Fluids 1
Fluids 1
Fluids 1
1 Equations of motion
1.1 Introduction
In this section we will derive the equations of motion for an inviscid fluid, that is a
fluid with zero viscosity. We begin by setting up the basic concepts that are needed to
describe the motion of a continuous medium in three dimensions, and the fundamental
kinematic equations relating the density to the deformation and the velocity. Next we
derive the momentum equation for an inviscid fluid. By assuming that the fluid has
constant density, we obtain a closed system of equations for the velocity and pressure,
known as the Euler equations. We introduce the concepts of vorticity and circulation,
and explain why it is reasonable to assume that most flows are irrotational. Finally, we
will show that incompressible irrotational flow is governed by Laplace’s equation.
It is worth emphasising that this course relies on familiarity with various concepts
from Mods, in particular vector calculus and manipulations of multidimensional inte-
grals. The basic material relevant to this section is collected in §1.6.
1.2 Kinematics
1.2.1 Preliminaries
The term kinematics refers to “the science of pure motion, considered without reference
to the matter or objects moved or the force producing or changing the motion.”1 In this
section, we will examine what can be said about the motion of any continuous medium,
although we will often use the word “fluid” to help fix ideas. We will later restrict our
attention to inviscid fluids in §1.3.
In a continuous medium, all state variables, such as density, velocity and pressure,
are assumed to be continuous functions of position x and time t; in fact in this course
we will assume that all dependent variables are continuously differentiable.
X x
x x
V (t)
V (0)
z z
Figure 1.1: Schematic of the deformation of a fluid occupying a volume V (t). The
highlighted particle has Eulerian coordinate x and Lagrangian coordinate X.
of each material element, while the current position vector x gives its Eulerian coor-
dinates. A deformation of the medium corresponds to a mapping from each element’s
initial position to its current position at time t, that is a vector-valued transformation
x = x(X, t). Our assumption that the medium is continuous implies that this mapping
should be continuous and one-to-one, so that each element in the reference configura-
tion is displaced continuously to a unique element in the deformed state. A sufficient
condition for this to be true is2
0 < J < ∞, (1.1)
Now, fixing X corresponds to tracking how f varies for a particular material element
that moves with the deforming fluid.
These two viewpoints prompt us to define two different time derivatives. We use
the usual partial derivative notation to denote the Eulerian time derivative, at a fixed
position x in space, that is
∂ ∂
:= = rate of change with x held constant. (1.4)
∂t ∂t x
D ∂
:= = rate of change with X held constant, (1.5)
Dt ∂t X
to denote the Lagrangian time derivative (the notation d/dt is also often employed).
With X held constant, D/Dt corresponds to the rate of change following an element
that convects with the fluid, and it is referred to as the convective derivative or the
material derivative or sometimes the derivative following the flow.
Df Dt ∂f Dx
(x, t) = (x, t) + · ∇f (x, t) (1.7)
Dt Dt ∂t Dt
and hence
Df ∂f
≡ + u · ∇f, (1.8)
Dt ∂t
for any differentiable function f . We can write this in operator form as
D ∂
≡ + (u · ∇), (1.9)
Dt ∂t
where (u · ∇) is shorthand for the directional derivative
∂ ∂ ∂
(u · ∇) ≡ u +v +w (1.10)
∂x ∂y ∂z
In particular, we can now compute the acceleration of the fluid, namely the material
rate of change of the velocity:
Du ∂u
≡ + (u · ∇)u. (1.11)
Dt ∂t
Note the way that the second term is grouped: (u · ∇) is a linear scalar differential
operator which can easily be applied to the vector u. Had we instead written this term
as u · (∇u) we would have faced the problem of defining the grad of a vector. This can
be done, but is to be avoided throughout this course.
x = Aes , y = Be−s , z = C,
where A, B and C are integration constants. These parametrise the hyperbolae xy = AB = const
in the (x, y)-plane, as illustrated in Figure 1.3.
Part A Fluid Dynamics & Waves Draft date: 21 January 2014 1–5
y0
-2
-4
-4 -2 0 2 4
x
Plotting streamlines is similar to plotting phase plane trajectories for plane au-
tonomous differential equations. The streamlines have a unique tangent vector equal to
u at each point in the flow. Hence they can only cross at so-called stagnation points,
where u = 0 and the fluid is locally stationary. In Example 1.1, there is just one
stagnation point at the origin, and it resembles a saddle point in a phase plane.
