (8.12)
mechanics, the particle distribution functions have been divided into
It is easy to show that the Oldroyd number, Od, is the same as the two sets: the first set of fifteen equations for the continuity equation and
Bingham number Bn = τyR/ηU. the equations of motion, and the second set of seven for the energy
The flow rate Q is given by equation. In connection with the latter set, it is worth pointing out that
1 So et al. [9] and Kam et al. [10] have incorporated the continuity
Q= π (3 − 4r0 + r04 ). equation, the equations of motion and the energy equation into a single
24 (8.13)
set of evolution equations to model the compressible gas flow problems
Turning to the Herschel–Bulkley fluid, the rigidity condition is still including viscous and thermal effects. While this reduces the number of
given by Eq. (8.2). When the fluid has yielded, its constitutive equation evolution equations, one has to modify and/or set a number of coeffi-
is given by cients to zero in modelling isothermal problems. If one adopts the
τ = kH [II (A1)]n − 1A1 + 2 τy A1, II (τ ) > τy, scheme proposed here, one can ignore the set of evolution equations for
(8.14)
the energy equation in modelling such flows, reducing the complexity
where kH is the viscosity parameter and n is an index such that of the numerical scheme.
0 < n ≤ 1. The characteristic velocity takes the form U = R (GR/ kH ) β , Finally, there is a fundamental difference between the LBE approach
where β = 1/ n. in [11,12] and the work presented here. In LBE, one usually begins with
Although a formal proof that the tensor Λ exists in this fluid is a set of finite difference equations to which forcing and source terms are
lacking, we shall assume that it does and proceed to perform ‘numerical added as needed to solve specific boundary/initial value problems in
experiments’, adopting the stance taken by Glowinski in [36]. Thus, the incompressible Newtonian fluids. In contrast, the aim of the present work
non-dimensional form of the constitutive equation becomes is to derive the equations of continuum mechanics at the utmost generality
for both compressible and incompressible fluids; numerical modelling fol-
τ = [II (A1)]n − 1A1 + 2 Od Λ. (8.15) lows subsequently, depending on the fluid and the problem being
Next, from Eq. (5.7) in [32], one notes that the velocity field in the considered. Thus, the methodology presented here is expected to be as
pipe of a circular cross-section take the form successful as the Finite Difference Lattice Boltzmann Method (FDLBM)
and Thermal Finite Difference Discrete Flux Method (TFDDFM) have
(1 − r0)1 + β /2 β (1 + β ), 0 ≤ r ≤ r0, been in modelling several flows, including thermal effects, of New-
w (r ) = ⎧
⎨[(1 − r0)1 + β − (r − r0)1 + β]/2 β (1 + β ), r0 ≤ r ≤ 1. tonian, power-law and viscoplastic fluids in Cartesian coordinates
⎩ (8.16)
[6–8,10,22,23]. Indeed, it has been shown by Kam et al. [10] in the
Once again, r0 = 2 Od, because the Herschel–Bulkly fluid has a constant simulation of thermoacoustic waves in two-dimensional enclosures that
shear stress τy. The flow rate is given by the following: FDLBM reduced the wall clock running time by half, when compared

128
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

with Direct Acoustic Simulation (DAS); see Table 1 in [10]. A direct Acknowledgments
comparison of the simulation of axisymmetric flows using the present
method and those performed by Huang and Lu [11] and Zhang et al. We would like to thank Professor E. Mitsoulis, National Technical
[12] would be desirable, along with the benchmark problem of the flow University of Athens, for his comments which were helpful in preparing
around a sphere in viscoelastic/viscoplastic fluids. The results of the the manuscript for publication. We are also extremely appreciative of
latter simulation will be available in the near future. the comments by the two reviewers which helped us to improve the
manuscript significantly.

