3.

APPROXIMATIONS AND SIMPLIFIED EQUATIONS SPRING 2010

3.1 Steady-state vs time-dependent

3.2 Two-dimensional vs three-dimensional
3.3 Incompressible vs compressible
3.4 Inviscid vs viscous
3.5 Hydrostatic vs non-hydrostatic
3.6 Boussinesq approximation for density (buoyancy-affected flow)
3.7 Depth-averaged / shallow-water equations
3.8 Reynolds-averaged equations (turbulent flow)
3.9 Potential flow
Examples

Fluid dynamics is governed by equations for mass, momentum and energy. The momentum
equation for a viscous fluid is called the Navier-Stokes equation; it is based upon:
• continuum mechanics;
• the momentum principle;
• a linear stress-strain relationship ( ∝ ∂U / ∂y ).

A fluid for which the last is true is called a Newtonian fluid; this is the case for almost all
fluids in civil engineering. However, there are important non-Newtonian fluids; e.g. blood,
paint, polymer solutions, mud, …. Specialised CFD techniques exist for these.

The full equations are time-dependent, 3-dimensional, viscous, compressible, non-linear and
highly coupled (i.e. the solution of one equation depends on the solution of others). However,
in most cases it is possible to simplify the analysis by adopting a reduced equation set. Some
common approximations are listed below.

Reduction of dimension:
• steady-state;
• 2-d (or axisymmetric).

Neglect of some physical feature:

• incompressible;
• inviscid.

Simplified forces:
• hydrostatic;
• Boussinesq density.

Approximations based upon averaging:

• Reynolds-averaging (turbulent flows);
• depth-averaging (shallow-water equations).

Use of derived variables:

• potential flow (inviscid, incompressible flow).

CFD 3–1 David Apsley

3.1 Steady-State vs Time-Dependent

Many flows are naturally time-dependent. Examples include waves, tides, reciprocating
pumps and internal combustion engines.

Other flows have stationary boundaries but become time-dependent because of an instability.
An important example is vortex shedding around cylindrical objects.

Some computational solution procedures rely on a time-stepping method to march to steady

state; an important example is transonic flow, because the mathematical nature of the
governing equations is different for Ma < 1 and Ma > 1.

Thus, there are three major reasons for using the full time-dependent equations:
• time-dependent problem;
• time-dependent instability;
• time-marching to steady state.

Numerical Consequences.
• Time-dependent equations are parabolic; they are 1st-order in time and are solved by
time-marching, only storing values at one or two time levels at any instant.
• Steady-state equations are elliptic (in incompressible flow) and are solved by implicit,
iterative methods for all points simultaneously.

3.2 Two- Dimensional vs Three-Dimensional

Geometry and boundary conditions may dictate that the flow is two-dimensional (at least in
the mean). Two-dimensional calculations require considerably less computer resources.

“Two-dimensional” may be extended to include “axisymmetric”. This is actually easier to

achieve in the laboratory, where side-wall boundary layers prevent true 2-dimensionality.

3.3 Incompressible vs Compressible

A flow (not a fluid, note) is said to be incompressible if flow-induced pressure or temperature

changes do not cause significant density changes.

For an incompressible flow:

• density is constant along a streamline (D /Dt = 0);
• conservation of mass translates to conservation of volume.

Note that incompressibility does not imply uniform density. Important flows driven by
density differences can arise from variations in salinity (oceans) or temperature (atmosphere).

All fluids are compressible to some degree. However, flow-induced density changes may be
neglected if:
• the Mach number, Ma u/c « 1; (a common rule of thumb is Ma < 0.3);
• absolute temperature changes are small.

CFD 3–2 David Apsley

This is usually the case in civil engineering; (an important exception is water hammer).

Consequences.
Compressible flow:
• transport equations for density and internal energy (or enthalpy);
• pressure derived from an equation of state (e.g., ideal gas law);
• solution by “density-based” methods.

Incompressible flow:
• internal energy is irrelevant; no thermodynamics;
• mass conservation leads to an equation for pressure;
• solution by “pressure-based” methods.

3.4 Inviscid vs Viscous

If viscosity is neglected, the Navier-Stokes equations become the Euler equations.

y
Consider streamwise momentum in a developing 2-d boundary layer: U
∂u ∂u ∂p ∂ 2u
(u +v ) =− +
∂x ∂y ∂x ∂y 2 viscous
Dropping the viscous term reduces the order of the highest space
derivative from 2 to 1 and hence there is one less boundary condition.

Viscous flows (real fluids) require a non-slip condition (zero velocity) y

at rigid walls – the dynamic boundary condition. U
Inviscid (ideal) flows require only the velocity component normal to
the wall to be zero – the kinematic boundary condition. The wall shear inviscid
stress is zero.

