Professional Documents
Culture Documents
Green Functions
Green Functions
kernel.
David Gurarie
In a similar way one handles the heat and wave propagators, e.g.
G (z; w; t) = G0 (z ¡ w; t) ¡ G0 (z ¡ w¤ ; t)
in terms of the free-space Gaussian G0 .
The corresponding Poisson kernels are computed from K; G; ::: via standard
relations, e. g.
P (z; w) = @n K (z; w)jw2¡ = @´ Kj´=0 Dirichlet
K (z; w) = K (z; w)j¡ = Kj´=0 Neumann
In particular for spherically-symmetric problem, like Laplacian/Helmholtz
potentials K0 = K0 (r); r = jz ¡ wj we get
q
2y
P (x ¡ »; y) = ¡ K00 (r) ; with r = (x ¡ »)2 + y2
r
In special cases 2D; 3D it yields familiar expressions, that could be also produced
by the Fourier transform
y
P = ¼[(x¡»)2 +y2 ]
2D
y
P = 3=2 3D
2¼[(x¡»)2 +y 2 ]
2(n¡2)y
P = n=2 nD
!n¡1 [(x¡»)2 +y2 ]
1
w
w*
K (x; y; a) = K0 (x ¡ a; y) ¡ K0 (x + a; y)
KN (x; y; »; ´)
= K0 (x ¡ »; y ¡ ´) + K0 (x ¡ »; y + ´) + K0 (x + »; y ¡ ´) + K0 (x + »; y + ´)
KD (x; y; »; ´)
= K0 (x ¡ »; y ¡ ´) ¡ K0 (x + »; y ¡ ´) ¡ K0 (x ¡ »; y + ´) + K0 (x + »; y + ´)
2
6
0.6
5
0.5
4 0.4
3 0.3
2 0.2
1 0.1
0 -2 2 4
-2 0 2 4
4
0.6
0.5
3
0.4
0.3
2
0.2
0.1
1
1 2 3 4 5
0
0 1 2 3 4 5
4
0.5
0.25
3
1 2 3 4 5
-0.25
2 -0.5
-0.75
1 -1
0
0 1 2 3 4 5
3
2 Conformal coordinate change in Laplacian
We take map Á : x ! y = Á(x) in Rn (or a region in Rn ), and denote by A(x) =
Á0 (x) its Jacobian matrix, and by J = det(A) - the Jacobian determinant. The
general change of variable formula for the Laplacian has the form
¡ ¢¡1
¢y = J1 rx ¢ J T AA rx (1)
-a consequence of the grad and div-transformations
x ! y = Á (x)
¡ ¢¡1
rx ! ry = T A rx
¡ ¢
rx ¢ ::: ! ry ¢ ::: = J rx JA¡1 ¢ :::
1
4
To get the Laplace’s equation for v the …rst and 0-th order terms of (3) must
vanish. Hence, ® = ¡ n¡2
2 and the conformal factor ½ solves a nonlinear PDE,
µ ¶ ¯ ¯
r½ ¯ r½ ¯2
r¢ + (n ¡ 2 + ®) ¯ ¯ ¯ =0
½ ½ ¯
We rewrite it for the log-derivative of ½ as,
n¡2
¢ (ln ½) + jr (ln ½)j2 = 0 (4)
2
The latter could be explicitly solved when ½, hence à = ln ½ are radial
functions à = à (r). Indeed, (4) becomes a radial ODE for Ã,
¡ ¢0 ¡ 0 ¢2
r1¡n rn¡1 Ã0 + n¡22 Ã =0 (5)
¡ 0 ¢2
d
transformed via change of variable rn¡1 dr = dzd
into Ã00 + n¡2
2 Ã = 0, and
solve by separation. The o¤-short is a general solution of (5) in the form
¡ ¢2=(n¡2)
à = C1 C2 + r2¡n (6)
C
A special family of solutions vanishing at 1 is given by à = r2
2.2 Inversion
Let us note that ½ = r12 is precisely the conformal factor of the inversion map,
Á : x ! jxjx 2 . Indeed, its Jacobian-matrix
( )
0 xi xj
A=Á = 1
jxj2
± ij ¡ 2 = r¡2 U
