# MTH-3D41 Fluid Dynamics Lecture Notes Dr Mark Blyth [1] D. J. Acheson Elementary ﬂuid dynamics, Oxford. [2] D. J.

Tritton Physical Fluid Dynamics, Oxford. [3] G. K. Batchelor An introduction to ﬂuid dynamics, Cambridge. To see the theory in action, the following is a book of photographs of ﬂuid motion: [4] M. Van-Dyke An album of ﬂuid motion, Parabolic Press.

Contents 0. Introduction 1. Preliminary ideas 2(a) Stress in ﬂuids 2(b) Strain in ﬂuids 2(c) Stress-strain relationship for a Newtonian ﬂuid 3. The Navier-Stokes equations 4. Vorticity dynamics and the stream function 5. Exact solutions of the Navier-Stokes equations 6. Stokes ﬂow 7. Lubrication theory 8. Boundary layers 9. Hydrodynamic stability Appendix: Invariants of a matrix under rotation.

1

0. Introduction The study of ﬂuid mechanics encompasses many varied applications in both natural and engineering processes. Some examples are Aerodynamics – ﬂow past wings; computation of drag and lift; D’Alembert’s paradox Biological ﬂuid dynamics – ﬁsh swimming; ﬂow in arteries and veins; blood ﬂow in vascular tumours; Korotkoﬀ sounds Geophysical ﬂuid dynamics (GFD) – oceanography and meteorology; climate and weather prediction Industrial ﬂuid dynamics – ﬂow past structures and buildings; coating ﬂows (e.g. deposition of photographic ﬁlms); spraying processes Magnetohydrodynamics (MHD) – ﬂow of conducting ﬂuids; Maxwell’s equations Physiochemical ﬂuid dynamics – ﬂow under surface tension in the presence of chemical contaminants Turbulence – transition to turbulence from laminar ﬂow; boundary layer receptivity All of these are active areas of research We will show that the Navier-Stokes equations for an incompressible ﬂuid are: ∇ · u = 0, where u is the velocity in the ﬂuid p is the pressure ρ is the density µ is the viscosity. If U and L are typical velocity and length scales for the problem in hand, we will also see that a given ﬂow may be characterised by the value of the dimensionless Reynolds number R= UL . ν ∂u + u · ∇u = −∇p/ρ + (µ/ρ)∇2 u, ∂t

Simpliﬁcations can be made if R ≪ 1 (see Section 6) or R ≫ 1 (see Section 8). 1. Preliminary ideas What is a ﬂuid? A simple ﬂuid cannot withstand an externally applied stress without deforming in some manner. This deﬁnition of a ﬂuid encompasses both liquids and gases. In order to develop a mathematical basis for the study of ﬂuid motion, we adopt the continuum hypothesis, 2

which states that the material under scrutiny is assumed to behave as if it is perfectly continuous. In real materials, ﬂuctuations at the molecular level mean that this hypothesis is inaccurate at very small scales. However, in most situations we are concerned with macroscopic observations of a given ﬂuid motion and are not troubled by the irregular variations apparent on molecular lengthscales. Focussing on a tiny volume of a given ﬂuid, we make the approximation that physical quantities, such as mass and momentum, are spread uniformly over that volume rather than unevenly distributed within it. Under this assumption we may deﬁne, for example, the ﬂuid velocity u(x, t), density ρ(x, t), or temperature T (x, t), valid at a point x and time t in the ﬂow ﬁeld and treat them as continuous, diﬀerentiable quantities over the domain of interest. A viscous ﬂuid is distinguished by its ability to resist shearing motions. A perfect or inviscid ﬂuid oﬀers no resistance to shear. The viscosity, µ, of a ﬂuid is a measure of its ability to withstand such shearing action. The kinematic viscosity ν of a ﬂuid is deﬁned to be the ratio between the viscosity and the density, so ν = µ/ρ. Describing a ﬂuid ﬂow There are two diﬀerent frameworks avaliable within which to describe a ﬂuid’s motion. These are the: Eulerian framework: In this approach, the ﬂuid is considered as a whole. Global vector and scalar variables such as velocity and temperature are deﬁned and solved for over the entire ﬂow-ﬁeld simultaneously. This is by far the most widely used. Lagrangian framework: In this formulation, individual ﬂuid particles are tracked. Each particle is labelled according to its known location at a given time. For example, a particular particle is labelled by its position x0 at time t = 0. Then, at a later time t > 0, it has position x(x0 , t), velocity u = (∂x/∂t)x0 , and acceleration (∂u/∂t)x0 , where the labels on the brackets indicate that the diﬀerentiations are performed for constant x0 , i.e. for this one particular particle. This description of the ﬂow tends to be rather cumbersome and is little used. Compressibility What role do compressibility eﬀects play in determining the motion of a ﬂuid? Air is clearly much more easily compressible than water, say, but to what extent is this important in a particular ﬂow? Bernoulli’s theorem states that in a steady, inviscid ﬂow with no body forces acting, the pressure is related to the velocity via 1 p + ρu2 = constant along a streamline. 2 It follows that a change in speed of a ﬂuid element from zero to U requires a pressure change 3

For water. changes in pressure occur with no corresponding change in density. By deﬁnition. 4 .5 × 103 ms−1 . 2(a). c2 = dp . 2. speciﬁcally. Examples: 1. dρ where c is the speed of sound. Long-range forces: Externally applied forces such as gravity. Stress in ﬂuids The forces acting on a ﬂuid in motion are divided into two categories: 1. Short-range forces: These refer to local dynamic stresses developing within the ﬂuid itself as it moves. 3. the eﬀects of compressibility can be safely ignored. U . c ≈ 340ms−1 . A consequence of this is that the ﬂuid feels the eﬀect of a local change everywhere in the ﬂuid instantaneously. 2. Thus a small change in density is related to a small change in pressure via dp dρ = 2. an imposed pressure gradient. In air. so would need U ≫ O(1500ms−1 ) for signiﬁcant compressibility eﬀects.29 and compressibility will not be important. the forces acting on a given element of ﬂuid by the surrounding ﬂuid. c on deﬁning the Mach number We can deduce that if M ≪ 1. or electromagnetic forces in the case of conducting ﬂuids. c ≈ 1. In this case dp/dρ is inﬁnite. ρ ρc and so dρ ρU 2 =O ρ ρc2 M= = O M2 .of O(ρU 2 ). so that for U = O(102 ms−1 ). M = 0. so the speed of sound in the incompressible medium is inﬁnite. For an incompressible ﬂuid.

3. as shown in the diagram. j both run over 1. n. z in a Cartesian frame. it may be acted upon by long-range forces such as gravity. t). In general.1) . This makes the stress on a surface easy to ﬁnd if the surface in question has normal pointing in one of the coordinate directions. From the deﬁnition of σij above. So. Suppose instead we have a solid boundary over which ﬂows a ﬂuid. we introduce the stress tensor σij . for the component of stress pointing in the x direction. the unit normal will not point in a coordinate direction. In general H will depend on n. we have σij = Hi (x. The unit normal to the patch is n and is chosen to point into ﬂuid 2. Einstein’s summation convention Einstein introduced the nifty convention of summing over a repeated index to save repeatedly writing the summation sign Σ (this shows up a lot in relativity!). of course. for example. corresponding to x. y. H = H(x. Shade on this surface a small patch of area dS. where i. Label the ﬂuid on one side of the dividing surface 1 and on the other side 2. Nj ) 5 (2. Stress within a ﬂuid As a ﬂuid moves. when n points into the ﬂuid on side 2. For example.2. We just pick out the relevant entry of σij . Consider an imaginary surface drawn in the ﬂuid. deﬁned as the i-component of stress acting on an element of surface with unit normal n in the j-direction. we need σ12 . and thereby write the previous formula as just σij nj where the summation over j is now understood.To describe those in the second category. We deﬁne the stress (force per unit area) acting on the patch on side 1 due to the ﬂuid on side 2 to be H (so the force is H dS). the stress on a plane wall at y = 0 is found by noting that the normal points in the y direction. We will adopt the same convention throughout this course.e. the component of stress in the z direction on a wall at x = 0 is σ31 . We show below that the force exerted at a point on the boundary by the passing ﬂuid is 3 σij nj j=1 where the unit vector n is normal to the boundary and points into the ﬂuid. Similarly. But there are also internal stresses active inside the ﬂuid. i. the position in the ﬂuid x and time t.

n) dS −Hi (x. 6 .1) is taken to be implicit. N3 ) dS 3        the force exerted by the ﬂuid outside dV on is . The dependence on t in (2. 2. x3 A3 dS1 n dS2 P A2 dS x2 x1 A1 dS3 We let   A1 A2 A3   P A2 A3  P A1 A3   P A1 A2            dS   dS       dS 1 n1 dS =  dS 2   n2 dS       dS 3 n3 dS        the area of be .) Now. N2 = j. N2 ) dS 2   −Hi (x.3) ( So the total force exerted on the tetrahedron is dS [Hi (x. N1 ) dS 1  −Hi (x. n2 .2) (Note: To ﬁnd the areas of each surface of the tetrahedron. consider a small tetrahedron of ﬂuid of volume dV as shown below. n3 )? To do this. Note: Here we are using the Einstein summation convention of summing over a repeated index.F n 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 1 0000000 2 dS where N1 =i. use the simple half-base times height rule. (2. So σij nj means σi1 n1 + σi2 n2 + σi3 n3 . N3 = k. How can we evaluate H for an arbitrary normal direction n = (n1 . n) − σij nj ].    dS      dS 1  dS 2      dS 3     Hi (x.

n3 = 1. To demonstrate this. if we let ε tend to zero.Newton’s second law states that Force = Mass × Acceleration. consider a small cube of ﬂuid of dimension ε as shown. and we conclude that and so Hi (x. 3 Symmetry of the stress tensor In the absence of external couples applied to the ﬂuid (so-called “body-couples”). The component in the x direction is F1 = σ12 . (2.4) produces the result that Force = 0. while the right hand side is O(dV ) = O(ε3 ).4) is O(dS) = O(ε2 ). Therefore. So Fi = σij nj = σi2 .5) Example 1: Find the stress in the x direction on a wall placed at y = 0.4) If the dimensions of the tetrahedron are of size O(ε). 1). In this case n1 = n2 = 0. |∇(x + y + z − 1)| 3 Then the stress in the z direction is F3 . with 1 1 1 F3 = σ3j nj = σ31 √ + σ32 √ + σ33 √ 3 3 3 1 = √ (σ31 + σ32 + σ33 ). n) = σij nj . σ22 ε 2 x2 x2 σ11 ε 2 σ21ε 2 x1 x3 ε ε 2 ε 2 σ12 ε 2 σ21 ε 2 σ11 ε 2 σ12 ε 2 σ22 ε 2 x1 We know that Moments = d {Angular momentum}. Example 2: Find the stress in the z direction on the upper side of the plane x + y + z = 1 due to the ﬂuid above it. (2. dt 7 . In this case n=+ 1 ∇(x + y + z − 1) = √ (1. the left hand side of (2. as we had above. 1. the stress tensor σij is symmetric. (2.

∂xj ∂xi 8 ξij = − 1 2 ∂uj ∂ui − . At time t. we deduce that σij = σji . u(x+dx. ε 2 ε2 [σ12 − σ21 ] = O(ε4 ) =⇒ σ12 = σ21 . on the face shown in the ﬁgure.7) .t) Expanding the second particle’s velocity in a Taylor series. j (2. 2 Following similar arguments. t). ∂xi ∂xj (2. moving with velocity u(x + dx. gravity) + surface forces due to surrounding ﬂuid O(ε4 ) O(ε3 ) while the right hand side is of size (cube volume) × (cube length) = O(ε4 ). t) = ∂ui ∂xj dxj + O(dx2 ). t) − ui (x.t) x+dx x u(x.6) x We now re-write the derivative on the right hand side like this ∂ui = eij + ξij ∂xj where eij = 1 2 ∂uj ∂ui + . with the ﬁrst at position x and moving with velocity u(x. t).g.The left hand side is made up of moments due to body forces (e. the particles are close together. Strain in ﬂuids Consider the relative positions of two individial ﬂuid particles under any particular ﬂuid motion. 2(b). and the second at x + dx. Therefore. we have the following expression for the relative velocity of the two particles: ui (x + dx.

To see why these two tensors are given their names. second order tensor. z z’ y’ v3 v2 y v1 x x’ A set of Cartesian axes (x. 9 . y ′ . ξij = −ξji ). v3 are aligned along the principal axes and form a basis. Writing both out as a matrix we have ij  e11 e12 e13 eij =  e21 e22 e23  e31 e32 e33  and  0 0 e′ 11 0  e′ =  0 e′ ij 22 0 0 e′ 33  The three eigenvalues of the new matrix are λ1 = e′ . in which eij is diagonal (see Appendix A). z) and a rotated set of principal axes (x′ . v1 . is e ·q = ′ i=1 αi e · vi = ′ λi αi vi . they are all real and have mutually orthogonal eigenvectors. we may rotate to a new set of axes. v2 . ij We denote it as e′ in the new frame. i=1 Therefore the vector q gets stretched or contracted in directions v1 . λ2 = e′ and λ3 = e′ and. The strain tensor eij Since eij is a symmetric. called principal axes. we examine their meanings in terms of the kinematics of the ﬂuid particles. ξij is sometimes called the vorticity tensor. The action of matrix e′ on a general vector. q = ij 3 3 i=1 αi vi . z ′ ) along which are aligned the three orthogonal eigenvectors of e′ . v2 . while ξij is antisymmetric (So eij = eji . eij is called the rate of deformation or strain tensor. 3 for constants αi . These eigenvectors. y. v3 according to whether each eigenvalue λi is positive or negative.Note that eij is symmetric. since e′ is 11 22 33 ij symmetric.

dx′ . Trace(e′ ) = Trace(eij ) = eii = ij ∂xi A ﬂuid is deﬁned to be incompressible if the local change in volume of a ﬂuid element such as the box just considered is zero. In other words ∇ · u = 0 for an incompressible ﬂuid. dx′ and volume V = dx′ dx′ dx′ . Similar arguments apply for the other sides so the 1 11 new edges are of lengths (1 + e′ dt) dx′ 11 1 (1 + e′ dt) dx′ 22 2 (1 + e′ dt) dx′ 33 3 Therefore. 22 33 ij 11 11 22 33 neglecting terms of size O(dt2 ). 10 . j. 1 2 3 1 2 3 z’ dx 3 y’ dx 1 dx 2 x’ After a short time dt. the trace of a matrix is invariant under a rotation to a new coordinate frame (see Appendix A). we introduce the alternating tensor deﬁned as   0 if any of i. k are equal εijk = 1 if i. j. The vorticity tensor ξij As a preliminary. the change in volume (divided by V dt) is equal to ∂ui = ∇ · u. Now. k are in cyclic order  −1 otherwise. So. the change in volume of the cube after time dt is equal to {(1 + e′ dt)(1 + e′ dt)(1 + e′ dt) − 1}V = (e′ + e′ + e′ )V dt = Trace(e′ )V dt.Local expansion of the ﬂuid Suppose we try the above ideas out on a small cube of ﬂuid aligned with the principal axes and with sides of length dx′ . the side dx′ will have become deformed so that its new length is 1 old length + (rate of stretching in the x′ direction)×dt 1 So its new length will be (1 + e′ dt) dx′ .

8) The factor −1/2 is included for convenience. ε133 = 0. Since ξij is antisymmetric. the antisymmetric contribution to the relative velocity of the two particles is 1 ξij dxj = − 2 εijk dxj ωk = 1 2 εikj ωk dxj 1 = ( 2 ω ×dx)i . εijk εlmk = δil δjm − δim δjl . 11 . 1 represents a rotation of one ﬂuid particle about the other with angular velocity 2 ω. 2 for some vector ω with components wk . This is of the form v = ω × r. ( 1 ω ×dx)i 2 w 1 _ |w| 2 dx Local rotation of two ﬂuid particles at angular rate 1 |ω| about an axis pointing 2 in the direction ω. To understand the kinematics associated with ξij . ωk .so. What are the components of ω in terms of the ﬂuid velocity? First we note the useful identity involving the alternating tensor. Exercise: Show that εijk aj bk is equivalent to a × b by writing out the components. where r is the position vector of a point relative to a suitably deﬁned origin.6). 2. 3. which describes rotation of a particle about at axis with angular velocity |ω|. Check for yourself that the right hand side of (2. The alternating tensor provides a convenient way of expressing a vector (cross) product in index notation. ε132 = −1.8) has only 3 independent components. k = 1. we reason as follows. ε123 = ε231 = 1. So. Referring back to the Taylor expansion (2. (2. it has only 3 independent components and may therefore be written quite generally as ξij = − 1 εijk ωk . for example.

eij corresponds to stretching/straining. ω = ∇ × u. and the stress is simply related to the static pressure p by σij = −p δij .10) is not in general the same as p in equation (2. Hence the stress exerted on an element with unit normal n is σij nj = −pni and has magnitude p in the −n direction. which consists of translation.where the Kronecker delta δij is 1 if i. we ﬁnd ωl = −εlij ξij = 1 εlij 2 ∂uj ∂ui − ∂xi ∂xj = εlij ∂uj ∂xi i. (2. we write σij = −p δij + dij . 12 (2.8) and using the deﬁnition (2. Any ﬂow for which ω = 0 is called irrotational. note that since the ﬂuid is now moving. So the pressure force exterted on a wall by a ﬂuid at rest acts in a direction perpendicular to that wall. The vector ω is called the vorticity. In other words. Now.9) So σii = −p δii = −3p and therefore p = −σii /3. how much is the ﬂuid being deformed by the stresses exerted by the surrounding ﬂuid? For a ﬂuid at rest. Hence εijk εljk = 2δil . where p is the dynamic pressure and dij is the deviatoric stress tensor. Recall that ξij corresponds to rigid rotation. rigid rotation and stretching or contracting. For a ﬂuid in motion. However. 2(c).7). the deviatoric stress depends only on eij and the dependence is linear.9). operating with εijl on both sides of (2. p in equation (2. We deﬁne dii = 0 so that once again σii = −3p and thus p = −σii /3. Stress-strain relationship for a Newtonian ﬂuid To proceed with the modelling of ﬂuid ﬂow. we need to understand the relationship between the stress in a ﬂuid and the local rate of strain. there are no viscous forces. Newtonian ﬂuid For a Newtonian ﬂuid.10) . j are equal and 0 otherwise.e. The deviatoric stress dij is due to the motion of the ﬂuid.

by (2. since dij = dji . visco-elastic ﬂuids can have a “memory” whereby the stress depends on the state of the ﬂuid at previous times.11) = 2µeij + λδij ekk . substituting dij into relation (2. In other words.10). where µ. We assume that Aijkl is isotropic.Notes 1. 2. ekk = 0 and so σij = −pδij + 2µeij 13 (2. For a Newtonian ﬂuid dij is therefore symmetric. 2 dii = 0 = [2µ + 3λ]eii =⇒ λ = − µ 3 Now. Non-Newtonian. The instantaneous stress at any point in the ﬂuid does not depend on the past history of the ﬂuid motion. which means that the stress dij generated on a ﬂuid element by a given straining motion is independent of the orientation of that element.14) Remember that ekk = ∇ · u represents the amount of dilatation or local expansion of the ﬂuid. Now.15) . the ﬂuid has the same properties in whichever direction you look. Aijkl = Ajikl Therefore. Therefore. we have 1 σij = −pδij + 2µ[eij − δij ekk ] 3 (2. (2. for an incompressible ﬂuid. dij = µ[δik δjl + δil δjk ] + λδij δkl ekl (2. The stress-strain relationship Since the deviatoric stress is a linear function of the strain rate for a Newtonian ﬂuid.12). µ′ and λ are constants. It is known that all isotropic tensors of even order can be written as sums of products of Kronecker deltas. we may write dij = Aijkl ekl where Aijkl is a constant fourth-order tensor. But dii = 0 by deﬁnition and so. Thus Aijkl = µδik δjl + µ′ δil δjk + λδij δkl .13) (2.12) and so µ′ = µ.

The full usefullness of this idea will become apparent later on when we apply Newton’s second law to derive the equations of ﬂuid motion. y(t). t) = dx dt Suppose now we wish to calculate the acceleration of this set of axes. z(t)) at time t. 3. z u y At rest x Consider a set of axes moving with the local ﬂuid velocity u and another set ﬁxed in space as shown in the diagram. If the origin of the moving set of axes is at x = (x(t). We assume equation (2.1) (3. The constant µ is called the viscosity of the ﬂuid.1 The convective derivative Our ultimate goal is to derive an equation describing the motion of a ﬂuid. we can write its velocity as u(x. The Navier-Stokes equations 3. t + dt) − u(x.2) . t) d2 x du . 2 = dt = lim dt dt dt→0 Using Taylor’s theorem. By deﬁnition. the acceleration is given by u(x(t + dt). we introduce the important concept of a derivative which tells us how things change as we move with the ﬂuid.This provides a constitutive relation between the stress and strain for an incompressible. To move towards this goal. 14 (3.15) throughout the rest of the course. x(t + dt) = x(t) + dt dx + ··· dt = x(t) + dx + · · · . Newtonian ﬂuid.

Dt 3.1) becomes du d2 x 2 = dt dt = = Thus du dt = ∂u 1 dt lim + dx · ∇u + · · · ∂t dt→0 dt ∂u dx + · ∇u ∂t dt ∂u + u · ∇u.4) ∂u + ··· . Dρ/Dt expresses the amount by which the density of a ﬂuid particle changes as it is convected along with the ﬂow. For example. ∂t (3. t) + dx · ∇u + dt Therefore. u(x + dx. Dt dt ∂t ∂x dt ∂y dt ∂z dt ∂t ∂x ∂y ∂z (3. z(t)). y(t).2 Lagrangian mappings 15 . we therefore have Dρ = 0.6) Df /Dt tells us how the function f changes as we move with the ﬂow.(dx/dt).3) The derivative on the left hand side of (3. Also by Taylor’s theorem. ∂t (3. we ﬁnd Df Dt = = ∂f + u · ∇f ∂t ∂f ∂f ∂f ∂f +u +v +w .4) is often written with a large D and referred to as the convective derivative (sometimes the material or substantial derivative).5) The convective derivative may be applied to any scalar or vector function. ∂t ∂x ∂y ∂z This is really just the statement of the chain rule: Df df ∂f ∂f dx ∂f dy ∂f dz ∂f ∂f ∂f ∂f ≡ = + + + = +u +v +w . consider the function f (x(t).on writing dx = dt. For example. t + dt) = u(x. For an incompressible ﬂuid. We will follow this convention and henceforth write ∂u Du = + u · ∇u Dt ∂t (3. Proceeding as above. (3.

7) and applying the chain rule we can write: dx(1) = ∂χt da1 . so a = (a1 .Working in the Lagrangian framework. we label all particles over the entire ﬂuid domain according to their position at time t = 0. a2 . at a later time. Consider now the diﬀerential vector (da1 . Using (3. In addition. Any position in the ﬂow-ﬁeld x may be regarded as a function of its label and time. to their current positions at time t. where a is the position vector of that particle at t = 0. a3 ). an individual ﬂuid particle is identiﬁed by its label. that is.8) Similar expressions follow for dx(2) and dx(3) .7) where χt (a) is a mapping from the initial particle positions at t = 0. Then a volume element in this space is given by dV l = da1 da2 da3 . we will refer to the state of the ﬂuid at time t as convected space as this is where all the particles have been carried by the ﬂow. as if it was carrying a ﬂag along with it as it moves in the ﬂow. a3 in labelling space.0. we consider the motion of individual point particles. We will refer to the initial conﬁguration of ﬂuid particles at t = 0 as labelling space as that is where all the labels sit. given a ﬂuid particle at a particular point in space at a particular time. Then. Moreover we can write x(a. 16 . we monitor how that particle travels through space as time progresses. a2 . The subscript on the mapping function indicates that the particles are being mapped from where they were at t = 0 to their location at time t. t) = χt (a) (3. Consider now the following important question: How is a small volume element in labelling space related to its counterpart at the later time t? Suppose we deﬁne axes a1 . We therefore ascribe to each particle a label a. Furthermore. To this end.t) time t=0 time t The mapping itself therefore changes in time.0) in labelling space and denote its counterpart in convected space as dx(1) . ∂a1 (3. χ0 (a) = a. χ 11 00 a t 1 0 1 0 x ( a.

we note that t ∂χt ∂χt ∂χt ∂χi t × = det · . where (3. we note that the expression dV c = (dx(1) × dx(2) ) · dx(3) represents a volume element in the convected space. (3. deﬁning a parcel of ﬂuid with volume Vc = Vc dV c = Vl J dV l . Therefore. Equation (3.11) we have immediately managed to change an integral deﬁned over a volume which is changing in time to one over a volume which is ﬁxed. ∂a1 ∂a2 ∂a3 ∂aj [see Exercise Sheet 1]. · ∂a1 ∂a2 ∂a3 Using the notation χi to denote the ith component of χt .At this point.8) we can write (dx(1) × dx(2) ) · dx(3) = ∂χt ∂χt ∂χt × da1 da2 da3 .9) ∂χ1 t ∂a1 J = det ∂χi t ∂aj = ∂χ1 t ∂a2 ∂χ2 t ∂a2 ∂χ3 t ∂a2 ∂χ1 t ∂a3 ∂χ2 t ∂a3 ∂χ3 t ∂a3 (3. Its usefulness is demonstrated by trying to work out the rate of change of volume of a parcel of ﬂuid as it is convected along with the ﬂow. Thus. The transformation between the two spaces is very useful as it allows us to relate objects which are varying in time (those in convected space) to objects which are static (those in labelling space).9) therefore shows how a volume element in labelling space is related to its counterpart in convected space. This will make diﬀerentiation a lot easier. using (3. dV c = J dV l . Convince yourself that this is true.10) ∂χ2 t ∂a1 ∂χ3 t ∂a1 is the Jacobian of the mapping function χt . This allows us to relate the two volume elements as follows. 17 .

(3.To ﬁnd the rate of change of the parcel’s volume. we form dV c d = dt dt J dV l = Vl Vl dJ dV l = dt Vl DJ dV l . the position in convected space. where n is the surface unit normal. and so it follows by the deﬁnition of the convective derivative that ∂J DJ dJ = + u · ∇J ≡ dt ∂t Dt Note that we were able to take the d/dt inside the integral in (3. (3.3 The Reynolds Transport Theorem 18 .12) and (3. the volume swept out by a small element dS c of its surface Sc in time dt (see ﬁgure) must equal u · n dS c . and thus dV c = dt Sc u · n dS c .12) since J is a function of x. Dt (3.13) Vc Vl Now.13) we deduce that DJ = J ∇ · u. Applying the divergence theorem and using (3. we ﬁnd dV c = dt ∇ · u dV c = ∇ · u J dV l .14) 3. we may argue as follows: Since the particles making up the parcel move with the 11 00 11 00 111111111111 000000000000 111111111111 000000000000 volume swept 11111111111111 00000000000000 11111111111111 00000000000000 out by surface patch V(t+dt) c V(t) c u n ﬂuid velocity u. comparing (3.9). Dt We will ﬁnd this useful in the proof of the following important theorem.12) since Vl is independent of time. Equally.

it follows immediately that the integrand on the right hand side is zero.9). Dt 19 . In fact. so Dρ + ρ ∇ · u = 0. we may convert the volume integral into one deﬁned over the equivalent volume in labelling space. since Vl is time-independent.15) where G(x. we can use (3. Dt (3.16) (3. Using the result (3.14). Vl .17) (3. Example If we identify the scalar G(x. Proof Using (3. t) is any scalar. this becomes d dt G dV V (t) = Vl J = Vl = V DG + GJ ∇ · u dV l Dt DG + G ∇ · u J dV l Dt DG + G ∇ · u dV . Dt (3. So. it states that d dt G dV = V (t) V (t) DG + G ∇ · u dV . Dt Since mass is conserved.This result tells us how an integral of any scalar quantity deﬁned over a time-dependent volume of ﬂuid V (t) changes in time. t). we have d dt G dV = V (t) Vl d (GJ) dV l = dt Vl D (GJ) dV l = Dt J Vl DG DJ +G Dt Dt dV l by the product rule of diﬀerentiation.15) to write d dt ρ dV = V (t) V (t) Dρ + ρ ∇ · u dV . ⋄. t) with the ﬂuid density ρ(x.18) using (3.9). d dt G dV = V (t) d dt G J dV l Vl Then.