Appendix A

First of all, in the review of the LBE, Zhang et al.[12] define the lattice speed vector c as the ratio of the lattice spacing, △x, and the time step △t.
In the present work, the lattice speed is given by σ, and has no relationship with the lattice spacing and the time step. Secondly, the lattice vectors in
the former approach are proportional to c, while they are proportional to σ here. Hence, ξαLBE = cξα / σ .
Next, in the LBE approach, two sets of products appear. The first is ξαLBE ·u . In connection with this, one finds that Eq. (2.6) appears in both the LBE
formulation [12] and in that due to Fu et al. [6,8], and Huilgol and Kefayati [4]. Thus, using Eq. (2.6), one finds that
14
c ⎛ ⎞
ξαLBE ·u = ξ · ∑ f eq ξ .
ρσ α ⎜ α = 0 α α ⎟ (A.1)
⎝ ⎠
Secondly, in LBE the equilibrium particle distribution function for a compressible Newtonian fluid, using a D2Q9 lattice, is given by [12]
2
⎡ ξ LBE ·u (ξ LBE ·u) u·u ⎤
f αeq, LBE = ρwα ⎢1 + α + α − .
RT 2(RT )2 2RT ⎥ (A.2)
⎣ ⎦
Leaving aside the precise values of the weight coefficients wα, the ideal gas constant R and the temperature of the fluid T, one notes that the dot
product u · u appears. This can be written as

1 ⎛ ⎛ ⎞
u·u = ∑ f eq ξ ⎞· ∑ f eq ξ .
ρ2 ⎜ α α α ⎟ ⎜ β β β ⎟
⎝ ⎠⎝ ⎠ (A.3)
Thus, an equilibrium distribution function in LBE, such as in Eq. (A.2), can be expressed in terms of the lattice vectors ξα and the equilibrium particle
distribution functions f αeq used here, although the algebra is laborious. Hence, any direct comparisons between the derivations in [12] and the
present work is difficult.

Appendix B

We collect here the various dyadic products required in the paper. First of all, we observe that the following matrices can be derived:

⎡1 0 0 ⎤
ξ1 ⊗ ξ1 = ξ3 ⊗ ξ3 = σ 2 ⎢ 0 0 0 ⎥,
⎣0 0 0⎦ (B.1)

⎡0 0 0⎤
ξ2 ⊗ ξ2 = ξ4 ⊗ ξ4 = σ 2 ⎢ 0 1 0 ⎥,
⎣0 0 0⎦ (B.2)

⎡0 0 0⎤
ξ5 ⊗ ξ5 = ξ6 ⊗ ξ6 = σ 2 ⎢ 0 0 0 ⎥.
⎣0 0 1 ⎦ (B.3)
Next,

⎡1 1 1⎤
ξ7 ⊗ ξ7 = ξ13 ⊗ ξ13 = σ 2 ⎢1 1 1⎥,
⎣1 1 1⎦ (B.4)

⎡1 −1 −1⎤
ξ8 ⊗ ξ8 = ξ14 ⊗ ξ14 = σ 2 ⎢−1 1 1 ⎥,
⎣− 1 1 1 ⎦ (B.5)

⎡1 1 −1⎤
ξ9 ⊗ ξ9 = ξ11 ⊗ ξ11 = σ 2 ⎢1 1 −1⎥,
⎣− 1 − 1 1 ⎦ (B.6)

⎡1 −1 1 ⎤
ξ10 ⊗ ξ10 = ξ12 ⊗ ξ12 = σ 2 ⎢−1 1 −1⎥.
⎣1 −1 1 ⎦ (B.7)
Finally, the following matrices can be obtained:

129
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

⎡1 1 1⎤
f 7eq ξ7 ⊗ ξ7 + f13eq ξ13 ⊗ ξ13 = 4σ 4 (C12 + C13 + C23) ⎢1 1 1⎥,
⎣1 1 1⎦ (B.8)

⎡1 −1 −1⎤
f 8eq ξ8 ⊗ ξ8 + f14eq ξ14 ⊗ ξ14 = 4σ 4 (−C12 − C13 + C23) ⎢−1 1 1 ⎥,
⎣− 1 1 1 ⎦ (B.9)

⎡1 1 −1⎤
f 9eq ξ9 ⊗ ξ9 + f11eq ξ11 ⊗ ξ11 = 4σ 4 (C12 − C13 − C23) ⎢1 1 −1⎥,
⎣− 1 − 1 1 ⎦ (B.10)