jxj2
5
Figure 5: Jacobian matrix of the inversion is the re‡ection about the normal
plane to x
1
Hence, remembering that K0 = c jx¡»j n¡2 (constant cn is the surface area
n
n
of the unit sphere in R ) we get the Green’s function of the ball
K (x; ») = K0 (jx ¡ »j) ¡ K0 (j(^
x ¡ jxj »)j) (8)
³¯³ ´¯´
¯ ^ ¯
= K0 (jx ¡ »j) ¡ K0 ¯ » ¡ j»j x ¯
Here x x
^ = jxj and ^» = j»j
»
denote normalized (unit) vectors in the direction
x and ». It is often convenient to rewrite (8) in the polar form, i.e. variables
r = jxj, ½ = j»j and angle µ between x and » (see …g.6),
³p ´ µq ¶
K(r; ½; µ) = K0 r2 + ½2 ¡ 2r½ cos µ ¡ K0 1 + (r½)2 ¡ 2r½ cos µ (9)
¯ ¯ s
¯ z¡w ¯ r2 + ½2 ¡ 2r½ cos µ
K(z; w) = 1
ln ¯¯ ¯= 1
ln (10)
2¼ ¹¯
1 ¡ zw 2¼
1 + (r½)2 ¡ 2r½ cos µ
in polar coordinates z = reiµ ,w = ½eiÁ .Exterior disk Green’s function has the
same form (10), but this time r; ½ lie outside the unit circle
6
Figure 6:
1
0.5
0
-0.5
-1
-1 0 1 2 3
7
2.4.2 3-D case:
1
Here free-space K0 = 4¼jx¡»j , hence
½ ¾
1 1 1
K(x; ») = 4¼ ¡
jx ¡ »j j(^ x ¡ jxj »)j
8 9
< 1 1 =
1
= 4¼ :
p ¡ q
r2 + ½2 ¡ 2r½ cos µ ;
1 + (r½)2 ¡ 2r½ cos µ
8
2.6 Other examples
2.6.1 Half-disk
Green’s function and the Poisson kernel in the upper half-disk are obtained by
a combination of the re‡ection and inversion:
½ ¯ ¯ ¯ ¹ ¯¯¾
¯ z¡³ ¯ ¯
K (z; ³) = 1
ln ¯¯ ¯ ¡ ln ¯ z ¡ ³ ¯
2¼
³¯
1 ¡ z¹ ¯ 1 ¡ z³ ¯
4 1
3 0.5
2
-1 -0.8-0.6-0.4-0.2 0 0.2 0.4 0.6 0.8
1 -0.5
-1
-3 -2 -1 0 1 2 3
9
Example 3 Map Á : z ! ez takes a complex strip f0 · Im(z) · 2¼g onto the
entire plane C.
4
4
3 2
2
-4 -2 0 2 4
1 -2
-4
-3 -2 -1 0 1 2 3
6
1
5
0.5
4
3 0
-1.5 -1 -0.5 0.5 1 1.5
2
-0.5
1
-1
0 0.2 0.4 0.6 0.8 1
Conformal maps allow to transfer Green’s functions from one region to an-
other. Namely, we let map z ! w = Á (z) take region D (with the known
Green’s function KD (z; ³)) onto region M, whose Green’s function KM is to be
computed. Assuming the boundary of D mapped onto the boundary of M , two
Laplacians are related via ¢M [u] ± Á = J1 ¢D [u ± Á] , where J = ½2 denotes the
Jacobian determinant. It follows then that the Green’s functions of two regions
are related by a change of variables
¡ ¢
KM (w; ´) = KD Á¡1 (w) ; Á¡1 (´)
In the above examples map w = z¡i z+i from the half-plane H onto the disk
D. The half-space Green’s function,¯ computed
¯ by the re‡ection (complex con-
1 ¯ z¡³ ¯
jugation) has the form: K = 2¼ ln ¯ z¡¹³ ¯ Substitution of the inverse map z =
Á¡1 (w) = i w+1
w¡1 and ³ = Á
¡1
(´) = ::: yields
¯ ¯
¯ w¡´ ¯
1
KD (w; ´) = 2¼ ln ¯¯ ¯
1 ¡ w¹́ ¯
that was produced earlier by the inversion method.