An alternative flow visualisation strategy, often used in experiments, is to insert tiny
tracer particles into the flow and follow their trajectories. Assuming that each particle
moves with the local flow velocity, its position vector x(t) must satisfy the differential
equation
dx(t)
= u x(t), t . (1.12)
dt
Solutions of this equation are called particle paths for the flow u(x, t).
If the flow is steady (i.e. u is independent of t), then we see that the streamline
pattern will not vary with time, and that the streamlines and particle paths will coincide.
For an unsteady flow, though, the streamline pattern will in general vary with time and
not coincide with the particle paths.
Example 1.2 For the two-dimensional unsteady flow u = (cos t, sin t, 0), the streamlines sat-
isfy
dx dy dz
= cos t, = sin t, = 0, (1.13)
ds ds ds
with t held fixed, and hence are given by
x = A + s cos t, y = B + s sin t, z = C,
where A, B and C are integration constants. These give a family of parallel straight lines in the
(x, y)-plane whose direction rotates as t varies, as shown in Figure 1.4.
1–6 OCIAM Mathematical Institute University of Oxford
2 2 2 2
y 0
y 0
y 0
y 0
-2 -2 -2 -2
-4 -4 -4 -4
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
x x x x
y0
-2
-4
-4 -2 0 2 4
x
dx dy dz
= cos t, = sin t, = 0, (1.14)
dt dt dt
x = A + sin t, y = B − cos t, z = C,
for some constants A, B, C. Particles therefore trace out unit circles in the (x, y)-plane; some
examples are shown in Figure 1.5.
Recall that the Jacobian relates infinitesimal volumes in the Eulerian and Lagrangian
frames, via dxdydz = JdXdY dZ. Hence we can interpret J as measuring the local ex-
pansion or contraction: the fluid is expanding if J is increasing with time or contracting
if J decreases with time. The identity (1.20) shows how this local expansion or con-
traction of the medium is related to the divergence of the velocity. A flow is said to
be incompressible or volume-preserving if it preserves infinitesimal volumes, that is if
DJ/Dt ≡ 0. From (1.20), we see that
To prove this important result, we transform the integral on the left-hand side into
Lagrangian variables to obtain
ZZZ ZZZ
I(t) := f dxdydz = f J dXdY dZ, (1.23)
V (t) V (0)
where J again denotes the Jacobian (1.2). In (1.23), the Lagrangian integral is over the
fixed initial domain V (0) corresponding to the moving volume V (t). We can therefore
differentiate through the integral to obtain
ZZZ
dI D
= (f J) dXdY dZ, (1.24)
dt V (0) Dt
where the time derivative is taken with the integration variables (X, Y, Z) held fixed.
Now we expand out the derivative in (1.24) and use Euler’s identity (1.20) to obtain
ZZZ ZZZ
dI Df Df
= + f ∇ · u J dXdY dZ = + f ∇ · u dxdydz. (1.25)
dt V (0) Dt V (t) Dt
The definition (1.9) of the convective derivative then leads to Reynolds’ Transport The-
orem (1.22).
u
D ∂D
Figure 1.6: Schematic of a region D, fixed in space, with fluid flowing in and out through
its boundary ∂D.
where ρ is the density, which may in general vary with both position and time. The net
rate at which mass flows out of D is given by
ZZ
ρu · n dS,
∂D
where n is the unit outward-pointing normal to ∂D. Since mass can neither be created
nor destroyed inside D, we must have
ZZZ ZZ
d
ρ dxdydz = − ρu · n dS. (1.26)
dt D ∂D
We can commute the differentiation and integration on the left-hand side to write
ZZZ ZZZ
d ∂ρ
ρ(x, t) dxdydz ≡ (x, t) dxdydz. (1.27)
dt D D ∂t
Note that, when we differentiate through the integral, the time derivative ∂/∂t is per-
formed while holding the integration variables (x, y, z) constant. Applying the Diver-
gence Theorem to the right-hand side of (1.26), we therefore obtain
ZZZ
∂ρ
+ ∇ · (ρu) dxdydz = 0. (1.28)
D ∂t
This result must hold for any fixed volume D, and it follows that (assuming it is con-
tinuous) the integrand must be zero. We therefore deduce the equation
∂ρ
+ ∇ · (ρu) = 0 (1.29)
∂t
relating the density and velocity in any continuous medium. We can expand out the
divergence here to write (1.29) in the equivalent form
Dρ
+ ρ∇ · u = 0, (1.30)
Dt
1–10 OCIAM Mathematical Institute University of Oxford
which demonstrates how the rate of change of the density and the divergence of the
velocity field are intimately related.
