fluids

Article
Derivation of the Adjoint Drift Flux Equations for
Multiphase Flow
Shenan Grossberg 1 , Daniel S. Jarman 2 and Gavin R Tabor 1, *
1 College of Engineering, Maths and Physical Sciences, University of Exeter, Harrison Building, North Park
Road, Exeter EX4 4QF, UK; s.grossberg@exeter.ac.uk
2 Hydro International Ltd.Shearwater House, Clevedon Hall Estate, Victoria Road, Clevedon BS21 7RD, UK;
djarman@hydro-int.com
* Correspondence: g.r.tabor@exeter.ac.uk




Received: 24 January 2020; Accepted: 7 March 2020; Published: 11 March 2020


Abstract: The continuous adjoint approach is a technique for calculating the sensitivity of a flow
to changes in input parameters, most commonly changes of geometry. Here we present for the
first time the mathematical derivation of the adjoint system for multiphase flow modeled by the
commonly used drift flux equations, together with the adjoint boundary conditions necessary to
solve a generic multiphase flow problem. The objective function is defined for such a system, and
specific examples derived for commonly used settling velocity formulations such as the Takacs and
Dahl models. We also discuss the use of these equations for a complete optimisation process.

Keywords: adjoint optimization; multiphase flow; computational fluid dynamics

1. Introduction
The adjoint method is currently attracting significant interest as an optimization process in CFD.
The objective of the adjoint approach is to calculate the sensitivity of the flow solution with respect
to changes in the input parameters, most commonly changes in the geometry. This can then in
principle be used as the basis for an iterative optimization algorithm based on gradient information
(the sensitivities) which can optimize the design with many fewer function evaluations than would be
the case for non-gradient-based approaches (such as genetic algorithms). Calculating the sensitivities
requires differentiating the governing equations with respect to the changes of the input parameters,
and since the governing equations for fluid flow are the Navier-Stokes equations (or equations derived
from these such as the Reynolds Averaged Navier Stokes equations), this is understandably very
challenging. There are two main approaches; the discrete adjoint approach, and the continuous
adjoint approach. In the discrete adjoint approach, the sensitivity matrix is calculated numerically
by evaluating the system for small changes in the inputs and applying standard finite difference
methods. In the continuous adjoint approach, the sensitivities are calculated mathematically using
lagrange multipliers. This is more elegant and provides an implementation which is easier to code,
requires fewer evaluations and can be made numerically consistent with the evaluation of the original
equation set. However it does require significant mathematical analysis in advance, and if the problem
formulation changes (different equations, boundary conditions etc) this has to be repeated. Examples
of the application of the continuous adjoint method for single phase flow can be found in a range of
areas [1,2] such as automotive [3–5], aerospace [6,7] and turbomachinery [8–10], and implementations
of the equations can be found in general purpose CFD codes such as STAR-CCM, ANSYS Fluent [11]
and Engys Helyx [4]. However the equations are complex to develop and application to multiphase
systems is only just starting [12]. In many cases, even just the evaluation of the sensitivities is valuable,
as they can be used to indicate possible changes to the design engineers. Beyond this the sensitivities

Fluids 2019, 5, 31; doi:10.3390/fluids5010031 www.mdpi.com/journal/fluids

Fluids 2019, 5, 31 2 of 21

can also be used as the basis for an optimization loop [2]. This of course necessitates the morphing of
the geometry through techniques such as volumetric B-splines [13] or Radial Basis Functions [14] and
consequent updating of the mesh [15].
Multiphase flow is the simultaneous flow of two or more immiscible phases in a system.
In dispersed multiphase systems, one or more of the phases exists as fluid particles small enough
not to be resolved in the simulation; examples include gas bubbles in water, emulsions (liquid
droplets in another immiscible liquid) and actual solid particles in gas or liquid. A wide variety
of different mathematical models have been derived over the years to describe dispersed multiphase
flow, including mixture models, lagrangian particle tracking, and eulerian n-fluid models [16,17].
Which is used depends on the exact physics of the problem, as well as factors such as available
computing resources and desired accuracy. In many physical systems, the density ratio between the
two phases is low, generally less than 2:1, and the drag force between them is high. Therefore, to a
good approximation, the two phases can be considered to respond to pressure gradients as a single
phase. Additionally, the slip (drift) between the phases is primarily due to the gravitational settling
of the dispersed phase. This might adequately describe solid particles in water or an emulsion of
immiscible liquids, and in these cases a commonly used mathematical model is the drift flux model.
Hence it is this set of equations we have decided to focus on.
In the drift flux model, the two phases are treated as one: the momentum and continuity equations
for both phases are summed to create a mixture-momentum and mixture-continuity equation, and the
transport of the dispersed phase is modelled using a drift equation. The three equations, collectively
called the drift flux equations, are listed below:

∂ρm vm

+ (vm · ∇)(ρm vm ) = −∇(ρm pm ) + ∇ · 2µm D(vm )
∂t  
α ρc ρd
−∇· vdj vdj + ρm g + F, (1a)
1 − α ρm
∂ρm
+ ∇ · (ρm vm ) = 0, (1b)
∂t  
∂α αρc
+ ∇ · (αvm ) = −∇ · vdj + ∇ · K ∇α, (1c)
∂t ρm

where:

• α is the dispersed-phase volume fraction,

• ρc is the continuum density,
• ρd is the dispersed-phase density,
• ρm is the mixture density, defined as αρd + (1 − α)ρc ,
• vm is the mixture velocity,
• pm is the mixture kinematic pressure,
• µm is the mixture viscosity, defined as the sum of the continuum, dispersed-phase and mixture
turbulent viscosities, µc + µd + µtm ,
D(vm ) = 21 ∇vm + (∇vm ) T is the mixture strain rate tensor,


•
• vdj is the dispersed-phase settling velocity,
• g is the acceleration due to gravity,
• F is the capillary force and
t = µtm
• K is the turbulent diffusion coefficient, defined as the mixture eddy diffusivity, νm ρm .

In summing the momentum equations, not only have the number of equations been reduced
from four to three, but the inter-phase momentum transfer terms have also been eliminated which
were numerically unstable [18]. Hence, a far more robust equation set has been produced and the
computational resources required to solve the system have been reduced. This also makes it a very
Fluids 2019, 5, 31 3 of 21

appropriate basis from which to develop an adjoint formulation suitable for applying to dispersed
multiphase flows in this regime. This is the challenge of the current paper. We focus in particular
on wall-bounded or ducted flows, in which there is no contribution to the objective function from
the interior of the domain, in other words, the performance of the system is entirely governed by the
boundary properties.
The paper is organized as follows. The optimization problem is stated in Section 2 and the adjoint
equations for the drift flux model are derived for the general case in Section 2.1. These equations
are then applied to the specific case of ducted or wall-bounded flows in Section 3, with the objective
function for this case being specified in Section 4, and different settling velocities in Section 5. Finally,
the conclusions follow in Section 6.

