Four approaches to hydrodynamic Green’s functions - the Oseen tensors

Maciej Lisicki∗
Institute of Theoretical Physics,
Faculty of Physics, University of Warsaw
ul. Hoza 69, 00-681 Warszawa, Poland

I. FORMULATION OF THE PROBLEM We have introduced the Kronecker delta to eliminate Fj

and therefore can write
We present several ways of solving the following prob-
Pbj G
b ij
lem: a force F is applied to a point particle immersed in − iki − k2 K
= −δij . (3)
space filled with viscous liquid. We are looking for the 8πµ 8π
fundamental solution (stokeslet) in case of incompress- The incompressibility condition reads
ible flow. The full system of flow equations in the low
Reynolds number regime (Re  1) has the form ki Gij = 0.
2
− ∇p(r) + µ∇ v(r) = −Fδ(r), (1) Multiplying the Stokes equation in the form as above
∇ · v = 0. and using the incompressibility relation we eliminate the
We demand that the flow field vanishes at infinity, i.e. velocity tensor part of the equation and get
r→∞
|v(r)| −→ 0. Then the solution is uniquely determined. Pbj ikj
By solving this problem we mean finding the form of =− 2
velocity and pressure fields in the fluid. 8πµ k
We can directly calculate the inverse Fourier transform
of this equation to get Pj
II. SOLUTION À LA KIM & KARILLA
Z
Pj i kj
=− dk 2 eik·r (4)
A very ’physical’ approach to this phenomenon can be 8πµ 3
(2π) R3 k
made by making use of the linearity of Stokes equations
1
and symmetries of the system. Here, we preform detailed Let us now consider a function φ = 4πr . One can show
−
calculations according to hints from Exercise 2.9 from that its Fourier transform equals k 2. Indeed, let us
the textbook [Kim & Karilla]. calculate the inverse transform. Choosing r to be parallel
As the Stokes equations are linear in p and v, the pres- to the x3 axis in the k-space, we can perform integration
sure field p(r) can be written as a scalar product of a in spherical coordinates (fig. 1)
certain vector field P(r) and the point force F and the
velocity field v(r) can be represented by a a tensor G(r) ∞ Z1
eikr cos θ
Z
−1 − 1 2
acting on a point force F with certain factors: F {k 2} = dkk d(cos θ) (5)
(2π)2 0 k2
F · P(r) G(r) · F −1
p(r) = , v(r) = . (2) 2
Z ∞
sin kr 1
8πµ 8πµ = d(kr) = ,
(2π)2 r 0 kr 4πr
Applying Einstein’s summation rule we can write these
equations using the vector components where we have used the (A1) integral (see App. A.2).
Pj Fj Gij Fj By taking the gradient of φ, we reproduce the Fourier
p(r) = , vi (r) = . terms from eq. (4). We can therefore explicitly write
8πµ 8πµ
down the solution for pressure:
Applying the Fourier transform (App. A.1) to Stokes
1
equations (1), we get
P(r) = −2µ∇ . (6)
r
p − µk 2 v
−ikb b = −F.
Inserting the expected field form to this equation we have Now we shall use expression for Pcj to construct the ve-
for component i: locity field. We insert P
cj into eq. (3) and get a closed
expression for the G tensor
Pbj G
b ij
−iki Fj − k 2 K
Fj = −Fi = −Fj δij . K
8πµ 8π G
b ij δij ki kj
= 2 − 4 . (7)
8π k k
Once again, the first term on the RHS is the transform
∗ Electronic address: mklis@fuw.edu.pl of the Green’s function for Laplace equation and after
 2

We therefore must have C1 + C2 = 0. From these two

conditions we can deduce the constants C1 = −C2 = 1
and sum up the results writing
K
δij ri rj 1 r ⊗ r
Gij = + 3 , or G= I+ 2 . (9)
r r r r
The pressure and velocity fields can be written as:

F·r F  rr 
p(r) = v(r) = · I+ 2 (10)
4πr3 8πµr r
Figure 1: Integration in the k–space with fixed orientation of
r = r · x̂3 .
III. SOLUTION À LA ZAPRYANOV &
k
δij TABAKOVA
inversion contributes 4πr . It remains to find the inverse
transform
nk k o This method has been proposed by
F −1 i j
(8) [Zapryanov & Tabakova]. It follows from the prop-
k4 erties of Fourier transform and fundamental solutions of
Now we can use the symmetry argument to deduce the Laplace and biharmonic equations. In fact, one does not
form of the inverted transform. We know that in the have to calculate any integral explicitly in this approach.
k-space Gb decreases as k −2 . This condition transforms We start by taking the divergence of eq. (1), obtaining
to the real space resulting in an asymptotic decay of G a Poisson equation for pressure
as r−1 . Moreover, the expression (8) is symmetric with
∇2 p(r) = ∇(Fδ(r)).
respect to index switching. Therefore we deduce the form
of the inverse transform to be We can now apply Fourier transform to the above equa-
K tion, which gives
nk k o
i j δij xi xj
F −1 4
= C 1 + C2
k 8πr 8πr3 k·F
The last task is to determine the values of numerical con- k 2 pb(k) = −ik · F =⇒ pb = −i
k2
stants C1 , C2 . Taking the trace of both sides, we get one
condition 3C1 + C2 = 2. Taking now i = j = 3 and again We can now apply the Fourier transform to Stokes equa-
fixing r = (0, 0, x3 ) in Cartesian coordinates, we perform tions and use the obtained expression to eliminate pres-
the integration in k–space. sure from the equations and get a closed expression for
the velocity field:
k2
nk k o Z
i j 1
F −1 4
= 3
dk 34 eik·r = [k3 = k cos θ] = k · F
k (2π) k ik + µk 2 v
b (k) = F.
R k2
Z∞ Z1
1 We find the expression for the velocity field in k-space:
dk d(cos θ) cos2 θeikr cos θ =
(2π)2 1 h  k · F i
0 −1 v
b (k) = F − k . (11)
Z∞ Z1 µk 2 k2
1
dk dξξ 2 eikrξ = (∗) Using the definition of the inverse Fourier transform, we
(2π)2
0 −1 can easily write down the expressions for pressure and
velocity fields
Performing integration by parts, we transfer the ξ–
integral into known definite integrals (A1,A2,A3). 1
Z
eik·r h  k · F i
v(r) = dk F − k (12)
Z∞ Z1 8µπ 3 k2 k2
1 h 2 sin kr 2 i R3
(∗) = dk − dξξeikrξ =
(2π)2 eik·r
Z
kr ikr i
0 −1 p(r) = dk (k · F). (13)
8π 3 k2
Z∞ Z1 R3
1 h 2 sin kr 4 cos kr 2 i
= dk − − dξ eikrξ
(2π)2 kr (kr) 2 (kr)2 We can proceed to solution of these equations. Basing on
0 −1 the knowledge of fundamental solutions of the Laplace
Z∞ equation (see App. A.1), we find useful expressions re-
1 h 2 sin u 4 cos u 4 sin u i
= du − − =0 lating derivatives of 1/r with their Fourier transforms.
(2π)2 r u u2 u3 We now note that by taking a scalar product of eq. (A6)
0
 3

with F, we get an integral which describes the pressure The usual terminology is the Oseen tensor for T and the
field. Inserting it into the relation (13), we get pressure vector for g.
Consider now an external force which is continuously
 1  F·r distributed over the entire fluid (i.e. there is a nonzero
p(r) = −F · ∇ = (14) external force density f (r0 ) in the fluid). The linearity of
4πr 4πr3
Stokes equations implies the superposition principle - the
In a similar manner, taking the scalar product of eq. (A7) fluid velocity in a certain point r is a vector sum of the
with F, we reproduce the integral present in the first part fluid velocity increments stemming from the forces acting
of the expression for velocity field (12), so that in every point of the fluid. We can therefore express this
sum as an integral
eik·r  k · F 
Z
F 1
v(r) = − dk k (15) Z
4πµr 8µπ 3 k2 k2 v(r) = dr0 T(r − r0 ) · f (r0 ).
R3