Using a Lagrangian approach, we would instead consider the mass of a material
volume V (t) that is convected by the flow, so that it always consists of the same fluid
elements. As above, we can write the net mass inside V in the form
ZZZ
ρ dxdydz.
V (t)
Now the integration region V varies with t, so we cannot directly differentiate through
the integral. Instead, we can use the Transport Theorem (1.22) to obtain
ZZZ ZZZ
d ∂ρ
ρ dxdydz ≡ + ∇ · (ρu) dxdydz. (1.31)
dt V (t) V (t) ∂t
Since the volume V (t) is defined to consist always of the same fluid elements, its mass
cannot change with time. This must be true for all material volumes, and, as above, we
deduce that the integrand on the right-hand side of (1.31) must be zero (assuming it is
continuous). Hence we reproduce the mass conservation equation (1.29).
We can use (1.29) to deduce the following useful corollary of the transport theorem.
If f = ρh in (1.22), where h is any continuously differentiable function, then
ZZZ ZZZ
d Dh
ρh dxdydz ≡ ρ dxdydz. (1.32)
dt V (t) V (t) Dt
In this course, we will ususally think of g as being the acceleration due to gravity,
although it might also incorporate other effects such as electromagnetic forces on a
liquid metal.
Part A Fluid Dynamics & Waves Draft date: 21 January 2014 1–11
Second there is the internal force exerted on each volume V by the surrounding fluid.
We suppose that this may be accounted for by a pressure, p, which acts in the inward
normal direction at each point, so the net internal force on V is
ZZ ZZZ
−pn dS = −∇p dxdydz,
∂V V
To calculate the left-hand side, we apply the transport theorem corollary (1.32) and
hence obtain ZZZ
Du
ρ + ∇p − ρg dxdydz = 0, (1.34)
V Dt
which must hold for all material volumes V . It follows that (assuming it is continuous)
the integrand must be zero, and we therefore obtain the momentum equation
Du
ρ = −∇p + ρg. (1.35)
Dt
∇ · u = 0, (1.36)
and we recall from (1.21) that the flow is therefore incompressible. The implication
does not quite go the other way. If the flow is incompressible, then (1.30) reduces to
3
cf B6a Viscous Flow
4
cf C6.1a Solid Mechanics, C6.2b Elasticity and Plasticity
5
cf B6b Waves & Compressible Flow
1–12 OCIAM Mathematical Institute University of Oxford
Dρ/Dt = 0, so that ρ is preserved following the flow, but need not be constant (this
can occur for example in stratified fluids). However, the term incompressible is often
slightly abused to refer to constant-density fluids, and we will assume that ρ is constant
throughout the remainder of this course.
By expanding out the convective derivative, we can write the momentum equation
(1.35) in the form
∂u 1
+ (u · ∇)u = − ∇p + g. (1.37)
∂t ρ
Now (1.36) and (1.37) amount to a closed system of four scalar equations for p and the
three components of u, known as the Euler equations.
If we assume that the body force is conservative, then it may be written as g = −∇χ
in terms of a potential χ. For example, a constant gravitational acceleration in the −z-
direction corresponds to g = −gez and hence χ = gz. We also note the vector identity
1 2
(u · ∇)u ≡ ∇ |u| + (∇ × u) × u, (1.38)
2
whose proof is a straightforward exercise.
We can therefore rearrange (1.37) to the alternative form
∂u p 1 2
+ (∇ × u) × u = −∇ + |u| + χ . (1.39)
∂t ρ 2
For steady flow, the first term on the left-hand side is zero. If we dot the whole equation
with u, then the second term also disappears, since we end up with a triple scalar
product [u, ∇ × u, u] with two repeated entries. It follows that
p 1 2
u·∇ + |u| + χ = 0 (1.40)
ρ 2
when ∂u/∂t = 0, and from this we deduce that
p 1 2
+ |u| + χ is constant along streamlines in steady flow. (1.41)
ρ 2
This is known as Bernoulli’s Theorem for steady flow. We will see shortly that
various different versions of Bernoulli’s Theorem may apply when the flow is not steady.