2. The Optimization Problem

If the performance of a device is measured by an objective function, J, and the residuals of the
primal (flow) equations are given by R, the optimisation problem can be stated as,

optimise J (x, y) subject to R(x, y) = 0, (2)

where x are the design parameters and y are the primal variables [19]. It can then be formulated as,
Z
L = J+ λ R dΩ, (3)
Ω

where L is the Lagrange function, λ are the Lagrange multipliers (also referred to as the adjoint
variables) and Ω is the flow domain. In this case, the primal equations are the steady state
drift flux equations, with the capillary force taken to be zero [18] and a Darcy term included
in the mixture-momentum equation. They are rearranged in terms of their residuals, R =
( R1 , R2 , R3 , R4 , R5 )T , as follows:

( R1 , R2 , R3 )T = (vm · ∇)(ρm vm ) + ∇(ρm pm ) − ∇ · 2µm D(vm )

 
α ρc ρd
+∇· v v − ρm g + ℵρm vm , (4a)
1 − α ρm dj dj
R4 = −∇ · (ρm vm ), (4b)
 
αρc
R5 = ∇ · (αvm ) + ∇ · v − ∇ · K ∇α, (4c)
ρm dj

where ℵ is the porosity, associated with the Darcy term. The variation of the Lagrange function with
respect to the primal variables, (vm , pm , α), and the design parameter, ℵ, is,

δ L = δvm L +δpm L +δα L +δℵ L , (5)

where, for example, δα L = L (α + δα) − L (α). We choose the adjoint variables, (u, q, β) =
(u1 , u2 , u3 , q, β), so that the variation with respect to the primal variables vanishes, i.e.

δvm L +δpm L +δα L = 0, (6)

and the Lagrange function now varies only with respect to the design parameter,
Z
δ L = δℵ L = δℵ J + (u, q, β)δℵ R dΩ. (7)
Ω
Fluids 2019, 5, 31 4 of 21

2.1. Derivation of the Adjoint Drift Flux Equations

The adjoint drift flux equations are derived by substituting Equation (3) into Equation (6), giving,

δvm J + δpm J + δα J
Z Z Z
+ (u, q, β)δvm R dΩ + (u, q, β)δpm R dΩ + (u, q, β)δα R dΩ = 0, (8)
Ω Ω Ω

which can be expanded to,

Z Z Z
δvm J + δpm J + δα J + dΩ u · δvm ( R1 , R2 , R3 ) T + dΩ qδvm R4 + dΩ βδvm R5
ZΩ ZΩ ZΩ
+ dΩ u · δpm ( R1 , R2 , R3 ) T + dΩ qδpm R4 + dΩ βδpm R5
Ω Ω Ω
Z Z Z
T
+ dΩ u · δα ( R1 , R2 , R3 ) + dΩ qδα R4 + dΩ βδα R5 = 0. (9)
Ω Ω Ω

The variation of R with respect to the primal variables can be determined as:

δvm ( R1 , R2 , R3 ) T = (δvm · ∇)(ρm vm ) + (vm · ∇)(ρm δvm ) − ∇ · 2µm D(δvm )

− ∇ · 2δvm µd D(vm ) + ℵρm δvm , (10a)
δvm R4 = −∇ · (ρm δvm ), (10b)
δvm R5 = ∇ · (αδvm ), (10c)
T
δpm ( R1 , R2 , R3 ) = ∇(ρm δpm ), (10d)
δpm R4 = 0, (10e)
δpm R5 = 0, (10f)
T


δα ( R1 , R2 , R3 ) = (ρd − ρc ) (vm · ∇)(δαvm ) + ∇(δαpm ) + δα(ℵvm − g)


− ∇ · 2δα µd D(vm ) + ∇ · δα (αρd vdj vdj ), (10g)
δα R4 = −(ρd − ρc )∇ · (δαvm ), (10h)
µtm
 
δα R5 = ∇ · (δαvm ) + ∇ · δα (αvdj ) − ∇ · ∇δα
ρc
µtm ρd µtm ρd
   
+ − 1 ∇ · (δα∇α) + − 1 ∇ · (α∇δα). (10i)
ρc ρc ρc ρc

Derivation of Equations (10a), (10g) and (10i) can be found in Appendies A–C, respectively,
where the variation of µtm has been neglected. This is correct only for laminar flow regimes.
For turbulent flows, neglecting this variation constitutes a common approximation, known as frozen
turbulence [19]. This may introduce errors into the optimisation [20], although there are cases in the
literature where the frozen turbulence assumption can be demonstrated to be acceptable [21].
Fluids 2019, 5, 31 5 of 21

With these variations, Equation (9) now reads,

δvm J + δpm J + δα J
Z 

+ dΩ u · (δvm · ∇)(ρm vm ) + (vm · ∇)(ρm δvm ) − ∇ · 2µm D(δvm )
Ω

 Z Z
− ∇ · 2δvm µd D(vm ) + ℵρm δvm − dΩ q∇ · (ρm δvm ) + dΩ β∇ · (αδvm )
Ω Ω
Z Z

+ dΩ u · ∇(ρm δpm ) + (ρd − ρc ) dΩ u · (vm · ∇)(δαvm ) + ∇(δαpm ) + δα(ℵvm − g)
Ω Ω
Z 
 Z

− dΩ u · ∇ · 2δα µd D(vm ) + dΩ u · ∇ · δα (αρd vdj vdj )
Ω Ω
Z Z   t 
µm
− (ρd − ρc ) dΩ q∇ · (δαvm ) + dΩ β ∇ · (δαvm ) + ∇ · δα (αvdj ) − ∇ · ∇δα
Ω Ω ρc
   t     t 
ρd µm ρd µm
+ −1 ∇· δα∇α + −1 ∇· α∇δα = 0. (11)
ρc ρc ρc ρc

Decomposing the objective function into contributions from the boundary, Γ, and interior, Ω,
of the domain, Z Z
J= JΓ dΓ + JΩ dΩ, (12)
Γ Ω
Equation (11) can be reformulated as,

Z  
dΓ n(u · ρm vm ) + u(ρm vm · n) + 2µm n · D(u) − qρm n + αβn + Γ
∂J
· δvm
Γ ∂v m
Z

− dΓ 2µm n · D(δvm ) · u + 2δvm µd n · D(vm ) · u − 2δvm µd n · D(u) · vm
Γ
Z 


+ dΩ − ∇u · (ρm vm ) − (ρm vm · ∇)u − ∇ · 2µm D(u)
Ω
 Z
+ ℵρm u + ρm ∇q − α∇ β + Ω · δvm −
∂J

dΩ ∇ · 2δvm µd D(u) · vm
∂vm Ω
Z   Z  
+ dΓ ρm u · n + Γ δpm + dΩ −∇ · ρm u + Ω δpm
∂J ∂J
Γ ∂pm Ω ∂pm
µt
Z 
+ dΓ (ρd − ρc ) u(vm · n) · vm + pm u · n − qvm · n + βvm · n + m (n · ∇) β