As the ﬂuid moves past. the rate of change of the mass of ﬂuid within V must equal the net amount of ﬂuid coming into (or out of) it. for an incompressible ﬂuid. Dt It therefore follows that.4. 3. Conservation of mass. 2. t). Conservation of mass Consider a closed volume V which is ﬁxed in space inside a moving ﬂuid. 3. ∇ · u = 0. d dt ρ dV + V S Then ρu · n dS = 0. We will also discuss how energy is dissipated by viscous forces in a moving ﬂuid and derive an equation for the conservation of energy in a parcel of ﬂuid. ∂ρ ∂ρ ∂ρ Dρ = +u +v =0 Dt ∂t ∂x ∂y so the condition for incompressibility is satisﬁed. ρ = ρ(z). This is a common mistake.4 Equations of ﬂuid motion Up to now. Newton’s second law of motion. We will make use of the Reynolds transport theorem when we derive the Navier-Stokes equations. For example. we have been concerned with the kinematics of a ﬂow due to a predetermined velocity ﬁeld u(x. consider the two-dimensional ﬂow with velocity and density ﬁelds u = u(x. 20 . then Dρ = 0. y)i + v(x.1. Note: It is important to realise that incompressible does not mean constant density. which is a restatement of the result ekk = 0 found before using the small cube of ﬂuid argument.We noted above that if the ﬂuid is incompressible. y)j. assuming there are no sources of ﬂuid within V . But how do we determine the underlying velocity ﬁeld in the ﬁrst place? To formulate the necessary equations governing the motion of a viscous ﬂuid we will apply the following fundamental principles: 1. Mathematically. but the density ﬁeld varies with z.

∂t (3. S is the volume surface.19) Equation (3. Let F represent some body force per unit mass acting on the ﬂuid (i. Newton’s second law In order to apply Newton’s second law of motion to a parcel of ﬂuid. We are now in a position to derive the equation of motion of a ﬂuid. 3.21) 21 . moving with the ﬂow. Dt ρ Fi dV + S(t) σij nj dS. We apply Newton’s second law of motion to a volume of ﬂuid V (t). Using the fact that V is ﬁxed in space and applying the divergence theorem. this becomes ∂ρ + ∇ · (ρu) dV = 0.where ρ is the ﬂuid density. Dt This is just the equation we found above using the Reynolds transport theorem. Using the deﬁnition of the convective derivative.2. and n is the normal to V pointing outwards. a long-range force such as gravity). and using (3. Thus. It is usually referred to as the continuity equation. the right hand side the forces acting on the ﬂuid. Applying the Reynolds transport theorem to the ﬁrst term and the divergence theorem to the last.e.20) we ﬁnd ρ V (t) Dui dV = Dt ρ Fi dV + V (t) V (t) ∂σij dV .4. we can re-express the continuity equation as Dρ + ρ∇ · u = 0. with surface area S(t) and unit outward normal n.19) therefore expresses the principle of mass conservation. which will frequently be abbreviated to cty equation.20) G dV = V (t) V (t) DG + G∇ · u dV . we will need to recall the Reynolds transport theorem introduced earlier: d dt for any scalar quantity G. The left hand side expresses the rate of change of momentum of the ﬂuid. d dt ρui dV = V (t) V (t) (3. ∂t V Since our choice of volume V was arbitrary. ∂xj (3. The last term represents the stresses acting over the surface of the chosen volume by the surrounding ﬂuid. it must be true that ∂ρ + ∇ · (ρu) = 0.

∂σij ∂p 2 ∂ ∂ =− − (µ∆) + 2 (µeij ). we have that ∆ = 0. we have Dui ∂p ∂ 2 ui ρ = ρ Fi − +µ 2 .Note that we have used the fact that Dui Dρ D(ρui ) =ρ + ui Dt Dt Dt and that Dρ/Dt = 0 since the ﬂuid is incompressible. we can write this as ρ ∂σij Dui = ρ Fi + .22) therefore reduces to ρ V (t) ∂ 2 uj ∂p Dui ∂ 2 ui − ρ Fi + −µ + Dt ∂xi ∂xi ∂xj ∂x2 j dV = 0. 2 σij = −p δij − µ∆δij + 2µeij .22) Since we are restricting our attention to incompressible ﬂuids. as the statement must be true for all possible volume choices. (3. we obtain ρ V (t) ∂p 2 ∂ ∂ Dui − ρ Fi + + (µ∆) − 2 (µeij ) dV = 0. Dt ∂xj 22 (3. If the ﬂuid is Newtonian. Accordingly. (3.24) . Dt ∂xi 3 ∂xi ∂xj (3. 3 where ∆ = ∇ · u = ekk represents the local expansion of the ﬂuid. ∂xj ∂xi 3 ∂xi ∂xj Substituting this relation into the above. Dt ∂xi ∂xj Alternatively.23) must vanish. Assuming µ is constant. Also.23) But ∂uj /∂xj ≡ ∇ · u = 0. since our original volume was chosen arbitrarily. Finally. (see the earlier discussion of the expansion of a small cube of ﬂuid). the integrand in (3. we may use the constitutive relation introduced earlier.

∂r 2 r ∂r r ∂θ ∂z If the body force. together with the continuity equation.e.21). In vector form. In cylindrical polar coordinates. 23 . r2 r2 2 v 1 = − pθ + Fθ + µ ∇2 v + 2 uθ − 2 .direct from (3. ρ( wt + uwx + vwy + wwz ) = −pz + Fz + µ∇2 w. ˆ where the velocity ﬁeld u = u ˆ + v θ + wk and r ∇2 = ∂2 1 ∂ 1 ∂2 ∂2 + + 2 2 + 2. we can write it as the gradient of a scalar potential. In Cartesian coordinates. ux + vy + wz = 0. where the velocity ﬁeld u = u i + vj + wk and ∇2 = ∂2 ∂2 ∂2 + 2 + 2. F. they are written in component form as ρ 1 v2 ut + uur + vuθ + wuz − r r uv 1 ρ vt + uvr + vvθ + wvz + r r 1 ρ( wt + uwr + vwθ + wwz ) r u 1 ur + + vθ + wz r r 2 u − vθ . r r r = −pr + Fr + µ ∇2 u − = −pz + Fz + µ∇2 w.25) These are the Navier-Stokes equations for an incompressible Newtonian ﬂuid. ∂x2 ∂y ∂z ρ( vt + uvx + vvy + wvz ) = −py + Fy + µ∇2 v. ρ Du = ρ F − ∇p + µ∇2 u. then it can be incorporated into the pressure term by writing ρ F − ∇p = −∇(ρ Ω + p) = −∇P. Dt (3. ∇ · u = 0. they are written in component form as ρ( ut + uux + vuy + wuz ) = −px + Fx + µ∇2 u. = 0. F = −∇Ω. is conservative. i.

3 Boundary conditions Suitable boundary conditions must be appended to the system (3. that is to say the local ﬂuid velocity at the wall is zero. 3.25). Example: Fluid ﬂowing over a plane wall at z = 0. For a viscous ﬂuid a further boundary condition is required to complete the ﬂow description due to the extra derivatives imported by the last term in (3.4. 24 . The latter is usually referred to as the no slip condition. ﬂuid cannot ﬂow into the wall. Experiments reveal that this supplementary condition should stipulate that the ﬂuid does not slip over the solid boundary.where we have deﬁned the modiﬁed pressure P to be P = p + ρ Ω. This states that u · n = 0. Mathematically this is stated as u · t = 0. we require u·n =0 and u·t=0 at the solid surface.e. i. where t is any vector tangential to the wall. In summary.25) in order to fully deﬁne the ﬂow problem in hand. w z Flow x u y v The no normal ﬂow condition states that w = 0. We divide them into the following categories: 1111111111111111111111 0000000000000000000000 1111111111111111111111 0000000000000000000000 1111111111111111111111 0000000000000000000000 1111111111111111111111 0000000000000000000000 1111111111111111111111 0000000000000000000000 11 00 Solid 1111111111111111111111 0000000000000000000000 11 00 t 1111111111111111111111 0000000000000000000000 11 00 n Fluid Fluid-Solid The usual inviscid no-penetration condition applies at a solid impermeable surface.

Note: We can use the continuity equation to deduce a further piece of information about the ﬂow at z = 0.4. we have Df = 0 at the surface. surf ace = where us . ∂x ∂y ∂x ∂y since u and v are both zero everywhere on z = 0. if the surface is convected along with the ﬂuid. ws are the velocity components of the ﬂuid at the surface. ∂w =0 ∂z 3. Therefore. z. Dt Thus. We know that ∂u ∂v ∂w + + = 0. Example: A ﬂat plate moves up and down in a surrounding ﬂuid so that at time t. t) = 0. its position is given by z = sin t.The no-slip condition states that u = v = 0. Df Dt ∂f ∂f ∂f ∂f + us + vs + ws = 0. We know from the deﬁnition of the convective derivative that the rate of change of f following the ﬂuid is Df /Dt. If the shape of the surface is described by the equation f (x. on z = 0. y.4 The kinematic condition This applies at any surface which is moving. it states that a particle sitting on the surface f = 0 must move along with the ﬂuid. Therefore. ∂t ∂x ∂y ∂z on z = 0. ∂u ∂v ∂v ∂u = = = = 0. vs . ∂x ∂y ∂z and. What is the kinematic condition on the moving wall? z x 25 .

Fluid 2 s τ ds Fluid 1 n τ t The height of the rectangular volume is assumed negligible in comparison with its length ds. Df Dt wall = ∂f ∂f ∂f ∂f + us + vs + ws ∂t ∂x ∂y ∂z =⇒ ws = cos t. Set f = z − sin t. t) must be determined. If s measures arc length along the interface. Using superscripts to denote variables in each of the ﬂuids we demand that u(1) · n = u(2) · n and u(1) · t = u(2) · t on the interface. Therefore. Surface tension τ (s) tugs on the volume at either end in the tangential direction. we consider a small rectangular volume of ﬂuid of length ds as shown in the ﬁgure below. These state that the normal and tangential components of the velocities are continuous across the interface.t)=0 Here the function f (x. the ﬂuid velocity at the wall is cos t in the z direction. y.The equation of the wall is z = sin t.y.z. To obtain this condition.4. Dynamic condition: We also need a condition expressing the balance of forces prevailing at the interface between the ﬂuids. Fluid 2 Fluid 1 n t f(x.5 Boundary condition at the interface between two ﬂuids This situation is more complicated as the position and shape of the interface are themselves unknown and must be found as part of the solution to the problem. The force exerted on the lower 26 . 3. the local unit normal and tangent vectors to the interface are denoted as n(s) and t respectively at position s. Kinematic conditions: These state that the ﬂuid velocity must be continuous across the interface. Then. = − cos t + ws = 0. z.

The ﬁrst two terms represent surface tension at either ends of the control volume. so. where σ (1) is the stress tensor in ﬂuid 1. at 20◦ C the surface tension between air and water is τ = 72. the force exerted on the top by ﬂuid 2 is given by −(σ (2) · n) ds. The minus sign is needed since the unit normal must point into the ﬂuid exerting the force. this yields dτ t σ(1) − σ(2) · n = κτ n − ds If the surface tension is constant.portion of the rectangular volume by ﬂuid 1 is given by (σ (1) · n) ds. This follows from the deﬁnition of the stress tensor given above. ds Rearranging. and deﬁning the local surface curvature κ so that dt κn = − . as it is generally speaking. ds Expanding the ﬁrst pair of square brackets and cancelling relevant terms. taking the limit as ds → 0. Since ds is small. N. the jump in stress is then written as ∆Fi ≡ [σij − σij ] nj = 2τ κ ni . In index notation. Refer to pages 64 and 69. we may expand in Taylor series.B. ds we obtain. τ (s) + ds dτ ds t(s) + ds dt + τ (s)t(s) + ds σ (1) · n − σ (2) · n + O(ds2 ) = 0. Applying Newton’s second law to the rectangular control volume (and disregarding its negligible inertia). Similarly. the last term disappears. ds τ dt dτ + ds t + ds σ (1) · n − σ (2) · n + O(ds2 ) = 0. 27 (1) (2) . dτ −κτ n + t + σ(1) · n − σ (2) · n = 0. we obtain the force balance τ (s + ds)t(s + ds) + τ (s)t(s) + ds σ(1) · n − σ(2) · n = 0. Surface tension is a very weak force. ds ds Dividing by ds. This states that the jump in stress ∆Fi at the interface is balanced by surface tension. A more general derivation valid for a two-dimensional curved surface can be found in Batchelor. For example.8 dyne/cm.

So. the pressure jump. since the vector on the right hand side is in the direction n. Euler’s equation holds: ρ Du = −∇p + ρF Dt 28 . If τ = 0. inviscid ﬂuid ﬂow. and then see how viscous eﬀects modify our observations later.5 The Energy Equation How much energy does a given ﬂuid ﬂow carry with it as it ﬂows? Suppose we track a chosen blob of ﬂuid . the jump in stress. Example: Consider a spherical bubble of radius a suspended motionless in air. We’ll start with the easier case of purely inviscid ﬂow. The interface between the tear ﬁlm on your eye and the surrounding air.5. The energy equation follows by mathematical manipulation of the equation of motion. ∆Fi = [σij − σij ] nj = [−p(1) δij + p(2) δij ] nj = −[p(1) − p(2) ] ni = 2τ κ ni . Since neither ﬂuid 1 nor 2 is moving. The tangential stress is still continuous. there is only a jump in the normal stress.how does the energy within the blob change with time? We will derive an energy equation to describe how energy changes within the ﬂow. In this case the ﬂuid stresses are continuous across the interface. The mean curvature of the spherical bubble is 1/a. In the absence of surface tension. 3. ∆p = p(1) − p(2) = −2τ /a =⇒ τ = − which is known as the Laplace-Young equation. Other examples: The interface between the atmosphere and the ocean.p (1) p (2) n Recall that 1 dyne = 10−5 N.1 Energy equation for inviscid ﬂow For an incompressible. a∆p 2 (1) (2) 3. τ = 0 and the right hand side vanishes.

we can read the equation as meaning that in each unit of time. d dt V 1 2 ρu dV = − 2 V ∇ · (pu) dV + V ρF · u dV . To proceed. of blob = work done by pressure force + work done by body force In other words. Recognising 1 2 ρu dV = K V 2 as the total kinetic energy in the blob. as we shall see in a moment. the second term on the RHS represents the rate of working of the body force on the blob. since −pn is the pressure force acting on the blob due to the surrounding ﬂuid. Firstly. V V V So. we take the dot product of this equation with the velocity u: ρu · Du = −u · ∇p + ρF · u. we have ﬁnally d dt 1 2 ρu dV = 2 (−pn) · u dS + ρF · u dV . we proceed as follows. 29 . Note that ∇ · (pu) = p∇ · u + u · ∇p = u · ∇p since ∇ · u = 0.26) V S V Now.where ρF is the body force per unit mass. we ﬁnd ρu · Du dV = − Dt u · ∇p dV + ρF · u dV . Using the divergence theorem. the ﬁrst term on the RHS represents the rate of working of the external pressure force on the blob. Dt Integrating over a volume V moving with the ﬂuid. V D Dt 1 2 ρu 2 dV = − V u · ∇p dV + V ρF · u dV . and so.E. the Increase of K. Similarly. on using the Reynolds Transport Theorem on the LHS. This remark does not hold true if viscosity is included. If instead we consider a volume V which is ﬁxed in the ﬂuid.26) applies for a volume V (t) moving with the ﬂuid. All work put in by the pressure force and the body force goes directly into increasing the kinetic energy of the blob. we recall that equation (3. (3. no energy at all is lost.

5. this becomes V ρu · ∇ 1 2 |u| 2 dV = − V u · ∇p dV + V ρu · ∇V dV . 2 u · ∇V = ∇ · (V u). So. we take the dot product of the Euler equation with the velocity u: ρu · Du = −u · ∇p + ρF · u. V V V For a steady ﬂow with a conservative body force such that F = −∇V . viscous ﬂuid ﬂow. since ∇ · u = 0. we have S 1 ρ|u2 |u · n dS = − 2 S pu · n dS + S ρV u · n dS.24). 30 . the Navier-Stokes equation holds: ρ Dui ∂σij = ρFi + Dt ∂xj where ρFi is the body force per unit mass (see equation 3. equation 1.7).2 Energy equation for viscous ﬂow For an incompressible. As above. S 1 ρ|u2 | + p + ρgy u · n dS = 0 2 is the statement of conservation of energy over a closed volume ﬁxed in space1 .As before. Dt Integrating over the ﬁxed volume V . we ﬁnd ρu · Du dV = − Dt u · ∇p dV + ρF · u dV . using the divergence theorem. Proceedings of the Royal Society of London. where we have used the fact that u · ∇u = ∇(u2 /2) for a divergence free velocity ﬁeld. Next we note that ρu · ∇ 1 2 |u| 2 =∇· 1 ρ|u2 |u . Skipping steps identical to before (albeit this time in index notation) we ﬁnd dK dt 1 = V ui ρFi + ∂σij ∂xj dV = V ρ u · F dV + ui V ∂σij dV . 3. 342. we take the dot product with u. Therefore. ∂xj Compare Longuet-Higgins (1975).

Note that the eﬀect of the pressure is here contained within the σij of the ﬁnal term since σij = −pδij + 2µeij This term therefore also incorporates the new eﬀect due to viscosity. V ∂uj 1 ∂ui ∂ui = σij + ∂xj 2 ∂xj ∂xi = σij eij . the rate of change of the volume’s kinetic energy dK = dt ρ u · F dV + (−pn) · u dS + 2µ 31 ui eij nj dS − 2µ e2 dV .27) V S For an incompressible ﬂuid. writing = S σij V ∂ui dV . ﬁnally. we have dK = dt ρ u · F dV + ui σij nj dS − σij eij dV .28) V S S V . ij (3. Now. and so σij nj = −pni + 2µeij nj . ij since eii = ∇ · u = 0. σij To summarize. (3. ∂(ui σij ) ∂σij ∂ui = − σij . Also. ∂xj 1 ∂ui ∂ui ∂ui . Thus. Therefore. ∂xj ∂xj ∂xj and so ui V ∂σij dV ∂xj = V ∂(ui σij ) dV − ∂xj ui σij nj dS − σij V ∂ui dV ∂xj (using divergence theorem) But. σij eij = −peii + 2µeij eij = 2µe2 . σij = −pδij + 2µeij . = + ∂xj 2 ∂xj ∂xj 1 ∂ui ∂ui ∂ui 1 1 1 ∂uj σij + σij = σij + σij 2 ∂xj 2 ∂xj 2 ∂xj 2 ∂xi we have σij ∂ui ∂ui 1 ∂ui = σij + ∂xj 2 ∂xj ∂xj = on swapping i and j in the second term and using the fact that σij is symmetric.

28).28) reduces to (3. 3 2 (∆ = eii ) ∆δij 3 . and energy is in general lost. The ﬁrst new term in (3.28). But where does this energy go to? 32 .6 Conservation of energy We saw above that shear forces dissipate energy in a viscous ﬂow. the preceding arguments do not work. −2µ V e2 dV ij is deﬁnitely negative and marks a reduction in kinetic energy. So some energy is being lost. The rate of loss of energy is represented by the dissipation function. . 2µ V ui eij nj dV is the rate of working of the deviatoric viscous stresses on the surface of the ﬂuid blob V . Summary No energy is lost in an inviscid ﬂow. But. We say is it lost due to viscous dissipation and deﬁne the dissipation function Φ..28) represent work done on the blob to increase its kinetic energy... where Φ = 2µeij eij . Energy is lost from a viscous ﬂow. This represents the rate at which energy is dissipated per unit volume by viscous action. This does not include such phenomena as hydraulic jumps and breaking waves .. If we set µ = 0.the second new term in (3.26). 3.in these examples discontinuities are introduced into the ﬂow.26) for inviscid ﬂow.Compare this with equation (3. as we would have expected. Φ. For a compressible ﬂuid we ﬁnd Φ = 2µ eij eij − (non-trivial step) = 2µ eij − ∆2 . Thus the viscous stress makes a contribution to the increase of kinetic energy in the blob. So far so good – all of the ﬁrst three terms on the RHS of (3. equation (3.

From equation (3. inside a moving volume of ﬂuid. so that dE = dt This means that ui σij nj dS. dE dt = = S dK dI + dt dt ui σij nj dS + dI − dt Φ dV V . It seems natural.Consider a ﬂuid moving with no body forces acting. The rate of change of the total energy E in the moving volume must be balanced by the rate of working of the local stresses on the volume surface. where K is the kinetic energy. The constant κ is called the thermal conductivity of the ﬂuid. to ask the following question: How does viscous dissipation aﬀect the temperature of a ﬂuid? Since we wish to focus on temperature.) When work is done on the volume by the surrounding ﬂuid. (Note there is no potential energy because we’ve left out body forces. S dI = dt Φ dV . V . where T is the local temperature. So. The total energy. we’ll generalise the argument a little to allow for the eﬀect of heat conduction in the ﬂuid.27). and I is the internal energy associated with the movement of the constituent molecules within the ﬂuid. we know that dK = dt ui σij nj dS − 2µe2 dV = ij ui σij nj dS − Φ dV . q = κn · ∇T per unit area of the volume’s surface. 33 . is given by E = K + I. Heat is conducted through the surface into the moving volume at a rate q given by Fick’s law. some of the work goes into increasing the kinetic energy (K) and some into increasing the internal energy (I). V and so viscous dissipation (in the form of heat) goes toward increasing the internal energy in the volume. then. E. V S V S (with the body force taken as zero).

28). according to the First Law of Thermodynamics (3. the internal energy of the volume is increased by (a) any work done. where c is the speciﬁc heat capacity of the ﬂuid and T is the temperature of the ﬂuid. Note: In general the viscosity µ depends on temperature. ∆E = Q + W. Using the Reynolds Transport Theorem. V The internal energy is related to the ﬂuid temperature by I= V ρ c T dV . With reference to the energy equation (3. since V was chosen arbitrarily.The First Law of Thermodynamics dictates that. V Hence. entering the volume2 . W . W = dt V 2µe2 dV = dt ij Φ dV . so µ = µ(T ). ρc V DT dV = Dt κ∇2 T + Φ dV . So. we have over the time interval dt. (3. Q = dt S q dS = dt κ S n · ∇T dS = dt κ V ∇2 T dV . Dt Conclusion: The rate of increase of temperature as we move with the ﬂuid is proportional to the rate of loss of energy due to viscous dissipation. Q. 2 34 . In symbols. the ﬁrst three terms on the RHS represent work done to increase the kinetic energy. and (b) any heat.29). dI d = dt dt ρ c T dV = V V κ∇2 T + Φ dV . and the fourth term represents work done to increase the internal energy. ρc DT = Φ + κ∇2 T. However. in this course we will assume that Φ is small and that changes in the temperature of the ﬂuid are insigniﬁcant. From above. on the volume. We will therefore take µ to be constant. Note that (a) does not include work done to increase the kinetic energy of the ﬂuid in the volume.29) where ∆E means the change in internal energy per unit time. in a unit of time.

3. carrying momentum with them. we now examine the ﬂux of momentum. Consider uniform inviscid ﬂow past an object as shown Suppose we now ask: what is the drag on the object due to the ﬂow? We assume that a long way from the object the ﬂow is the uniform stream U i in the x 3 4 This is obtained by integrating Newton’s second law.30) In turn: The term on the left hand side is the rate of change of momentum inside the ﬁxed volume. Newton’s second law states that the force acting on the ﬁxed volume V is equal to rate of change of its momentum.8 D’Alembert’s Paradox Use may be made of the momentum integral equation to determine the drag on an object placed in an inviscid ﬂow. To this end. we then have ρ u(u · n) + pn dS = 0.3. See Appendix B This is the same as saying σij = −pδij 35 . The third term on the right hand side expresses the total of the surface forces acting on S due to the surrounding ﬂuid. it is useful to consider a volume of ﬂuid which is ﬁxed in space. For a steady ﬂow in the absence of a body force. The ﬁrst term on the right hand side expresses the ﬂux of momentum through the surface S. (3. where I is the identity matrix4 . Hence3 . Since the volume is ﬁxed. as we did for the conservation of mass argument above. However. instead of considering the mass ﬂux into and out of the ﬁxed volume.7 The momentum integral equation It is sometimes convenient to consider the balance of forces acting on a given parcel of ﬂuid. ∂ (ρ u) dV = − ∂t ρ u(u · n) dS + ρF dV + V S V S σ · n dS. S This relation may also be obtained by integrating the Euler equations directly over a volume V and using appropriate vector identities. some ﬂuid particles are entering and some are leaving the volume. The second term on the right hand side expresses the total body force acting on the ﬁxed volume. For an inviscid ﬂuid. σ = −pI.

since gravity is a conservative force.30) over the volume bounded by the rectangular box of height H as shown with faces Front (F ). P ρF dV . Back (B).30). U . and writing F = ∇ϕ. To determine the drag. ϕ = −gz. Let D be the total force on the object. L and P . (3.31) S Assuming that the only body force acting is gravity. we apply the momentum integral equation (3.31) gives 0=− ρ u(u · n) dS − p n dS + S Ω ρϕ n dS. the volume integral becomes a surface integral courtesy of the divergence theorem. (3. and the object surface P . Then D=− where P is the surface of the object. Upper (U ) and Lower (L). U n P Ui F n n n B H n L Let Ω denote the union of F . F = −gk. To determine D. S where n is the unit outward normal to the bounding surface S. B. 0=− ρ u(u · n) dS − p n dS + S V p n dS.Ui direction at constant pressure P .32) Ω 36 . ρF dV = V V ρ∇ϕ dV = S ρϕ n dS = − ρgz n dS. we use the inviscid steady form of the momentum integral equation (3. Equation (3.

To determine the second integral.Assuming that the bounding box is suﬃciently large that the disturbance caused by the object has decayed and the ﬂow has settled to a uniform stream. u · n = 0 on P . 1 p + ρu2 + ρgy = C. along a streamline. since the ﬂuid can’t penetrate the object. Also. Letting the size of the box approach inﬁnity we see that pU − pL = −ρgH where pU and pL denote the pressures on U and L respectively. with a similar notation. where A is the area of the upper (or lower) face and VP is the volume occupied by the object. Also. So (3. pB = pF .32) gives 0 = D − ρgH k + ρg[HA − VP ] k. we note that Bernoulli’s equation in the ﬂuid states that. and thus D = ρgVP k. So the second integral gives − p n dS = − =D− p n dS − p n dS = D − pL k dS + L U pU k dS S P Ω pU + ρgH k dS + U U pU k dS = D − ρgHA k where A is the area of U . Recalling that φ = −gz. 37 . and u · n = U on F and u · n = −U on B.32) gives − Ω ρ u(u · n) dS = − ρ U 2 i dS + F F ρ U 2 i dS = 0. So the ﬁrst integral in (3. the third integral gives (assuming constant density) ρϕ n dS = Ω B ρgz i dS − = U ρgz i dS + F U ρgz k dS − ρgz k dS + L P ρ∇φ dV ρgz k dS − U ρg(z − H) k dS + ρ∇φ dV P = ρgHA k − ρgVP k = ρg[HA − VP ] k. then u · n = 0 on U and L. 2 for constant C.

F = −gk. deﬁned as U . for example. T U H z x P W The wall is at z = 0 and the undisturbed height of the water is H. ρ. If the Froude number. is constant. and we predict zero drag on the object! This is D’Alembert’s Paradox. T . The previous argument applies for an object placed in a uniform stream. may develop waves as shown in the ﬁgure. Let us calculate the drag on the object P for this ﬂow.So the total force just involves the Archimedes upthrust of the solid object in the vertical direction. It states that an object of an shape placed in a uniform stream in an inviscid ﬂuid experiences zero drag. Taking the horizontal component in the direction of the ﬂow.9 Wave resistance Consider the inviscid two-dimensional ﬂow of a stream of water over an object P as shown in the ﬁgure. Assume the density of the water. and writing F = −∇(gz) as in the previous section. F = (gH)1/2 is less than 1 then the free surface of the stream. 3. Assuming that the only body force acting is gravity. the equation becomes 0=− ρ u(u · n) dl − p n dl − ρgz n dl. D · i = 0. The inviscid steady form of the momentum integral equation (3. F . It is important to realise that D’Alembert’s paradox does not mean than any object placed in any inviscid ﬂow experiences zero drag. as is explored in the next section. an object such as a ship may experience drag to wave resistance. If there is a free surface. C over an area A enclosed by the closed contour C of incremental arc length dl.33) C C 38 . C (3.30) in two-dimensions is 0=− ρ u(u · n) dC − p n dl + C A ρF dA.