Taking a scalar product of eq. (A10) with F, we get the For pressure the same arguments hold and we can write
second term of the RHS of velocity field equation. We Z
have used the identity kk · F = (F · k)k. These relations p(r) = dr0 g(r − r0 ) · f (r0 ).
imply
F F  r 
In this structure one immediately sees the role of Green’s
v(r) = − ·∇⊗∇ (16)
4πµr µ 8π functions for linear problems. Once the force field is
specified, knowing the Green’s functions for a particu-
r
Due to the fact that ∇r = r and ∇ ⊗ r = I - the unit lar geometry, the pressure and velocity are easily found
tensor, we can write by integration. Deriving the Green’s functions needs an
’inversed’ reasoning. It involves solving the Stokes equa-
1 1 r 1 rr I
∇ ⊗ (∇r)) = ∇⊗ =− + , tions with a specific force field, namely a point force (con-
8π 8π r 8π r3 8πr centrated in one point, what is represented by a Dirac
where we have used the Leibniz chain rule for calculation delta function). Let us substitute eq. (18) and (19) into
of ∇⊗ rr . We can always write a vector in a form F = IF, the Stokes equations . We get
so that in the end we can write down the expression for Z
velocity field dr0 [∇ · T(r − r0 )] · f (r0 ), = 0 (20)
Z
F  rr 
v(r) = · I+ 2 (17) dr0 [∇g(r − r0 ) − µ∇2 · T(r − r0 ) − Iδ(r − r0 )] · f (r0 ) = 0,
8πµr r
where we have used an identity F = dr0 f (r0 )δ(r − r0 ).
R
Since the external force density is arbitrary, the expres-
IV. SOLUTION À LA DHONT
sions in brackets must vanish and hence we get the equa-
tions for the Green’s functions
A very physical approach, basing on the linearity of
Stokes equations has been presented by [Dhont]. In this ∇ · T(r − r0 ) = 0, (21)
approach we clearly see the Green’s functions role in the ∇g(r − r0 ) − µ∇2 · T(r − r0 ) − Iδ(r − r0 ) = 0. (22)
solution of the problem of external force density in the
fluid and illustration of the superposition principle. In this equation we have now tensor quantities and we
Consider an external force F acting on a fluid only in shall in fact write ∇ ⊗ g instead of ∇g. An usual opera-
a single point r0 , so that F(r) = Fδ(r − r0 ). Since the tion is now to take the divergence of the second equation
Stokes equations (1) are linear, the fluid flow velocity in and use the incompressibility condition to get a Poisson
a point r somewhere in the fluid is proportional to the equation fo the pressure vector:
force and has to depend on the direction of the force
and the distance to the point where the force is exerted. ∇2 g = −∇ · Iδ(r) = δ(r)
Moreover, this relation has to be a linear transformation
(which can be represented as a matrix). Hence, in a We can now recall the fundamental solution of the
natural way, we can write Laplace equation (A4), which yields the form of g

v(r) = T(r − r0 ) · F.
 1 
(18)
g = −∇ + G(r),
4πr
Similarly, pressure is linearly related to the force by a
vector quantity as we can alway add a vector G(r) which satisfies the
laplace equation, i.e. ∇2 G = 0. One can show that if we
p(r) = g(r − r0 ) · F. (19) demand an asymptotic decay so that G → 0 as r → ∞,
 4

1
this implies G ≡ 0 (see App. A.4). Hence, we find the −∇( 4πr . Substituting the delta function and the pres-
pressure vector sure function to the Stokes equations, we get
 1  1 r 1 1
g = −∇ = (23) µ∇2 v = − g(∇∇ − I∇2 ) . (27)
4πr 4π r3 4π r

Now we can substitute (23) to eq. (22). Noting that We can introduce a scalar function H and express the
∇ ⊗ rr3 = rI3 − r⊗r
r 5 and eliminating the Dirac delta by velocity field in the form
substituting the fundamental solution of Laplace equa-
tion, one gets 1
v= g · (∇∇ − I∇2 )H. (28)
h 1 I i 1 h rr Ii µ
∇2 − µT(r) = 3 5 − 3 (24)
4π r 4π r r
One can prove that such an operation can always be pre-
We now choose, basing on the RHS form of the above formed and H can be found. By replacing v in eq. (27),
equation, the appropriate form of the LHS to be we arrive at a closed expression for H (if we discard an
h 1 I arbitrary constant g):
i 1 1 rr
∇2 − µT(r) = α0 n I − α1 m 2 , (25)
4π r r r r !
2 1 2
where α0 , α1 , m and n are constants. These con- (∇∇ − I∇ ) ∇ H + = 0.
4πr
stants can be chosen in such a way that this Ansatz pro-
vides a solution for the Oseen tensor decaying at inifinity
(T(r) → 0 as t → ∞) and after some algebra we arrive This equality can be surely satisfied by any solution of
1
at the known result Poisson’s equation ∇2 H = − 4πr . We therefore find (by
applying the Laplace operator to this condition) that H
1 rr  satisfies the biharmonic equation ∇4 H = δ(r). We know