Γ ρc
t   
β(n · ∇)α − α(n · ∇) β + Γ δα
µ ρd  ∂J
+ m −1
ρc ρc ∂α
Z

+ dΓ u · n δα (αρd vdj vdj ) + β δα (αvdj ) · n − 2δα µd n · D(vm ) · u + 2δα µd n · D(u) · vm
Γ
µt
Z   
ρd
− dΓβ m 1 − α −1 (n · ∇)δα
Γ ρc ρc
Z   
+ dΩ (ρd − ρc ) − (vm · ∇)u · vm − pm ∇ · u + u · (ℵvm − g) + (vm · ∇)q
Ω
µtm
   
ρd  ∂JΩ
− (vm · ∇) β + − 1 ∇ · (α∇ β) − ∇α · ∇ β − ∇ · ∇ β + δα
ρc ρc ∂α
Z 

− dΩ ∇ · u δα (αρd vdj vdj ) + δα (αvdj ) · ∇ β + ∇ · 2δα µd D(u) · vm = 0. (13)
Ω
Fluids 2019, 5, 31 6 of 21

Derivation of Equation (13) can be found in Appendix D. In order to satisfy Equation (13) in
general, the integrals must vanish individually. The adjoint drift flux equations are deduced from the
integrals over the interior of the domain:

−∇u · (ρm vm ) − (ρm vm · ∇)u − ∇ · 2µm D(u) = −ρm ∇q + α∇ β − ℵρm u − Ω

∂J
∂vm
 
∂µd
+∇· 2 D( u ) · v m , (14a)
∂vm

∇ · (ρm u) = Ω ,
∂J
(14b)
∂pm
µtm ρd µtm
      
∂ ρd
− vm + (αvdj ) + − 1 ∇α · ∇ β = ∇ · 1−α −1 ∇β
∂α ρc ρc ρc ρc

+ S1 + S2 − Ω ,
∂J
(14c)
∂α
where:
 
∂µd

S1 = (ρd − ρc ) (vm · ∇)(u · vm − q) − u · (ℵvm − g) + ∇ · 2 D( u ) · v m , (15a)
∂α
 
∂
S2 = ( ρ d − ρ c ) p m + (αρd vdj vdj ) ∇ · u, (15b)
∂α

and the boundary conditions for the adjoint variables are deduced from the surface integrals:
Z 
dΓ n(u · ρm vm ) + ρm vm · nu + 2µm n · D(u) − qρm n + αβn + Γ
∂J
Γ ∂v m
 Z
∂µd

+2 n · D( v m ) · u − D( u ) · v m · δvm − dΓ2µm n · D(δvm ) · u = 0, (16a)
∂vm Γ
Z  
dΓ ρm un + Γ δpm = 0,
∂J
(16b)
Γ ∂pm
Z 
∂ µ t 
ρd
 
dΓ vm · n + (αvdj ) · n + m − 1 (n · ∇)α β
Γ ∂α ρc ρc
µmt  
ρd

∂JΓ

+ 1−α −1 (n · ∇) β + C1 + C2 + δα (16c)
ρc ρc ∂α
µt
Z   
ρd
− dΓβ m 1 − α −1 (n · ∇)δα = 0,
Γ ρc ρc

where:
 
∂µd

C1 = (ρd − ρc )(u · vm − q)vm + 2 D( u ) · v m · n, (17a)
∂α
  
∂ ∂µd

C2 = (ρd − ρc ) pm + (αρd vdj vdj ) u − 2 D( v m ) · u · n (17b)
∂α ∂α

and un = u · n is the normal component of the adjoint velocity. This is the general form of the
adjoint equation system for the steady state drift flux equations with Darcy porosity term and frozen
turbulence.

3. Application to Wall Bounded Flows

Thus far in the paper we have presented the optimisation problem in as generic a way as possible.
To proceed further with the derivation we now need to derive expressions for the boundary conditions,
objective function and slip velocity. We will examine these for the case of wall-bounded or ducted
Fluids 2019, 5, 31 7 of 21

flows, for which there is no contribution to the objective function from the interior of the domain. So,
in the cases where the objective function only involves integrals over the surface of the flow domain
rather than over its interior, the adjoint equations reduce to:


−∇u · (ρm vm ) − (ρm vm · ∇)u − ∇ · 2µm D(u) = −ρm ∇q + α∇ β − ℵρm u
 
∂µd
+∇· 2 D( u ) · v m , (18a)
∂vm
∇ · (ρm u) = 0, (18b)
µtm µt
      
∂ ρd ρd
− vm + (αvdj ) + − 1 ∇α · ∇ β = ∇ · m 1 − α −1 ∇β
∂α ρc ρc ρc ρc
+ S1 . (18c)

These equations no longer depend on the objective function, so when switching from one
optimisation objective to another, they remain unchanged and only the boundary conditions have to
be adapted to the specific objective function. Note that as a result of Equation (18b), ∇ · u = 0 [22] and,
therefore, S2 = 0.
For the adjoint boundary conditions, the terms in Equation (16a) involving µm can be rewritten as,
Z
Z

dΓ 2µm n · D(u) · δvm − D(δvm ) · u = dΓ µm (n · ∇)u · δvm − (n · ∇)δvm · u (19)
Γ Γ

(Ref. [19]) and therefore the adjoint boundary conditions, Equation (16), reduce to:

Z 
dΓ n(u · ρm vm ) + ρm vm · nu + µm (n · ∇)u − qρm n + αβn + Γ
∂J
Γ ∂vm
 Z
∂µ
+2 d n · D(vm ) · u − D(u) · vm · δvm − dΓµm (n · ∇)δvm · u = 0,


(20a)
∂vm Γ
Z  
∂JΓ
dΓ ρm un + δpm = 0, (20b)
Γ ∂pm
µtm ρd
Z     
∂
dΓ vm · n + (αvdj ) · n + − 1 (n · ∇)α β
Γ ∂α ρc ρc
µt
   
(n · ∇) β + C1 + C2 + Γ δα
ρd ∂J
+ m 1−α −1
ρc ρc ∂α
Z
µ t  
ρd

− dΓβ m 1 − α −1 (n · ∇)δα = 0. (20c)
Γ ρc ρc

In order to determine the boundary conditions of the adjoint variables, the boundary conditions
imposed on the primal variables are listed in Table 1. We will derive expressions for the three main
boundary conditions.