⎡1 −1 1 ⎤
f10eq ξ10 ⊗ ξ10 + f12eq ξ12 ⊗ ξ12 = 4σ 4 (−C12 + C13 − C23) ⎢−1 1 −1⎥.
⎣1 −1 1 ⎦ (B.11)
We shall demonstrate through a direct derivation that
∂ (ρ u)
+ ∇ ·M = O (ɛ)
∂t (B.12)
delivers the correct form of the equations of motion in cylindrical coordinates. In order to show that this is valid, it is necessary to realise that one
knows the expansion of ∇ · T, where T is the symmetric stress tensor, in cylindrical coordinates. Hence, one can write down ∇ · M, for the latter tensor
is also symmetric.
The Table 3 is required in the derivation of the equations of motion in cylindrical coordinates.
Now,

∂ (ρu2 − Trr ) 1 ∂ (ρuv − Trθ ) ∂ (ρuw − Trz ) ρ (u2 − v 2) − (Trr − Tθθ ) ⎤

∇·M = ⎡ + + + e
⎢
⎣ ∂r r ∂θ ∂z r ⎥ r
⎦
∂ (ρuv − Trθ ) 1 ∂ (ρv 2 − Tθθ ) ∂ (ρvw − Tθz ) 2
+⎡ + + + (ρuv − Trθ ) ⎤ eθ
⎢
⎣ ∂r r ∂θ ∂ z r ⎥
⎦
∂ (ρuw − T ) ∂ (ρvw − T ) ∂ (ρw 2 − T )
⎡ rz 1 θz zz 1 ⎤
+ + + (ρuw − Trz ) ez .
⎢
⎣ ∂r r ∂θ ∂z r ⎥
⎦ (B.13)
Note that the acceleration terms are given by
∂u ∂u v ∂u ∂u v2
ar = +u + +w − ,
∂t ∂r r ∂θ ∂z r (B.14)

∂v ∂v v ∂v ∂v uv
aθ = +u + +w + ,
∂t ∂r r ∂θ ∂z r (B.15)

∂w ∂w v ∂w ∂w
az = +u + +w .
∂t ∂r r ∂θ ∂z (B.16)
Using the continuity equation
∂ρ ∂ (ρu) ρu 1 ∂ (ρv ) ∂ (ρw )
+ + + + = 0,
∂t ∂r r r ∂θ ∂z (B.17)
one finds that
∇ · M = ρ u · ∇u − ∇ · T , (B.18)
where T is the stress tensor in cylindrical coordinates.
Thus, to within O(ε),
∂ (ρ u)
+ ∇ ·M = O (ɛ)
∂t (B.19)
delivers the correct form of the equations of motion in cylindrical coordinates.
The Table 4 is required in the derivation of the equations of motion in spherical coordinates.
The acceleration terms in spherical coordinates are
∂u ∂u v ∂u w ∂u v2 + w2
ar = +u + + − ,
∂t ∂r r ∂θ r sin θ ∂ϕ r (B.20)

∂v ∂v v ∂v w ∂v uv w2 cot θ
aθ = +u + + + − ,
∂t ∂r r ∂θ r sin θ ∂ϕ r r (B.21)

∂w ∂w v ∂w w ∂w uw vw cot θ
aϕ = +u + + + + .
∂t ∂r r ∂θ r sin θ ∂ϕ r r (B.22)
The divergence of the tensor M in spherical coordinates is given by

130
R.R. Huilgol, G.H.R. Kefayati Journal of Non-Newtonian Fluid Mechanics 251 (2018) 119–131

∂ (ρu2 − Trr ) 1 ∂ (ρuv − Trθ ) 1 ∂ (ρuw − Trϕ)