Also. the horizontal component of (3. Noting that u · n = 0 on the object. and the free surface up to one of the crests T . We also note that H+d H − z nx dl = F. we take p = 0 on the free surface.B.B 0 z dz − 0 1 z dz = [(H + d)2 − H 2 ]. on the free surface T . ny ). where n = (nx . Moreover.F. Note that there is no direction associated with the contour as we are integrating with respect to arc length. B F.Take S to be the dotted contour shown in the ﬁgure below. z nx dl = T T 1 1 f f ′ dx = [f 2 ]X = [(H + d)2 − H 2 ] −∞ 2 2 if z = f (x) is the shape of the free surface. z nx dl = P P ηη ′ dx = 1 2L [η ]−L = 0 2 if z = η(x) describes the bubble shape. T d n F n n n P W X The contour comprises the ends F and B.T is the hydrodynamic force on the object. and on the wall W . Without loss of generality. u(u · n) dl = u2 dl − u2 dl.B p nx dl − ρg z nx dl. 2 where d is the height of a wave crest over and above the undisturbed stream height (see ﬁgure). the wall W . T . and that nx is zero on the wall. the object P . P.B u(u · n) dl − F. Now.33) gives −Dx = −i · D = −ρ where D=− p n dl P n B F.B F 39 .

Eng.610). we get Dx = B (p + ρu2 ) dl − (p + ρu2 ) dl = 0 F and so there is no drag on the object (see Forbes. one can obtain waves trapped over the obstacle with the free surface becoming ﬂat at the same height on either side like this: d U U In this case. this becomes − C pB n ds − γ 40 κn ds. 3. V.171-180). JFM 1980. C . The force on the bubble is given by − pn ds. −Dx = ρ = B B u2 dl − ρ F u2 dl − 1 1 p nx dl + ρg[(H + d)2 − H 2 ] − ρg[(H + d)2 − H 2 ] 2 2 F. Hence the drag on an object is non-zero if on one side there is a train of waves extending to inﬁnity. Under some circumstances. we have the Laplace-Young equation p = pB + γκ. Using the Laplace-Young equation.Math V 16. p. p. J. we note that at the surface of the bubble. To see this. since the conditions are the same a long way upstream or downstream. 1982.10 D’Alembert’s Paradox for a bubble It is interesting to note that a two-dimensional bubble at equilibrium within an inviscid ﬂow ﬁeld experiences zero net force regardless of the nature of the velocity ﬁeld around it.96.So.B (p + ρu2 ) dl − (p + ρu2 ) dl F which in general is non-zero. Longuet Higgins shows how this expression may be written in terms of the potential energy of the wave train at inﬁnity downstream (see Vanden-Broeck. C where C is the bubble contour and s is arc length around the bubble contour.

it can be shown that n dS = 0. and γ is the surface tension. with the result that the bubble experiences no drag regardless of the nature of the surrounding ﬂow. p. So. κn = − dt . for any choice of S (see Blackmore & Ting. 41 . we have p = pB + γκ. where pB is the constant pressure inside the bubble. In fact.The ﬁrst integral is easily shown to be zero by the divergence theorem. we do not need a free stream at inﬁnity as in section 3. 27(4). A spherical bubble placed within an inviscid ﬂow also experiences zero no force whatever the far-ﬁeld ﬂow (i. By the Laplace-Young equation given above.e. D=− B where B is the closed surface of the bubble. ds where t is the unit tangent to the bubble contour pointing in the direction of increasing arc length. For the second integral. SIAM Review. a closed three-dimensional bubble of any shape experiences zero net force in an inviscid ﬂuid since it can be shown that κn dS = 0 S where κ is the mean curvature at a point on the surface S. The force becomes −γ dt ds = −γ ds dt = [t]C = 0 C C since the bubble is closed.569-572). then κ is a constant and the second integral also vanishes.5). the resultant force acting upon it must be zero (see also the force balance on an interfacial segment in section 3. and κ is the bubble curvature.8).4. B Since the bubble is spherical. Hence the bubble experiences no net force in any direction. The force on the bubble is pn dS. we note that. pn dS = pB B B n dS + γ B κn dS But. by the deﬁnition of curvature. using the divergence theorem. This is consistent with physical thinking: since the bubble interface has zero inertia.

Conservation of mass in a ﬂuid. Summary We have derived equations governing 1. 348. This means that we cannot have a bubble of ﬁxed shape in a ﬂuid with gravity.. which must sum to zero to give a total overall force of zero on the bubble. since then the total force on the bubble is − p+ n dS + S S p− n dS where p+ and p− are the pressures on the inside of the bubble respectively. from the visous drag on the bubble. u · ∇u = ∇( 1 |u|2 ) + ω × u. Conservation of energy in a ﬂuid. Other vector identities are also available.So the total force on the bubble is zero. Vector identities The following vector identities are useful in ﬂuid mechanics: ∇2 u = ∇(∇ · u) − ∇ × (∇ × u). 42 where ω = ∇ × u. This comes. ∇ × (φu) = φ∇ × u + (∇φ) × u. In a static ﬁeld. 3. The dynamic motion of a ﬂuid. in which case the total force acting is a combination of the pressure force and the normal viscous stress. 2 ∇ · (φu) = φ∇ · u + u · ∇φ. 2. for example. So there must be another force to balance this buoyancy force. the total force is − = V V ∇p+ dV + V ∇p− dV ρ+ g dV + V ρ− g dV = (ρ+ − ρ− )gV. But this must be zero by the previous argument. p = −ρgy + B for constant B. See Acheson p. . So. on using the divergence theorem.

To this end. ∂t which may be written more compactly as Dω − ω · ∇u = ν∇2 ω. Dt Finally.1) (4.6) (4.5) (4. we note that. Making use of the vector identity ∇ × (a × b) = (b · ∇)a − (a · ∇)b + a(∇ · b) − b(∇ · a) we obtain the vorticity transport equation ∂ω + u · ∇ω − ω · ∇u = ν∇2 ω.4) (4. ∂t It is convenient to use two vector identities to rewrite (4. we have ∇ · ω = 0. We can derive these by taking the curl of the Navier-Stokes momentum equation. ∂u + u · ∇u = −∇p/ρ + ν∇2 u. Taking the curl of equation (4. It is sometimes useful to understand the dynamics of a ﬂow by working with the vorticity rather than the velocity ﬁeld.1) as ∂u 1 + ∇ u2 + ω × u = −∇p/ρ − ν∇ × (∇ × u) ∂t 2 = −∇p/ρ − ν∇ × ω. Physical interpretation 43 (4.4.3) .2) gives ∂ω + ∇ × (ω × u) = −ν∇ × (∇ × ω) ∂t = ν∇2 ω. we would like to have a set of governing equations for ω rather than u. since ω = ∇ × u.2) (4. Vorticity dynamics and the stream function 4.1 Vorticity dynamics We have deﬁned the ﬂuid vorticity ω =∇×u to be the local rotation of ﬂuid elements.

5) represents the convection of vorticity with the ﬂow. The ν∇2 ω in (4. The term Dω/Dt in (4. The term ω · ∇u in (4. The vorticity is ω = ∇ × u = −k. so the vortex lines point along the z axis as shown in the diagram. thin vortex tubes experience greatly intensiﬁed vorticity at the pole and the rate of local spinning of ﬂuid particles increases.5) represents stretching of the vortex lines. Therefore.5) represents viscous diﬀusion of vorticity. Example: Snow clearing at telegraph poles pole x y z Wind blows in the x direction over ﬂat ground at y = 0 with velocity u = yi. The ﬂow encounters a telegraph pole at some x. 44 . Notes 1. the vorticity |ω 1 | is approximately constant inside the tube.w1 w2 Consider a thin vortex tube made up of vortex lines pointing in the direction of the vorticity vector ω 1 as shown in the left of the ﬁgure above. Then the fact that ∇ · ω = 0 means that |ω 2 | < |ω 1 |. Since the tube is thin. The vortex lines wrap themselves around this obstacle. since the “ﬂux of vorticity” through the tube is conserved. There is no vortex stretching in two-dimensional ﬂows. becoming very stretched close to the pole. This mechanism is responsible for the cleared areas you sometimes see at the foot of telegraph poles when it snows in the winter. Suppose that a while later the tube has expanded a little and the vorticity inside is now |ω 2 |. since ∇ · ω = 0.

2 ω(r) = v = v(r) 1 d(rv) . 2πr The Burger’s vortex described above is a steady and. This is an example of an axisymmetric ﬂow. Note that none of the components depend on θ.This is because if u = u(x. We ﬁnd v= Γ 2 (1 − e−kr /4ν ).4). for some chosen constant k.2 The Prandtl-Batchelor theorem This pertains to the vorticity in regions of two-dimensional ﬂow with closed streamlines when the viscosity is very weak. r dr (4. 45 . the vorticity is constant. this ﬂow is given by w = kz. exact solution of the Navier-Stokes equations.10) for some constant Γ. and viscous diﬀusion of vorticity. y)i + v(x. ∂z ω = ω(r)k. It is now possible to ﬁnd v(r) by integrating the last of equations (4.8). vortex stretching. It states that within such regions.9) Solving this for ω. and assuming ω → 0 as r → ∞. 4. y)j. remarkably. In terms of cylindrical polar coordinates. (4.7. the vorticity is given by ω= kΓ −kr2/4ν e k. 4πν (4.8) ∂ u = 0. It involves a perfect balance between the three mechanisms of convection of vorticity. So most of the vorticity is concentrated in a ‘core’ region of radius of order O( ν/k). 1 u = − kr.7) (4. we obtain k d2 ω dω d(ωr 2 ) + 2ν r 2 + dr dr dr = 0.8) into the vorticity transport equation (4. Substituting (4. then ω = (vx − uy )k = ωk and so ω · ∇u = w Burger’s vortex A famous example of a ﬂuid ﬂow which exhibits all of the features described above is Burger’s vortex. 4.

If the kinematic viscosity ν is very small. C C So unless by ﬂuke this line integral happens to be zero. (u · ∇)ω = 0. C Inserting (4. so that ω = ωk. Consider the steady version of the vorticity transport equation for two-dimensional ﬂow.12) into this result yields (since ω is constant along the streamline C) ω ′ (ψ) u · dx = ω ′ (ψ) u · dx = 0.11) This means that the vorticity. ∇ × ω = ∇ × (ωk) = ∂ω ∂ψ ∂ψ ∂ω i− j = ω ′ (ψ) i− j = ω ′ (ψ) u. is constant along streamlines and hence just depends on the streamfunction ψ. then the line integral (∇ × ω) · dx = 0. to a good approximation. 4. i. ω = ω(ψ). ∂y ∂x ∂y ∂x (4. 46 . usually represented by the symbol ψ.3 The stream function It is sometimes convenient to describe a ﬂuid in terms of a stream function.12) We can show (see Problem Sheet 2) that. Now.e. (4. Therefore in a region of closed streamlines the vorticity is constant. if there exists a closed streamline C in steady viscous ﬂow with vorticity ω. Recall that in 2-D the term ω · ∇u is identically zero. ω. which assumes constant values on streamlines. we must have ω ′ (ψ) = 0. we may write.Illustration of Burger’s vortex. We can see how and when this is possible as follows.

If we pick A = ψ(x. ∇ · A = 0. ∂y ∂x ∂ψ . For example. ∂x ∂ψ = 0. i.13).13) Examples i) Two-dimensional ﬂow In this case. the choice of A is not unique and we can add onto it any so-called gauge function ∇q. Now. ∂y ∂ψ ∂ψ i− j = u i + v j. if we take A′ = A + ∇q. y) j. 47 ∂ψ . then ∇ × A′ = ∇ × (A + ∇q) = ∇ × A + ∇ × ∇q = ∇ × A. Due to the aforementioned non-uniqueness. However. y) i + v(x. The general solution of this equation is u = ∇ × A. the vorticity ω = ∇ × u = ∇ × (∇ × A) = −∇2 A + ∇(∇ · A). for some vector A. y)k.e. (4. Thus ω = −∇2 A. we are at liberty to insist that A is solenoidal. u = u(x. using one of the vector identities. then ∇ · (ψk) = Thus we can represent the ﬂow velocity as u = ∇ × (ψk) = So u= and the vorticity is given by (4. ∂z v=− .The equation of conservation of mass (continuity) states that ∇ · u = 0.

b) In spherical polar coordinates. the velocity is ˆ u(r. z). but equivalent. θ) ˆ + v(r. r 2 ∂θ u=− So The representation of a ﬂow in terms of a stream function can therefore be done in diﬀerent. where ψ = ψ(r. ˆ If we choose A = (ψ/r) θ. If we choose A= ψ(r. ways. θ) ˆ φ. θ) θ. r r r r ∂r ∂z 1 ∂ψ . r ∂r u=− ∂ψ . r sin θ 48 . z) then ˆ ∇ · A = ∇ · (ψ/r) θ = Thus we can represent the ﬂow velocity as u=∇× So w= with the vorticity given by (4. where ψ = ψ(r.ii) Axisymmetric ﬂow a) In cylindrical polar coordinates. ˆ Equally. z) ˆ + w(r. θ. z) k r and so does not depend on the azimuthal coordinate. r ∂θ and ∂ψ 1 ∂ ˆ (rψ) k − ˆ = w k + u ˆ. Then 1 ∂ψ ˆ ∇ · A = ∇ · (ψ θ) = = 0. r ∂z 1 ∂ψ = 0.13). the velocity is u = u(r. ∂z ψˆ 1 ∂ψ ∂ψ θ = k− ˆ = w k + u ˆ. we could have taken A = ψ θ. r r u = ∇ × (ψ θ) = r ∂r ∂z w= 1 ∂ (rψ). r ∂r 1 ∂ψ . r so there is no dependence on the coordinate φ.

θ) ˆ ˆ φ = 2 ˆ−r r θ = u ˆ + v θ. So the gradient of ψ in direction u is zero.then ∇·A =∇· ψ(r. r sin θ r sin θ ∂φ Thus we can represent the ﬂow velocity as u=∇× So u= ∂ψ 1 ∂ψ ˆ ψ(r. Now. using the identity ∇ × (φA) = φ∇ × A + ∇φ × A. r r sin θ r sin θ ∂θ ∂r 1 r 2 sin θ ∂ψ . r sin θ ∂r v=− with the vorticity given by (4. we concentrate on 2-D ﬂow. ∂θ 1 ∂ψ .13). θ) ˆ ∂ψ 1 φ = 2 2 = 0. Physical interpretation For simplicity. Let ψ and ψ + dψ be two neighbouring streamlines: y dy dx ψ+ d ψ x ψ The ﬂux of ﬂuid ﬂowing between ψ and ψ + dψ in unit time is u dy − v dx = ∂ψ ∂ψ dy + dx = dψ. 2-D equation of motion in terms of a stream function 49 . ψ is constant along a streamline. the diﬀerence in ψ measures the ﬂux per unit time. ∂y ∂x Therefore. In other words. So u = ∇ × ψk. we can write u · ∇ψ = (∇ × ψk) · ∇ψ = (∇ψ × k) · ∇ψ = k · (∇ψ × ∇ψ) = 0.

∇2 ψ) − = ν∇4 ψ.Start with the vorticity transport equation ∂ω + u · ∇ω − ω · ∇u = ν∇2 ω.14) is ∂ ∂ψ ∂ ∂ψ ∂ ∂∇2 ψ ∂(ψ. y) where the Jacobian ∂ψ ∂∇2 ψ ∂ψ ∂∇2 ψ ∂(ψ. We study the function φ(x0 ) = V ρ(x) dV (x). Consider now ∇2 0 φ(x0 ). y) 4. and ρ is some scalar ﬁeld. So. ∇2 ω = −∇4 ψk. where x ∂2 ∂2 ∂2 ∇2 0 ≡ + 2 + 2. and so ω · ∇u = 0. ∂t ∂(x. ∂t For two-dimensional ﬂow.4 Deducing the velocity ﬁeld from a known vorticity distribution.15) where V is volume of ﬂuid surrounding the ﬁeld point x0 . Recall that in 2-D. y) ∂x ∂y ∂y ∂x ω = −∇2 ψ. using (4. x ∂x2 ∂y0 ∂z0 0 50 . ∇2 ψ) k. ∂y v=− ∂ψ . Therefore the left hand side of (4. ∂x w=0 (4.13). ∂(x. ∇2 ψ) = − . (−∇2 ψk) + (−∇2 ψk) − (−∇2 ψk) = − − ∂t ∂y ∂x ∂x ∂y ∂t ∂(x. Can we deduce u from ω? We start with a bit of potential theory. u= and ω = ωk. there is no vortex stretching.14) The equation of motion is therefore. ∂∇2 ψ ∂(ψ. r (4. Suppose we know the vorticity ﬁeld ω = ∇ × u. ∂ψ .

we have ∇2 0 φ(x0 ) = ∇2 0 x x ρ(x) dV (x) = r ∇2 0 x 1 ρ(x) dV (x). ∇ × u = ∇ × (∇ × B) = −∇2 B = ω. Now consider the vorticity distribution ω where ω = ∇ × u.Then.16) for as yet unknown velocity ﬁeld u. Hence φ(x0 ) satisﬁes the Poisson equation ∇2 φ(x0 ) = −4πρ(x0 ) Stated another way. We can do this since G = 1/r is the Green’s function for the x singularly forced Laplace’s equation. by the properties of the delta function. x In other words. a solution to this Poisson equation is B(x0 ) = 1 4π ω(x) dV (x).15) is a solution to the Poisson equation (4. Let’s express u as the curl of its vector potential as follows. (4. From the above. ∇·u=0 (4. ∇2 0 φ(x0 ) = −4π x δ(ˆ) ρ(x) dV (x). r V V But recall that we can express the delta function δ(ˆ). So. So. u=∇×B where we have chosen ∇ · B = 0. G = 1/r is a solution to this equation. x V = −4πρ(x0 ).16). ∇2 B = −ω. Then. where x = x − x0 . ∇2 G = −4π δ(ˆ ). r V 51 . in the form x ˆ δ(ˆ ) = − x 1 2 1 ∇ 4π r where r = |ˆ| = |x − x0 |.

So. Then the velocity distribution is given by u= 1 4π ∇ x0 × ω(x) dV (x). we have ∇x0 · B(x0 ) = 1 4π 1 ω(x) · n(x) dS(x). r Using the divergence theorem.We need to check ∇ · B = 0. ∂t (5. Most problems are far too complicated to expect to be able to write down the solution in a simple way. Keep in mind that we are always trying to ﬁnd solutions to the Navier-Stokes equations. 86 for a discussion). r V 5. Often these are obtained by reducing the full equations using some symmetry or other special property of the problem. Exact solutions of the Navier-Stokes equations Exact solutions to the Navier-Stokes equations which may be written down in closed form are very rare. Unidirectional ﬂows 52 . So we need ω · n = 0 on the boundary S (see Batchelor p.1) with boundary conditions appropriate to the particular problem in hand. r S where S is the surface containing the volume V . We begin with a discussion of ﬂows which only move in one direction. ∇x0 · B(x0 ) = 1 ∇x · 4π 0 = = 1 4π 1 4π V ω(x) 1 dV (x) = r 4π ω(x) · ∇x0 ∇x · V ∇ x0 · ω(x) dV (x) r V 1 dV (x) r V ω(x) dV (x). However. ∂u + u · ∇u = −∇p/ρ + ν∇2 u. there are a number of very well-known examples of exact solutions which can be found fairly easily. Instead approximate solutions to the equations have to be found numerically on a computer. ∇ · u = 0.

The continuity equation gives us some useful information about w straightaway. 0 = −pz + µ(wxx + wyy ). ∂u ∂v ∂w + + = 0.These are very special types of ﬂow for which there is only one component of velocity. Now we need to solve u = wk.2) together with appropriate boundary conditions. linear problem to solve. So the nonlinear term vanishes! This leaves us with a much simpler.1). ∂u = −∇p/ρ + ν∇2 u. In component form. (5. In Cartesians. w = w(x. y). i. We can write u = wk so that the ﬂuid is moving with speed w in the z direction only. we have to solve u = wk. y). Hence w can only be a function of x and y.3) . 0 = −px + µ(uxx + uyy ). 0 = −∇p + µ∇2 u. To summarize. for a unidirectional ﬂow. ∇·u = ∂x ∂y ∂z but for this unidirectional ﬂow. 53 w = w(x.e. Consider the nonlinear term in the Navier-Stokes equations (5. This observation leads to the following remarkable property of unidirectional ﬂows. u · ∇u = (wk · ∇)(wk) ∂ = w (wk) ∂z = 0 since neither w nor k depend on z. 0 = −py + µ(vxx + vyy ). So we have ∂w =0 ∂z which states that w does not change as we move along in the z direction. w(x. u = v = 0. ∂t (5. Steady examples of unidirectional ﬂows We start with the simpler case of steady ﬂow. y).

i) Plane Poiseuille Flow (PPF) Consider two-dimensional (y. These are set by applying the boundary conditions: w(d) = 0 =⇒ Ad + B = w(−d) = 0 =⇒ −Ad + B = Solving for A and B we obtain the solution. B. dp = −G. Then.5) 1 2 Gd . we need to solve 0=G+µ d2 w . Integrating (5. where G > 0. Note that the normal ﬂow condition v = 0 at y = ±d is satisﬁed automatically since v ≡ 0 everywhere. w= G 2 (d − y 2 ). 2µ 54 (5. z) ﬂow in a channel with centre line z and with walls placed a distance 2d apart at y = ±d. In this case u = 0 and there is no dependence on x. i.4) twice with respect to y. the ﬁrst two of these equations tells us that px = py = 0.Since u = v = 0. dy 2 w(d) = w(−d) = 0. 2 1 2 Gd 2 . we ﬁnd Ay + B = 1 2 Gy + µw 2 for arbitrary constants A.4) The boundary conditions are those of no-slip at the walls y = d and y = −d. y d z −d Suppose that ﬂuid is driven along the channel by a pressure gradient of size −G.e. dz We try looking for a unidirectional solution with w = w(y) in the z direction and v = 0 in the y direction. So there are no pressure gradients in the x or y directions. from (5.3). (5.

So the ﬂux d Q= −d w dy = = G 2µ d −d (d2 − y 2 ) dy 2Gd3 . 3µ This represents the volume of ﬂuid passing through a vertical line across the channel in unit time.y w It is customary to plot the velocity proﬁle with the velocity on the horizontal axis like this: The maximum velocity wmax = Gd2 /2µ occurs on the centre line y = 0. In that case. Thus A= U . Flux One feature of the ﬂow we might be interested in is the volume ﬂow rate or ﬂux through the channel. We expect a unidirectional solution of the form w = w(y). equations (5. B. This follows from integrating the velocity proﬁle across the channel width. d 55 B = 0. dy 2 . The boundary conditions are w(0) = 0 and w(d) = U .3) imply 0= Integrating twice. y d U 0 z In this case the ﬂuid ﬂow is driven purely by the upper wall motion. d2 w . we have w = Ay + B. ii) Plane Couette Flow Consider ﬂow in a channel whose upper wall is moving at speed U in the z-direction. for constants A.

So w = U y/d. For convenience we use cylindrical polar coordinates.6) . So now (5. ∂r r r ∂θ ∂z ∂w = 0. The continuity equation states that ∇·u = But. since w = wk and u = v = 0.3) requires us to solve 0=G+µ d2 w 1 dw . G>0 ∂u u 1 ∂v ∂w + + + = 0. and the velocity proﬁle looks like this: y d U w iii) Circular Poiseuille Flow (CPF) This is also known as Hagen-Poiseuille ﬂow after the work of Hagen (1839) and Poiseuille (1840). (5. ∂z Therefore we just have w = w(r. + dr r dr 56 w(a) = 0. dz as the pressure gradient driving the ﬂuid ﬂow. In this case we consider pressure driven ﬂow through an inﬁnite tube of circular cross-section of radius a. θ). Assuming no variation in θ we take w = w(r). taking z r=a dp = −G.

we once more integrate the velocity over the cross-section. so 1 B = Ga2 . a 0 r(a2 − r 2 ) dr πGa4 8µ . So.7) To prevent w from diverging at r = 0. 4 So the solution for CPF is w= G 2 (a − r 2 ). The constant B is ﬁxed by requiring w(a) = 0. Flux To compute the ﬂux through the circular tube.r dA=rdrdθ w The boundary condition at the wall is that of no-slip. A= for constant A. 2π a Q= 0 0 w rdrdθ = 2π = 57 G 4µ . we need to set A = 0. And again. We will need to apply another condition at r = 0 later on. This is the second boundary condition alluded to above. Recall that an element of area in the cross-section is dA = rdrdθ. A log r + B = 1 2 Gr + µw. 4µ (5.8) The maximum velocity wmax = Ga2 /4µ occurs on the centre line r = 0. 2 dr µ d dw r . r dr dr (5. We can re-write (5. 4 dw 1 2 Gr + µ r .6) as 0=G+ Integrating once with respect to r.

r Encyclopaedia of Physics. for constants a. at the wall. vol. We know that erz = ezr and we are given that erz = 1 ∂w ∂u + 2 ∂r ∂z Gr . ed. In Cartesians the domain of ﬂow lies inside the circle (x − a)2 + y 2 = a2 . umean = Shear stress To calculate the axial shear stress at the wall. and outside the circle x2 + y 2 = b2 . z) direction. 2 (5. we deﬁne the mean velocity.Furthermore. b. In this case the unit normal at the wall pointing into the ﬂuid is n = −ˆ. = x-sec area πa2 8µ (in this case) = − Therefore. where (r. Flugge. Springer 58 .e. Speciﬁcally. So the stress in the z direction at the r wall is Fz = −σzr = −2µezr . the domain of ﬂow in the cross section lies in b < r < 2a cos θ. The solution which satisﬁes the Poisson equation ∇2 u = −1 is 1 u = (r 2 − b2 ) 4 5 2a cos θ −1 . iv) Indented circular Poiseuille Flow Berker (1963)5 discusses an exact solution of the Navier-Stokes equations corresponding to pressure-driven unidirectional ﬂow along a circular pipe whose cross-section is a circle indented with part of another circle. Note that r = 2a cos θ describes a circle of radius a with centre at x = a. θ) are polar coordinates.9) which is the shear stress exerted by the ﬂuid on the wall in the downstream (i. 4µ Flux Q Ga2 = . Fz r=a = −2µerz |r=a = Ga . VIII/2. S. recall that Fi = σij nj .

8 Notes 1.where the ﬁrst factor in brackets may be identiﬁed as the solution for Poiseuille ﬂow in a circular pipe.8 0 0.9) shows that. Experimentally.2 1. 0.3. In this sense the solution resembles that for undirectional ﬂow in an equilateral triangular pipe where each factor is zero of one the three walls. for example. one might set up CPF as shown: A circular tube of length L has one L p p CPF 2 1 Reservoir 59 . the pressure gradient needs to be 16 times larger to produce the same ﬂux! 3. The shear stress is important because of its link with the onset of atherosclerosis.2 0 −0.8 0.2 0. this states that build-up of fatty deposits on the arterial walls due to an unhealthy lifestyle rapidly increases the amount of eﬀort required of the heart to drive the same amount of blood through the body. the disease responsible for about a third of deaths in Western society. Note that the ﬁrst factor is identically zero on the smaller circle of radius b and the second factor is identically zero on the larger circle of radius a.6 0.4 −0.4 0.6 1.4 0. Applying CPF to blood ﬂow. If the artery’s diameter is halved. the driving pressure gradient G ∝ a−4 .2 −0.6 0. 2. The result (5. for a ﬁxed ﬂux through the tube.6 −0.8 1 1. CPF is often used by medical workers as a model of blood ﬂow in arteries.4 1. A contour plot showing lines of constant velocity u is shown in the ﬁgure below for the case a = 1 and b = 0.

the CPF solution (5. The pressure at the exit. He injected a blob of dye at the entrance to the pipe and it stretched out in a perfectly straight line. where ∇2 = µ ∂x2 ∂y 60 . L say. Reynolds found that for suﬃciently low ﬂow rates Q. as this solution predicts. the start of the turbulent region moved further and further upstream. Reynolds found that at some distance along the pipe the laminar CPF ﬂow eventually broke up and thereafter became turbulent. As Q was increased. His original apparatus is kept in the Engineering department at Manchester University. x z y f(x. as Q was increased.end in a reservoir of water so that the entry pressure p2 is held ﬁxed. The reason for the break-up of the ﬂuid ﬂow in a circular pipe into turbulence is still not fully understood! Steady unidirectional ﬂow in pipes of arbitrary cross-section Unidirectional solutions exist in pipes of general cross-section.y)=0 With a pressure gradient −G in the z direction driving the ﬂow. streak of dye However. The pressure gradient is then G=− p2 − p1 L =− ∆p . This experiment was ﬁrst performed by Osborne Reynolds in 1883. the problem to be solved is ∂2 ∂2 G + 2 ∇2 u = − . is also constant.8) was indeed observed. p1 .

with u=0 on f (x, y) = 0.