Although its direct influence in the interior

of the flow is tiny, viscosity can have a
global influence out of all proportion to its
separation size. The most important example is flow
separation, where the viscous boundary
layer required to satisfy the non-slip
condition is first slowed and then reversed
rec
irc by an adverse pressure gradient. When this
flo ulati occurs there is substantial disturbance to the
w ng
flow and fore-aft pressure differences yield
a pressure drag many times larger than the
viscous drag.

Consequences of the inviscid approximation

• Different boundary condition.
• Can’t be used for viscous phenomena – notably, separated flow.
• Can’t calculate viscous drag, heat transfer or sediment transport.

CFD 3–3 David Apsley

For non-separated flows it is possible to
adopt a flow decomposition using outer
Prandtl’s boundary-layer hypothesis: f lo w
bound
idealise the flow as an inviscid outer region a ry l
ayer
where the flow is driven by pressure
gradients, matched with a thin inner layer
(across which the pressure is effectively
constant) to satisfy the no-slip condition.
The inner-layer solution is often obtained by forward-marching in space (parabolised Navier-
Stokes equations) in a manner similar to time-marching.

This “viscid-inviscid interaction” technique is widely used for aerofoils:

outer flow → pressure distribution and hence “form” lift and drag;
inner flow → viscous drag (usually small).

The boundary-layer hypothesis is OK if the boundary layer is thin, slowly-developing and

doesn’t separate from the surface. Using a potential-flow method in the outer layer offers
considerable computational savings. However, matching-type calculations are seldom used in
general-purpose codes because:
• they limit the class of flows which can be computed;
• the matching region is difficult to establish a priori;
• they don’t work when the flow separates.

3.5 Hydrostatic vs Non-Hydrostatic

The vertical momentum equation can be written:

Dw ∂p
= − − g + viscous forces
Dt ∂z
For large horizontal scales the vertical acceleration Dw/Dt is much less than g and the viscous
forces are small. The balance of terms is the same as the hydrostatic law in a quiescent fluid:
∂p
≈− g i.e. pressure forces balance weight.
∂z

Numerical Consequences.
With this approximation (in constant-density flows
z
with a free surface) the pressure is determined
everywhere by the position of the free surface:
patm
p = p atm + g (h − z ) , where h = h( x, y )
This results in a huge saving in computational time h-z
because the position of the surface automatically
determines the pressure field without the need for a
separate pressure equation. h(x) p = p atm + g (h − z )

The hydrostatic approximation is widely used in

conjunction with the depth-averaged shallow-water x
equations (see below).

CFD 3–4 David Apsley

3.6 Boussinesq Approximation for Density (Buoyancy-Affected Flow) 1

For constant-density flows, pressure and gravitational forces in the vertical momentum
equation can be combined through the piezometric pressure p* = p + gz. Gravity need not
explicitly enter the flow equations unless:
• pressure enters the boundary conditions (e.g. at a free surface, where p = patm);
• there is variable density.
Density variations may arise even at low speeds because of changes in temperature or
humidity (atmosphere), or salinity (water).
stable boundary layer
Temperature variations in the atmosphere, brought
about by surface (or, occasionally, cloud-top) heating ρ u
or cooling, are responsible for significant changes in
airflow and turbulence. mixing depth
• On a cold night the atmosphere is stable.
Cool, dense air collects near the surface and
vertical motions are suppressed; the boundary convective boundary layer
layer depth is 100 m or less.
• On a warm day the atmosphere is unstable. ρ u
Surface heating produces warm; light air near
the ground and convection occurs; the
boundary layer may be 2 km deep.
The vertical dispersion of pollution is very different in the two cases.

If density is a function of some scalar (typically, temperature or, in the oceans, salinity),
then the relative change in density is proportional to the change in ; i.e.
= or = 0 + 0 ( − 0 )
0
where 0 and 0 are reference scalar and density and is the coefficient of expansion; (the
sign convention adopted here is that for salinity, where an increase in salinity lead to a rise in
density – the opposite would be true for temperature-driven density changes).

The Boussinesq approximation for density amounts to retaining density variations in the
gravitational forces but disregarding them in the inertial (mass × acceleration) term; i.e.
Dw ∂p
=− − g
∂z
0
Dt
or, in terms of θ, with the weight resulting from the constant reference density 0 subsumed
into a modified pressure p*:
Dw ∂p *
=− − 0 g where p* = p + 0 gz
∂z
0
Dt 1
424 3
buoyancy force

The approximation is justified if relative density variations are not too large; i.e.
«1
0
This condition is usually satisfied in the atmosphere and oceans.