10
2.6.3 Green and Poisson kernels on strip
1. . We use conformal map: z ! ez from strip f0 · Im z · ¼g onto half-
space fIm w ¸ 0g. Then
¯ ¯
1
¯ z
¯ e ¡ ew ¯
¯
1 ¯ sinh ¡ z¡w ¢ ¯
¯ ¢ ¯¯
G (z; w) = ¡ ln ¯¯ z ¯ = ¡ ln ¯ ¡ 2
2¼ e ¡ ew¹ ¯ 2¼ ¯ sinh z¡2w¹ ¯
8 ³ ´ ¡ ¢9
1 < cosh2 x¡» 2 ¡ cos2 y¡´ 2
=
= ¡ ln ³ ´ ¡ ¢
4¼ : cosh2 x¡» ¡ cos2 y+´ ;
2 2
½ ¾
1 cosh (x ¡ ») ¡ cos (y ¡ ´)
= ¡ ln
4¼ cosh (x ¡ ») ¡ cos (y + ´)
1 sin y
P (z; ») = @´ G (z; w)j´=0 =
2¼ cosh (x ¡ ») ¡ cos (y)
3
2.5
2
1.5
1
0.5
0
-3 -2 -1 0 1 2 3
Function P (x; y) along with its contour map near zero, in strip 0 < y < ¼
2.6.4 Problems
1. Show that complex map w = cosh(z) takes strip P =f0 · Im(z) · 2¼g
onto the upper half-plane H, so that the boundaries of P:fIm(z) = 0g and
fIm(z) = 2¼g go into the boundary of H. Use this map and the known
Green’s function of H to construct the Green’s function and the Poisson
kernel of P.
2. Do the same exercise with the fractional power map Á (z) = z ®=¼ that
takes the upper half-plane onto a sector of angle ®, and derive the Green’s
function and the Poisson kernel of the sector.
11
¡ ¢
Eulerian velocity u = ¡Ãy ; Ãx is given by a complex derivative u = i@z¹Ã.
Irrotational ‡ow r £ u = 0 corresponds to harmonic stream-…eld: ¢Ã = 0.
Equivalently u has velocity potential u = rÁ. In fact, it could be described
by a complex analytic potential w (z) = Á + iÃ, and complex velocity w0 (z) =
u ¡ iv. Indeed, the relations u = rÁ = rÃ? are nothing but Cauchy-Riemann
equations for w.
Let us also remark that steady Euler ‡ow obeys the Bernoulli relation:
½
2 ju j2 + p = Const - the hydrostatic pressure (14)
where ½ an p are ‡uid density and pressure. The latter follows from the mo-
mentum conservation
³ ´
u t + r ½2 ju j2 + p ¡ u £ ! = 0
12
½ - ‡uid density. The coe¢cients of the quadratic form E = E (U ) make up the
so-called virtual mass-tensor
ZZ I
mij = ½ rAi (~r) ¢ rAj (~r) = ½ Ai (@n Aj ) (18)
Rn nD
¡
I I
= ½ xi @n P¡ [xj ] = ½ xj @n P¡ [xi ]
¡ ¡
De…nition (18) resembles the inertia tensor for a rotating H xjsolid of mass-
xi
density ½r distributed over surface ¡. It’s not however, equal ½ r , as Poisson
¡
P¡ makes important di¤erence. Though Ai coinsides with a linear function
xi j¡ , its gradient rAi and normal derivative @n Aj on ¡are not e i and Ni
© 2 i-the
(the component
ª of n ) respectively. A simple illustartion would be a disk
x
x + y 2 · 1 , where A1 = x2 +y 2 has rA1 j¡ = (cos 2µ; sin 2µ) as opposed to
rxj¡ = (cos µ; 0). Let us also remark that tensor mij is a geometric invariant
of surface ¡, independent of its position in Rn , as one could easily verify, using
translational/ rotational symmetries of solution (16), and volume-integral1 form
(18) .