For very special cross-sectional geometries (a circle, an ellipse, a triangle – see Sheet 3) the solution can be found in closed form. Otherwise a solution can be sought using a complex variable - conformal mapping technique. Alternatively, we can seek a series solution, as in the next example. Flow in a rectangular channel
y

2b

x

2a

z

Consider steady undirectional ﬂow in a channel of rectangular cross-section −a ≤ x ≤ a, We need to solve uxx + uyy = − G , µ with u=0 on x = ±a, y = ±b. (5.10) −b ≤ y ≤ b,

driven in the z direction by a pressure gradient dp/dz = −G.

If we look for a separable solution of type u(x, y) = X(x)Y (y), we ﬁnd that the velocity can be expressed in the series form
∞ ∞

u=
m=0 n=0

Amn cos (2m + 1)

πy πx cos (2n + 1) , 2a 2b

(5.11)

where the Amn are constants to be found. Exercise 1: Determine the Amn as follows. Substitute (5.11) into (5.10). Then, multiplying both sides by πx πy cos (2m + 1) cos (2n + 1) , 2a 2b integrate over the rectangular cross-section. The result is: Amn = (2m + 1)2 (2n + 1)2 64G(−1)m+n + (2m + 1)(2n + 1) a2 b2 61
−1

.

Exercise 2: Show that the channel solution above reduces to plane Poiseuille ﬂow (PPF) in the limit b/a → 0. Film ﬂow down an inclined plane Consider the two-dimensional ﬂow of a ﬁlm of ﬂuid of height h under gravity down an inclined plane:

α

y

g

h
x

Fix Cartesian axes as shown. The body force, F, due to gravity is given by F = (Fx , Fy ) = g(cos α, − sin α). The pressure at y = h is equal to atmospheric, pa . We seek a steady unidirectional solution with u = u(y), v = 0. First, check that continuity is satisﬁed: ∂u ∂v + = 0 + 0 = 0. ∂x ∂y From before, we know that for unidirectional ﬂow u · ∇u = 0. The momentum equations are therefore, 0 = − ∂p d2 u + ρg cos α + µ 2 , ∂x dy ∂p 0 = − − ρg sin α. ∂y

Integrating the second, p = −ρgy sin α + f (x). But p = pa at y = h and so p = pa + (h − y)ρg sin α. From this we can see that ∂p = 0. ∂x 62

Therefore, the ﬁrst momentum equation becomes 0 = ρg cos α + µ Integrating twice, u=− for constants β, γ. No-slip at y = 0 =⇒ γ = 0. At y = h there is a ﬂuid-ﬂuid interface between the water (say) in the ﬁlm and the air above it. The stress condition at y = h is given by (assuming negligible surface tension) [σij where n = (0, 1, 0) = j. So, −pa ni + 2µ(air) eij i.e. p|y=h = pa , since e22 = ∂v/∂y = 0, and 2µ(air) e12 |y=h = 2µ e12 |y=h . But, since air is much less viscous than water, µ(air) ≪ µ, the last equation becomes just e12 |y=h = 0, i.e. du dy
y=h (air) (air) (air)

d2 u . dy 2

ρgy 2 cos α + βy + γ, 2µ

− σij

(water)

] nj = 0,

|y=h nj = −p|y=h ni + 2µ eij |y=h nj ,

= 0.

Exercise: Follow a similar argument to above to derive the shear stress condition at the surface of a liquid ﬁlm ﬂowing down the outside of a cylindrical rod of radius a. Applying the condition of zero surface stress, we ﬁnd 0=− ρgh cos α + β, µ 63

64 . umax = The ﬂux. h ρgh2 cos α. 2µ ρg 2 h − (h − y)2 cos α. dF The jump travels downstream at a ﬁnite speed. so that. given a ﬂux of ﬂuid F and slope angle α. 1 Poiseuille flow 2 The maximum speed. 2µ F = 0 u dy = ρgh3 cos α . dh = dh dF .and so u= which can be re-written u= ρgy (2h − y) cos α. we have the picture: Continuity requires that net ﬂux before jump = net ﬂux beyond jump. V . 2µ This is the Nusselt solution. It represents ‘half’ of plane Poiseuille ﬂow. a steady unidirectional motion is possible for a ﬁlm of height 1 3µF 3 . we expect a corresponding increase in the layer thickness. 3µ ρg cos α 2µ h 0 (2hy − y 2 ) dy = In other words. with axes ﬁxed in the jump. h= ρg cos α What happens if F is suddenly increased by a small amount dF ? Since dF is small.

t). Unsteady unidirectional flows For an unsteady. Thus. 2µ dF . and taking the limit. unidirectional ﬂow. Rearranging.2). t). y. we have to solve (5. the speed of the jump is twice the maximum speed of the ﬂuid.h+ dh y jump F+dF h F h+ dh x V stationary F+dF V h F x i. ∂t As in the steady case. 65 . u = wk.e. w = w(x. (F + dF ) − (h + dh)V = F − hV. V = From above. ρgh2 cos α. dh 1 3 . the ﬁrst and second momentum equations imply that px = py = 0 =⇒ p = p(z. ∂u = −∇p/ρ + ν∇2 u. we know that umax = and so V = 2umax . dF ρgh2 9gF 2 cos α = cos α = dh µ ν But.

This motivates us to seek a solution with the same property. y Fluid z For t < 0. The problem for w is thus: νwyy − wt = 0. p = 0 for all z and t. accordingly. the transformation y = αˆ. t) and. w(y. W = f (η). There is no variation in z so all z derivatives are zero. y > 0. t).1. we try looking for a solution in the form y . ∂/∂z ≡ 0. y ˆ t = α2 t leaves the equations and boundary conditions unchanged. this becomes νWyy − Wt = 0. t) → 0 We now make the observation that. w(0. the wall is at rest. Impulsively started ﬂows . y → ∞. Since there is no imposed pressure gradient. where η = 2(νt)1/2 66 . W (0. t) → 0 as 0 for If we write w = U W (y. t) W (y. So w = w(y. rigid wall. t) w(y. 0) = = 1 0 for t > 0. W (y. for constant α. Accordingly. 0) = = U for t > 0. continuity is satisﬁed. y → ∞. At t = 0 it instantaneously acquires a speed U in the z direction. We note that the grouping y t1/2 is unchanged by the α-transformation. as for y > 0.Rayleigh layers a) Unbounded ﬂow Consider a semi-inﬁnite mass of ﬂuid at rest above a plane.

So Then. 2 where erf is the Error function. √ π . The (y. f =B+A 0 f ′ = Ae−η . This means that they are just re-scaled versions of each other.This is called a similarity solution. e−ξ dξ. 1η ∂ ∂ 1 ∂2 1 ∂2 ∂ ∂ =− . which means ∞ A 0 e−ξ dξ = −1. since W = 1 at y = 0. it follows that f = 1 at η = 0 and so B = 1. For example. Applying the chain rule. = . Also. f ) proﬁles shown in the ﬁgure above are self-similar. 67 . we have f → 0 as η → ∞. = . This is called a Rayleigh layer. It has the advantage of reducing the partial diﬀerential equation to an ordinary one. 2 for η constant A. Plotting y against f : y O( νt 1/2) t increasing 1 f Thus there is a layer of thickness O((νt)1/2 ). for reasons which will become clear below. 2 for constant B. since W → 0 as y → ∞. ∂t 2 t ∂η ∂y ∂y 2 4νt ∂η 2 2(νt)1/2 ∂η Thus the equation becomes f ′′ + 2ηf = 0. 2 2 It is known that 0 ∞ e−ξ dξ = Thus η 0 2 2 f =1− √ π e−ξ dξ = 1 − erf(η) = erfc(η). in which the ﬂuid is most aﬀected by the wall motion. But.

i= i=− 1/2 dη 1/2 ∂y π 2(νt) (νt) 1 U 2 √ e−y /(4νt) i. but retains its original shape. ω = ∇ × u = ∇ × (wk) = So. Thus the dimensions of the LHS are L2 1 1 [ν] = = 2] 2 [y T L T and those of the RHS are 1 1 = . Thus the vorticity (shown as a shaded area) diﬀuses away from the wall in time. the exponential is not small and ω will have an appreciable value. as time progresses. the exponential is very small indeed and ω is approximately zero. In this case. L. y O( t 1/2) O( t 1/2) z later time y z Thus. A similarity solution is possible in a problem like this when there is no natural lengthscale. the velocity proﬁle is simply rescaled. the vorticity behaves like heat diﬀusing away from an impulsively heated plate at y = 0. if y 2 ≫ 4νt. Any proﬁle at any given time may be simply rescaled to lie on top of any other proﬁle. Similarity solutions We can now see why we called the previous solution a similarity solution. the plate is inﬁnite and so we cannot do this. However. 1/2 π (νt) Now. As time changes. Consider the governing equation: νWyy = Wt . ω=− dW 1 U U ∂w 2 √ e−η i. measurable values of vorticity lie in the approximate y range [0. We know that ν has dimensions [ν] = L2 T −1 . 2(νt)1/2 ]. If the plate were ﬁnite we could choose its length. This range extends in time like the square root of t. say. So. [t] T 68 . Considering the vorticity of this impulsively-driven ﬂow. as a basis for non-dimensionalization. if y 2 ∼ 4νt.we could multiply the ﬁrst proﬁle by a suitably-chosen and get it to lie directly on top of any of the other proﬁles.

it is reasonable to assume that the pressure gradient G = 0 and that the ﬂow is unidirectional. Then W (y) satisﬁes νW ′′ − iωW = 0. a similarity solution with similarity variable η. with w = w(y. So continuity is satisﬁed and the problem is νwyy = wt . 69 . where ℜ means ‘take the real part’. we expect the frequency of the ﬂuid velocity w also to be ω. with w → 0 as y → ∞. So the solution must feature y. We can conclude that the ﬂuid velocity must be expressed in the form w = W f (η). t and ν grouped only in this way. [y 2 ] So the only possible dimensionless grouping of the variables is y (νt)1/2 or any power thereof.Stokes layers Consider a similar ﬂow to the above. t). Oscillating flows . Since the frequency of the wall motion is ω and the problem is linear.A dimensional balance therefore gives [ν][t] = 1. Thus we seek a solution by writing w = W0 ℜ W (y)eiωt . (νt)1/2 2. η= y . w = W0 cos(ωt) on y = 0. y z W0 cos(ω t) Again. but with the wall at y = 0 now oscillating in its own plane at a ﬁxed frequency.

2ν According to the exponentially decaying term in (5. Note that the Stokes layer thickness does not grow in time. Thus. This contrasts with the growing Rayleigh layer studied above. This region is called a Stokes layer. to satisfy the inﬁnity condition. Therefore there is a phase diﬀerence of π/4 between the velocity w and the wall shear. W →0 as y → ∞. for constants A. y+ 2ν 4 (5. ν √ i = (1 + i)/ 2 and so. W = Aeσy + Be−σy . The vorticity ω = ∇×u = ∂w i = −W0 ∂y ω −√ ω y 2ν cos ωt − e ν π ω . The higher the frequency. W = B exp − (1 + i)y/(2ν/ω)1/2 .12) ω y . the thinner the Stokes layer. we need to set A = 0. Other exact solutions 70 . √ iω . y Stokes layer 1/2 O( (ν/ω) ) z Note: The wall shear stress is proportional to ∂w ∂y y=0 ∝ cos ωt − sin ωt = √ 2 cos(ωt + π/4).12).W =1 So on y = 0. most of the vorticity is conﬁned in a region of thickness O( ν/ω). and so √ω w = e− 2ν y cos ωt − W0 This is called a Stokes shear wave. B and where σ= Now.

we note that the continuity equation is satisﬁed straightaway. Speciﬁcally. since ux + vy = f ′ − f ′ = 0. By analogy with the inviscid solution. Clearly this solution does not satisfy the no slip condition at y = 0. v = −ky. we seek a solution to the viscous ﬂow problem of a similar form. ∂x the inviscid. With this choice. 71 . u = kx. The boundary condition is that of no normal ﬂow at y = 0. Stagnation-point (Hiemenz) ﬂow. x Consider the ﬂow of a viscous ﬂuid towards a plane wall as shown in the diagram. namely v = 0 on y = 0 or ψ = 0 on y = 0. Thus. for some constant k. where f is a function to be determined. irrotational ﬂow is governed by Laplace’s equation ∇2 ψ = 0. we need to incorporate viscous eﬀects. we see that the solution is ψ = kxy. we write u = xf ′ (y). v = −f (y).y 111111111111111111111111111111 000000000000000000000000000000 1. First consider the ﬂow as if it were perfectly inviscid and irrotational. ∂y v=− ∂ψ . To manage to do this. By inspection. Introducing a twodimensional stream function ψ. deﬁned by u= ∂ψ .

1 uvx + vvy = − py + ν[vxx + vyy ]. we need to solve the x-momentum equation. we have that νf ′′′ + f f ′′ − f ′2 = β. (5. y =⇒ p = p(x) + ρ 0 Q(ξ) dξ. consider the y-momentum equation. f must satisfy the system νf ′′′ + f f ′′ + k2 − f ′2 = 0. we ﬁnd px = ρx[νf ′′′ + f f ′′ − f ′2 ]. Thus we require f ′ → k as y → ∞. in which case u → ky as y → ∞. we don’t yet know the form of the pressure. 2 for constant p0 . for some constant β. ρ However. we deduce that p= 1 2 ′′′ ρx [νf + f f ′′ − f ′2 ] + p0 . for function of integration p(x). py = −ρ[f f ′ + f ′′ ] ≡ Q(y). Letting t → ∞ in (5. as y → ∞. Rearranging the x momentum equation. f′ → k as y → ∞. ρ which becomes. we ﬁnd that β = −k2 . Since p = p(x).13) The ﬁrst two boundary conditions are those of no normal ﬂow and no slip at the wall. To resolve this issue. Therefore. Combining these results. we expect this solution to approach the inviscid ﬂow calculated above.To determine f .13). 1 uux + vuy = − px + ν[uxx + uyy ]. 72 . say. f = f ′ (0) = 0 on y = 0. Now.

Exact Navier-Stokes solutions in cylindrical polar coordinates If we express the Navier-Stokes equations in cylindrical polar coordinates (see sheet). The solution for f must be found numerically. bluﬀ body. This means that nothing changes with θ. we can make considerable simpliﬁcations by assuming that the ﬂow in question is axisymmetric. The stagnation point ﬂow on a plane wall is also relevant to that occuring on a curved.5 0 x 0. For small x.e. ∂ ≡ 0.5 1 Notes 1. x y In the neighbourhood of the stagnation point. 2. i. the external inviscid ﬂow may be expanded in a power series U (x) = u0 x + u1 x2 + u2 x3 + · · · . the body may be considered locally ﬂat. ∂θ 73 . The streamlines for this ﬂow look like this: 3. and so locally U ∝ x.70 60 50 40 y 30 20 10 0 -1 -0.

t) θ + w(r. The inner cylinder rotates at angular speed Ω1 . ∂t ∂ ≡ 0. t) ˆ + v(r. ∂θ and since it is just swirling around. r We have already seen an example of an exact axisymmetric solution to the Navier-Stokes equations. Steady axisymmetric swirling flow between coaxial cylinders Fluid ﬁlls the annular gap between two inﬁnite coaxial circular cylinders. r=a Ω1 r=b Ω2 Since the ﬂow is steady and axisymmetric. The equations of motion are − 1 ∂p v2 =− . dr . ˆ u = u(r. with no dependence on the z coordinate. Thus the streamlines for this ﬂow are concentric circles.So. z. ∂ ≡ 0. z. t) k. z. ˆ u = v(r) θ. t) ˆ + w(r. t) k. the velocity ﬁeld is given by u = u(r. This was circular Poiseuille ﬂow (CPF). Next we look at an axisymmetric swirling ﬂow. for an axisymmetric ﬂow. r For an axisymmetric ﬂow with swirl. z. and the outer at Ω2 . z. r ρ ∂r 0= 74 d2 v 1 dv v 2 + r dr − r 2 .

r2 Hence. τa = 4πµa2 b2 75 . b2 − a2 B = a2 b2 b2 Ω2 − a2 Ω1 k. r2 2µB . b r = a =⇒ aΩ1 = Aa + Ω2 − Ω1 . 1 d v eθr = r 2 dr r Thus.Solving the second of these. We know that (see sheet). v = Ar + for constants A. we will need σθr at r = a. b2 − a2 What is the couple on the inner cylinder? [Recall that couple (or torque or moment) means the turning eﬀect of a force acting at a distance from a given point. B. r Thus the solution is a combination of a rigid-body rotation ˆ u = r θ. and a line vortex u= The boundary conditions are v = aΩ1 v = bΩ2 Thus.] To determine this. the couple or torque per unit length in the z direction is given by 2π τa = 0 (aσθr |r=a ) a dθ = −4πµB. on on B . B . r b2 Ω2 − a2 Ω1 . a2 − b2 1 ˆ θ. A= The vorticity ω = ∇×u = 2 and is therefore constant. Ω2 − Ω1 . σθr = − =− B . a B r = b =⇒ bΩ2 = Ab + . b2 − a2 So.

1 ∂(rv) k = 0. Again. r but still everything is independent of θ. Stationary outer cylinder: Circular Couette ﬂow If Ω2 = 0. the couples on the inner and outer cylinder both vanish. Note that in this case. v = rΩ2 . Steady radial axisymmetric flow with swirl between coaxial cylinders We can generalize the previous problem to include a radial component of ﬂow. b2 − a2 Letting b → ∞. 76 . 3. Equal rotation rates In this case Ω1 = Ω2 and so B = 0. the couple on the outer cylinder is found to be τb = −4πµa2 b2 Special cases: 1. We continue to assume the ﬂow is also independent of z. r b − a2 r 2 v= This ﬂow is irrotational since ω= a2 Ω1 . r Ω2 − Ω1 .Similarly. No inner cylinder Now a = 0 and so we must choose B = 0 to avoid a singularity at r = 0. This means the ﬂuid swirls round as if it were just a rotating solid body. r ∂r Exercise: Show that taking the double limit a → ∞ and b → ∞ such that the diﬀerence (b − a) is held constant. A = Ω1 . v = rΩ1 . we recover plane Couette ﬂow (PCF). so ˆ u = u(r) ˆ + v(r) θ. we get a2 Ω1 b2 v = 2 −1 . 2. we just have solid body rotation.

and so Suppose that u = −U at r = a. we can re-write the v equation as v 1 dv d2 v 2 + (1 + R) r dr − (1 − R) r 2 = 0. R = U a/ν. n has a repeated root of −1. In the special case of R = 2. once v is determined. u=− r and. 1 d (ru) = 0. aU u=− . we get n(n − 1) + n(1 + R) − (1 − R) = n2 + Rn + (R − 1) = (n + 1)(n + [R − 1]) = 0. dr Seeking a solution of the form v = kr n . Then aU . 2 + 1+ ν r dr ν r2 dr This is an equation for v(r). Since u is already known. r dr ru = constant. and so n = −1. b so there is injection of ﬂuid at the outer cylinder. so there is suction of ﬂuid at the inner cylinder. d2 v U a 1 dv Ua v − 1− = 0. we can in principle solve the r-momentum equation to obtain the pressure p. Deﬁning a Reynolds number based on the suction velocity. + dr r r dr r 2 dr 2 Substituting for u in the second equation. with U > 0.The continuity equation implies. The r-momentum equation states that u 1 dp u d2 u 1 du du v 2 − =− +ν − 2 . + dr r ρ dr r dr r dr 2 while the θ-momentum equation has u dv uv v d2 v 1 dv + =ν − . 77 . (1 − R). at r = b.

the only signiﬁcant swirl occurs very near to the inner cylinder. v = aΩ1 v = bΩ2 on on r = a. Then. a Consider what happens when the Reynolds number becomes large. r 1 − (b/a)R−2 C + Db(1−R) . a very narrow region over which there is a very rapid change in the solution. Spin down in a circular cylinder 78 . Then aΩ1 = 0= Then. We consider the special case Ω2 = 0 and R = 2. v≈ a2 Ω1 a r r R−2 = aΩ1 a r R−1 Since r > a and R is very large. R = 2. This becomes more severe as R gets larger. We will deal with these in more detail later in the course. R ≫ 1.e. The constants C and D come from the boundary conditions. This is our ﬁrst example of a boundary layer. In other words. v = = = a2 Ω1 b2−R − r 1−R b2−R − a2−R r a2 Ω1 1 − (r/b)2−R r 1 − (a/b)2−R a2 Ω1 1 − (b/r)R−2 . this is essentially zero unless r is very close to a. i. The ﬁgure shows that the solution goes from near zero to one over a very short distance. r = b. b C + Da(1−R) .So there are two possibilities: v = v = C + Dr (1−R) r C ln r +D r r if if R = 2.

5 r 2 Make a cup of coﬀee and stir it round a few times. ∂t ∂r 2 r ∂r r 2 with v = 0 on r = a for t > 0. v = Ω r for t < 0.a=1. we can set . the right hand side of t. At t = 0 the cylinder is suddenly stopped. This is the special case discussed above of rigid body rotation with no inner cylinder. Thus.8 0. We can seek a separable solution to this problem.2 0 1 1. r r ˙ 1 1G H ′′ 1 H ′ + − 2 = . writing v = G(t)H(r). b=2. So. that of how long it takes swirling ﬂuid in an inﬁnite cylinder to stop once the forcing is removed.6 v(r) 0.4 0. Omega1=1 1 R=10 R=100 0. Suppose viscous ﬂuid occupies the cylinder r = a and suppose that both the cylinder and the ﬂuid are initially rotating with angular velocity Ω. The ensuing unsteady ﬂuid motion is governed by the θ-momentum equation v ∂ 2 v 1 ∂v ∂v =ν − + . so ˆ u = v θ. H rH r νG H ′′ 1 H ′ 1 + − 2 = −λ2 . How long does it take for the coﬀee to come to rest? We consider a simpliﬁed version of this problem. H rH r 79 The left hand side is purely a function of r. so 1 1 ˙ GH = ν GH ′′ + GH ′ − 2 GH .

83. To satisfy the boundary condition at r = a. β2 ≈ 7.2 Label the zeroes βn . r r r 2 H ′′ + rH ′ + (λ2 r 2 − 1)H = 0.2 0 2 4 6 8 10 x 12 14 16 18 20 –0. Thus. β3 ≈ 10. This is Bessel’s equation of order 1 (see MTH-2C23). Now Y1 is singular at r = 0. Therefore pick values of λ equal to βn /a. which has solution ˙ G = −νλ2 G. Y1 are Bessel functions. B are constants and J1 . The ﬁrst few are: β1 ≈ 3. we need H(a) = 0 =⇒ J1 (λa) = 0.4 0. The Bessel function J1 (x) looks like this: 0.6 0. so we must reject it by setting B = 0.for separation constant λ. 2 This leaves or 1 1 H ′′ + H ′ + λ2 − 2 H = 0. disregarding the trivial solution provided by A = 0. 80 . G = e−νλ t . It has the general solution H(r) = AJ1 (λr) + BY1 (λr).17.02. where A.

Deﬁning λn = βn /a. a 2 0 r[J1 (λn r)] dr We can use the particular properties of Bessel functions to simplify this to An = − 2aΩ . So. t) = −2aΩ where β1 ≈ 3. β J (β ) n=1 n 0 n ∞ When t > 0.83. the general solution is given by ∞ v= n=1 An J1 (βn r/a) e−νβn t/a . Multiplying both sides by rJ1 (λn ). using the general solution. t) = −2aΩ J1 (βn r/a) −νβn t/a2 2 e . ∞ Ωr = n=1 An J1 (λn r). v(r. J1 (β1 r/a) −νβ1 t/a2 2 . We know that a rJ1 (λn r)J1 (λm r) dr = 0 if 0 n = m. An = Ω a 2 0 r J1 (λn r) dr . We can see that v will have dropped to 1/e ≈ 0. and using the orthogonality property above. We can exploit the orthogonality of the Bessel functions to ﬁnd the An . Thus. we ﬁnd a Ω 0 r 2 J1 (λn r) dr = An 0 a r[J1 (λn r)]2 dr. the inﬁnite sum is dominated by its ﬁrst term. The initial condition is that v = Ωr at t = 0 and so. 2 2 for constants An . integrating from 0 to a. v(r. e βn J0 (β1 ) 81 . βn J0 (βn ) where J0 is a Bessel function of zeroth order.37 of its initial value when 2 t = te = a2 /(νβ1 ).

p= µU ∗ p . since the ﬂow is very sluggish and so the magnitude of the velocity at any point is small. we would expect it to be smaller than the term on the right hand side as a reasonable approximation. Suppose that characteristic length and velocity scales are L and U respectively. we may wish to model a bacterium swimming at speed U . which seems rather long. Stokes ﬂow (creeping ﬂow) We have shown that the steady ﬂow of a Newtonian.So by this time. ∇ · u = 0. if we take a = 4cm and ν = 10−6 m2 s−1 (the value for water). where the length of the bacterium is L. Therefore. describes the motion of an extremely viscous or sticky ﬂuid such as treacle or golden syrup. The discrepancy lies in the crucial role played by a boundary layer (see later lectures) of ﬂuid on the bottom of the cup (see Acheson pp. (6. 165.2) We may derive these equations more formally as follows. we nondimensionalize (i. we get te ≈ 2 minutes. we may approximate the ﬂuid motion using the Stokes equations ∇ · u = 0. y = L y∗. incompressible ﬂuid of kinematic viscosity ν and density ρ satisﬁes the Navier-Stokes equations (written here without a body force). Then. the swirling ﬂuid has signiﬁcantly slowed. 0 = −∇p + µ∇2 u. 284).1) Stokes ﬂow. remove all dimensions from the problem) by writing u = U u∗ . 82 . For example. 6.e. Under these conditions. since the left hand side is quadratic in the small velocity. (6. 45. L x = L x∗ . u · ∇u = −∇p/ρ + ν∇2 u. Going back to the cup of coﬀee. v = U v∗ . sometimes referred to as creeping ﬂow. we might expect |u · ∇u| ≪ |∇2 u|. Experiment suggests that in fact the coﬀee slows to 1/e of its initial value in about 15 seconds. Equations of Stokes ﬂow Under the aforementioned conditions.