Table 1. Primal boundary conditions.

um α pm
Inlet fixed value fixed value zero gradient
Wall zero zero gradient zero gradient
Outlet zero gradient zero gradient zero
Fluids 2019, 5, 31 8 of 21

3.1. Adjoint Boundary Conditions at the Inlet

At an inlet, the primal velocity and dispersed-phase volume fraction are usually fixed, so,

δvm = 0 and δα = 0. (21)

The first integrals in Equations (20a) and (20c) therefore go to zero and Equation (20) reduces to:
Z
dΓµm (n · ∇)δvm · u = 0, (22a)
Γ
Z  
∂JΓ
dΓ ρm un + δpm = 0, (22b)
Γ ∂pm
µt
Z   
ρd
dΓβ m 1 − α −1 (n · ∇)δα = 0. (22c)
Γ ρc ρc

When both fluids are incompressible, ∇ · vm = 0 [22], and as δvmt = 0 along the inlet, (n ·
∇)δvm = (n · ∇)δvmt [19], where vmt is the tangential component of the mixture velocity. Hence,
Equation (22) reduces to:
Z
dΓµm (n · ∇)δvmt · ut = 0, (23a)
Γ
Z  
∂JΓ
dΓ ρm un + δpm = 0, (23b)
Γ ∂pm
µt
Z   
ρd
dΓβ m 1 − α −1 (n · ∇)δα = 0, (23c)
Γ ρ c ρc

where ut is the tangential component of the adjoint velocity, from which we deduce the boundary
conditions for the adjoint variables at the inlet to be:

ut = 0, (24a)
1 ∂JΓ
un = − , (24b)
ρm ∂pm
β = 0 ⇐⇒ µtm 6= 0. (24c)

Note that these derivations do not impose a condition for q. Since q enters the adjoint drift flux
equations in a manner similar to the way pm enters the primal drift flux equations, the zero gradient
boundary condition of pm at the inlet is applied to q as well,

(n · ∇)q = 0. (25)

3.2. Adjoint Boundary Conditions at the Wall

At a wall, typical primal conditions are zero velocity and zero gradient of the dispersed-phase
volume fraction. Therefore, we have,

vm = 0, δvm = 0 and (n · ∇)δα = 0. (26)

Fluids 2019, 5, 31 9 of 21

The first integral in Equation (20a) and the second integral in Equation (20c) therefore go to zero
and the terms in the first integral in Equation (20c), containing vm , go to zero. Equation (20) therefore
reduces to:
Z
dΓµm (n · ∇)δvm · u = 0, (27a)
Γ
Z  
dΓ ρm un + Γ δpm = 0,
∂J
(27b)
Γ ∂pm
Z 
∂ t 
µm ρd
 
dΓ (αvdj ) · n + − 1 (n · ∇)α β
Γ ∂α ρc ρc
µtm
   
ρd ∂JΓ
+ 1−α −1 (n · ∇) β + C2 + δα = 0. (27c)
ρc ρc ∂α

As at the inlet, the primal velocity does not diverge and δvmt = 0 along the wall, so Equation (27)
reduces to:
Z
dΓµm (n · ∇)δvmt · ut = 0, (28a)
Γ
Z  
∂JΓ
dΓ ρm un + δpm = 0, (28b)
Γ ∂pm
µtm ρd
Z     
∂
dΓ (αvdj ) · n + − 1 (n · ∇)α β
Γ ∂α ρc ρc
µt
   
(n · ∇) β + C2 + Γ δα = 0,
ρd ∂J
+ m 1−α −1 (28c)
ρc ρc ∂α

from which we deduce the boundary conditions for the adjoint variables at the wall to be:

ut = 0, (29a)
1 ∂JΓ
un = − , (29b)
ρm ∂pm
µt
   
∂ ρd
(αvdj ) · n + m − 1 (n · ∇)α β
∂α ρc ρc
t   
(n · ∇) β = −C2 − Γ .
µ ρd ∂J
+ m 1−α −1 (29c)
ρc ρc ∂α

Equation (29c) is used to determine β and, as at the inlet, Equation (25) applies.

3.3. Adjoint Boundary Conditions at the Outlet

At an outlet, typical primal conditions are zero pressure and zero gradient of velocity and
dispersed-phase volume fraction. Therefore, we have,

δpm = 0, (n · ∇)δvm = 0 and (n · ∇)δα = 0. (30)

Fluids 2019, 5, 31 10 of 21

The second integral in Equations (20a) and (20c) therefore goes to zero and, with δpm = 0,
Equation (20b) is identically fulfilled. The remaining terms in Equation (20) are the first integrals in
Equations (20a) and (20c), which can be made to go to zero by enforcing the integrands to vanish:

∂JΓ
n(u · ρm vm ) + ρm vm · nu + µm (n · ∇)u − qρm n + αβn +
∂vm
∂µd
−2 n · D(u) · vm = 0, (31a)
∂vm
µt
   
∂ ρd
vm · n + (αvdj ) · n + m − 1 (n · ∇)α β
∂α ρc ρc
t   
(n · ∇) β + C1 + C2 + Γ = 0.
µ ρd ∂J
+ m 1−α −1 (31b)
ρc ρc ∂α

Note that the term containing D(vm ) = 0, because (n · ∇)vm = 0. Decomposing Equation (31a)
into its normal and tangential components yields:

∂JΓ
ρm u · vm + ρm un vm · n + µm (n · ∇)un − ρm q + αβ +
∂vm · n
∂µd
−2 n · D(u) · vm · n = 0, (32a)
∂vm
∂JΓ ∂µ
ρm vm · nut + µm (n · ∇)ut + − 2 d n · D(u) · vmt = 0. (32b)
∂vmt ∂vm

Equations (31b), (32a) and (32b) are used to determine β, q and ut , respectively. Since un is
prescribed at the inlet, the adjoint continuity equation, Equation (18b), is used to calculate un at the
outlet, ΦΓ . The boundary conditions for the adjoint variables at the outlet are summarised as:

n · D(u) · vmt = − Γ ,
∂µd ∂J
ρm vm · nut + µm (n · ∇)ut − 2 (33a)
∂vm ∂vmt
un = Φ Γ , (33b)
µt
   
∂ ρd
vm · n + (αvdj ) · n + m − 1 (n · ∇)α β
∂α ρc ρc
µt
  
(n · ∇) β = −C1 − C2 − Γ ,
ρd ∂J
+ m 1−α −1 (33c)
ρc ρc ∂α
αβ 1 ∂JΓ
u · vm + un vm · n + νm (n · ∇)un + +
ρm ρm ∂vm · n
2 ∂µd
− n · D(u) · vm · n = q, (33d)
ρm ∂vm
µm
where νm = is the mixture kinematic viscosity. A summary of the boundary conditions for the
ρm
adjoint variables is presented in Table 2.