1
Note that several other very-different approximations are also referred to as the Boussinesq
approximation – e.g. shallow-water equations or eddy-viscosity turbulence models.

CFD 3–5 David Apsley

Actually, whilst the Boussinesq approximation may be necessary to obtain analytical
solutions, it is not particularly important in general-purpose CFD because the momentum and
scalar transport equations are solved iteratively and any scalar-dependent density variations
are easily incorporated in the iterative procedure.

3.7 Depth-Averaged / Shallow-Water Equations

This approximation is used for flow of a constant-density fluid with a free surface, where the
depth of fluid is small compared with typical horizontal scales.
z
In this “hydraulic” approximation, the fluid can be
regarded as quasi-2d with: h(x,t)
• horizontal velocity components u, v;
• depth of water, h. u

Note that h may vary due to changes in both free-

surface and bed height.
x
By applying mass and momentum principles to a vertical column of constant-density fluid of
variable height h, the depth-integrated equations governing the motion can be written (for the
one-dimensional case and in conservative form) as:
∂h ∂ (uh)
+ =0
∂t ∂x
∂ (uh) ∂ (u 2 h) ∂ ( 1 gh 2 ) 1
+ =− 2 + ( surface − bed )
∂t ∂x ∂x
The 12 gh 2 term comes from (1/ times) the hydrostatic pressure force per unit width on a
water column of height h; i.e. average hydrostatic pressure ( 12 gh ) × height (h). The final
term is the net effect of surface stress (due to wind) and the bed shear stress (due to friction).

The resulting shallow-water (or Saint-Venant) equations are mathematically similar to those
for a compressible gas. There are direct analogies between
• hydraulic jumps (shallow flow) and shocks (compressible flow);
• critical flow over a weir (shallow) or gas flow through a throat (compressible).

In both cases there is a characteristic wave speed ( c = gh in the hydraulic case, c = p/

in compressible flow). Depending on whether this is greater or smaller than the flow velocity
determines whether disturbances can propagate upstream and hence the nature of the flow.
The ratio of fluid speed to wave speed is known as:
u
Froude number: Fr = in shallow flow
gh
u
Mach number: Ma = in compressible flow
c

CFD 3–6 David Apsley

3.8 Reynolds-Averaged Equations (Turbulent Flow)

The majority of flows encountered in engineering are turbulent. Most, however, can be
regarded as time-dependent, three-dimensional fluctuations superimposed on a much simpler
mean flow. Generally, however, we are only interested in the mean quantities, rather than
details of the time-dependent flow.

The process of Reynolds-averaging (named after

Osborne Reynolds2, first Professor of Engineering at
Owens College, later to become the University of
Manchester) decomposes each flow variable into mean mean fluctuation
and turbulent parts:
u = u + u′
mean fluctuation

The “mean” may be a time average (this is usually what is measured in the laboratory) or an
ensemble average (a hypothetical statistical average over a large number of identical
experiments).

When the averaging process is applied to the Navier-Stokes equation, the result is:
• an equivalent equation for the mean flow,
except for
• turbulent fluxes, − u ′v ′ etc. (called the Reynolds stresses) which provide a net
transport of momentum.

For example, the viscous shear stress

∂u
visc =
∂y
is supplemented by an additional turbulent stress (see Section 7):
′ ′
turb = − u v

In order to solve the mean-flow equations, a turbulence model is required for these turbulent
stresses. Popular models exploit an analogy between viscous and turbulent transport and
employ an eddy viscosity t to supplement the molecular viscosity. Thus,
∂u ∂u
= − u ′v ′ → ( + t )
∂y ∂y
This is readily incorporated into the mean momentum equation because it simply requires a
(position-dependent) effective viscosity. However, actually specifying t is by no means
trivial – see the lectures on turbulence modelling (Sections 7 and 8).

2
Osborne Reynolds’ experimental apparatus – including that used in his famous pipe-flow experiments – is on
display in the basement of the George Begg building at the University of Manchester.

CFD 3–7 David Apsley

3.9 Potential Flow

The angular momentum of a fluid element can only change due to the action of viscous forces
(since pressure forces always act perpendicular to a surface and cannot impart rotation). For
an inviscid, incompressible fluid the flow can be regarded as irrotational. In these
circumstances it can be shown that the velocity u can be related to a velocity potential via:
∂φ ∂φ ∂φ
u= , v= , w= (abbreviated as u = ∇φ )
∂x ∂y ∂z

∂u ∂v ∂w
Substituting in the continuity equation, + + = 0 , produces a Laplace equation:
∂x ∂y ∂z
∂ 2φ ∂ 2φ ∂ 2φ
+ + =0 (abbreviated as ∇ 2 φ = 0 )
∂x 2 ∂y 2 ∂z 2

Important consequence.
The entire 3-d flow field is completely described by a single scalar field φ. Moreover, since
Laplace’s equation occurs in many branches of physics, many good solvers exist.