Figure 1.7: A closed curve C(t) in flows with (i) zero circulation, (ii) positive circulation,
(iii) negative circulation.
The circulation is thus the net flow along the closed curve C(t). Figure 1.7 shows
schematically how circulation is related to rotation in the flow. In diagram (i) there is
no rotation. The net clockwise and anticlockwise flows around C will cancel, resulting
in a net circulation of zero. In diagram (ii), there is an anti clockwise rotation in the
flow, resulting in a positive circulation about C. Finally, in diagram (iii) we see that a
clockwise-rotating flow leads to a negative circulation about C.
We can also relate circulation to vorticity, since Stokes’ Theorem implies that
ZZ ZZ
Γ(t) = (∇ × u) · n dS = ω · n dS, (1.48)
S(t) S(t)
where S is any surface spanning C. This reinforces the connection between vorticity
and rotation in the flow alluded to in §1.4.1.
Kelvin’s Circulation Theorem states that Γ is independent of t, and we will
prove it by showing that dΓ/dt is zero. To differentiate Γ, it is helpful to transform the
integral to Lagrangian variables, using the chain rule:
I X I X ∂xi
Γ= ui dxi = ui dXj . (1.49)
C(t) C(0) ∂Xj
i i,j
With respect to Lagrangian variables, the integral is taken around the fixed initial curve
C(0). We can therefore now differentiate through the integral to obtain
I I
dΓ d X ∂xi D X ∂xi
= ui dXj = ui dXj , (1.50)
dt dt C(0) ∂Xj C(0) Dt ∂Xj
i,j i,j
holding the integration variables X constant when performing the time derivative D/Dt.
We expand out the derivative in the integrand, using the fact that D/Dt commutes with
∂/∂Xj , to obtain
dΓ
I X Dui ∂xi ∂ui
= + ui dXj
dt C(0) i,j Dt ∂Xj ∂Xj
I X Dui I X ∂ui
= dxi + ui dxj . (1.51)
C(t) Dt C(t) ∂xj
i i,j
The second integrand may be rewritten as ∂j ( 12 |u|2 ), and we use (1.37) to substitute
for the acceleration in the first integral. Swapping i and j in the second integral and
combining it with the first integral gives
dΓ
I X ∂ p 1 2
p 1 2
= − − χ + |u| dxi = − − χ + |u| , (1.52)
dt C(t) ∂xi ρ 2 ρ 2 C(t)
i
where [·]C(t) denotes the change in · as the closed loop C is traversed. Since p, χ and
u are all single-valued functions of position, we deduce that the right-hand side is zero
and, hence, that Γ is constant.
Part A Fluid Dynamics & Waves Draft date: 21 January 2014 1–15
Now, we can use this property to show that, if the vorticity is initially zero, then it
remains zero for all time. Suppose for contradiction that ∇ × u = 0 at t = 0 but that
∇ × u is nonzero at some later time t. By (1.48), we can thus find a closed loop C(t)
such that the circulation Γ(t) is nonzero. Since Γ is independent of t, Γ(0) must likewise
be nonzero, which is impossible because ∇ × u was supposed to be zero initially.
At first glance, this might seem like a far-fetched assumption. However, we note that
the trivial solution ω ≡ 0 is consistent with the vorticity equation (1.46). Furthermore,
we have just argued from Kelvin’s Circulation Theorem that an initially irrotational
flow must remain irrotational for all time. Therefore, it is difficult to create vorticity in
an inviscid fluid, and actually rather likely that a flow will be irrotational.
For an irrotational flow, the momentum equation (1.39) simplifies to
∂u p 1 2
= −∇ + |u| + χ . (1.54)
∂t ρ 2
If the flow is steady, so the left-hand side is zero, we see that the bracketed term on the
right-hand side must be constant and therefore deduce that
p 1 2
+ |u| + χ is constant everywhere in steady irrotational flow. (1.55)
ρ 2
This is Bernoulli’s Theorem for steady irrotational flow, and should be compared
with the weaker version (1.41) that holds for general steady flow.
u ≡ ∇φ. (1.56)
C2
C1 − C2
C1
S C1 − C2
Figure 1.8: Schematic of two paths C1 and C2 joining the origin 0 to a point x, along
with the closed path C1 − C2 formed by joining them, spanned by a surface S.
where S is any surface spanning C1 − C2 , and this is zero since the flow is assumed to
be irrotational.