L2 x x Multiplying by L/U 2 and dropping the asterisks for convenience. ux + vy = 0. Deﬁning the Reynolds number R such that R= and multiplying through by R. we have R(uux + vuy ) = −px + uxx + uyy . UL ν (6. ∇ · u = 0.3) Formally taking the limit R → 0. Or. we obtain (using the chain rule to convert the derivatives): U2 ∗ ∗ u ux∗ + v ∗ u∗∗ y L 2 U ∗ ∗ u∗ vx∗ + v ∗ vy∗ L ∗ u∗ ∗ + vy∗ x = − νU ∗ p ∗+ L2 x νU = − 2 p∗ ∗ + L y = 0. Substituting in. in vector form.Then all quantities with an asterisk superscript are dimensionless. UL ν ( −py + vxx + vyy ) . y L2 x x νU ∗ ∗ (v ∗ ∗ + vy∗ y∗ ). = UL = 0. R(uvx + vvy ) = −py + vxx + vyy . 83 . we derive the Stokes equations of creeping ﬂuid motion: 0 = −py + vxx + vyy . as we had above. we have uux + vuy = uvx + vvy ux + vy ν ( −px + uxx + uyy ) . νU ∗ (u ∗ ∗ + u∗∗ y∗ ). In component form. ux + vy = 0. 0 = −∇p + µ∇2 u. the momentum and continuity equations for a 2D steady ﬂow with no body force acting are: 1 uux + vuy = − px + ν(uxx + uyy ). 0 = −px + uxx + uyy . ρ 1 uvx + vvy = − py + ν(vxx + vyy ). ρ ux + vy = 0.

we may re-write the Stokes equations (6. ∇2 p = 0. we may use the Stokes equations (6. taking the divergence of the second of equations (6. UL . So Stokes ﬂow arises if • The characteristic velocity scale is suﬃciently small (e. a bubble moving through treacle) • The characteristic length scale is suﬃciently small (e. p satisﬁes Laplace’s equation.6) .4) Instead.. a swimming spermatozoa) • The kinematic viscosity.g. where ω = ∇ × u is the vorticity. we see that.4). ∇ · ω = 0. 0 = ∇2 ω.g. ν.2) like this. Alternative forms of the Stokes equations Remembering the identity ∇2 u = ∇(∇ · u) − ∇ × (∇ × u). and so the pressure is a harmonic function. The great advantage of doing this lies in the fact that the Stokes equations are linear.. By the deﬁnition of R. Taking the curl of the second of equations (6. L.The key step is taking the limit R → 0.4). linear problems are easier to solve than nonlinear ones. which means we are considering ﬂows for which the Reynolds number is small enough for the nonlinear term u · ∇u to be neglected. we need U to be small.5) ∇p = −µ∇ × ω (6. is large (e..2) to approximate the ﬂuid ﬂow. golden syrup) If the scales U . or for ν to be large. or L to be small. and ν combine to make R small. (6. R= ν This means that for R ≪ 1. In general. ∇ · u = 0. for a Stokes ﬂow.g. we ﬁnd 0 = ∇ × (∇ × ω). and hence we derive the equations of vorticity transport for a Stokes ﬂow. 84 (6.

Laplace’s equation ∇2 φ = 0 at all points x inside some bounded region D with boundary curve C. by deﬁnition. (x0 − x) · n = (x0 − x) · (x0 − x) = a a 85 . and dl(x) is the increment of arc length along C at the point x. since C is a circle. Then (6. Choose C to be a circle of radius a centred about the point x0 . say. Ω. But ∇2 φ = 0. n = (x0 − x)/a. since p is a harmonic function.1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 xo 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1 0 V 1111111111111111111111111111111111 0000000000000000000000000000000000 Ω 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 1111111111111111111111111111111111 0000000000000000000000000000000000 Points of maximum/minimum pressure An interesting consequence of this is as follows. its maxima and minima can only occur on the boundary of Ω. Consider Stokes ﬂow in a closed region.7) gives φ(x0 ) = ln a 2π ∇φ(x) · n(x) dl(x) + 1 2πa2 (x0 − x) · n(x) φ(x) dl(x). D C where dS is an element of surface area. at a point x0 in D. φ(x0 ) = ln a 2π ∇2 φ(x) dS(x) + 1 2πa2 (x0 − x) · n(x) φ(x) dl(x). Also. Then. φ(x0 ) = 1 2π ln r ∇φ(x) · n(x) dl(x) + 1 2π 1 (x0 − x) · n(x) φ(x) dl(x). r2 (6.7) C C where n(x) is the unit normal at a point x on the curve C. Extrema of harmonic functions Consider a harmonic function φ(x) satisfying. Thus a2 1 = a. [Sketch region D] Green’s third identity in two-dimensions states that. C C Using the divergence theorem on the ﬁrst integral.

we may introduce a stream function deﬁned by ∂ψ ∂ψ . since the unit vector k is constant. In fact it is possible to describe a Stokes ﬂow entirely by the values the various ﬂow variables assume on the boundary.5). it is clear that there must be at least one point on any surrounding circle (lying within the domain) where the value of φ is higher or lower than φ(x0 ) in order for the mean value to be φ(x0 ). by equation (6. This leads to a very eﬃcient means of computing Stokes ﬂows.6) over any small volume V of surface area S with outward unit normal n around this point. Hence extrema can only arise on the boundary. It follows that p must be increasing or decreasing arbitrarily close to the point x0 . v=− . For more information on this advanced procedure.8) This is Gauss’ mean value theorem for harmonic functions in two-dimensions.Hence we are left with φ(x0 ) = 1 2πa φ(x) dl(x). contradicting the supposition that it is either a maximum or a minimum. Using (6. C (6. Then. we ﬁnd that the stream function satisﬁes the biharmonic 86 . If we integrate equation (6. we obtain ω = ∇ × u = −∇2 ψ k.8). Therefore the points of maximum or minimum pressure in a closed-region Stokes ﬂow can only occur on the boundary. known as the boundary integral method. we ﬁnd 0= V ∇2 p dV = S ∇p · n dS (by the divergence theorem. suppose that a maximum or minimum of p exists at a point x0 within Ω. ⋄ Nifty move: This observation hints at a more profound characteristic of Stokes ﬂows. This contradicts the original assumption.) (6. It states that φ at a point x0 is the average of the φ values over a surrounding circle of radius a. Two-dimensional Stokes ﬂow If the ﬂow in question is two-dimensional with velocity ﬁeld u = ui + vj.9) Clearly ∇p · n cannot be single-signed over the arbitrary surface S. Proof Assume that an extremum is located at x0 inside the domain of interest. u= ∂y ∂x Calculating the vorticity. Cambridge (1992). ‘Boundary Integral and Singularity Methods’. Proposition: Maxima or minima of harmonic functions in a bounded region can only occur on the boundary. see Pozrikidis. To see this.

equation ∇4 ψ = 0. e′ be the stress tensors associated with (u. (6. p) and (u′ . ∂xj V (by divergence theorem) = S V (ui − u′ ) i ∂eij dV ∂xj (since ui = u′ on S) = − i V 87 .10) We will make use of this feature of two-dimensional Stokes ﬂows later when we study ﬂow in corners. and ∇ · u′ = 0. then the equations of Stokes ﬂow have at most one solution. (eij − e′ )eij dV ij = 1 2 ∂ ∂ (ui − u′ ) + (uj − u′ ) eij dV i j ∂xj ∂xi µ∇2 u′ = ∇p′ µ∇2 u = ∇p V V (since eij symmetric) = V 1 ∂ 1 ∂ (ui − u′ ) eij + (ui − u′ ) eij dV i i 2 ∂xj 2 ∂xj eij ∂ (ui − u′ ) dV i ∂xj (ui − u′ ) i ∂eij dV ∂xj = V = V ∂ [(ui − u′ )eij ] dV − i ∂xj (ui − u′ )eij nj dS − i (ui − u′ ) i ∂eij dV . Let eij . p′ ) respectively. If u is prescribed on the boundary S. Proof Suppose that there exist velocity ﬁelds u and u′ satisfying ∇ · u = 0. ij Then. Uniqueness theorem for Stokes ﬂow Consider Stokes ﬂow in a region V bounded by a closed surface S with unit outward normal n. with u = u′ on the boundary S.

by a similar line of reasoning. QED ⋄. e′ ij u′ Minimum dissipation theorem for Stokes ﬂow The rate of viscous dissipation in a Stokes ﬂow is less than or equal to that in any other incompressible ﬂow satisfying the same boundary conditions. The proof does not work for full Navier-Stokes ﬂow. V (ui − u′ ) i (ui − u′ ) i (ui − u′ ) i ∂ ∂uj 1 ∂ 2 ui + 2 ∂xj ∂xj ∂xi ∂xj 1 ∂ 2 ui dV 2 ∂x2 j 1 ∂p dV 2µ ∂xi dV V V 1 2µ 1 2µ V ∂ [p(ui − u′ )] dV i ∂xi p(ui − u′ )ni dS i S V (eij − e′ )eij dV = 0. ij (6. However. V Therefore eij = and so u and diﬀer only by a constant. but assume that µ∇2 u′ = ∇p′ .2) in a bounded region V with u prescribed on the boundary surface S.12) V Adding (6. i Hence. p satisfy the Stokes equations (6. (e′ − eij )e′ dV = 0. It is here we use the fact that we are dealing with a Stokes ﬂow. 88 .= − (by continuity) = − (♣) (since µ∇2 u = ∇p) = − (by continuity again) = − (by divergence theorem) = − (since ui = u′ on S) = 0. Proof Let u. p′ satisfy ∇ · u′ = 0. we know that u = u′ on S and so it follows that u = u′ everywhere in V .12). ij ij (e′ − eij )2 dV = 0. Note the crucial step (♣) in the above proof.11) and (6.11) Moreover. ij (6. let u′ . In addition.

p) is Φ = 2µ V eij eij dV = 2µ V e2 dV . the equations of motion are ∇ · u = 0. 89 (6. they are therefore minimum dissipation ﬂows. (u′ . From above. The viscous dissipation for ﬂow (u. Φ′ − Φ = 2µ V (e′ − eij )2 dV ≥ 0. 2. Navier-Stokes ﬂows always dissipate more energy than Stokes ﬂows.e. p′ ) is deﬁned as Φ′ = 2µ V e′ e′ dV . Since plane (PPF) and circular Poiseuille ﬂow (CPF) happen to satisfy the Stokes equations.13) . ij since µ∇2 u′ = ∇p′ . p) does not represent a Stokes ﬂow. Including a body force F. ij QED ⋄ Notes 1. ij Hence. ij ij We may re-write this in the following way. (eij − e′ )eij dV = 0. we know that. Φ′ = 2µ V (e′ − eij )(e′ − eij ) + 2(e′ − eij )eij + eij eij dV ij ij ij (e′ − eij )2 + e2 dV ij ij (since Clearly V (eij − e′ )eij dV = 0) = 2µ ij V Φ′ ≥ 0.i. 0 = −∇p + ρF + µ∇2 u. V but now V (e′ − eij )e′ dV = 0 ij ij Recall that the viscous dissipation for the ﬂow (u′ . Reversibility of Stokes ﬂow One very striking feature of creeping ﬂows is that they are reversible.

∇ × (∇ × ω) = 0. it can’t. θ. For this problem. r U 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 θ 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 11111111111111 00000000000000 ϕ Fix axes in the sphere and adopt spherical polar coordinates (r. If the boundary conditions involve u we must also switch the sign in these. as the nonlinear term u · ∇u scuppers this sign-reversal process. It if did. Puzzle: There is an interesting consequence of this feature of Stokes ﬂows. φ). A solid sphere moves slowly through a viscous ﬂuid parallel to a plane wall. The ﬂuid velocity ˆ ˆ u = u ˆ + v θ + w φ. ∂φ and also w ≡ 0. r sin θ ∂r . Consider ﬂuid ﬂowing at low Reynolds number past a solid sphere of radius a. p to −p and F to −F. It follows that if we reverse the velocity on the boundary of a Stokes ﬂow. But the symmetry of the problem makes this ridiculous. Therefore the force acting on the sphere perpendicular to the wall must be zero. ∇ · u = 0. we also reverse the ﬂow velocity throughout the ﬂuid. The oncoming ﬂuid moves at speed U and the entire ﬂow is taken to be steady. Does the sphere feel any force repelling it away from or attracting it towards the wall? Answer: No. r 2 sin θ ∂θ 90 v=− 1 ∂ψ . when we reversed the ﬂow it would naturally feel the same force in the opposite direction. we can assume that there are no changes in φ. we get another equally valid solution. ∂ ≡ 0. We can therefore introduce a stream function ψ (see earlier section on these). which is ﬁxed inside an unbounded ﬂuid. Creeping ﬂow past a solid sphere This is the classical Stokes ﬂow problem. deﬁned by u= 1 ∂ψ .so if we change u to −u. it is convenient to work with the Stokes equations in the form since ∇2 ω = ∇ × (∇ × ω). r Exploiting the axisymmetry of the sphere. This property is not valid for ﬂows at non-zero Reynolds number.

the latter requires u → U cos θ. D2 ψ . u → free stream velocity as r → ∞. + 2 2 ∂r r ∂θ sin θ ∂θ ˆ φ. ∇×ω = 1 ∂ 1 ∂ ˆ (ωφ sin θ) ˆ − r (rωφ ) θ r sin θ ∂θ r ∂r ∂ 1 1 ∂ ˆ = − 2 (D 2 ψ) ˆ + r (D 2 ψ) θ. where ωφ = − Now. 1 1 r sin θ i. r ∇×F = So the vorticity. r 2 sin θ ∂θ r sin θ ∂r So. r sin θ ∂θ r sin θ ∂r − 1 ∂ 1 ∂ψ 1 ∂ψ 1 ∂ − 2 sin θ ∂θ r ∂r sin θ ∂r r ∂θ r ˆ ω = ωφ φ. r sin θ D2 ψ ≡ 1 ∂ψ ∂ 2 ψ sin θ ∂ .ˆ ˆ In spherical polars.e. straightaway we can write down ∇ × (∇ × ω) = ∇ × = − Hence ∇ × (∇ × ω) = 0 =⇒ D 2 (D2 ψ) = 0 =⇒ D 4 ψ = 0. But this has the same form as u= 1 ∂ψ 1 ∂ψ ˆ ˆ− r θ. v → −U sin θ 91 as r → ∞. for vector function F = Fr ˆ + Fθ θ + Fφ φ. 1 ∂ 1 ∂ 1 ∂Fr 1 ∂ ˆ (Fφ sin θ) ˆ − r (rFφ ) θ + (rFθ ) − r sin θ ∂θ r ∂r r ∂r r ∂θ ˆ φ. r 2 sin θ ∂ 1 ∂ ˆ (−D 2 ψ) ˆ − r (−D 2 ψ) θ ∂θ r sin θ ∂r ˆ D2 (−D 2 ψ) φ. ω =∇×u= or. . The boundary conditions are u=0 on r = a..

the boundary conditions become f = f ′ = 0 on In this case. 2. D2 ψ = = ∂2 1 ∂ sin θ ∂ + 2 2 ∂r r ∂θ sin θ ∂θ f ′′ f − 2 2 sin2 θ r f (r) sin2 θ r = a.These mean that ψ→ and 1 2 2 U r sin θ 2 as r → ∞. D4 ψ = = = 0. In this case. where H(r) = f ′′ − 2 f . r2 H ′′ − 2 H sin2 θ r2 2 2 d2 f sin2 θ. 2 − r2 dr 2 d2 2 − r2 dr i.14) ∂ψ ∂ψ = =0 ∂r ∂θ on These boundary conditions suggest that we seek a solution of the form ψ = f (r) sin2 θ. r 92 =⇒ (n2 − 5n + 4)(n2 − n − 2) = 0 . = H(r) sin2 θ. We try solutions of the form f = r n . 2 rn = 0 d2 2 r n−2 [n(n − 1) − 2] = 0 2 − r2 dr so r n−4 [(n − 2)(n − 3) − 2][n2 − n − 2] = 0 so (n − 4)(n − 2)(n − 1)(n + 1) = 0. (6.e. Therefore n = 1. 4 or −1. So. Thus. Hence f= A + Br 2 + Cr + Dr 4 . r = a.

The inﬁnity condition (6.for constants A. a 2 A = 0 =⇒ − 2 + U a + C = 0. r=a = 0 =⇒ r=a A U a2 + + Ca = 0. 2r Now we know ψ. 4 giving A= Thus. ψ = U r 2 sin2 θ 1 − 2 2r 2r ψ= 1 2 2 a U r sin θ 1 − 2 r 2 or 1+ a . r 2 sin θ ∂θ 2r 2r 1 ∂ψ 3a a3 = −U sin θ 1 − − 3 . we make use of the alternative form of the momentum equation ∇p = −µ∇ × ω. 2r 3a a3 1 ∂ψ = U cos θ 1 − + 3 .14) implies that B= while ∂ψ ∂θ ∂ψ ∂r U . D. 1 ∂ 1 ∂ ˆ (ωφ sin θ) ˆ − r (rωφ ) θ . a U a3 . 2 D = 0. 4 3 C = − U a. 3a a3 1 + 3 . u = f sin2 θ r2 (= −D 2 ψ) 3 Ua sin2 θ 2 r 3U a = − 2 sin θ. = −µ r sin θ ∂θ r ∂r 93 . B. we can compute the velocity and vorticity components: r sin θ ωφ = − f ′′ − = − =⇒ ωφ Also. C. r sin θ ∂r 4r 4r v = − To calculate the ﬂuid pressure.

94 . its value at inﬁnity. p=− p=− Notes: 1. If we impose a velocity −U on the solution. Exercise: Use Maple to plot the streamlines around the sphere. The solution for ψ is symmetric about θ = π/2. 1 1 ∂p µ r ∂θ 1 ∂ (rωφ ) r ∂r 3U a sin θ = r3 3U a = − 2 cos θ + f (r). 2r = = − 3U a 1 − 2 sin θ cos θ r sin θ r 3U a = − 2 cos θ + g(θ). 2r =⇒ Combining these. 3µU a cos θ + const. arbitrary g. 2r 2 2. 2r 2 Letting r → ∞. p µ arbitrary f .So. 1 ∂p µ ∂r p =⇒ µ and. we can obtain results for a sphere moving through a ﬂuid at rest. So. 3U a cos θ + p∞ . Therefore the streamlines are symmetric fore and aft the sphere. p → p∞ .

0]i . Then θ r x θ we get ˆ (F · x) = Fi xi = σij nj xi ˆ ˆ = −p cos θ − µ ωφ sin θ 3U a = −(p∞ − cos θ) cos θ − µ ωφ sin θ. the drag force will only have a component in the direction of motion. 0. 0]i xi ˆ ˆ . Here the subscript i means the ith Cartesian component. Recalling the deﬁnition of the stress tensor. 4 3r Exercise: Use Maple to plot the streamlines for this ﬂow. σij = −pδij + 2µeij . By symmetry. we need to take the dot product of F with i. 2r 2 95 = −p ni xi − µ [0. To get the desired component. −ωφ . 0) in a Cartesian set of axes. 0] × [0. Denote i as x = (1. Drag on a moving sphere We can use the previous calculation to compute the drag on a solid sphere moving through a viscous ﬂuid at low Reynolds number. Fi = σij nj = −pni + ∂uj ∂ui + nj ∂xj ∂xi = −pni − µ(n × ω)i (see Problem Sheet 2). 0. 0. But n = ˆ and so r Fi = σij nj = −pni − µ([1. the unit vector ˆ pointing along the line of motion. −ωφ .1 So Ψ = ψ − U r 2 sin2 θ 2 a2 3 = − U ar sin2 θ 1 − 2 . ωφ ])i = −pni − µ[0.

we examine the order of magnitude of the terms. r 2 sin θ ∂θ 2r 2r 3a a3 1 ∂ψ = −U sin θ 1 − − 3 .15) since the surface element dS = a2 sin θdθdφ in spherical polar coordinates.Thus. v = − and so. the total force in the x direction is: 3µU . ∂2u Ua ∼O ν 3 2 ∂r r 96 . we can deﬁne a dimensionless drag coeﬃcient CD = Ftot 1 2 2 2 ρU πa = 12 . ˆ F · x|r=a = −p∞ cos θ + ˆ Therefore. Consistency analysis Is the above solution for the sphere everywhere a consistent approximation to the correct solution of the full Navier-Stokes equations when R ≪ 1? In other words. We have u = 1 ∂ψ 3a a3 = U cos θ 1 − + 3 . Ftot = 6µπaU. R 1 The agreement with experiment is good for R < 1 and very good for R < 2 . on r = a. and so the total drag on the sphere. r sin θ ∂r 4r 4r u ν ∂u U 2a ∼O ∂r r2 for large r. So the total x force is π 3µU a2 sin θ cos θ dθ − 2π p∞ 4πa2 2a 0 The integral in the second term is zero. Taking the Reynolds number to be R = U a/ν. are the approximations leading to the Stokes equations valid for all values of r? To answer this question. for large r. 2a sphere 3µU − p∞ cos θ dS 2a 2π 0 0 π = 4πa2 3µU − p∞ 2a a2 sin θ cos θ dθ dφ (6.

The outer ﬂuid cannot mix with that in the drop. Outer fluid r U θ Drop z The axes are ﬁxed in the moving drop. u · ∇u. we can still ﬁnd r suﬃciently large such that the terms neglected are as important. where the Stokes solution doesn’t work at all (see Problem Sheet 4)! The diﬃculty in the far-ﬁeld was partially tied up by Oseen in 1910. or more important. from the Navier-Stokes equations. 97 . For this problem we have to consider the ﬂows inside and outside of the drop. So. although the Stokes solution is consistent near the sphere for 0 < R ≪ 1. no matter how small. So. than the retained viscous terms. we assume that the conditions of Stokes ﬂow are satisﬁed. it breaks down as a valid approximation when a . ∇ × (∇ × ω) = 0. the equations of motion are ∇ · u = 0. so the outer ﬂuid appears to be ﬂowing towards it at speed U . terms neglected terms retained U 2 a/r 2 νU a/r 3 Ur = O ν Rr . and later more fully by Proudman & Pearson in 1957..Recall that we neglected the nonlinear terms.e. R. but at least it’s not as bad as the equivalent ﬂow past a circular cylinder. = O a ∼ O i. In both cases. for any ﬁxed Reynolds number. r=O R This is a major problem. Thus. Motion of a liquid drop Consider a spherical liquid drop of radius a and viscosity µd moving at speed U through another ﬂuid of viscosity µo at low Reynolds number.

r sin θ ∂r Outside the drop Label the stream function outside the drop ψo . r Following the results for the solid sphere. C. D4 ψ = d2 2 2 − r2 dr 2 f sin2 θ = 0.14) implies that B=− U . we assume the ﬂows are axisymmetric. ∂ ≡ 0. Thus. we try solutions of the form f = r n . + 2 2 ∂r r ∂θ sin θ ∂θ 1 r 2 sin θ ∂ψ . 2 98 D = 0. from before. deﬁned by u= for velocity ﬁeld ˆ u = u ˆ + v θ. We can therefore introduce a stream function ψ. B. ∂φ and also w ≡ 0. Then. 1 ψo → − U r 2 sin2 θ as r → ∞ (cf. ∂θ v=− 1 ∂ψ . The inﬁnity condition (6. A + Br 2 + Cr + Dr 4 .As for the solid sphere calculation above. where D 2 ψ ≡ 1 ∂ψ ∂ 2 ψ sin θ ∂ . 2 This suggests we look for a solution of the form ψo = f (r) sin2 θ. solid sphere). r . ﬁnding that f= for constants A. At inﬁnity the ﬂow consists of a uniform stream moving at speed U in the −z direction. the equation of motion reduces to D 4 ψ = 0. D. As above.

and so f=

A 1 2 − U r + Cr. r 2

Before we can ﬁnd A and C we need to consider the motion within the drop. Inside the drop Label the stream function inside the drop ψd . Since the two ﬂuids are immiscible, there must be no ﬂow across the ﬂuid-drop interface. Thus we require the normal velocities ud = uo = 0 on r = a =⇒ ψd = ψo = 0 on r = a. (6.16)

Also we will need the tangential velocities to be continous at the interface: vd = vo = 0 on r = a =⇒ ∂ψd ∂ψo = ∂r ∂r on r = a. (6.17)

In addition, we need the tangential stress at the surface to be continuous (see previous discussion on ﬂuid-ﬂuid boundary conditions). Thus we want 2µd ed = 2µo eo rθ rθ where κ = µd /µo since erθ = is the viscosity ratio, on r = a =⇒ κ ∂ 1 ∂ψd ∂r r 2 ∂r = ∂ 1 ∂ψo , ∂r r 2 ∂r (6.18)

r ∂ v 1 ∂u + 2 ∂r r 2r ∂θ in spherical polars, and u is everywhere zero on the drop’s surface. The form of these conditions at the interface suggest that inside the drop we also look for a solution of the form ψd = f (r) sin2 θ. Thus, f= for constants E, F, G, H. Now, using the deﬁnition of ψ, we have u= 2 cos θf (r) , r2 1 v = − sin θf ′ (r). r E + F r 2 + Gr + Hr 4 , r

Therefore, to avoid a singular velocity at r = 0, we need E = G = 0. 99

Condition (6.16) implies that F a2 + Ha4 = while condition (6.17) demands
′ ′ fd (a) = fo (a) =⇒ 2aF + 4a3 H = −

A 1 2 − U a + Ca = 0, a 2

A − U a + C, a2

and condition (6.18) requires κ which implies κ − 1 ′ f r2 d

=

1 ′ f r2 o

on

r=a

2F 4A U 2C + 4H = 5 + 2 − 3 . 2 a a a a

Solving for the various constants we ﬁnd A=− Thus, we have ψo = − ψd = U r 2 sin2 θ a(2 + 3κ) a3 κ 1− + 3 2 2r(1 + κ) 2r (1 + κ) outside the drop, in the drop. a3 U κ , 4(1 + κ) C= aU (2 + 3κ) , 4(1 + κ) F = U , 4(1 + κ) H=− U . 4a2 (1 + κ)

U r 2 sin2 θ r2 1− 2 4(1 + κ) a

The ﬂow inside the drop is called Hill’s spherical vortex. Following a similar procedure to that for the rigid sphere, we can compute the pressure within each ﬂuid and evaluate the drag on the drop as it moves with the ﬂuid. Doing this calculation, we ﬁnd that the drag D = 4πµU a 1 + 3κ 2 . 1+κ

Exercise: Verify the above formula for the drag on the moving drop. As the ﬂuid in the drop becomes more and more viscous, we would expect this result to approach that for the sphere. Indeed, taking the limit κ → ∞, we obtain
κ→∞

lim Ddrop = 6πµU a = Dsphere.

100

Low Reynolds number swimming The reversibility of Stokes ﬂows has an interesting consequence with regard to the swimming of very small organisms, such as a spermatozoa. Since the lengthscale of such organisms is very small, the Reynolds number is likely to be very small and the conditions of Stokes ﬂow apply.

If this idealised little spermatozoa swishes its tail up and down as shown, since the ﬂow around it is reversible, it will never get anywhere! We will look at an idealised model of a swimming motion which overcomes the reversibility problem.

Motion of a thin ﬂexible sheet
y1
0.8 0.6 0.4 0.2 0 -0.2 -0.4

λ=2π/k

(x s ,ys )
sheet

-0.6 -0.8 -1 0 2 4 6

x

8

10

12

A thin ﬂexible sheet ‘swims’ in a viscous ﬂuid under conditions of Stokes ﬂow. A point on the sheet (xs , ys ) moves according to xs = x, ys = a sin(kx − ωt),

i.e. the wave moves from left to right with speed c = ω/k. The wavelength λ = 2π/k. Assume the ﬂuid above the sheet (y ≥ ys ) is in two-dimensional motion, satisfying the equation ∇4 ψ = 0, 101

ys ) = (0. 0) + · · · = 0. where us = (xs . ψ = kψ ∗ /(aω). From now on we will assume that ε = ka ≪ 1. t = t∗ /ω. 0) + · · · = cos X. v = −ψx . −ωa cos(kx − ωt) ). ∂y ∂y ∂y and ∂ψ ∂ψ ∂2ψ (x. we require u to be bounded. Then ∇4 ψ = 0 becomes and the boundary conditions. ε sin X) = (x. The boundary condition at the sheet is that u = us . Dropping all the asterisks for convenience. ε sin X) = 0. ∂x ∂x ∂x∂y In view of these boundary conditions. ∂y ∂x Expanding in Taylor series. 0) + ε sin X (x. (x. y = y ∗ /k. 102 . ˙ ˙ As y → ∞. k3 a ω ∂2 ∂2 + ∗2 ∂x∗2 ∂y 2 ψ ∗ = 0. ε sin X) = (x. ∂y ∗ −v ∗ = ∂ψ ∗ = cos(x∗ − t∗ ) on ∂x∗ y ∗ = ka sin(x∗ − t∗ ). ∂ψ ∂2ψ ∂ψ (x. 0) + ε sin X 2 (x. ε sin X) = cos X. Non-dimensionalize by writing x = x∗ /k. and setting X = (x−t). we seek a solution in the form of the expansion ψ = ψ0 + εψ1 + ε2 ψ2 + · · · . the boundary conditions are ∂ψ ∂ψ (x.where u = ψy . and so we obtain ∇4 ψ0 + ε ∇4 ψ1 + · · · + εn ∇4 ψn + · · · = 0. u∗ = ∂ψ ∗ = 0.