Table 2. Adjoint boundary conditions for ducted flows.

ut un β q
Inlet zero Equation (24b) zero zero gradient
Wall zero Equation (29b) Equation (29c) zero gradient
Outlet Equation (33a) Equation (33b) Equation (33c) Equation (33d)
Fluids 2019, 5, 31 11 of 21

4. Objective Function
The objective function is related to the dispersed-phase mass-flow rate at the boundaries of
the domain,
JΓ = αρd vd · n, (34)

where αρd is the dispersed-phase mass fraction and vd · n is the dispersed-phase velocity normal to
the boundary. Since the phase fraction at the inlet is specified, the objective function is defined as the
mass-flow rate of solid at the outlet, and Equation (12) becomes,
Z
J= JΓo dΓo , (35)
Γo

where o refers to the outlet. The derivatives of the objective function, Equation (34), with respect to the
primal variables are:

∂JΓ
= αρd , (36a)
∂vm · n
∂JΓ
= 0, (36b)
∂vmt
 
∂JΓ ∂vdj
= ρd vm + vdj + α · n, (36c)
∂α ∂α
∂JΓ
= 0. (36d)
∂pm

Derivation of Equation (36c) can be found in Appendix E. Using these derivatives, the adjoint
boundary conditions at an inlet reduces to:

ut = 0, (37a)
un = 0, (37b)
β = 0 ⇐⇒ µtm 6= 0, (37c)
(n · ∇)q = 0. (37d)

At a wall, there is no contribution from the objective function, so:

ut = 0, (38a)
un = 0, (38b)
µtm
   
∂ ρd
(αvdj ) · n + − 1 (n · ∇)α β
∂α ρc ρc
µt
  
ρd
+ m 1−α −1 (n · ∇) β = 0, (38c)
ρc ρc
(n · ∇)q = 0. (38d)
Fluids 2019, 5, 31 12 of 21

Note that C2 = 0 because u = 0. At an outlet, to satisfy the adjoint continuity equation,

Equation (18b), un = 0, so:

2 ∂µd
vm · nut + νm (n · ∇)ut − n · D(u) · vmt = 0, (39a)
ρm ∂vm
un = 0, (39b)
µt
   
∂ ρd
vm · n + (αvdj ) · n + m − 1 (n · ∇)α β
∂α ρc ρc
µt
  
(n · ∇) β = −C1 − Γ ,
ρd ∂J
+ m 1−α −1 (39c)
ρc ρc ∂α
αβ 1 ∂JΓ 2 ∂µd
u · vm + νm (n · ∇)un + + − n · D(u) · vm · n = q. (39d)
ρm ρm ∂vm · n ρm ∂vm

Note that C2 = 0 because un = 0 and D(vm ) = 0 when (n · ∇)vm = 0. A summary of the adjoint
boundary conditions, using the objective function defined in Equation (34), is presented in Table 3.

Table 3. Adjoint boundary conditions, using objective function Equation (34).

ut un β q
Inlet zero zero zero zero gradient
Wall zero zero Equation (38c) zero gradient
Outlet Equation (39a) zero Equation (39c) Equation (39d)

5. Settling Velocity
The equations thus far have been derived for the most general case in which the settling (drift)
velocity has not been specified. Of course the settling velocity is key to the behaviour of the drift
flux model, and incorporates much of the physics of the multiphase system. Here we will derive the
appropriate additional equations for two common settling velocity models, vis. the Dahl [23] and
Takacs [24] models.

5.1. Dahl Model

In this formulation vdj is modelled using,

vdj = v0 10−kα , (40)

where v0 is the maximum theoretical settling velocity and k is a settling parameter, and its partial
derivative with respect to α is given by,

∂vdj
= −k ln 10 vdj . (41)
∂α

5.2. Takacs Model

In this formulation vdj is modelled using,
 
vdj = v0 e−a(α−αr ) − e− a1 (α−αr ) , (42a)
0 6 vdj 6 v00 , (42b)

where:

• a is the hindered settling parameter,

• a1 is the flocculent settling parameter,
Fluids 2019, 5, 31 13 of 21

• αr is the volume fraction of non-settleable solids at the inlet and

• v00 is the maximum practical settling velocity,

and its partial derivative with respect to α is given by,

∂vdj  
= v0 − ae−a(α−αr ) + a1 e−a1 (α−αr ) . (43)
∂α
For both models the partial derivative of αvdj with respect to α is given by,

∂ ∂vdj
(αvdj ) = vdj + α . (44)
∂α ∂α

6. Conclusions
In this paper we have derived, for the first time, the adjoint equations based on the drift flux
model for dispersed multiphase flow. In addition to the adjoint drift flux equations themselves we
have presented the adjoint boundary conditions for the common boundary conditions (Inlet, Outlet,
Wall), as well as a treatment of the generic objective function, and specific formulations corresponding
to the common settling velocity models proposed by Dahl [23] and Takacs [24]. From these elements
a full adjoint set of equations can be derived for any specific ducted flow problem, and of course
implemented in an appropriate numerical code. This of course also presents many, largely numerical
and coding, challenges, and this will be the subject of subsequent papers.
Author Contributions: Formal analysis, S.G.; Funding acquisition, D.J.; Investigation, S.G. and G.R.T.;
Methodology,D.J.; Project administration, G.R.T.; Writing–original draft, S.G.; Writing–review and editing, D.J.
and G.R.T. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the STREAM CDT under grant number EP/L015412/1 and InnovateUK
under KTP grant KTP010621.
Conflicts of Interest: There is no conflict of interest for Hydro International Ltd. in the publication of the paper.
Hydro International holds interests in the implementation and application of the methods discussed in the paper
for specific scenarios and geometries, but this work will remain private and unpublished.

Appendix A. Derivation of Equation (10a)

The variation of ( R1 , R2 , R3 ) T with respect to vm is calculated as,

δvm ( R1 , R2 , R3 ) T = δvm (vm · ∇)(ρm vm ) + ∇(ρm pm ) − ∇ · 2µm D(vm )

  
α ρc ρd
+∇· v v − ρm g + ℵρm vm
1 − α ρm dj dj


= (δvm · ∇)(ρm vm ) + (vm · ∇)(ρm δvm ) − ∇ · 2µm D(δvm )


− ∇ · 2δvm µm D(vm ) + ℵρm δvm .r (A1)

As stated above, µm is defined as the sum of the continuum, dispersed-phase and mixture
turbulent viscosities,
µm = µc + µd + µtm , (A2)

where µc is constant, µd is a function of vm and α, and µtm is obtained from turbulence modelling.
Equation (A1) can now be rewritten as,

δvm ( R1 , R2 , R3 ) T = (δvm · ∇)(ρm vm ) + (vm · ∇)(ρm δvm ) − ∇ · 2µm D(δvm )

− ∇ · 2δvm µd D(vm ) + ℵρm δvm , (A3)

where δvm µtm has been neglected.