∇·M = ⎧ ⎡ + +
⎨ ⎢ ∂ r r ∂ θ r sin θ ∂ϕ
⎩⎣
1
+ [2(ρu2 − Trr ) − (ρv 2 − Tθθ ) − (ρw 2 − Tϕϕ) + cot θ (ρuv − Trθ )] ⎫ er
r ⎬
⎭
∂ (ρuv − Trθ ) 1 ∂ (ρv 2 − Tθθ ) 1 ∂ (ρvw − Tθϕ) 1
+ ⎧⎡ + + + [3(ρuv − Trθ ) + cot θ [(ρv 2 − Tθθ ) − (ρw 2 − Tϕϕ)]] ⎫ eθ
⎨ ⎢ ∂r r ∂θ r sin θ ∂ϕ r ⎬
⎩⎣ ⎭
∂ (ρuw − Trϕ) 1 ∂ (ρvw − Tθϕ) 1 ∂ (ρw 2 − Tϕϕ) 1
+ ⎧⎡ + + + [3(ρuw − Trϕ) + 2 cot θ (ρvw − Tθϕ)] ⎫ eϕ .
⎨⎢ ∂r r ∂θ r sin θ ∂ϕ r ⎬ (B.23)
⎩⎣ ⎭
The continuity equation is given by
∂ρ ∂ (ρu) ρu 1 ∂ (ρv ) ρv 1 ∂ (ρw )
+ +2 + + cot θ + = 0.
∂t ∂r r r ∂θ r r sin θ ∂ϕ (B.24)
Using the above, a long calculation shows that once again,
∇ ·M = u ·∇ (ρ u) − ∇ ·T. (B.25)
That is, to within O(ε), we recover the equations of motion in spherical coordinates through
∂ (ρ u)
+ ∇ ·M = O (ɛ).
∂t (B.26)