4. At the next ‘order’ we pick out all terms of size O(ε1 ). We can continue in this vein for all orders O(εn ). = [f0 103 .The boundary conditions become ∂ ψ0 + εψ1 + · · · ∂y and ∂ ψ0 + εψ1 + · · · ∂x + ε sin X ∂2 ψ0 + εψ1 + · · · ∂y 2 + · · · = 0. + ε sin X ∂2 ψ0 + εψ1 + · · · ∂x∂y + · · · = cos X. ∂x ∂ψ0 =0 ∂y on y = 0. Then. with conditions ∂ψ1 ∂ 2 ψ0 + sin X = 0. namely. n = 2. 3. since ∂ ∂X ∂ ∂ = = . ∂x ∂x ∂X ∂X ′′ ∇2 ψ0 = (f0 − f0 ) sin X = F (y) sin X. . ∇4 ψ0 = ∇2 (∇2 ψ0 ) = ∇2 (F (y) sin X) = (F ′′ − F ) sin X. we get: So. . say. ∇4 ψ1 = 0. as far as we like. (iv) ′′ − 2f0 + f0 ] sin X. . We look for a solution to the O(ε0 ) problem by trying ψ0 = f0 (y) sin X. = ′′ ′′ (f0 − f0 )′′ − (f0 − f0 ) sin X. with conditions ∂ψ0 = cos X. Taking the limit ε → 0 yields all the terms in the equations of size O(ε0 ): ∇4 ψ0 = 0. ∂x ∂x∂y ∂ψ1 ∂ 2 ψ0 + sin X =0 ∂y ∂y 2 on y = 0.

∂x And. Accordingly. on y = 0.(− sin X) = sin2 X = (1 − cos 2X). The boundary conditions suggest that we look for a solution of the form ψ1 = f11 (y) + f12 (y) cos 2X. We need A0 = B0 = 0 to satisfy the boundedness as y → ∞ condition. ∂ψ0 = 0 =⇒ −C0 + D0 = 0 =⇒ D0 = 1. But. again on y = 0. ∂y Therefore the solution is ψ0 = (1 + y)e−y sin X. −1. ∂y 2 ∇4 ψ1 = 0. ψ0 = (A0 + B0 y)ey + (C0 + D0 y)e−y sin X. we have f0 (iv) ′′ − 2f0 + f0 = 0. 1. Then ′′ ′′ ∇2 ψ1 = f11 + (f12 − 4f12 ) cos 2X. ∂ψ1 = 0. ∂ψ0 = cos X =⇒ C0 = 1. −1. 104 .So Writing f0 = emy . m4 − 2m2 + 1 = (m2 − 1)2 = 0 =⇒ m = 1. so ψ0 = (C0 + D0 y)e−y sin X. ∂x ∂ψ1 1 = − sin X. The problem for ψ1 now becomes with conditions on y = 0.

2 1 ′ f12 (0) = − . For f12 we shall also require A12 = B12 = 0 in order for the velocities to be bounded at inﬁnity. we need A11 = B11 = 0. and so f12 = (A12 + B12 y) e2y + (C12 + D12 y) e−2y . Trying f12 = emy . f11 = A11 y 3 + B11 y 2 + C11 y + D11 . we obtain the auxiliary equation (m2 − 4)2 = 0 =⇒ m = 2. −2. (iv) on y = 0 =⇒ f12 (0) = 0.So. Now. 2 2 Therefore ψ1 = f11 (y) + f12 (y) cos 2X = 1 1 y + D11 − y e−2y cos 2X. ′′ ′′ ∇4 ψ1 = f11 + (f12 − 4f12 ) cos 2X − 4(f12 − 4f12 ) cos 2X ′′ = f11 + (f12 − 8f12 − 16f12 ) cos 2X (iv) (iv) (iv) (iv) = 0. because the velocities are bounded as y → ∞. 2 2 105 . ′ y = 0 =⇒ f11 (0) = and on 1 . Also. Then f12 (0) = 0 =⇒ C12 = 0. ′ f12 (0) = − 1 1 =⇒ D12 = − . −2. 2. ∂ψ1 =0 ∂x ∂ψ1 1 = (1 − cos 2X) ∂y 2 (iv) ′′ and f12 − 8f12 − 16f12 = 0. 2 Hence C11 = 1/2 and. We therefore require f11 = 0 Thus.

2 Thus. Bacteria.Moﬀatt eddies 106 . 1 ψ1 = y(1 − e−2y cos 2X). 2 where c = ω/k is the wave speed. So. similar ‘non-symmetric’ motions have to be adopted at low Reynolds numbers to avoid the reversibility problem. u = −ωkaye−ky sin(kx − ωt) + εωa 1 + (2ky − 1) e−2ky cos 2[kx − ωt] + · · · . If we take a set of axes moving at this streaming velocity. Flow in corners . sending helical waves down to the tip. 2 1 1 + (y − ) e−2y cos 2X + · · · 2 2 1 1 + (y − ) e−2y cos 2[x − t] + · · · . the sheet ‘swims’ from right to left. which often have several tails (or ﬂagellae) adopt similar non-symmetric strategies for swimming. 2 Hence ψ = ψ0 + εψ1 + · · · 1 = (1 + y)e−y sin X + εy(1 − e−2y cos 2X) + · · · . 2 2 So we see that a steady horizontal component of velocity is generated: u = ω εaω = ε2 2 2k (since ε = ka) 1 = ε2 c. In real life.We can take D11 = 0 without loss of generality (since it is just adding a constant to a streamfunction). This overcomes the problems with reversibility and allows it to move forwards. u = ψy = −ye−y sin X + ε = −ye−y sin(x − t) + ε In dimensional form. A more realistic picture of the spermatozoa has it wiggling its tail in a rotary motion. the horizontal velocity component.

v=− . and so 1 ∂ 1 ∂2 ∂2 + + 2 2 ∂r 2 r ∂r r ∂r We seek a class of solutions of the form ψ = r λ f (θ). For example. where f and λ are to be determined. 107 λ=0 2 ψ = 0. we will need β 2 λ2 f + (λ2 + β 2 )f ′′ + f (iv) = 0. ψ.19). (λ − 2)2 λ2 f + [λ2 + (λ − 2)2 ]f ′′ + f (iv) = 0. Thus. The ﬂuid satisifes no slip and no normal ﬂow at θ = ±α.Now we will look at slow ﬂow of viscous ﬂuid in a sharp corner or crevice of angle 2α. U θ=α r 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 111111111111111111111111111111111111111111111 000000000000000000000000000000000000000000000 θ=−α The ﬂow is two-dimensional and satisﬁes the Stokes equations. we ﬁnd ∇2 ψ = r λ−2 [λ(λ − 1) + λ]f + f ′′ = r λ−2 λ2 f + f ′′ = r β F (θ). and can be thought of as driven by some agency a distance from the corner r = 0. deﬁned by 1 ∂ψ ∂ψ u= .19) . and so ∇4 ψ = ∇2 (∇2 ψ) = r β−2 β 2 F + F ′′ = r β−2 β 2 [λ2 f + f ′′ ] + [λ2 f ′′ + f (iv) ] . uniform ﬂow over a sharp indentation will drive a ﬂow consisting of eddies inside the crevice. to satisfy (6. r ∂θ ∂r ˆ for velocity ﬁeld u = u ˆ + v θ. We use a stream function.19). Substituting into (6. since β = λ − 2. Thus. (6. say. r The equation of motion is ∇4 ψ = 0.

Seeking a solution of the form f ∝ emθ , we ﬁnd m4 + 2(λ2 − 2λ + 2)m2 + λ2 (λ − 2)2 = (m2 + λ2 )(m2 + [λ − 2]2 ) = 0 =⇒ m = λi, −λi, (λ − 2)i, −(λ − 2)i. Hence, assuming λ = 1 and λ = 2, f (θ) = A cos λθ + B sin λθ + C cos(λ − 2)θ + D sin(λ − 2)θ. If we now concentrate on solutions which are symmetric about θ = 0, we can set B = D = 0 to leave f (θ) = A cos λθ + C cos(λ − 2)θ. The boundary conditions are u = v = 0 at θ = ±α, which translate into ψ= and thus f = f ′ = 0 on Hence, A cos λα + C cos(λ − 2)α = 0, Aλ sin λα + C(λ − 2) sin(λ − 2)α = 0, or, cos λα cos(λ − 2)α A C =0 (6.20) θ = ±α. ∂ψ = 0 on ∂θ θ = ±α

λ sin λα (λ − 2) sin(λ − 2)α

So, for non-trivial solutions, we need the determinant of the matrix to vanish: λ tan λα = (λ − 2) tan(λ − 2)α. Exercise: Show that this can be re-written as sin x sin 2α =− . x 2α where x = 2(λ − 1)α. Fixing 2α, we can read B = (2α)−1 sin 2α from the graph (upper dotted line). For the example shown we get B = 0.217. Then we ﬁnd where the solid curve corresponds to −B = −0.217. This is shown as the lower dashed line. 108 (6.21)

1

0.8

0.6

sin(x)/x

0.4

0.2

0

-0.2 0 5 10 x 15 20

If we choose 2α = 2.55, we get B = 0.217, and −0.217 just touches the global minimum of the solid curve, as shown. So if α < 2.55, there will be no real solutions. In degrees, this means that there are no real solutions if 2α < 146.3◦ . Let’s consider the two possibilities in turn: i) 2α > 146.3◦ . In this case, (6.21) has real solutions λ. Thus the θ velocity component v = −λr λ−1 f (θ) is single signed for all r and so the ﬂuid moves in and out of the corner something like this:

ii) 2α < 146.3◦ . Now the solutions λ are complex, so write λ = p + iq. In general A and C will also be complex. Since the equation of motion ∇4 ψ = 0 is linear, using an overbar to denote a complex conjugate, both r λ f (θ) and r λ f (θ),

109

are clearly solutions to the equation of motion ∇4 ψ = 0, and since this equation is linear we can add them together, so ψ= is a solution. Suppose we ﬁx θ = 0, so we are focussing on the centre line. Then f (0) is just some complex number, call it c. Since λ = p + iq, the azimuthal velocity component v|θ=0 = ℜ{−λ c r λ−1 } = ℜ{−(p + i q) c r (p−1)+iq }. Writing r iq = eiq log r = cos(q log r) + i sin(q log r), we have v|θ=0 = a r p−1 cos(q log r + δ), for real constants a, δ, which we can ﬁnd in terms of p, q and c. So the azimuthal velocity can now change sign with r and the corner ﬂow must consist of re-circulating eddies, something like this: (6.22) 1 λ r f (θ) + r λ f (θ) = ℜ{r λ f (θ)} 2

Intriguingly, it is clear from (6.22) that there must be an inﬁnite number of these eddies, as v will repeatedly change sign as r → 0. These eddies have been visualized in experiments, although the ones very close to the corner are far too weak to be observed in practice.

7. Lubrication theory Lubrication theory is the study of thin ﬁlms of ﬂuid. The key feature of such problems is that one dimension is much larger than the other. For example, you can get a coin to “stick” to a surface by putting a small amount of water between the two. A surprisingly large force is required to pull the coin away from the table. Compared to the diameter of the coin, the height of the ﬁlm of water is very small (h ≪ L in the ﬁgure below) and as such is amenable to lubrication analysis. We shall deal with this problem later. 110

and if δ= h L then δ ≪ 1. y) and vertical (z) dimensions respectively. using the fact that we are dealing with a thin ﬁlm of ﬂuid. ∂x2 L with similar estimates for v and its derivatives. we write down the Navier-Stokes equations. u and v must change by an amount of order U as z varies over a distance h. Now. ∂z 2 h U ∂u ∼ . we work on the assumption that if h and L are typical horizontal (x. L These arguments suggest that we can assume |u · ∇u| ≪ |∇2 u| if 111 δ2 UL ≪ 1. ∂z h ∂2u U ∼ 2.F z F 11111111111 00000000000 11111111111 00000000000 h x L To start the analysis. the continuity equation ∇·u= So we can estimate u · ∇u ∼ O and ∇2 u ≈ ∂2u U U U ∼ O 2 i + O 2 j + O δ 2 k. Let U be a typical horizontal ﬂow speed. Speciﬁcally. u · ∇u = −∇p/ρ + ν∇2 u and make some simpliﬁcations. ∇ · u = 0. since u = 0 at the solid walls z = 0 and z = h. ∂x L ∂2u U ∼ 2. L L L2 ∂u ∂v ∂w ∂w U + + = 0 =⇒ ∼O ∂x ∂y ∂z ∂z L =⇒ w ∼ O Uh . Then. ν . ∂z 2 h h h U2 U2 U 2h i+O j+O k. We can estimate sizes of the terms in the equations using these type of arguments. In the ﬁlm we expect U ∂u ∼ .

(Try not to blink for longer than that!) This is also the a subject of current research. ∂y ∂z 0=− ∂p ∂2w +µ 2. Some examples: 1. when part of the vitreous humour in the back of the eye has liqueﬁed. 3. ∂x ∂z 0=− ∂p ∂2v + µ 2. such as that shown above. h and L (and generally speaking we are interested in the case h ≪ L). this means that if Rm ≪ 1. In some situations. Peeling of the retina away from the choroid during retinal detachment. only Rm ≪ 1. we may neglect the nonlinear terms and approximate the thin ﬁlm ﬂow with the lubrication equations: 0=− ∂2u ∂p +µ 2.2) 2. A lubrication model predicts that this ﬁlm will break-apart after approximately 15 seconds. ∂z ∂z (7. ∂x ∂y ∂z Note that we do not require R ≪ 1 for this approximation. Blood ﬂow in slowly-tapering arteries. Flow in a slider-bearing (see below). a lubrication model is appropriate. Applications The applications of the lubrication approximation are manifest. Formal derivation of the lubrication equations y Solid h x L Fluid Rigid wall V U In a typical conﬁguration. Modelling the thin ﬁlm of ﬂuid drawn over the eye during a blink.1) ∂u ∂v ∂w + + = 0. It is therefore sensible to 112 . 4. we have two lengthscales.Deﬁning the Reynolds number R = U L/ν as usual. (7. and introducing the modiﬁed Reynolds number Rm = δ2 R. This is a topic of current research.

and the modiﬁed Reynolds number Rm = δ2 R. assuming ∂2u ∂p ∼ µ 2. ∂x ∂y we estimate p U UL ∼ µ 2 =⇒ p ∼ µ 2 . In other words. A typical scale for the pressure comes from considering an order of magnitude balance between the pressure gradient and the largest occuring term in the momentum equation. p 2 ˆ v = U (h/L)V . 0 = −ˆY p ˆX + VY = 0. p = 0. we expect h U v .. we obtain the a reduced set of equations. ˆ u = UU. we non-dimensionalize using these expressions. p . x = LX. ˆ U 113 ˆ 0 = −ˆX + UY Y . since the layer is thin. i. L h h 2 p suggesting that we write p = µ(U L/h )ˆ. R= UL . y = hY p = µ(U L/h )ˆ. this suggests we take u = U U . ν In the limit δ → 0 and Rm → 0.e. ˆ If the solid wall is moving with horizontal speed U . Substituting into the two-dimensional Navier-Stokes equations we ﬁnd ˆˆ ˆˆ Rm U UX + V UY ˆˆ ˆˆ Rm U V X + V V Y ˆ ˆ UX + VY ˆ ˆ = −ˆX + UY Y + δ2 UXX .non-dimensionalize by setting x = LX and y = hY. p ˆ ˆ = −ˆY + δ2 VY Y + δ4 VXX . For an order of magnitude balance of terms in the continuity equation. Assuming steady ﬂow (so ∂/∂t ≡ 0). =⇒ v ∼ O U ∼ L L h ˆ and so this suggests we write v = U (h/L)V . where δ = h/L.

4). ∂x u dy − U dh . The vertical velocity component.5) ux + vy = 0. given by 0 = −px + µ uyy . dx (7. For the next bit. 2µ h Note that the ﬁrst term in this expression corresponds locally to plane Poiseuille ﬂow while the second represents plane Couette ﬂow. and so we can integrate (7. according to (7.3) (7. We notice that p = p(x). the above analysis shows that we can use an approximate form of the Navier-Stokes equations.4) (7. comes from integrating the continuity equation as follows. with boundary conditions u = v = 0 on y = 0. ∂ ∂x h(x) h(x) f (x. for a lubricating ﬂow. y v = − 0 ∂u dy ∂x (so v = 0 on y = 0 is satisﬁed). ∂x dx Demanding that v = V on y = h we ﬁnd h(x) V thus Meanwhile. (7. 0 = −py u = U. h) .3) twice with respect to y to obtain 1 2 Uy u= y − yh px + .8) . y) dy = 0 0 ∂f dh dy + f (x.6) The approximation is valid in the limit δ → 0 and Rm → 0. (7.7) h(x) 0 −V u dy = 0 = and so ∂ ∂x h(x) 1 2 Uy y − yh px + 2µ h 0 3 Uh h 1 + px − 2µ 6 2 h3 ∂ 1 px − ∂x 2µ 6 114 + U dh . 2 dx h dy. u dy = 0 (7. v = V. the ﬂux h(x) = − = ∂ ∂x 0 ∂u dy. on y = h(x). v.Thus. we will need to use the following result (see Problem Sheet 4).

We usually solve (7. d/dx = αd/dh. p = p2 at x = x2 .9) also applies when U and V are functions of time. we obtain the Reynolds lubrication equation: ∂p 1 dh 1 ∂ h3 =− U + V. For convenience we change variable from x to h and so. provided that ∂2 ∂ ≪ ν 2. if there is a thin ﬁlm of ﬂuid in between the bodies. which we are at liberty to set to zero.Combining (7. The pressure on either side of the bearing is atmospheric. the tangential resistance may be a lot smaller than the normal. A good example of this is a slider bearing moving over a solid wall at y = 0 with ﬂuid in between (see diagram). The height function h is normally given. 2 .8) with (7. However.9) subject to p = p1 at x = x1 . Flow in a slider bearing When two solid bodies slide against each other.9) This equation involves a second order derivative in x and so two boundary conditions are required. Then the Reynolds Lubrication Equation becomes: dp 1 2 d h3 α 12µ dh dh 115 1 = − U α. Note: The lubrication equation (7.7). ∂t ∂y We will make use of this fact later. 12µ ∂x ∂x 2 dx (7. t. pressure force. since h ≈ αx + const. y pa h 1 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 U 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 α pa x h(x) h2 L Fluid Assume that the angle α ≪ 1. the frictional resistance and normal reaction force are comparable in magnitude.

we ﬁnd p= E= Then. σ12 dx = µ LW ∂u ∂v + ∂y ∂x y=0 dx = LW µ ∂u ∂y y=0 dx. h dp U ∂u =− + . F. And so. E. h2 .E 6µU + F − 2. If the system is closed and the forces at both ends are negligible. α (h1 + h2 ) (7. 1 6µU . ∂y 2µ dx h =⇒ D = µ U α h2 h1 6h1 h2 4 − h (h1 + h2 )2 h2 dh. 4 log α h1 h1 + h2 116 . p= 6µU (h2 − h)(h − h1 ) > 0 and h2 α(h1 + h2 ) 12µ U h1 h2 . const αh 2h Choosing E. D=µ 6(h1 − h2 ) h2 U − . α h1 + h2 F =− Integrating. F such that p = 0 at h = h1 . then a force balance shows that the lift and drag forces on the bearing are equal and opposite to those on the wall. as shown: y 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 L 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 D 1111111111111111111 0000000000000000000 1111111111111111111 0000000000000000000 D L x For simplicity. Drag force: D= Now.10) U > 0 for h1 < h < h2 . we therefore compute D and L on the lower wall (LW).

We have already seen how a thin ﬂuid layer can lead to normal forces which are much larger than tangential ones. ∂t ∂y so the ﬂow is steady. h1 h1 + h2 6µU α2 But h2 − h1 is of O(α). Therefore. 117 . =⇒ Adhesive problems We now return to the coin problem mentioned in the introduction to this section. D In other words. D α 1 + O(α) = L 3 L ∼ 1/α and so L ≫ D if α ≪ 1.Now the lift force. L: L = (since ∂v ∂y −σ22 dx = − h2 h1 LW − p + µ{ ∂v ∂v + }y=0 dx ∂y ∂y = − ∂u = 0 on y = 0) = ∂x = p dh α log 2(h2 − h1 ) h2 − . with one of the walls moving away from the other as shown: y h moving wall p=0 −a h(t) a p=0 x Fixed wall L We assume that Fluid ∂2 ∂ ≪ ν 2. An object sitting on top of a thin ﬁlm of ﬂuid on a table can be surprisingly tricky to pick up. Consider two parallel plane walls. the normal force is much larger than the tangential force and the bearing can withstand heavy loads.

We know that a a F = − = −a σ22 |y=0 dx. U ∼ V . We will assume that the horizontal velocity components of the surface are of similar size. resisting the motion. Furthermore. We can thus compute the resistance by calculating minus the force exerted on the lower wall. . a −a (since vy |y=0 = 0 from cty) =− ˙ 8µa3 h . then. A (possibly curved) surface sits on a thin ﬁlm of ﬂuid as shown in the diagram. by analogy with the two-dimensional case. As for the slider bearing. Thus (7. we have. ∼O ∼O ∂x ∂y L ∂z h 118 . dx dx and so. 12µ ∂x ∂x 2 dx together with the boundary conditions u=v=0 on y = 0. ∂ ∂ 1 1 ∂ ≪ . 1 ∂ ∂p 1 dh h3 =− U + V. we expect changes in the z direction to dwarf those in the x and y directions. As usual we assume h ≪ L and the modiﬁed Reynolds number Rm ≪ 1. but this time in three dimensions. h3 = −a p dx ˙ 6µh h3 (x2 − a2 ) dx Note that force is negative so the ﬂuid is trying to pull the moving wall back downwards. on integration. ˙ In this case. The most interesting thing about this result is that F ∼ O(h−3 ) as h → 0.11) becomes dp d ˙ h3 = 12µh. This is closer to the coin problem mentioned in the introduction. ˙ v = h on y = h(t). which implies a huge resistance force from a very thin ﬁlm! An adhesive problem in 3-D Consider a similar problem to the above. h3 We want to calculate the force. F . p= 6µ ˙ 2 h(x − a2 ).e. we assume the forces at the ends are negligible so the force exerted by the ﬂuid on the stationary wall is equal and opposite to that exerted on the moving wall. U = 0. u = 0. V = h and h is independent of x. i.We need to solve the Reynolds lubrication equation.

Restoring the dimensions to the reduced set of equations. which we use to obtain the estimate h w∼O U . 7.11. we non-dimensionalize the equations and take the limits δ → 0 and Rm → 0. (7. with boundary conditions u = v = w = 0 on y = 0. 0 = −px + µ uzz . 0 = −pz 0 = −py + µ vzz .t) W 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 V 1111111111111111111111111111111 0000000000000000000000000000000 U 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 h L x Fluid y It seems sensible to deﬁne u = U u. y. y.y. 2µ h 119 . p Following exactly the same procedure as in the 2-D case. there is no vertical pressure gradient. ∂x ∂y which suggests that p ∼ µU L/h2 . we have. t).11) (7.12) ux + vy + wz = 0. L As before. Note: As for the two-dimensional problem. v = V. ˆ w= Uh w.12) with respect to z twice. w=W on y = h(x. so we can write p = (µU L/h2 )ˆ. t). ˆ L where the vertical velocity scale comes from the continuity equation ux + vy + wz = 0.z z=h(x. Since p = p(x. ˆ v = V v. We ﬁnd u= Uz 1 px z 2 − zh + . the pressure scale is ﬁxed by the expectation that ∂p ∂2u ∼ µ 2. u = U. we can integrate (7.

v= 1 Vz py z 2 − zh + . ˙ ˙ On the surface of the sphere u = v = 0 and w = h. In other words. ∂x 12 ∂x ∂y 12 ∂y Alternatively. we obtain the three-dimensional version of the Reynolds lubrication equation: ∂ 3 ∂p ∂ 3 ∂p h + h = 6µ − U hx − V hy + 2W ∂x ∂x ∂y ∂y We make use of this equation in the following example. it squeezes a thin layer of ﬂuid between itself and the ﬂat surface as shown.14) . we use the continuity equation to ﬁnd the vertical velocity component: −w = Thus. (7. and integrating. 0 −W so −W = 0 ux + vy dz ∂ ∂x h = u dz + 0 ∂ ∂y h 0 v dz − U hx − V hy . Close to the moment of impact. Reynolds lubrication equation states that ∂ h3 ∂p ∂ h3 ∂p + = µW. In this case. 12 120 (7. we may write this as ∇· h3 ∇p = µW. we assume that ∂/∂t ≪ ν∂ 2 /∂z 2 . so U = V = 0 and W = h. Using the above results for u and v. h ux + vy dz.13) Impact of a sphere on a plane A sphere falls slowly through a viscous ﬂuid towards a plane wall. 2µ h z Again. Sphere Fluid h x y z We assume that the sphere falls slowly enough to justify a steady lubrication analysis. at z = h.

dr Now. equation (7. (7. ∇ · (Aˆ) = (1/r)∂(rA)/∂r. For the lubrication approximation to hold. but not large enough to destroy the validity of the lubrication approximation. we have 121 . Thus. we expect that p = p(r). we found above that h = h0 + r 2 /2a. the angle φ ≪ 1 and 1 h = h0 = a − a cos φ ≈ aφ2 . since h is ﬁnite when r → 0.14) becomes r 1 d dp rh3 = 12µW. a approximately. when r is large.15) But we must choose c = 0 else. θ r x y z Recalling that in these coordinates. F . where r 2 = x2 + y 2 .Appealing to the symmetry of the problem. we note that the force in the z direction on the wall. It is therefore more sensible to discuss the problem in cylindrical polar coordinates. This will be true when r is small. ∞ F = 0 2πrp(r) dr = πr 2 p(r) ∞ 0 ∞ − πr 2 0 dp dr. To calculate the force on the sphere. Integrating by parts. Suppose that the sphere’s radius is a. pr → ∞ as r → 0. is equal and opposite to that on the sphere. within the lubrication limit. r dr dr (integrating) rh3 pr = 6µW r 2 + c. we need h ≪ a. So pr = 6µW r/h3 . 2 So h − h0 = r2 2a r = sin φ ≈ φ. ϕ a r ho h(r) Near r = 0.