Fluids 2019, 5, 31 14 of 21

Appendix B. Derivation of Equation (10g)

The variation of ( R1 , R2 , R3 ) T with respect to α is calculated as,

δα ( R1 , R2 , R3 ) T = δα (vm · ∇)(ρm vm ) + ∇(ρm pm ) − ∇ · 2µm D(vm )

  
α ρc ρd
+∇· v v − ρm g + ℵρm vm
1 − α ρm dj dj



= δα (vm · ∇)(ρm vm ) + δα ∇(ρm pm ) − δα ∇ · 2µm D(vm )


+ δα ∇ · A + δα ρm (ℵvm − g) , (A4)

where
α ρc ρd
A= v v (A5)
1 − α ρm dj dj
and

ρm = αρd + (1 − α)ρc
 
α ρd
= (1 − α ) ρ c 1 + . (A6)
1 − α ρc

Substituting Equation (A6) into Equation (A5) and rewriting the parentheses as binomial
expansions,

(1 − α ) −1 α ρ d −1
 
−1
A = α (1 − α ) ρ c ρ d 1+ vdj vdj
ρc 1 − α ρc
   
1 ρd
= αρd 1 + α 2 − + · · · vdj vdj . (A7)
1 − α ρc
 
As α  1 and ρd ≈ 2ρc =⇒ 2 − 1−1 α ρdc  < 1 and ignoring terms containing squared and higher
 ρ 

powers of α,
A ≈ αρd vdj vdj . (A8)

Substituting Equations (A2) and (A8) into Equation (A4),

δα ( R1 , R2 , R3 ) T ≈ (ρd − ρc ) (vm · ∇)(δαvm ) + ∇(δαpm ) + δα(ℵvm − g)

− ∇ · 2δα µd D(vm ) + ∇ · δα (αρd vdj vdj ), (A9)

where δα µtm has been neglected.

Appendix C. Derivation of Equation (10i)

Similarly, the variation of R5 with respect to α is calculated as,
   
αρc
δα R5 = δα ∇ · (αvm ) + ∇ · v − ∇ · (K ∇α)
ρm dj
= δα ∇ · (αvm ) + δα ∇ · B − δα ∇ · (K ∇α), (A10)

where
αρc
B= v . (A11)
ρm dj
Fluids 2019, 5, 31 15 of 21

Substituting Equation (A6) into Equation (A11) and rewriting the parentheses as binomial
expansions,

(1 − α ) −1 α ρ d −1
 
B = αρc 1+ vdj
ρc 1 − α ρc
   
1 ρd
= α 1+α 1− + · · · vdj . (A12)
1 − α ρc
 
1 ρd 
As α  1 and ρd ≈ 2ρc =⇒ 1 − < 2 and ignoring terms containing squared and higher

1− α ρ c 
powers of α,
B ≈ αvdj . (A13)

As stated above, K is defined as the mixture eddy diffusivity,

t µtm
νm = . (A14)
ρm

1
From Equation (A6), ρm can be written as,

α ρ d −1
 
1 1 −1
= (1 − α ) 1+
ρm ρc 1 − α ρc
  
1 ρd
= 1+α 1− , (A15)
ρc ρc

ignoring terms containing squared and higher powers of α.

Substituting Equations (A14) and (A15) into the term containing K in Equation (A10),

µtm
 
δα ∇ · (K ∇α) = δα ∇ · ∇α
ρm
 t    
µm ρ
= δα ∇ · 1+α 1− d ∇α
ρc ρc
 t    
µm ρ
= ∇· 1 + (α + δα) 1 − d ∇(α + δα) (A16)
ρc ρc
 t    
µm ρd
−∇· 1+α 1− ∇α
ρc ρc
 t     t 
µm ρd µm
= ∇· δα 1 − ∇α + ∇ · ∇δα
ρc ρc ρc
 t   
µm ρ
+∇· α 1 − d ∇δα ,
ρc ρc

ignoring the term containing δα∇δα, because when substitued into Equation (9) becomes terms
containing squared powers of δα. Equation (A10) now becomes,
 t 
µm
δα R5 ≈ ∇ · (δαvm ) + ∇ · δα (αvdj ) − ∇ · ∇δα
ρc
µ t 
ρd

µ t 
ρd

+ m − 1 ∇ · (δα∇α) + m − 1 ∇ · (α∇δα), (A17)
ρc ρc ρc ρc

where δα µtm has been neglected.

Fluids 2019, 5, 31 16 of 21

Appendix D. Derivation of Equation (13)

Decomposing the objective function into contributions from the boundary and interior of the
domain, according to Equation (12), the terms in Equation (11) can be written as follows. The variations
of the objective function can be written as,
Z Z
∂JΓ ∂JΩ
δvm J = dΓ · δvm + dΩ · δvm , (A18)
Γ ∂vm Ω ∂vm
Z Z
δpm J = dΓ Γ δpm +
∂J ∂JΩ
dΩ δpm (A19)
Γ ∂pm Ω ∂pm
and Z Z
∂JΓ ∂JΩ
δα J = dΓ δα + dΩ δα. (A20)
Γ ∂α Ω ∂α
Applying the product rule, divergence theorem and continuity equation, and using the Einstein
notation for clarity, the terms containing u, vm and ∇ can be written as,

Z Z
∂
dΩ u · (δvm · ∇)(ρm vm ) = dΩ uk δvmi (ρ v )
Ω Ω ∂xi mk mk
∂(uk δvmi )
Z Z
∂
= dΩ (u ρ v δv ) − dΩ ρmk vmk
Ω ∂xi k mk mk mi ∂xi Ω
Z Z
∂uk
= dΓ ni uk ρmk vmk δvmi − dΩ ρmk vmk δvmi (A21)
Γ Ω ∂xi
Z
∂δvmi
− dΩ ρmk vmk uk
Ω ∂xi
Z Z
= dΓ n(u · ρm vm ) · δvm − dΩ ∇u · (ρm vm ) · δvm ,
Γ Ω

Z Z
∂
dΩ u · (vm · ∇)ρm δvm = dΩ uk vmi (ρ δv )
Ω Ω ∂xi mk mk
Z Z
∂ ∂
= dΩ (u v ρ δv ) − (u v ) dΩ ρmk δvmk
Ω ∂xi k mi mk mk
Ω ∂xi k mi
Z Z
∂u
= dΓ ni uk vmi ρmk δvmk − dΩ ρmk δvmk vmi k (A22)
Γ Ω ∂xi
Z
∂vmi
− dΩ ρmk δvmk uk
Ω ∂xi
Z Z
= dΓ u(ρm vm · n) · δvm − dΩ (ρm vm · ∇)u · δvm
Γ Ω

and
Z Z
∂
dΩ u · (vm · ∇)(δαvm ) = dΩ uk vmi (δαk vmk )
Ω Ω ∂xi
Z Z
∂ ∂
= dΩ (u v δα v ) − (u v ) dΩ δαk vmk
Ω ∂xi k mi k mk
Ω ∂xi k mi
Z Z
∂u
= dΓ ni uk vmi δαk vmk − dΩ δαk vmk vmi k (A23)
Γ Ω ∂xi
Z
∂v
− dΩ δαk vmk uk mi
Ω ∂xi
Z Z
= dΓ u(vm · n) · vm δα − dΩ (vm · ∇)u · vm δα.
Γ Ω
Fluids 2019, 5, 31 17 of 21