References [19] J. Blazek, Computational Fluid Dynamics: Principles and Applications, Elsevier,
Amsterdam, 2001, pp. 347–350.
[20] T. Cebeci, J.P. Shao, F. Kafyeke, E. Laurendeau, Computational Fluid Dynamics for
[1] X.J. Fan, R.I. Tanner, R. Zheng, Smoothed particle hydrodynamics and its appli- Engineers, Springer-Verlag, New York, 2005.
cation to non-Newtonian moulding flow, J. Non-Newton. Fluid Mech. 165 (2010) [21] R.R. Huilgol, G.H.R. Kefayati, Natural convection problem in a Bingham fluid using
219–226. the operator-splitting method, J. Non-Newt. Fluid Mech. 220 (2015) 22–32.
[2] M.R. Hashemi, R. Fatehi, M.T. Manzari, SPH Simulation of interacting solid bodies [22] G.H.R. Kefayati, R.R. Huilgol, Lattice Boltzmann method for simulation of mixed
suspended in a shear flow of an Oldroyd-B fluid, J. Non-Newton. Fluid Mech. 166 convection of a Bingham fluid in a lid-driven cavity, Int. J. Heat Mass Transf. 103
(2011) 1239–1252. (2016) 725–743.
[3] J.J. Monaghan, Smoothed particle hydrodynamics, Rep. Prog. Phys. 68 (2005) [23] G.H.R. Kefayati, R.R. Huilgol, Lattice Boltzmann method for the simulation of the
1703–1759. steady flow of a Bingham fluid in a pipe of square cross-section, Eur. J. Mech/B
[4] R.R. Huilgol, G.H.R. Kefayati, From mesoscopic methods to continuum mechanics: Fluids 65 (2017) 412–422.
Newtonian and non-Newtonian fluids, J. Non-Newton. Fluid Mech. 233 (2016) [24] S.V. Patankar, D.B. Spalding, A calculation procedure for heat, mass and mo-
146–154. mentum transfer in three-dimensional parabolic flows, Int. J. Heat Mass Transf. 15
[5] Z. Guo, B. Shi, C. Zheng, A coupled BGK model for the Boussinesq equations, Int. J. (1972) 1787–1806.
Numer. Methods Fluids 39 (2002) 325–342. [25] S.V. Patankar, A calculation procedure for two-dimensional elliptic situations,
[6] S.C. Fu, R.M.C. So, Modeled lattice Boltzmann equation and the constant density Numer. Heat Transf. 4 (1981) 409–425.
assumption, AIAA J. 47 (2009) 3038–3042. [26] E.F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics: A Practical
[7] S.C. Fu, R.M.C. So, W.W.F. Leung, A discrete flux scheme for aerodynamic and Introduction, 2nd, Springer-Verlag, New York, 1999.
hydrodynamic flows, Commun. Comp. Phys. 9 (2011) 1257–1283. [27] P.D. Lax, B. Wendroff, Systems of conservation laws, Comm. Pure Appl. Math. 13
[8] S.C. Fu, R.M.C. So, W.W.F. Leung, Linearized–Boltzmann-type-equation-based finite (1960) 217–237.
difference method for incompressible flow, Comput. Fluids 6 (2012) 67–80. [28] P. Neofytou, A 3rd order upwind finite volume method for generalised Newtonian
[9] R.M.C. So, S.C. Fu, R.C.K. Leung, Finite difference lattice Boltzmann method for fluid flows, Adv. Eng. Softw. 36 (2005) 664–680.
compressible thermal fluids, AIAA J. 48 (2010) 1059–1071. [29] E.J. Dean, R. Glowinski, Operator-splitting methods for the simulation of Bingham
[10] E.W.S. Kam, R.M.C. So, S.C. Fu, One-step simulation of thermoacoustic waves in visco-plastic flow, Chin. Ann. Math. 23B (2002) 187–204.
two-dimensional enclosures, Comput. Fluids 140 (2016) 270–288. [30] P. Saramito, N. Roquet, An adaptive finite element method for viscoplastic fluids in
[11] H. Huang, X.-Y. Lu, Theoretical and numerical study of axisymmetric lattice pipes, Comput. Methds Appl. Mech. Eng. 190 (2001) 5391–5412.
Boltzmann models, Phys. Rev. E 80 (2009) 016701. [31] M. Moyers-Gonzalez, I. Frigaard, Numerical solution of duct flows of multiple visco-
[12] L. Zhang, S. Yang, Z. Z, L. Yin, Y. Zhao, J.W. Chew, Consistent lattice Boltzmann plastic fluids, J. Non-Newt. Fluid Mech. 127 (2004) 227–241.
methods for incompressible axisymmetric flows, Phys. Rev. E 94 (2016) 023302. [32] R.R. Huilgol, Z. You, Application of the augmented Lagrangian method to steady
[13] R.R. Huilgol, N. Phan-Thien, Fluid Mechanics of Viscoelasticity, Elsevier, pipe flows of Bingham, Casson and Herschel–Bulkley fluids, J. Non-Newt. Fluid
Amsterdam, 1997, pp. 81–83. Mech. 128 (2005) 126–143.
[14] Q. Li, Y.L. He, G.H. Tang, W.Q. Tao, Improved axisymmetric lattice Boltzmann [33] R.R. Huilgol, Fluid Mechanics of Viscoplasticity, Springer, Berlin-Heidelberg, 2015.
scheme, Phys. Rev. E 79 (2010) 056707. [34] R.R. Huilgol, On the definition of pressure in rheology, Rheol. Bull. 78 (2) (2009)
[15] J.G. Zhou, Axisymmetric lattice Boltzmann method revised, Phys. Rev. E 84 (2011) 12,14–15, 29.
036704. [35] G. Duvaut, J.-L. Lions, Inequalities in Mechanics and Physics, Springer, New York,
[16] R.S. Rivlin, J.L. Ericksen, Stress deformation relations for isotropic materials, J. 1976.
Ration. Mech. Anal. 4 (1955) 323–425. [36] R. Glowinski, Finite Element Methods for Incompressible Viscous Flow, in:
[17] R.I. Tanner, Engineering Rheology, 2nd, Oxford University Press, Oxford, 2000. P.G. Ciarlet, J.-L. Lions (Eds.), Handbook of Numerical Analysis, vol. IX, North-
[18] Q. Zou, X. He, On pressure and velocity boundary conditions for the lattice Holland, Amsterdam, 2003.
Boltzmann BGK model, Phys. Fluids 9 (1997) 1591–1598.

131