Since h > 0 for all time. Integrating once. it follows that pr ∼ O We can thus set πr 2 p(r) ∞ 0 r h3 ∼O 1 r5 for large r. dξ = 3 ho (1 + ξ) 2 0 F = πr 2 An interesting consequence of this is that the sphere never reaches the wall! F a mg ho ˙ Taking W = h0 . h0 = H0 at t = 0. Using the result above for the pressure. So it takes an inﬁnite amount of time for h0 to reduce to zero and for the sphere to reach the wall. we ﬁnd ˙ h0 + κ log ˙ ˙ on taking h0 = H0 . the right hand side approaches negative inﬁnity. the equation of motion of the sphere is ¨ mh0 = −6µπa2 This can be written as ˙ h0 − mg. h0 H0 ˙ = −gt + H0 . ˙ As t → ∞. h0 d ¨ h0 + κ (log h0 ) = −g dt for positive constant κ. this must be balanced by h0 → 0 as t → ∞ on the left hand side. 122 . = 0 and deduce that ∞ ∞ dp r3 dr = −6µπW dr dr [h0 + r 2 /2a]3 0 0 ξ 12µπa2 W ∞ dξ (setting r 2 = 2ah0 ξ) = − ho (1 + ξ)3 0 ∞ ξ 6µπa2 W 1 So F = − since .h ∼ O(r 2 ).

ii) the tangential stress must equal zero.Flow of a thin ﬁlm down a sloped wall A thin ﬁlm of viscous ﬂuid spreads along a plane wall which is inclined at an angle α to the horizontal. 7. Since we are implicitly assuming a rather slender ﬁlm. the normal vector points almost vertically. Integrating the second equation. 0 = −pz − ρg cos α. t). we need to add body force terms to account for gravity. As a ﬁrst estimate. we ignore this fact and proceed regardless. Note that the tangential direction is almost in the x direction. a not uncommon feature of lubrication-type analyses. and is taken exactly in the x direction to within a consistent approximation. ux + wz = 0.5). so p = ρg[h − z] cos α + pa . for some function f . (7. i) the pressure must be atmospheric. 7.3. More accurate calculations can be made taking into account the curvature at the nose. t). This is clearly not true near the nose of the ﬁlm where the surface is highly curved. z w u 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111111111111111111 000000000000000000 111 000 111111111111111111 000000000000000000 111 000 111111111111111111 000000000000000000 111 000 111111111111111111 000000000000000000 111 000 111111111111111111 000000000000000000 111 000 x N g α x Starting with the two-dimensional lubrication equations (7.16) where z is now playing the role of the vertical coordinate and w the vertical ﬂuid speed. pa . p = −ρgz cos α + f (x. we take the normal to point exactly vertically6 and require ∂u ∂w =0 + 2µexz = µ ∂z ∂x as the zero tangential stress condition. The results turn out to be surprisingly accurate. as shown in the diagram. We therefore write 0 = −px + µ uzz + ρg sin α. Consistent with the approximations already made in deriving the lubrication equations. The nose of the ﬁlm is at xN (t). 6 123 . At the free surface z = h(x. but these are beyond the scope of the current course.4.

we obtain − g sin α ∂h 2 ∂h gh2 sin α ∂h h = + . 2ν Lastly we need a kinematic condition at the free surface z = h (see the section on boundary conditions earlier in the course). ν g sin α w = − hx z 2 . Applying no slip at z = 0 and the zero tangential stress condition at z = h we therefore ﬁnd g sin α 1 u= (hz − z 2 ). ∂t ν ∂x This equation has a solution ν h= g sin α 124 1 2 1 (7.But in the lubrication approximation. ∂z ∂x and so we need ∂u/∂z = 0 on z = h. This leaves νuzz = −g sin α. (7. w = +u ∂t ∂x on z = h(x. we can ﬁnd the vertical velocity component by integrating the continuity equation: wz = − (using w = 0 at z = 0) =⇒ g sin α hx z. we can drop hx as it is expected to be small. ν 2 As usual. νuzz = −g sin α + ghx cos α. Dt ∂h ∂h i. we need: D (z − h) = 0. (ν = µ/ρ). Again.17) x2 t2 1 . evaluated at z = h. Substituting for the pressure. appealing to the fact that the ﬁlm is thin. we have the evolution equation for h. ∂h g sin α 2 ∂h + h = 0. ∂w ∂u ≫ . 2ν ∂x ∂t 2ν ∂x Rearranging. which is easily integrated.e. t).16) is now. . Substituting for u and w. In fact.

vol. The resulting Marangoni stress will drive a ﬂow in the liquid layer.18) = Therefore. however. Disregarding this issue. dt So the front of the ﬁlm spreads more slowly as time increases.19) t2 g sin α 4ν 1/3 xN (t) = 9A2 So the speed of the nose is given by t1/3 . t) dx = A. that certain ﬂow features such as eddies are not necessarily precluded by the restrictions placed upon a lubrication analysis. Solving the lubrication equations in the liquid layer. dxN ∝ t−2/3 . we permit the surface tension at the air-liquid interface to vary so that γ = cos 2θ. say. ν 2 3 g sin α . 7 See L. we ﬁnd that the ﬂow looks like this7 . 7. (7. consider an annular liquid layer at rest on the inside of a cylindrical tube as shown. It is interesting to note. 173. 557-594. the prediction for the rate of spread agrees well with experiments (see Huppert (1986). For example. This journal is available in the Library). pp. t) dx = 0 1 2 3 2 xN 1 ν g sin α 1 2 x2 t2 1 1 dx (7. In practice such a ﬂuid layer would become longitutdinally unstable along the tube axis due to the Rayleigh instability and break up to form a sequence of liquid collars. we ﬁnd xN (t) xN A = 0 h(x. Note: Despite neglecting eﬀects associated with the curvature around the nose of the spreading ﬁlm. 0 must be constant. Since the total volume of the ﬁlm.5 Limitations of lubrication theory We have seen above that lubrication theory has its limitations and may become invalid in part of the ﬂow ﬁeld. University of Nottingham. Band. Journal of Fluid Mechanics. 2007 125 . xN (t) h(x.which is only expected to be valid at large times. PhD thesis.

u passes through zero. Yet as an eddy turns a corner. L. The presence of the eddies might seem surprising at ﬁrst. Boundary layers The Reynolds number. 126 . How can the ﬂuid turn a corner and not contravence the restrictions of the lubrication analysis? This stipulates that while the streamwise velocity (here the azimuthal component) u is O(1).wall liquid air The thick-lined arrows indicate the direction of the driving Marangoni force γθ = −2 sin 2θ. u is zero and v is comparatively large. the requirements of lubrication theory are still met. Careful thought shows that the restrictions are not in fact contravened. Notice that four viscous eddies are established circling the interior of the ﬁlm. the vertical component (here the radial component) should be O(δ). deﬁned as R = U L/ν. 8. as θ increases. All that is happening is that. for typical velocity and lengthscales U . Provided u still varies slowly with respect to θ.

R (8. it would appear that we can just drop the viscous term ∇2 u/R on the grounds that it is very small.1) tend to inﬁnity.1) When R is very small we have Stokes ﬂow. Experiments of high Reynolds number ﬂows past a circular cylinder. predict a drag on the cylinder which is most certainly not zero.appears in the non-dimensional form of the Navier-Stokes equations: ∇ · u = 0. D’Alembert’s Paradox tells us that the drag on the object is precisely zero! But this is clearly at odds with experience. In dimensional form these are ∇ · u = 0. What happens when R is very large? Letting the Reynolds number in equations (8.2) . The constant a is positive and the parameter ε is assumed to be very small. Suppose we ask the question: what is the drag on an object placed in a high Reynolds number ﬂow? Ui From section ??. u · ∇u = −∇p + 1 2 ∇ u. with boundary conditions f (0) = 0. What is going on? Notion of a boundary layer Consider the following model problem εf ′′ + f ′ = a. ρu · ∇u = −∇p. f (1) = 1. (1 − e−1/ε ) 127 f = f (y). This would leave us with Euler’s equations governing an inviscid ﬂuid. for example. The exact solution satisfying the boundary conditions is f = ay + (1 − a) (1 − e−y/ε ). (8. which we have already discussed.

say. The equation becomes fY Y + fY = εa. which satisﬁes the boundary condition at y = 1. It is noticeable that this function satisﬁes the condition at y = Y = 0 but not that at y = 1. so now fi = (1 − a)(1 − e−Y ). To summarize. fo . to a ﬁrst approximation we may set the right hand side to zero (as we did to obtain f0 (y) above). Suppose we now focus on what happens when y is small. ‘inviscid’ solution. Since ε ≪ 1. however. f ∼ ay + 1 − a = f0 (y). = fi . not small. To this end. y will be small. y→0 We then ﬁnd that A = (1 − a). Crucially. It can be seen to be the solution of equation (8. and an outer solution. we have an inner solution fi . 128 . it satisﬁes the condition at y = 1 but not that at y = 0. say.e. valid in a region away from y = 0. so that if Y is of order one. leaving fY Y + fY = 0. we introduce the change of variables y = εY . which satisﬁes the boundary condition at y = 0.For y = 0(1).2) if we na¨ ıvely set ε to zero on the grounds that it is small. i. We can tie these solutions together by using the following matching procedure Y →∞ for constant A = A(1 − e −y/ε ) lim fi = lim f0 . So it is not the complete solution. This can be thought of as an outer. Integrating this equation yields f = A(1 − e−Y ).

6 0. In this very thin region.The following graphs illustrate these solutions when ε = 0.2 0.4 y 0.6 0. There we may indeed neglect the terms 1 2 ∇ u R 129 .8 0.4 y 0.2 outer inner exact 0 0 0.6 f 0. Something similar is happening with the high speed ﬂow problem discussed above.4 0.2 0. The key feature of the above toy problem is the clear development of a region (here near y = 0) in which there is a rapid change of the solution.6 f 0.01. it is not appropriate to drop the term εf ′′ in the equation. The thin region near y = 0 is an example of a boundary layer. The outer solution has only a slight slope.4 0.8 1 Clearly the inner solution only performs well near y = 0.1 and ε = 0.1 1 0. while the outer works well away from y = 0. near y = 0. However. the inner solution takes over and exhibits a dramatic change over a very short distance.8 0.8 1 Results when eps=0.2 outer inner exact 0 0 0.01 1 0. The agreement between the approximate inner and outer solutions and the exact solution improves as ε → 0. Results for eps=0.

If the plate is ﬁnite. We can therefore estimate ∂ 1 ∼ . a balance of terms in the continuity equation suggests that V δ U ∼ =⇒ V ∼ U. We aim to derive a set of approximate equations valid in the boundary layer.e. L δ L 130 . The continuity equation states ∂u ∂v + = 0.over most of the ﬂow ﬁeld. ν Since u varies from zero at the plate (no-slip) to U in the external stream. Changes in the boundary layer are expected to occur on a lengthscale of typical size L parallel to the plate and δ perpendicular to it (i. we anticipate that the viscous term R−1 ∇2 u will only be important in a very thin boundary layer of height δ(x) close to the plate. Following the above discussion. ∂x ∂y UL ≫ 1. we can say that u ∼ U. typically. However. L can be chosen to be its length. Later we will go through a more formal derivation. The Reynolds number is R= and so is assumed to be very large. there is much more rapid change in the y direction than the x direction). The external stream consists of inviscid ﬂow moving with velocity u = U i. High Reynolds number ﬂow past a ﬂat plate Consider the classical problem of ﬂuid streaming at high speed past a ﬂat plate: y U External stream δ (x) viscous layer x The ﬂow is taken to be steady. ∂x L ∂ 1 ∼ . very close to the surface of the body there is a boundary layer in which this term is no longer negligible. Let L be a characteristic horizontal lengthscale. ∂y δ If V is a typical size for the vertical velocity v. with U constant. We will ﬁrst do this using an argument based on the estimation of the order of magnitude of all the terms.

we can estimate uvy ∼ U and so these two terms balance. Prandtl’s Hypothesis This states that the inertial term. this layer is very thin. ρ 131 V U δ/L U2 =U = ∼ uux .e. δ ν ∼ L UL which can be re-stated as δ ∼ LR−1/2 . Continuing. we can estimate ∂2 1 ∼ 2. .Now consider the x-momentum Navier-Stokes equation 1 uux + vuy = − px + ν[uxx + uyy ]. we estimate that uux ∼ U . νuyy . So viscous eﬀects are only expected to be important in a layer of thickness R−1/2 . should balance the viscous one. we expect uyy to be a lot bigger than uxx . ρ Following the above arguments. we obtain the following estimate of the boundary layer thickness. L δ Simplifying. uux . δ2 U2 . Since R is very large. ∂x2 L ∂2 1 ∼ 2. ∂y 2 δ So. δ δ L 1 2 . L The next step is crucial and is due to Prandtl (1905). inside the boundary layer. U2 U uux ∼ νuyy =⇒ ∼ ν 2. If we choose a scale for the pressure so that 1 px ∼ uux =⇒ p ∼ ρU 2 . In fact uyy ∼ Also. i.

ρ In the y-momentum equation. ρ we have the estimates py ∼ ρ νvxx ∼ ν U2 . LL L δ V δ U 1 δ U2 U2 V 1 U2 =ν = ≪ ρ . Collecting everything together. we have used the fact that R = U L/ν and the Prandtl hypothesis that Rδ2 ∼ 1. The boundary conditions include no-slip at the ﬂat plate. the boxed terms in the x-momentum equation are comparable for large R. Therefore.then. (8. in summary. the x-momentum equation reduces to 1 uux + vuy = − px + νuyy . we must match to the external stream velocity U i. 132 u→U as y → ∞. the y-momentum reduces to just py = 0. retaining the dominant terms. δ uvx ∼ vvy ∼ U δU δU 2 U2 = 2 ≪ρ . u = v = 0 on y = 0. only the boxed term is important uvx + vvy = − 1 py + ν(uxx + uyy ). ρ So. 1 uvx + vvy = − py + ν(vxx + vyy ). we have the Prandtl boundary layer equations 1 uux + vuy = − px + νuyy . ux + vy = 0. ρ py = 0.e. Thus..3) . uux + vuy = − 1 px + ν(uxx + uyy ). L2 L L2 RL L δ δ Rδ2 δ In the above. νvyy ∼ 2 = δU 2 ∼ δU 2 ≪ ρ . The second of these implies that p = p(x). This states that the pressure does not vary across the boundary layer. Also. for this equation. ρ i.

allows us to ﬁnd the pressure gradient term px by relating it to the external stream velocity. Note that we have chosen ρU 2 as a suitable scale for the pressure. v = U v∗ . and we can write the x-momentum boundary layer equation (8. Let’s just consider the x-momentum equation: ∂u 1 ∂p ∂u +v =− + ν∇2 u u ∂x ∂y ρ ∂x So. The above all follows through if U is a function of x. This means that we introduce length. L ∂x∗ L ∂x∗2 ∂y Dividing by U 2 and multiplying by L we ﬁnd u∗ ∂u∗ ∂p∗ ν ∂ 2 u∗ ∂ 2 u∗ ∂u∗ + v∗ ∗ = − ∗ + + ∗2 . velocity and pressure scales with which to deﬁne dimensionless variables. we obtain 1 U Ux + 0 = − px + 0. Hence. namely that py = 0.3). we obtain. ∂x∗ ∂y ∂x U L ∂x∗2 ∂y 133 . px = −ρU Ux . If we let y → ∞ in equation (8. The special feature of the boundary layer ﬂow. Check for yourself that ρU 2 has dimensions of pressure. that is if the ﬂow at inﬁnity (the external stream) is non-uniform. Formal derivation of the boundary layer equations We begin by non-dimensionalizing the Navier-Stokes equations. Here a ∗ denotes a dimensionless variable. we may write x = Lx∗ . ρ The ﬁrst zero follows since uy → ∂U/∂y = 0 and the second for a similar reason. We now enter these deﬁnitions into the Navier-Stokes equations. Proceeding. y = Ly ∗ . p = ρU 2 p∗ .3) as uux + vuy = U Ux + νuyy . since U and L provide typical velocity and length scales for the current problem. since u → U (x) in general.No condition on v can be imposed as y → ∞ since the term vyy has been neglected. ∂u∗ U 2 ∗ ∂u∗ u + v∗ ∗ L ∂x∗ ∂y =− νU ∂ 2 u∗ ∂ 2 u∗ U 2 ∂p∗ + 2 + ∗2 . u = U u∗ .

Y ) + R−1/2 u1 (X. we have the two-dimensional Navier-Stokes equations in dimensionless form: 1 uux + vuy = −px + ∇2 u. Y ) + · · · } (8. In the boundary layer. y on L. and x. v have been scaled on U . 134 . Y ) + · · · ∂u0 ∂u0 + v0 ∂X ∂Y v = R−1/2 {v0 (X. we expect ∂ ∼ R−1/2 ∂y for y ≪ 1. R 1 uvx + vvy = −py + ∇2 v. ∂X ∂Y 2 ∂p0 . Y ) = (x.To save writing. ∂x ∂y ∂x R ∂x2 ∂y UL ν Deﬁne new variables (X. R1/2 y) and let u = u0 (X. Y ) + R−1/2 p1 (X. ∂u ∂p 1 ∂2u ∂2u ∂u +v =− + + 2 . and for the continuity equation. we obtain u0 ∂p0 ∂ 2 u0 + + O R−1/2 ∂X ∂Y 2 ∂p0 0 = − + O R−1/2 ∂Y = − = O R−1/2 ∂u0 ∂v0 + ∂X ∂Y Now let R → ∞ to give u0 ∂u0 ∂u0 + v0 ∂X ∂Y ∂u0 ∂v0 + ∂X ∂Y = − 0 = ∂p0 ∂ 2 u0 + . Y ) + · · · p = p0 (X. R ux + vy = 0.4) Subsituting these into the Navier-Stokes equations. ∂Y = 0. with R = U L/ν and where u. we now drop the asterisks and write u where R= is the Reynolds number. A similar procedure follows for all of the momentum equations. p on ρU 2 . Y ) + R−1/2 v1 (X. Finally.

y measuring distance along and normal to the surface respectively. Restoring the dimensions. so u → U (x) as which leads to −px = U Ux . ρ ∂x ∂y ∂p . 135 . y → ∞. Boundary conditions i) No slip requires u=v=0 or. on y = 0. so ψy → U as y → ∞. ∂y v=− ∂ψ ∂x 1 ψy ψxy − ψx ψyy = − px + νψyyy . where u= We ﬁnd: ∂ψ . y ∼ R−1/2 . ρ py = 0. it may be convenient to write the boundary layer equations in terms of a stream function ψ. we have u ∂u ∂u +v ∂x ∂y = − 1 ∂p ∂2u +ν 2. The equations apply on a curved surface. 2) Outside the layer. Notes 1. equivalently. ∂y 0 = ∂u ∂v + ∂x ∂y = 0. They can be viewed as the leading order asympotic approximation to the Navier-Stokes equations when R → ∞. u matches to the external stream velocity.These are the boundary layer equations in dimensionless form. ψ = ψy = 0 on y = 0. with x. where the external stream velocity U is calculated from inviscid theory. Since the ﬂow is two-dimensional.

permitting information to travel upstream. and diﬀerent sets of equations apply in each layer. The vertical velocity v = ∂ψ/∂x is ﬁxed on solving the equations. Famously. does a boundary layer negotiate a small bump on the wall? Common experience suggests that the streamlines will start to move upwards as they near the bump on the upstream side. the lower deck interacts with the upper deck. then ψ is determined for x > x. This means that if we specify ψ(= f (x). Brieﬂy. Therefore we can only apply 3 boundary conditions on ψ. see Acheson. So. some experimentalists showed that upstream eﬀects may occur when a shock wave impinges on a boundary layer in a supersonic ﬂow. x0 This opens up many interesting questions about boundary layer ﬂows. say) at x = x. for example. 1111111111 0000000000 1111111111 0000000000 But note 3 says that this is not allowed since the bump may not aﬀect the ﬂow upstream. There can be no upstream inﬂuence. The equations are parabolic in x. This means that the ﬂow at location x = x1 has no eﬀect on that at x = x0 if x0 < x1 . Messiter 1969). only the shaded region downstream is aﬀected. How. 1111111111111111111111111111111111111111 0000000000000000000000000000000000000000 Shock wave U Upstream influence Boundary layer 1111111111111111111111111111111111111111 0000000000000000000000000000000000000000 1111111111111111111111111111111111111111 0000000000000000000000000000000000000000 Attempts to understand how such upstream inﬂuence might be possible in a boundary layer led to the development of triple-deck theory (Stewartson. if we introduce a small disturbance (say an imperfection in the wall) at x = x0 . 136 .y x Body 2. The stream function version of the equations are 3rd order in y. the boundary layer gets divided up into three separate superposed decks or layers. For more information. 3. Crucially.

Since ∂U0 /∂x ≡ 0. inside the boundary layer. the oncoming ﬂow is unaware that the plate is of length L. From the above arguments.e.e. we expect δ to depend on ν. T [x] = L then. it must appear as the combination δ∼ which is independent of L. We seek a similarity solution to this problem. at any x station in the boundary layer the ﬂow cannot be aﬀected by what is happening further downstream. To motivate this.5) = 0. We take the case when the external stream. U . δ provides a typical scale for y so y∼δ∼ i. In other words. ∂y 2 (8. at the plate). say. = U0 . ν At a general x station. there is no streamwise pressure gradient and the boundary layer problem to be solved is as follows u ∂u ∂u +v ∂x ∂y ∂u ∂v + ∂x ∂y = ν ∂2u .Blasius’ solution The solution of the boundary layer equations for high speed ﬂow past a ﬂat plate was given by Blasius in 1908. consider brieﬂy the ﬂow when the ﬂat plate is of ﬁnite length. T [U ] = L . L. with conditions for x > 0 (i. U and x (but not L). 137 νx U 1/2 νx U 1/2 . u → U0 as y → ∞. is constant. y ∼ x1/2 . We used a scaling argument to show that the typical boundary layer thickness is δ ∼ LR−1/2 = νL U 1/2 . Since [ν] = L2 . u=v=0 on y = 0. since R = UL . Any solution we try to construct should therefore reﬂect this property. since [δ] = L. Now.

η= Note that η ∂η =− . we seek a solution of the same form. U0 ∂2ψ . y) = (2νU0 x)1/2 ψ(x. For the inﬁnite plate problem.5) becomes ∂ 2 ψ ∂ψ ∂ψ ∂ 2 ψ ∂2ψ ∂2ψ . = ∂y 2 2νx ∂η The x-momentum equation (8.This motivates us to try looking for a solution which is a function of the order one variable η= U νx 1/2 y. u = Uf′ U 2νx 1/2 does not involve L. η). 2νx The factor of 2 is included purely for convenience. as we shall see below. We write the stream function as ψ(x. This agrees with our expectation that the solution should not know where the end of the plate is. ∂x 2x ∂η U0 = ∂y 2νx 1/2 where . This is another example of a similarity solution. +ψ = 2x − ∂η 3 ∂η 2 ∂η ∂η∂x ∂η 2 ∂x 138 . We then write u = U f ′ (η) for the horizontal velocity. ∂η v = − ∂ ∂ ∂ψ (2νU0 x)1/2 ψ =− + ∂x ∂x ∂η η ∂ψ ψ ∂ψ + − = −(2νU0 x)1/2 ∂x 2x 2x ∂η ∂u ∂u ∂η U0 = = ∂y ∂η ∂y 2νx 1/2 Furthermore. Note that. The velocity components are u= ∂ψ U0 ∂ψ = ∂y ∂η 2νx 1/2 = U0 ∂ψ . U0 1/2 y. ∂η 2 ∂3ψ U0 ∂2u U0 3 .

The existence and uniqueness of f was proved by Weyl in 1942. The boundary conditions suggest that we look for a solution which is purely a function of η. for example. but we don’t know f ′′ (0). starting at η = 0 we step the solution forward in η with the aim of satisfying the condition at inﬁnity. Thus. and u → U0 as y → ∞ which becomes ∂ψ →1 ∂η as η → ∞. In this procedure. f is called the Blasius function. step forward from η = 0 and continue stepping until η is large enough for us to have reached ‘inﬁnity’. 3. letting ψ = f (η). we need three values at η = 0 to step forward from. 139 . Numerical calculation of the Blasius function In order to solve the system for f (η). f′ → 1 as η → ∞. We can then keep adjusting our initial guess for f ′′ (0) and shooting forwards until we obtain the desired behaviour for f . 2.The boundary conditions are ψ= ∂ψ = 0 on ∂η η=0 [u = v = 0 on y = 0]. We can do the adjusting of the value of f ′′ (0) using Newton-Raphson iteration.e. In terms of f . in other words far enough for the large η behaviour of f to have become apparent. u = U0 f ′ (η). Shooting and matching Since the equation for f is third order. the equation is discretized to produce a rule to step the solution forward a small distance in η. We know f (0) and f ′ (0). we ﬁnd that f satisﬁes f ′′′ + f f ′′ = 0. i. we integrate the equation forwards in η using a technique called ‘Runge-Kutta’ integration. Thus. We can guess this latter value. η = 0. f′ → 1 as η → ∞. f must be calculated numerically. with conditions f = f ′ = 0 on Notes 1.

f′ → 1 as η → ∞. To see this. So if we choose a = κ−1/2 . then f (η) = κ−1/2 F (κ−1/2 η) satisﬁes f ′′′ + f f ′′ = 0. to the Blasius problem straightaway by taking f (η) = κ−1/2 F (κ−1/2 η) The computed solution to the Blasius problem looks like this: η 1 u/U The correct value of f ′′ at the plate is f ′′ (0) ≈ 0. and we get the solution. which will not in general (unless we’re really lucky) equal one. Furthermore. suppose we integrate the system F ′′′ + F F ′′ = 0. and f (0) = f ′ (0) = 0.A neat trick In fact it turns out that we can get away with only having to do one integration over η. f (0) = f ′ (0) = 0. Then f satisﬁes f ′′′ + f f ′′ = 0. with conditions F = F ′ = 0 and F ′′ (0) = 1 on η = 0. This produces a solution F (η) with the property F ′ → κ as η → ∞. 140 . Now let f (η) = aF (aη) for constant a. f . f ′ (∞) = a2 F ′ (∞) = a2 κ. for some constant κ.4696. So all we need to do is to solve the system for F .

Displacement thickness The displacement thickness provides a measure of the thickness of the boundary layer in terms of the amount by which the oncoming streamlines have been displaced. µ ∂u ∂y y=0 dx = 2µ U0 2ν 1/2 f ′′ (0)L1/2 . the displacement thickness is a measure of the upwards displacement of the streamlines of the ﬂow by the boundary layer. To do this. Physical interpretation As stated above. 141 . U0 where u = U0 f ′ . y u(y) U0 u The deﬁcit in ﬂux from a purely uniform stream is (shaded area) ∞ 0 U0 − u dy = U0 δ1 . we will need to know σ12 |y=0 = µ Then the drag D= 0 L ∂u ∂y y=0 =µ U0 2νx 1/2 f ′′ (0).4696 and so we predict D = 0.664µ U0 ν 1/2 L1/2 . From numerical calculation. It is deﬁned to be δ1 = 0 ∞ 1− u dy.Shear stress on the plate Suppose we wish to calculate the drag in the x direction on a portion of the plate of length L. we have f ′′ (0) = 0. This result agrees well with experimental observations.

Example: For a Boeing 747.1cm. but can we say anything more about the manner in which f ′ approaches this unit value for large η? Since f ′ → 1 as η → ∞.y δ 1 0000000000000 1111111111111 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 1111111111111 0000000000000 U0 u where δ1 is the corresponding ‘thickness’ of the uniform stream needed for the same ﬂux level. where g ≪ 1. For the ﬂat plate. numerical integration yields δ1 = 1. Suppose we now write f → η − η + g(η) as η → ∞. We aim to determine the function g. 142 . The shaded areas in these two ﬁgures are the same. typically δ1 = 0.72 νx U0 1/2 . In addition. Asymptotic form of f We know that f ′ → 1 as η → ∞. U0 for the displacement thickness. U0 U0 It is sometimes possible to calculate both the displacement and momentum thicknesses in closed form. Rearranging. we have δ1 = 0 ∞ 1− u dy. we have f → η − η as η → ∞ for constant η. we may deﬁne the ‘momentum thickness’ ∞ θ1 = 0 u u 1− dy. Substituting into the equation f ′′′ + f f ′′ = 0 we have g′′′ + (η − η)g′′ + gg′′ = 0.

which is assumed to be given. for constant λ. The boundary layer equations to be solved are ψy ψxy − ψx ψyy = U Ux + νψyyy . Hence. constant 2 /2 A. consider a general external stream velocity U (x). g(x) U(x) g(x) This form is a more general version of that presented for the Blasius solution. since g is small.331 . we have 2 /2 . let ψ = g′′ .But. 143 . and g = λη. Clearly both of these must be rejected as they don’t satisfy the boundary conditions. then φ′ + (η − η)φ = 0. g(x) here represents the ‘thickness’ of the layer. f ∼ η − η + Ae−(η−η) . where η = y . (η − η)2 Falkner-Skan solutions The Blasius solution is a special case of a broader class of similarity solutions to the boundary layer equations known as Falkner-Skan solutions. In fact. we would ﬁnd that. more precisely the constant A should be replaced by − 0. By analogy with that solution. Two of the possible solutions are g = const. φ = Ae−(η−η) So. as η → ∞. We need solutions to this equation to satsify g → 0 as η → ∞. we can neglect the nonlinear term and just take g′′′ + (η − η)g′′ = 0. To motivate these. To ﬁnd the third solution. We seek solutions of the boundary layer equations of the form ψ = U (x)g(x)f (η). had we not neglected the nonlinear terms above.