Applying the tensor-vector identity [25], the divergence theorem and a property of the colon
product, demonstrated below,

1 
∇u : D(δvm ) = ∇u : ∇δvm + (∇δvm )T
2
1 
= ∇u : ∇δvm + ∇u : (∇δvm )T
2
1 
= ∇u : ∇δvm + (∇u)T : ∇δvm (A24)
2
1 
= ∇u + (∇u)T : ∇δvm
2
= D(u) : ∇δvm ,

the term containing µm can be written as,

Z
Z
Z
dΩ u · ∇ · 2µm D(δvm ) = dΩ ∇ · 2µm D(δvm ) · u − dΩ ∇u : 2µm D(δvm )
Ω
ZΩ Z Ω

= dΓ 2µm n · D(δvm ) · u − dΩ 2µm D(u) : ∇δvm

ZΓ ZΩ

= dΓ 2µm n · D(δvm ) · u − dΩ ∇ · 2µm D(u) · δvm
Γ Ω
Z

+ dΩ ∇ · 2µm D(u) · δvm (A25)
Ω
Z Z
= dΓ 2µm n · D(δvm ) · u − dΓ 2µm n · D(u) · δvm
Γ Γ
Z

+ dΩ ∇ · 2µm D(u) · δvm .
Ω

Similarly, the terms containing µd can be written as,

Z
Z Z
dΩ u · ∇ · δα µd D(vm ) = dΓ δα µd n · D(vm ) · u − dΓ δα µd n · D(u) · vm
Ω Γ Γ
Z

+ dΩ ∇ · δα µd D(u) · vm (A26)
Ω

and
Z
Z Z
dΩ u · ∇ · δvm µd D(vm ) = dΓ δvm µd n · D(vm ) · u − dΓ δvm µd n · D(u) · vm
Ω Γ Γ
Z

+ dΩ ∇ · δvm µd D(u) · vm . (A27)
Ω

Applying the product rule and divergence theorem, the remaining terms in Equation (11) can be
written as,
Z Z Z
dΩ q∇ · (ρm δvm ) = dΩ ∇ · (qρm δvm ) − dΩ ∇q · ρm δvm
Ω
ZΩ Z Ω

= dΓ qρm n · δvm − dΩ ρm ∇q · δvm , (A28)

Γ Ω

Z Z Z
dΩ β∇ · (αδvm ) = dΩ ∇ · ( βαδvm ) − dΩ ∇ β · αδvm
Ω
ZΩ Z Ω

= dΓ αβn · δvm − dΩ α∇ β · δvm , (A29)

Γ Ω
Fluids 2019, 5, 31 18 of 21

Z Z Z
dΩ u · ∇(ρm δpm ) = dΩ ∇ · (uρm δpm ) − dΩ ∇ · uρm δpm
Ω
ZΩ Z Ω

= dΓ ρm u · nδpm − dΩ ∇ · ρm uδpm , (A30)

Γ Ω

Z Z Z
dΩ u · ∇(δαpm ) = dΩ ∇ · (uδαpm ) − dΩ ∇ · uδαpm
Ω
ZΩ Z Ω

= dΓ u · nδαpm − dΩ ∇ · uδαpm , (A31)

Γ Ω

Z Z

dΩ u · ∇ · δα (αρd vdj vdj ) = dΩ ∇ · uδα (αρd vdj vdj )
Ω Ω
Z
− dΩ ∇ · uδα (αρd vdj vdj ) (A32)
Ω
Z
= dΓ u · nδα (αρd vdj vdj )
Γ
Z
− dΩ ∇ · uδα (αρd vdj vdj ),
Ω

Z Z Z
dΩ q∇ · (δαvm ) = dΩ ∇ · (qδαvm ) − dΩ ∇q · (δαvm )
Ω
ZΩ Z Ω

= dΓ qvm · nδα − dΩ (vm · ∇)qδα, (A33)

Γ Ω

Z Z Z
dΩ β∇ · (δαvm ) = dΩ ∇ · ( βδαvm ) − dΩ ∇ β · (δαvm )
Ω
ZΩ Z Ω

= dΓ βvm · nδα − dΩ (vm · ∇) βδα, (A34)

Γ Ω

Z Z
Z
dΩ β∇ · δα (αvdj ) = dΩ ∇ · βδα (αvdj ) − dΩ ∇ β · δα (αvdj )
Ω
ZΩ Z Ω

= dΓ βδα (αvdj ) · n − dΩ δα (αvdj ) · ∇ β, (A35)

Γ Ω

µt µt
Z Z Z 
dΩ β m ∇ · ∇δα = m dΩ ∇ · ( β∇δα) − dΩ ∇ β · ∇δα
Ω ρc ρc Ω Ω
µtm
Z
= dΓ βn · ∇δα
ρc Γ
Z Z 
− dΩ ∇ · (δα∇ β) + dΩ δα∇ · ∇ β (A36)
Ω Ω
µt
Z
= m dΓ β(n · ∇)δα
ρc Γ
Z Z 
− dΓ δα(n · ∇) β + dΩ δα∇ · ∇ β ,
Γ Ω
Fluids 2019, 5, 31 19 of 21

µtm µt
Z    Z
ρd ρd
dΩ β − 1 ∇ · (δα∇α) = m −1 dΩ ∇ · ( βδα∇α)
Ω ρc ρc ρc ρc Ω
Z 
− dΩ ∇ β · (δα∇α) (A37)
Ω
µtm
 Z
ρd
= −1 dΓ βδα(n · ∇)α
ρc ρc Γ
Z 
− dΩ δα∇α · ∇ β
Ω

and

µtm µt
Z    Z
ρd ρd
dΩβ − 1 ∇ · (α∇δα) = m −1 dΩ ∇ · ( βα∇δα)
Ω ρc ρc ρc ρc Ω
Z 
− dΩ ∇ β · (α∇δα)
Ω
µtm
 Z
ρd
= −1 dΓ βαn · ∇δα (A38)
ρc ρc Γ
Z Z 
− dΩ∇ · (αδα∇ β) + dΩδα∇ · (α∇ β)
Ω Ω
µtm
 Z
ρd
= −1 dΓ αβ(n · ∇)δα
ρc ρc Γ
Z Z 
− dΓ αδα(n · ∇) β + dΩδα∇ · (α∇ β) .
Γ Ω

Equation (11) can now be reformulated and rearranged as Equation (13).