Velocity components u, v and w are obtained by differentiating φ. Pressure is then

recoverable from Bernoulli’s theorem:
p + 12 U 2 = constant (along a streamline)
where U is the magnitude of velocity.

The potential-flow assumption often gives an adequate description of the flow and pressure
fields for real fluids, except very close to solid surfaces where viscous forces are significant.
It is useful, for example, in calculating the lift force on aerofoils and in wave theory.
However, in ignoring the effects of viscosity it implies that there are neither tangential
stresses on boundaries nor flow separation, which leads to the erroneous conclusion that an
object moving through a fluid experiences no drag.

CFD 3–8 David Apsley

Examples

Q1.
Discuss the circumstances under which a fluid flow can be adequately approximated as:
(a) incompressible;
(b) inviscid.

Q2.
A velocity field is given by the velocity potential φ = x 2 − y 2 .
(a) Calculate the velocity components u and v.
(b) Calculate the acceleration.
(c) Calculate the corresponding streamfunction, .
(d) Sketch the streamlines and suggest a geometry in which one might expect this flow.

Q3.
For incompressible flow in a rotating reference frame the force per unit volume f is the sum
of pressure, gravitational, Coriolis and viscous forces:
f = −∇p + g − 2 ∧ u + ∇ 2u
where g = (0, 0, –g) is gravity and is the angular velocity of the rotating frame.

(a) If the density is uniform, show that pressure and gravitational forces can be combined
in a piezometric pressure (which should be defined).

(b) If density variations arise from a scalar field, describe the “Boussinesq”
approximation (in this context) and give an application in which it is used.

(c) Show how the momentum equation (with Boussinesq approximation for density) can
be non-dimensionalised in terms of Froude number, Rossby number and Reynolds
number:
U0 U U L
Fr = , Ro = 0 , Re = 0 0 0
( / 0 ) gL0 L0
where 0, L0, U0 are characteristic density, length and velocity scales, respectively,
and is a typical magnitude of density variations.

CFD 3–9 David Apsley

Q4.
(a) In flow with a free surface, by taking a control volume as a column of (time-varying)
height h(x,y,t) and horizontal cross-section x × y, assuming that density is constant,
13 and 23 are the only significant stress components, the horizontal velocity field
may be replaced by the depth-averaged velocity (u, v) and the pressure is hydrostatic,
derive the shallow-water equations for continuity and x-momentum in the form
∂h ∂ (hu ) ∂ (hv)
+ + =0
∂t ∂x ∂y
∂ (hu ) ∂ (hu 2 ) ∂ (hvu ) ∂z ( surface) − 13 (bed )
+ + = − gh s + 13
∂t ∂x ∂y ∂x

(b) Provide an alternative derivation by integrating the continuity and horizontal

momentum equations for incompressible flow:
∂u ∂v ∂w
+ + =0
∂x ∂y ∂z
∂ ( u ) ∂ ( u 2 ) ∂ ( vu ) ∂ ( wu ) ∂p ∂
+ + + = − + 13
∂t ∂x ∂y ∂z ∂x ∂z
over a depth h = zs – zb.

For part (b) you will need the boundary condition that the top and bottom surfaces
z = z s ( x, y ) and z = z b ( x, y ) are material surfaces:
D ∂z ∂z ∂z
(z − zs ) = 0 or w − s − u s − v s = 0 on z = zs
Dt ∂t ∂x ∂y
and similarly for zb, together with Leibniz’ Theorem for differentiating an integral:
⌠ ∂f
b( x) b( x)
d ⌠ db da
′
 f ( x ) dx =  ′ dx ′ + f (b) − f (a)
dx ⌡a ( x ) ⌡a ( x ) ∂x ′ dx dx

Note: this is easily extended to consider additional forces such as Coriolis forces and
other stress terms (“horizontal diffusion”). For undergraduates Professor Stansby will
cover this in the second part of the Computational Hydraulics course.

CFD 3 – 10 David Apsley

Selected Answers

Q2. (a) (u , v) = (2 x,−2 y )

(b) ( a x , a y ) = ( 4 x, 4 y )
(c) = 2 xy (+ constant )
(d) Flow in a rectangular corner

Q3. (a) − ∇p + 0 g = −∇( p + 0 gz )

(b) Boussinesq approximation: ignore density variations in all terms except the
buoyancy force ( – 0) g; often applied in geophysical flows.
Du 2 1 2
(c) = −∇p − 2 e z − e ∧u+ ∇u
Dt Fr Ro Re

CFD 3 – 11 David Apsley