In this case g2 U = ν(2α − β)(x − x0 ). for constant x0 . f will satisfy an ordinary diﬀerential equation provided that α and β are constants. assume that 2α = β. we assume α and β are constant. the x momentum equation becomes f ′′′ + αf f ′′ + β(1 − f ′2 ) = 0. 2U g (2α − β)x x (2α − β) where n is a real number. 2U g C Such forms of U (x) are of no particular physical interest. then (g2 U )x = 0. Consider the expression ν(2α − β) = 2g(U g)x − g2 Ux = (g2 U )x . By a shift of origin we can take x0 = 0. νβ = g2 Ux . β n β g2 Ux = = =⇒ n = . then. g 1 (U g)x f ′ − U gx (ηf ′′ + f ′ ) . Now. Instead. y = 0. u = ψy = g(x)U (x)ηy fη = U f ′ . in which case g2 U = constant = C. g ψyyy = since U ′′′ f . g2 ηx = −ηgx /g. 144 . Then g2 Ux 2α = =⇒ U = Be2αx/C .with ψ = ψy = 0 on Now. U ′′ f . Otherwise the equation depends on x and η. ψy → U (x) as y → ∞. If we ﬁrst suppose that 2α = β. So henceforth. say. −v = ψx = (U g)x f − ψyy = ψxy = ηU ggx f ′ . g With these. for constant B. where να = g(U g)x .

we have β= and the Falkner-Skan solutions are given by U = U0 x L ψ= where f satisﬁes n x L n . 145 . η= 1/2 1/2 U νx 1/2 y. Notes 1. then n = 1 and we have Hiemenz stagnation point ﬂow Here. 2 n+1 (U νx)1/2 f (η). External ﬂows with U ∝ xn arise for inviscid ﬂow over a wedge of half-angle φ = nπ/(n + 1). 3. g= Taking α = 1. . ν(2α − β)x U (x) 1/2 = ν(2α − β)L U0 2n n+1 n+1 2 1/2 x L 1−n 2 . with f (0) = f ′ (0) = 0. f ′ (∞) = 1. If ψ = π/2.Therefore. 2. In the special case n = 0 we recover the Blasius solution discussed above. Also. with f (0) = f ′ (0) = 0. f ′′′ + f f ′′ + β(1 − f ′2 ) = 0. x n x U 2φ f ′ (∞) = 1. U = U0 for constant U0 . f ′′′ + f f ′′ + β(1 − f ′2 ) = 0. L.

Here the ﬂow encounters the Goldstein singularity and the boundary layer solution cannot be continued for x > xs . If U Ux < 0 so that px > 0 and the pressure gradient is adverse (so that the pressure increases as you move downstream). xs The boundary layer equations break down at the separation point xs . For some values of n there is more than one solution. Flow in a viscous jet Consider the motion of a high speed jet of viscous ﬂuid. We assume that the jet is very thin.x stagnation point In this case g(x) is a constant. 11 00 y 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 x The jet shoots out of a narrow slot in a wall into another ﬂuid which is at rest. eventually the boundary layer will tend to separate from the wall. 4. The pressure at inﬁnity in the stationary ﬂuid is asummed to be constant. and so the boundary layer at the wall has constant thickness. 5. so that the variation of velocity across the jet is 146 . This solution was discussed before. equal to p0 . when we noted that it formed an exact solution of the Navier-Stokes equations.

ux = xα+β−1 {(α + β)f ′ + βηf ′′ }. we seek a similarity solution of the jet equations in the form ψ = xα f (η). uyy = xα+3β f ′′′ . Note: there is no pressure gradient in the momentum equation. −∞ . Introducing a 2-D stream function. Simplifying.xα+β−1 [(α + β)f ′ + βηf ′′ ] − xα−1 [αf + βηf ′ ]xα+2β f ′′ = νxα+3β f ′′′ . we need u → 0 as we leave the jet. where η = xβ y. Thus ∂u ∂u ≫ . Furthermore. Since the surrounding ﬂuid is at rest. Thus.7) (8. (8. u = ψy = xα+β f ′ .6) −v = ψx = xα−1 {αf + βηf ′ }. xα−β−1 [(α + β)f ′2 − αf f ′′ ] = νf ′′′ .extremely rapid. The jet is assumed symmetric about y = 0. ρ ux + vy = 0. So we need α = β + 1. we integrate the momentum equation across the jet as follows. the equations and boundary conditions are: 1 uux + vuy = − px + νuyy . as usual. Thus. ∞ ∞ ∞ uux dy + −∞ −∞ vuy dy = ν uy 147 . the momentum equation becomes xα+β f ′ . with uy = 0 and v = 0 y → ±∞ on y=0 [symmetry of jet] u → 0 as [match to ﬂuid at rest]. we assume that so that the boundary layer equations are appropriate inside the jet. To obtain another relation between α and β. Thus. ∂y ∂x |v| ≪ |u|. ψ. uy = xα+2β f ′′ .

8) and thus we require 2α + β = 0. dη = ∂y we ﬁnd ∞ x2α+β −∞ f ′2 dη = M. 148 . 3νf ′′ + f f ′ = A. ∞ −∞ ∞ vuy dy = [vu]∞ − −∞ ∞ uvy dy −∞ (since u(±∞) = 0) (using cty) Using this result. the right hand side vanishes. Combining this with (8. ∂η dy. Now. The momentum equation (8. we have α = 1/3. integrating by parts. we have ∞ = − = uvy dy −∞ ∞ uux dy.6) thereby reduces to 3νf ′′′ + f f ′′ + f ′2 = 0. β = −2/3. (8. −∞ and thus. −∞ 2 −∞ uux dy = d dx ∞ u2 dy = 0. Applying the condition u → 0 as y → ±∞. we ﬁnd A = 0.7).Applying the boundary condition that u → 0 as y → ±∞. Inserting the form of the similarity solution and noting that. −∞ for constant M . since the integration is performed at a ﬁxed x value. for constant A. This can be integrated to yield. ∞ u2 dy = M.

9 ν f ′2 dη = −∞ Thus. B2 1 1 + B+f B−f 3ν B+f . 6ν Bη B2 . B= Finally. we can relate B to M . and the vertical velocity 2 1 v = −x−2/3 { f − ηf ′ } 3 3 B −2/3 By By By = − x − sech2 tanh 3 6νx2/3 3νx2/3 6νx2/3 149 . But f (0) = 0 =⇒ C = 0.8). and so f (η) = B tanh Therefore f ′ (η) = and so ∞ Bη . 6ν The left hand side equals 3ν B We know that tanh−1 x = Hence. we have df = − f2 df = dη. Rearranging. 2 2 for constant B.Integrating again. 1 B2 3νf ′ + f 2 = . ln B B−f 1 1+x . tanh−1 B B for constant C. ln 2 1−x 6ν f = η + C. applying (8. the horizontal velocity u= By B2 sech2 6νx1/3 6νx2/3 9M ν 2 1 3 . sech2 6ν 6ν 2 B3 .

1 u 0. 9.Horizontal velocity profile 0. where the velocity components are signiﬁcant in magnitude. that is.05 0 -40 -20 0 y 20 40 We observe that at a ﬁxed x station. Thus. x B Thus the jet thickness grows like x2/3 as we move downstream. in general. We noted that at low Reynolds numbers the ﬂow is always unique. ﬂuid is being entrained into the edges of the jet: 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 The ‘width’ of the jet corresponds to where the argument of sech2 is O(1).2 0. there may be more than one solution to the Navier-Stokes equations at a any given ﬁnite Reynolds 150 . as y → ∞. However. where y∼O 6ν 2/3 . while v→− B −2/3 x . 3 So. Hydrodynamic stability We have calculated quite a wide variety of solutions to the Navier-Stokes equations.15 0. u → 0.

U0 Rc No solns 2 solns R As R increases from zero. we have: U a = R − Rc . . it is instructive to draw an analogy with the solution of some simple algebraic equations. At this point. (ii) Now consider the following cubic equation for u: au − u3 = 0. A ﬂuid ﬂow may undergo similar bifurcations as the Reynolds number increases. u = u0 • R > Rc : Two solutions. the number of possible solutions jumps from none to two.number. Model problems (i) Consider the following simple quadratic equation to be solved for u: a − (u − u0 )2 = 0. On this point. There are three possible cases: • R < Rc : No solutions for u • R = Rc : One solution. The solution is u = u0 ± (R − Rc )1/2 . 151 a = R − Rc . u = u0 ± (R − Rc )1/2 If we plot u versus R. U is said to go through a saddle-node bifurcation as R passes through the critical value Rc .

The solution is u = 0. we have: U o Rc 1 soln 3 solns R As R increases from zero. written here in dimensionless form: 1 ∇ · u = 0. If we plot u versus R. At this point. A ﬂuid ﬂow may undergo similar bifurcations as the Reynolds number increases.1) R 152 . or if it changes dramatically into a diﬀerent kind of ﬂow altogether. we take the ﬂow to be governed by the Navier-Stokes equations. Physically. the number of possible solutions jumps from one to three. As usual. Stability theory The basic idea is to take a solution of the equations and perturb it a little bit. This means that we will only consider inﬁnitesimal disturbances. We will only consider linear stability. (9. • R ≤ Rc : The unique solution is u = 0 • R > Rc : There are three solutions: u = 0 and u = ±(R − Rc )1/2 . we can think of adding a small disturbance into a ﬂow and observing whether the ﬂow settles back down to its original state. In the event of multiple solutions to the Navier-Stokes equations. U is said to go through a pitchfork bifurcation as R passes through the critical value Rc . ut + u · ∇u = −∇p + ∇2 u. we need to assess a solution’s stability. the question arises: Which solution would you observe in a laboratory experiment? To answer this question. We call the ﬂow whose stability is to be assessed the basic ﬂow or basic state. There are two cases to consider: u = ±(R − Rc )1/2 .

writing the new. we obtain ˆ ∇ · U + ε∇ · u = 0. y. To this ﬂow we add a small time-dependent disturbance. p Substituting this. which satisﬁes the equations ∇ · U = 0. perturbed ﬂow as u = U + εˆ .3) 1 ˆ ˆ εut + U · ∇U + εˆ · ∇U + εU · ∇ˆ + ε2 u · ∇ˆ = −ε∇ˆ + ε ∇2 u. writing p = PB + εˆ. from (9. We perturb the pressure ﬁeld in a similar way. together with (9.2) p = pB . 0 = −∇pB + 1 2 ∇ U R (9. t)j. t)i + v (x. ˆ u p εut + U · ∇U + εˆ · ∇U + εU · ∇ˆ + ε2 u · ∇ˆ = −∇PB − ε∇ˆ + u u But. into the Navier-Stokes equations (9.for a suitably deﬁned Reynolds number. u u u p R 153 .1). R R ˆ ˆ where u = u(x. So. y A parallel flow 11111111111111111111111111111111 00000000000000000000000000000000 11111111111111111111111111111111 00000000000000000000000000000000 x The basic state is u = U (y)i = U. u and ε ≪ 1. ∇ · U = 0. while ﬂow in a Blasius boundary layer is not. ˆ (9. 0 = −∇pB + 1 2 ∇ U R 1 1 2 ˆ ∇ U + ε ∇2 u.3). plane Poiseuille ﬂow is an example of a parallel ﬂow. Stability of steady parallel ﬂows A parallel ﬂow past a solid wall is one whose streamlines are parallel to that wall.2). y. Thus.

ˆ The inviscid limit R → ∞. We therefore write u = ℜ{v(y)eik(x−ct) }. Always k > 0. ℜ{eik(x−ct) } = ℜ{eik(x−cr t) ekci t } = ekci t cos(k[x − cr t]). In general. So the forms (9. the disturbance will grow so large. R (9. or remaining constant in amplitude. so it follows that if ci > 0. The disturbances u and v are called modes and this whole procedure is referred to as the ˆ ˆ method of normal modes. the functions u(y) and v(y) are complex the parameter k is real the complex wave speed c = cr + ici is a complex number. the amplitude of the wave will grow.5) represent travelling waves moving with speed cr of amplitude ekci t . it will completely disrupt the basic ﬂow. To summarise. Rayleigh’s criterion 154 . the ﬂow is neutrally stable. v manifest themselves in the ˆ ˆ form of travelling waves. Neglecting the term of O(ε2 ) as being smaller than the others.Also. the disturbance will get bigger. Eventually. v = ℜ{v(y)eik(x−ct) }. ˆ where ℜ means “take the real part”. We write the pressure disturbance in a similar way.4) It is usual at this stage to assume that the disturbances u. we are left with ˆ ∇ · u = 0. if if if ci > 0 ci < 0 ci = 0 the ﬂow is linearly unstable. exponentially decaying. the ﬂow is linearly stable. Since ik(x − ct) = ik(x − cr t) + kci t. ˆ (9. ˆ ˆ ut + u · ∇U + U · ∇ˆ = −∇ˆ + u p 1 2 ˆ ∇ u. U · ∇U = 0 since U(y) represents a unidirectional ﬂow (see previous work on unidirectional ﬂows). p = ℜ{P (y) eik(x−ct) }.5) as then the wave is either exponentially growing.

and assume that it is all right to drop the ﬁnal term. Example 1 Consider now the speciﬁc problem of parallel ﬂow in a channel 0 ≤ y ≤ 1: y 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 U ′′ − k2 v = 0. we seek values of c such that this problem has a solution. (c − U ) with v = 0 on y = 0. (9. multiplying the second by −ik and adding. Then.6) This is an eigenvalue problem for c. using the third equation to note that iku + v ′ = 0 and to write u′ = −v ′′ /ik. (c − U ) x In this case. given a particular disturbance of frequency k. iku + v ′ = 0. Therefore we have v ′′ + U ′′ − k2 v = 0. It is convenient to try to eliminate the pressure from these equations. 155 . −ik(c − U )u + U ′ v = −ikP. we have. 1. we obtain −ik(c − U )(u′ − ikv) + U ′′ v + U ′ (iku + v ′ ) = 0. writing the equations in component form.4) in the limit R → ∞. ˆ ˆ ˆ ˆ ∇ · u = 0. we have −ik(c − U )(−v ′′ + k2 v) + ikU ′′ v = 0.We ﬁrst consider equation (9. and so v ′′ + This is Rayleigh’s equation. ut + u · ∇U + U · ∇ˆ = −∇P . That is. u Substituting in the normal modes. U (y) = y(1 − y) and we need the conditions u = v = 0 on y = 0. Diﬀerentiating the ﬁrst with respect to y. 1. −ik(c − U )v = −P ′ .

(c − U ) Integrating by parts. This gives a necessary but not a suﬃcient condition for instability of the ﬂow. 156 . 1]. However. so 1 1 0 1 − 0 |v ′ |2 dy + 1 0 U ′′ − k2 |v|2 dy = 0.7) and so the imaginary part of (9. v v′ But v(0) = v(1) = 0. We are interested in the response of the ﬂuid to this disturbance. ˆ ∇ · u = 0. This means we can no longer neglect the ﬁnal term. It is known as Rayleigh’s inﬂection point theorem. the only way for this equation to hold is if U ′′ changes sign at some point in the interval [0.Physically. we get 1 1 v v ′′ dy + 0 0 U ′′ − k2 |v|2 dy = 0.7) gives. ˆ ˆ ut + u · ∇u + u · ∇ˆ = −∇ˆ + u p 1 2 ˆ ∇ u. c−U cr + ici − U |c − U |2 U ′′ |v|2 dy = 0.4). So U must have an inﬂection point in that interval. The ribbon is vibrated so that it produces ﬂuctuations of ﬁxed wavenumber k. we might think of a ﬂexible ribbon being placed within the channel. The system (9. Returning to equations (9. If we multiply (9. (c − U ) − Now 0 |v ′ |2 dy + 1 0 U ′′ − k2 |v|2 dy = 0. General size R: The Orr-Sommerfeld equation. namely v. (c − U ) (9. R (9. |c − U |2 ci 0 So if ci > 0.6) can be solved numerically. As above. some analytical progress is still possible.6) by the complex conjugate of v.8) we now consider O(1) values of R. and then integrate over the channel width. 1 1 1 cr − U − ici = = . we write c = ci + ici and say that the ﬂow is stable if ci < 0 and unstable if ci > 0.

y 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 x The ﬂow is driven by a constant negative pressure gradient dp/dx = −G.9) to write iku = −v ′ . 1.11) we obtain. writing u = ℜ{u(y)eik(x−ct) }. with G > 0 and the basic solution is U (y) = y(1 − y). We can now recognise this as Plane Poiseuille Flow discussed above in the Exact Solutions section.12) This is the Orr-Sommerfeld equation. we ﬁnd −ikcv ′ + ikU v ′ − ikU ′ v = −k2 p + 1 [−k2 v ′ + v ′′′ ].10) by ik and using (9.10) (9. R (9. 2 d2 1 d2 − k2 v = (U − c) − k2 v − U ′′ v. 1 −ikcu + ikU u + vU ′ = −ikp + [−k2 u + u′′ ] R 1 ′ −ikcv + ikU v = −p + [−k2 v + v ′′ ]. The disturbance ﬂow is governed by the Orr-Sommerfeld equation (9. ˆ Therefore we have iku + v ′ = 0. Using equation (9. ikR dy 2 dy 2 (9.11) v = ℜ{v(y)eik(x−ct) }. 1.9) these can be re-expressed as v = v ′ = 0 on y = 0.12) with boundary conditions u = v = 0 on y = 0.As before we use the method of normal modes. ˆ p = ℜ{P (y) eik(x−ct) }. R Multiplying (9. ˆ Diﬀerentiating this with respect to y and then eliminating p′ between the result and (9.9) (9. Consider now the channel ﬂow of Example 1 at ﬁnite R. 157 .

Modes corresponding to values of c lying on the curve are known as neutral modes. i. As we can see from the picture.13) Solving for c is generally a numerical task for a computer. if if if ci > 0 ci < 0 ci = 0 the ﬂow is linearly unstable. i. an unstable ﬂow. writing c = cr + ici . R.e. Computing c in this way for many values of k and R we can plot the neutral curve. Note: Both of the tails of the hoop asymptote to the R axis as R → ∞. there is a critical Reynolds number beneath which there are no unstable modes. the ﬂow is neutrally stable.For this example we know that U = y(1 − y). As before. which delineates the boundary between stable and unstable solutions. the ﬂow is linearly stable. ci = 0. For the current example the neutral curve looks like this: k cI < 0 kc stable cI > 0 unstable Rc R Every point inside the hoop corresponds to a value of c with positive imaginary part. a stable ﬂow. (9. 1. Every point outside has ci < 0. We now select a wavenumber k (so we restrict attention to disturbances of this wavenumber) and a Reynolds number. y = 0. The corresponding wavenumber at the very nose of the neutral curve is found to be kc = 102. ikR with v = v ′ = 0 on for c. Then we have to solve the problem 1 ′′′′ [v − 2k2 v ′′ + k4 v] = (y − y 2 − c)(v ′′ − k2 v) + 2v.e. It is computed to be Rc ≈ 5774. 158 . On the curve itself.

e.A surprise We have just looked at the stability of ﬂow in a channel. pick a wavenumber k and Reynolds number R and solve for c. The resolution of this problem is still not complete. This ﬂow (see above) is linearly stable at any Reynolds number. This ﬂow (see above) is linearly unstable at suﬃciently large Reynolds numbers Plane Couette Flow Flow in a channel driven by the motion of the upper wall. the basic ﬂow is now. θ) direction). Expressing the disturbance ﬂow as ˆ ˆr ˆ u = uˆ + wk. (assuming there is no ﬂow component in the azimuthal (i. 159 . In cylindrical polars. Circular Poiseuille ﬂow Flow in a circular pipe driven by a constant pressure gradient. In other words. the ﬂow is stable at any value of the Reynolds number. this contradicts experiments which show that the ﬂow becomes turbulent at a critical Reynolds number. substitute in for U . Suppose we instead look at that in a circular pipe. but it is generally believed that the turbulence results from perturbations of ﬁnite rather than inﬁnitesimal amplitude. Plane Poiseuille ﬂow y 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 1111111111111111111111111111111 0000000000000000000000000000000 x Flow in a channel driven by a constant pressure gradient. which is the circular Poiseuille ﬂow we calculated earlier in the course. This ﬂow (see above) is linearly stable at any Reynolds number. we can write down the corresponding form of the Orr-Sommerfeld equation for cylindrical polars. U = 1 − r2. As was mentioned earlier. Somewhat surprisingly we ﬁnd that ci is always negative. We now summarize known stability results for the classic ﬂows discussed in this course.

According to geometry. we perturb the interface so that now r = f (x. f = a and κ = 1/(2a).14) where p0 is the pressure in the gas (e.15) (9. gas r perturbed interface liquid x Fix a frame of reference moving with the jet. In the basic state. of radius a. Equation (9. κ is the mean curvature of the interface. t) = a + εa1 (t) cos kx. We assume that the ﬂow in the liquid is incompressible and irrotational.14) gives (since ∂/∂θ ≡ 0) g′ ∂ 2 φ1 ∂ 2 φ1 1 ∂φ1 = g′′ + − k2 g = 0. t) cos kx. for the basic ﬂow. This means that we can model it using Laplace’s equation: ∇2 φ = 0 with u = ∇φ the ﬂuid velocity. then 1 1 + f ′2 − f f ′′ . the jet is perfectly cylindrical. κ= 2 f (1 + f ′2 )3/2 So. atmospheric pressure) and P is the pressure in the liquid. Fix cylindrical polars in the jet as shown in the diagram.Capillary instability of a liquid jet Lord Rayleigh (1879) conducted a famous study of the stability of a jet of liquid shooting into an ambient gas.g. which is given by P − p0 = 2 γ κ (9. We assume that the ﬂow is axisymmetric. if the surface is given by r = f (x). so that in this frame the jet appears stationary. To assess the jet’s stability. We assume that we can expand the potential like this: φ = φ0 + εφ1 = φ0 + εg(r. But φ0 = 0 since the ﬂuid is stationary in the moving frame. At the interface between the liquid and the gas there is a pressure jump due to surface tension γ. for some small parameter ε. + + ∂x2 ∂r 2 r ∂r r 160 .

So, r 2 g′′ + rg′ − k2 r 2 g = 0. (9.16)

This is a modiﬁed form of Bessel’s equation. Recall that Bessel’s equation of zeroth order looks like this: r 2 g′′ + rg′ + r 2 g = 0. Equation (9.16) is called the modiﬁed Bessel’s equation of zeroth order. It has linearly independent solutions I0 (kr), K0 (kr). We can thus write the solution as g(r, t) = α(t) I0 (kr) + β(t) K0 (kr). In fact K0 (r) → ∞ as r → 0, so to get a solution which is regular on the axis, we need β = 0, leaving φ1 = α I0 (kr) cos kx. At the interface we have the kinematic condition (see section 3.4) D (r − f ) = 0 Dt on r = f.

This states the the interface at r = f (x, t) must move along with the liquid. Expanding, D Dr Df ∂f ∂f (r − f ) = − = u− −w , Dt Dt Dt ∂t ∂x on r = f . But u = ∇φ and so u= ∂φ ′ = αkI0 (kr) cos kx, ∂r w= ∂φ = −αkI0 (kr) sin kx. ∂x (9.17)

Therefore, on r = f , (9.17) becomes
′ ε αkI0 (ka) −

da1 dt

cos kx + O(ε2 ) = 0,

giving

da1 ′ = αkI0 (ka). dt

To ﬁnd α, we need to consider the pressure jump at the surface, given by (9.15). To ﬁnd the pressure in the liquid we apply Bernoulli’s equation (see Hydrodynamics I, II) in its unsteady version: p ∂φ 1 + |∇φ|2 + = c(t) ∂t 2 ρ for some function c(t). 161 (9.18)

Assuming we can write the pressure as p = P + εˆ cos kx p then (9.18) gives 1 ∂ εαI0 (ka) cos kx + (P + εˆ(a, t) cos kx) + O(ε2 ) = c(t) p ∂t ρ on r = f . So at O(ε), dα + p(a, t) = 0, ˆ dt which gives an equation for α in terms of the interfacial pressure. ρI0 (ka) From above, da1 ′ = αkI0 (ka). dt
′ kI0 (ka) d2 a1 =− p(a, t). ˆ ρI0 (ka) dt2

So

Now, since r = f = a + εa1 cos kx, the curvature 2κ = = But the pressure jump p − p0 = (P − p0 ) + εˆ(a, t) cos kx p = 2γ κ = γ So P = p0 + γ/a and p(a, t) = −γ ˆ Combining this with the previous result, − ˆ we have, ﬁnally, setting k = ak, ˆ ′ ˆ γ kI0 (k) d2 a1 ˆ a (1 − k2 ) = 0, 2 − ρa3 ˆ 1 dt I0 (k) 162
′ d2 a1 kI0 (ka) p(a, t) = ˆ , ρI0 (ka) dt2

1 1 + f ′2 − f f ′′ a1 = − ε 2 − k2 a1 cos kx + O(ε2 ) ′2 )3/2 a a f (1 + f a1 1 − ε 2 1 − a2 k2 cos kx + · · · . a a (9.19)

1 a1 − ε 2 (1 − a2 k2 ) cos kx . a a a1 (1 − a2 k2 ). a2

or

d2 a1 2 2 − σ a1 = 0, dt

with

σ2 =

ˆ ′ ˆ γ kI0 (k) ˆ (1 − k2 ) ˆ ρa3 I0 (k)

This equation tells us how the amplitude a1 of the disturbance to the jet varies in time. The solution is a1 = c1 exp (σt) + c2 exp (−σt). If k > 1/a, σ is imaginary and the solution is bounded. If k < 1/a, σ is real and the solution grows indeﬁnitely. Thus, the jet is unstable if the wavenumber of the disturbance k< or if the wavelength λ = 2π/k, λ > 1/a. So disturbances of large enough wavelength destabilize the interface. NOTE: If a suitable wavenumber destabilizes the jet, the amplitude of the disturbance will keep on growing. Eventually it will grow so large that the above theory is invalidated. Remember that we had to insist that ε is small. At this stage the disturbance interacts nonlinearly with the basic ﬂow and disrupts it drastically. Eventually, the jet pinches into a sequence of satellite drops like this: 1 a

Appendix A: Invariants of a matrix under rotation In section 2(b) we noted that it is possible to rotate from one set of axes to a second in which the strain tensor is a diagonal matrix with non-zero entries along its diagonal and zeros everywhere else. Under this action, the trace of the matrix is invariant. In this appendix, we will see why this is the case. First, we investigate rotations of a vector. Deﬁne a transformation matrix R such that its action on a general vector x is to map it to another vector y of the same length. Accordingly, yT y = xT x. 163

R represents a pure rotation of that vector through some determined angle. R is an orthonal matrix whose transpose is equal to its inverse. Expanding the equation. Also. Now. I3 the invariants of the matrix A. and ij A′ has components a′ has written with respect to a second frame rotated with respect to ij the ﬁrst. A′ = RT AR. yT y = (Rx)T (Rx) = xT RT Rx = xT x. Since x is arbitrary. we have λ3 − I1 λ2 + I2 λ − I3 = 0. Rotating to a system in which A is diagonal. RT R = I. it can be shown that a′ = Rki Rlj akl . T a′ = Rik Rlj akl . and the characteristic equation. det(A′ − λI) = det(A − λI). det(A − λI) is independent of the choice of coordinate system. Thus. I2 . where the coeﬃcients I1 . We can I1 . we deduce that xT (RT R − I)x = 0. I3 are.Furthermore. we exploit the fact that we can rotate to any coordinate system we like without changing their values. det(A′ − λI) = det(RT AR − λRT R) = det(RT [A − λI]R) = det(A − λI). if a matrix A has components a′ written with respect to one Cartesian frame. since detR = 1. In other words. ij and so. I2 and I3 must also be independent of the choice of coordinate system. we ﬁnd easily that det(A − λI) = (λ − λ1 )(λ − λ2 )(λ − λ3 ) = λ3 − (λ1 + λ2 + λ3 )λ2 + (λ1 λ2 + λ2 λ3 + λ1 λ3 )λ − λ1 λ2 λ3 . So. I2 . Consider now the determinant We can see that det(A′ − λI). since the length of the vector does not change. 164 . To work out what I1 . ij Or.

as in (3. since I1 = trA.Hence. I1 = trA. 8 Note that uu is an example of a dyadic product. we state Newton’s second law for a ﬁxed volume of constant density ﬂuid V (t). since ∇ · u = 0. Or. as is 1 (trA)2 − tr(A2 ) . 2 I3 = detA. Note that8 u · ∇u = ∇ · (uu). 1 (λ1 + λ2 + λ3 )2 − (λ2 + λ2 + λ2 ) 1 2 3 2 So. 165 . Thus. In index notation. First. Similarly. it means the matrix ui uj . ∂ (ρ u) dV = − ∂t ρ u(u · n) dS + ρF dV + V S V S σ · n dS. Using the divergence theorem then. we have ρ V ∂u dV + ∂t S ρ u(u · n) dS = ρF dV + V S σ · n dS. the determinant of a matrix A is invariant under rotation.30) is derived as follows. 2 Appendix B: The momentum integral equation The momentum integral equation (3. I1 = λ1 + λ2 + λ3 I2 = λ1 λ2 + λ2 λ3 + λ1 λ3 = I3 = λ1 λ2 λ3 . ρ V ∂u dV + ρ ∂t V u · ∇u dV = ρF dV + V V ∇ · σ dV . I2 = 1 (trA)2 − tr(A2 ) . the trace of a matrix A is invariant under rotation.30).