Appendix E. Derivation of Equation (36c)

Decomposing the dispersed-phase velocity into the mixture and dispersed-phase diffusion
velocities, Equation (34) can be rewritten as,

JΓ = αρd (vm + vdm ) · n

 
ρc
= αρd vm + v · n, (A39)
ρm dj

where vdj is defined in terms of vdm , the dispersed-phase velocity relative to the mixture velocity,
ρ
as vdj = ρmc vdm . Substituting Equation (A6) into Equation (A39) and rewriting the parentheses as
binomial expansions,

α ρ d −1
  !
−1
JΓ = αρd vm + (1 − α) 1+ vdj · n (A40)
1 − α ρc
   
α ρd
= αρd vm + (1 + α + · · · ) 1 − + · · · vdj · n (A41)
1 − α ρc
     
1 ρd
= αρd vm + 1 + α 1 − + · · · vdj · n. (A42)
1 − α ρc
 
1 ρd 
As α  1 and ρd ≈ 2ρc =⇒ 1 − < 2 and ignoring terms containing squared and higher

1− α ρ c 
powers of α,
JΓ ≈ αρd (vm + vdj ) · n. (A43)
Fluids 2019, 5, 31 20 of 21

Applying the product rule,

∂JΓ ∂vdj
= ρd (vm + vdj ) · n + αρd ·n
∂α ∂α
 
∂vdj
= ρd vm + vdj + α · n. (A44)
∂α

References
1. Soto, O.; Löhner, R.; Yang, C. An adjoint-based design methodology for CFD problems. Int. J. Num. Meth.
Heat Fluid Flow 2004, 14, 734–759.
2. Giles, M.; Pierce, N. An introduction to the adjoint approach to design. Flow Turb. Combust. 2000, 65, 393–415.
3. Othmer, C. Adjoint methods for car aerodynamics. J. Math. Ind 2014, 4, 1–23.
4. Karpouzas, G.K.; Papoutsis-Kiachagias, E.M.; Schumacher, T.; de Villiers, E.; Giannakouglou, K.C.; Othmer,
C. Adjoint Optimisation for Vehicle External Aerodynamics. Int. J. Automot. Eng. 2016, 7, 1–7.
5. Papoutsis-Kiachagias, E.M.; Asouti, V.G.; Giannakoglou, K.C.; Gkagkas, K.; Shimokawa, S.; Itakura, E.
Multi-Point Aerodynamic Shape Optimization of Cars Based on Continuous Adjoint. Struct. Multidiscip.
Optim. 2019, 59, 675–694. doi:10.1007/s00158-018-2091-3.
6. Reuther, J.; Jameson, A.; Farmer, J.; Martinelli, L.; Saunders, D. Aerodynamic Shape Optimization of Complex
Aircraft Configurations Via an Adjoint Formulation; Paper 94; AIAA: Reston, VA, USA, 1996.
7. Kroll, N.; Gauger, N.; Brezillon, J.; Dwight, R.; Fazzolari, A.; Vollmer, D.; Becker, K.; Barnewitz, H.; Schulz, V.;
Hazra, S. Flow simulation and shape optimization for aircraft design. J. Comput. Appl. Math 2007,
203, 397–411.
8. Campobasso, M.; Duta, M.; Giles, M. Adjoint calculation of sensitivities of turbomachinery objective
functions. J. Propul. Power 2003, 19, 693–703.
9. D. Wang, L.H. Adjoint aerodynamic design optimization for blades in multistage turbomachines; part I:
methodology and verification. J. Turbomach. 2010, 132, 021011.
10. Wang, D.; He, L.; Li, Y.; Wells, R. Adjoint aerodynamic design optimization for blades in multistage
turbomachines–part II: validation and application. J. Turbomach. 2010, 132, 021012.
11. Shape Optimization for Aerodynamic Efficiency using Adjoint Methods; ANSYS White Paper; ANSYS: Canonsburg,
PA, USA, 2016.
12. Alexias, P.; Giannakoglou, K.C. Shape Optimisation of a Two-Fluid Mixing Device using Continuous Adjoint.
Fluids 2020, 5, 11.
13. Martin, M.J.; Andres, E.; Lozano, C.; Valero, E. Volumetric B-splines shape parametrization for aerodynamic
shape design. Aerosp. Sci. Technol. 2014, 37, 26–36.
14. Jakobsson, S.; Amoignon, O. Mesh deformation using radial basis functions for gradient-based aerodynamic
shape optimisation. Comput. Fluids 2007, 36, 1119–1136.
15. Kapellos, C.; Alexias, P.; De Villiers, E. The adjoint mehod for automotive optimisation using a sphericity
based morpher. In Proceedings of the International Association for the Engineering Modelling, Analysis and
Simulation Adjoint CFD Seminar (NAFEMS Adjoint CFD Seminar), Wiesbaden, Germany, 23–24 October
2016.
16. Drew, D.A. Mathematical Modeling of Two-Phase Flow. Ann. Rev. Fluid Mech. 1983, 15, 261–291.
17. Balachandar, S.; Eaton, J. Turbulent Dispersed Multiphase Flow. Ann. Rev. Fluid Mech. 2010, 42, 111–133.
18. Brennan, D. The Numerical Simulation of Two-Phase Flows in Settling Tanks. Ph.D. Thesis, Imperial College
London, London, UK, 2001.
19. Othmer, C. A continuous adjoint formulation for the computation of topological and surface sensitivities of
ducted flows. Int. J. Numer. Methods Fluids 2008, 58, 861–877.
20. Kavvadias, I.S.; Papoutsis-Kiachagias, E.M.; Dimitrakopoulos, G.; Giannakoglou, K.C. The continuous
adjoint approach to the k − ω SST turbulence model with applications in shape optimisation. Eng. Optim.
2015, 47, 1523–1542.
21. Schramm, M.; Stoevesandt, B.; Peinke, J. Optimisation of Airfoils using the Adjoint Approach and the
Influence of Adjoint Turbulent Viscosity. Computation 2018, 6, 23.
Fluids 2019, 5, 31 21 of 21

22. Ubbink, O. Numerical prediction of two fluid systems with sharp interfaces. Ph.D. Thesis, Imperial College
London, London, UK, 1997.
23. Dahl, C. Numerical Modelling of Flow and Settling in Secondary Settling Tanks. Ph.D Thesis, Aalborg
University, Aalborg, Denmark, 1993.
24. Takacs, I.; Patry, G.G.; Nolasco, D. A dynamic model of the clarification-thickening process. Water Res. 1991,
25, 1263–1271.
25. Clarke, D.A. A Primer on Tensor Calculus. Unpublished manuscript, Department of Physics, Saint Mary’s
University, Halifax, NS, Canada, 2011.

c 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).