Hence φ is well defined by (1.57), up to the arbitrary function φ0 (t), and it is a simple
exercise to show that φ then satisfies (1.56). Then the incompressibility condition (1.36)
gives us
∇ · u ≡ ∇ · (∇φ) ≡ ∇2 φ = 0, (1.60)
so that φ satisfies Laplace’s equation. This is very much easier to solve than the nonlinear
Euler equations: given suitable boundary conditions, all the standard techniques such
Part A Fluid Dynamics & Waves Draft date: 21 January 2014 1–17
as separation of variables, transforms, etc. can be used to solve for φ and hence the
velocity field u.
The pressure may be found a posteriori from the momentum equation (1.35). This
final step may be simplified as follows. With ∇ × u ≡ 0 and u ≡ ∇φ, equation (1.39)
becomes
∂∇φ p 1 2
= −∇ + |∇φ| + χ . (1.61)
∂t ρ 2
Since the t-derivative commutes with ∇, we can rearrange this to
∂φ 1 2 p
∇ + |∇φ| + + χ = 0. (1.62)
∂t 2 ρ
∂φ p 1
+ + |∇φ|2 + χ = F (t) in irrotational flow. (1.63)
∂t ρ 2
∂ φ̃ p 1
+ + |∇φ̃|2 + χ = F (t) − f 0 (t). (1.65)
∂t ρ 2
Hence the function F (t) may be chosen arbitrarily by simply absorbing a suitable
function of t into φ. For example, we can obtain (1.63) with F (t) ≡ 0 by choosing
f 0 (t) = F (t).
where we define x1 ≡ x, x2 ≡ y, x3 ≡ z, e1 ≡ i, e2 ≡ j, e3 ≡ k.
First we recall that the grad of a scalar function φ is defined by
∂φ ∂φ ∂φ X ∂φ
grad φ ≡ ∇φ := i +j +k ≡ ek ; (1.67)
∂x ∂y ∂z ∂xk
k
∇ × (∇φ) ≡ 0, ∇ · (∇ × u) ≡ 0. (1.70)
Directional derivative The directional derivative of a scalar function φ(x) along the
vector u is given by
u · (∇φ) ≡ (u · ∇)φ, (1.72)
where (u · ∇) denotes the differential operator
∂ ∂ ∂ X ∂
(u · ∇) := (i · u) + (j · u) + (k · u) ≡ (ek · u) . (1.73)
∂x ∂y ∂z ∂xk
k
In this way, we can make sense of the directional derivative of a vector field v(x) without
addressing the problem of defining the grad of a vector:
X ∂v
(u · ∇)v := (ek · u) . (1.74)
∂xk
k
Differentiation under the integral Given a function f (x, t), the integral
Z b
I(t) = f (x, t) dx
a
is a function of t alone. When computing the derivative of I(t), the order of integration
and differentation may be reversed, so that
Z b Z b
d ∂f
f (x, t) dx ≡ (x, t) dx. (1.76)
dt a a ∂t
Note that the partial derivative ∂f /∂t is performed while holding the integration variable
x constant.
Equation (1.76) is a special case of Leibnitz’ rule: if the limits a and b also depend
on t, then
Z b(t) Z b(t)
d ∂f
f (x, t) dx ≡ (x, t) dx + f b(t), t ḃ(t) − f a(t), t ȧ(t). (1.77)
dt a(t) a(t) ∂t
Line integrals A space curve C may be described using a single parameter, ξ say, by
x = x(ξ) (i.e. x = x(ξ), y = y(ξ), z = z(ξ)), where ξ lies in some interval, say [a, b].
Then, given a scalar field φ(x) and vector field u(x), we define
Z Z b
dx(ξ)
φ dx :=φ x(ξ) dξ, (1.78a)
C a dξ
Z Z b
dx(ξ)
u · dx := u x(ξ) · dξ, (1.78b)
C a dξ
Z Z b
dx(ξ)
φ ds := φ x(ξ) dξ. (1.78c)
C a dξ
H
The notation is sometimes used to distinguish integrals around closed curves, i.e.
those where x(a) = x(b).
where J is the Jacobian of the transformation from (ξ, η, ζ) to (x, y, z), that is
For any surface S spanning a simple closed curve C, Stokes’ Theorem states that
ZZ I
(∇ × u) · n dS ≡ u · dx, (1.84)
S C
where the orientation of the normal n is chosen such that C rotates around it in a
right-handed sense.