T(r) = · I+ 2 . (26)
8πµr r the form of the fundamental solution ((see App. A.3) so
that
r
V. SOLUTION À LA POZRIKIDIS H=−
8π
The fourth method in our review was proposed by Substituting this result into eq. (28), we arrive (after
[Pozrikidis]. In this case we consider Stokes equations some algebra, the same as in eq. (16) at our result
with arbitrary force F, concentrated on one point. Be-
cause pressure is a harmonic function, and replacing the
1
delta function on the RHS of eq. (1) basing on the fun- vi (r) = Sij gj ,
damental solution of Laplace equation, we can set (to 8πµ
balance the dimensions of pressure)
where the Oseen tensor Sij is defined as follows:
1 1
p=− g·∇ .
4π r δij xi xj
Sij (r) = + 3
This can be obtained by taking the divergence of Stokes r r
equation and then replacing the delta function with

[Kim & Karilla] Kim S., Karilla S., Microhydrodynamics. Appendix A: MATHEMATICAL ADDENDUM
Principles and Selected Applications, Dover, 2005.
[Zapryanov & Tabakova] Zapryanov Z., Tabakova S., Dy- 1. Fourier transform
namics of Bubbles, Drops and Rigid Particles, Kluwer -
Dordrecht, 1999.
[Dhont] Dhont J.K.G.,An Introduction to Dynamics of Col- We define the Fourier transform pair in the following
loids, Elsevier Science, 1996. manner
[Pozrikidis] Pozrikidis C., Boundary Integral and Singularity Z
Methods for Linearized Viscous Flow, Cambirdge Univ. F{f } = fb(k) = drf (r)e−ik·r ,
Press, 1992. R3
Z
−1 1
F {f } = f (r) = dkfb(k)eik·r .
(2π)3 R3
 5

2. Useful integrals monic equation lead to

 r  Z
4 1
We have the following integrals ∇ = −δ(r) = − dkeik·r , (A8)
8π (2π)3 R3
Z∞  r  1
Z
1
sin k π = − dk 4 eik·r , (A9)
dk = (A1) 8π (2π)3 R3 k
k 2  r  Z
0 1 kk
Z∞ ∇∇ = dk 4 eik·r . (A10)
1 − cos k π 8π (2π)3 R3 k
dk = (A2)
k2 2 The last term contains tensor product of two vectors:
0
r r
Z∞ ∇∇( 8π ) ≡ ∇ ⊗ [∇( 8π )] and kk ≡ k ⊗ k. This notation
k − sin k π defines a second–rank tensor which can be represented as
dk = (A3)
k 3 4 a matrix. In terms of matrix components one can write
0 (kk)ij = ki kj . We have created a tensor on the LHS of
the last equation so that the character of the RHS is the
The second and third integral actually can be obtained
same in the k–space.
from integration by parts of the first integral.

3. Fundamental solutions 4. Laplace equation solutions’ properties

We introduce fundamental solutions for Laplace and We encounter the problem of showing that a function
biharmonic equations. These are such functions ψ and φ G(r) satisfying the following conditions:
that satisfy the corresponding equations with Dirac delta
functions: ∇2 f (r) = 0 on R3 ,
f (r) → 0 for r → ∞,
∇2 φ(r) = −δ(r)
∇4 ψ(r) = −δ(r) is identically equal to 0.
For this, we use the Green’s integral formula for two
It appears that in 3D the fundamental solutions have the scalar fields φ, ψ:
form Z Z Z
1 r dr(φ∇2 ψ) = φ(n · (∇ψ)dS − dr(ψ∇2 φ)
φ= ; ψ= (A4) V ∂V V
4πr 8π
Taking ψ = f (r0 ) and φ = |r−r 1
0 | , the boundary integral
Finding these solutions involves inverting Fourier trans-
forms of the corresponding equations and involves inte- vanishes as r → ∞ and we get
gration in complex plane. Z
1
Z
1
0 0 02
By differentiating the fundamental solution of Laplace dr f (r )∇ = − dr0 ∇02 f (r0 ) = 0,
equation we get the corresponding terms in Fourier space |r − r0 | |r − r0 |

 1  1
Z as f satisfies the Laplace equation. But we know the
∇2 = −δ(r) = − dkeik·r , (A5) fundamental solution of the Laplace equation, so that
4πr (2π)3 R3
 1  Z ∇02 |r−r
1 0
0 | = −4πδ(r − r ). From the above equation we
i k
∇ = 3
dk 2 eik·r , (A6) deduce
4πr (2π) R3 k
Z
 1  1 1 4πf (r) = 0 =⇒ f (r) ≡ 0.
= dk 2 eik·r . (A7)
4πr (2π)3 R3 k
We have proved that a harmonic function decreasing to
Analogical operations on fundamental solutions of bihar- 0 at infinity is identically 0 everywhere.