In a similar fashion one computes the total momentum of the induced ‡ow
0 1 0 1
ZZ I X P :::
~p = ½ u =½ Ni @ Uj xj A = @ mij Uj A (19)
Rn nD j ::::
¡
The rate of the change of the momentum gives the ‡uid reaction force,
exerted on the body. So the resulting equation of motion take the
dU d~
p
M + = F -external force
dt dt
RR
where M = ½0 is the body mass. Remembering the exact form of ‡uid
DP
dU
momentum (19) j (M ± ij + mij ) dtj = Fi . Thus the e¤ect of ‡uid on the
moving body is to replace the standard mass by the virtual mass-tensor in the
kinetic energy. The components of the reaction force ¡ d~ p
dt along U , and its
?
normal U gives the drag and buoyancy (lift) forces
X X
Fd = ¡ mij U_ i Uj ; Fb = ¡ mij U_ i Uj?
j j
13
Remark 6 Notice that the reaction force depends on acceleration U, _ so there
is no net drag or lift in a uniform ‡ow (d’Alembert paradox). This is not
surprising, particularly for the drag-force, as its presence would require either
energy dissipation or a non-zero energy ‡ux to 1. Both are absent in the ideal
‡uid.
Z2¼
0
à = P (r; µ ¡ ¿ ) U a sin ¿ (21)
0
ikµ imµ
¡ a ¢convolution identity e ¤e
Direct evaluation of (21) using = 2¼± km yields
Ã0 = ¡Ua rsin µ ¢= Uray
2 = U Im z . The latter is also an obvious consequence
z a
of Im a + z = 0 on the ³ circle2 ´of radius jzj = a. We plot stream-lines of
¡z a¢
à = U a Im a + z = U y 1 ¡ ar2
1.5
0.5
-0.5
-1
-1.5
-2 -1 0 1 2
14
H
evaluating pressure gradient along the circle F = C
rpds. Thus we get
µ ¶
a2
à (r; µ) = U r ¡ sin µ -stream
r
(µ ¶2 µ ¶2 )
1 @Ã (r; µ) 2 @Ã (r; µ)
p (r; µ) = ¡ 2 +r -pressure
@r @µ
µ ¶ µ ¶
@p (r; µ) sin µ @p (r; µ)
f (r; µ) = r cos µ ¡ -horizontal force
@r r @µ
¡¢Ã = ¡± (z ¡ ³)
Ãj@D = Const
¯ ¯
¯ ¯
For the cylinder D = fjzj < ag we have à (z; ³) = ¡
2¼ ln ¯ a(z¡³)
a ¡z ³ ¯,
2 ¹ and the corre-
sponding velocity
i¡ n 1 ³
o i¡ a2 ¡ j³j2
u = i@z¹Ã = + = ¡ ¢
4¼ z¹¡³¹ a2 ¡¹
z³ 4¼ z¹ ¡ ¹
³ (a2 ¡ z¹³)
Due to rotational symmetry we could place source on the real axis ³ = b > 0
call z = reiµ and compute pressure
¡ 2 ¢
¡2 a ¡ b2
p=¡ (22)
2 (4¼)2 jr ¡ beiµ j2 jbr ¡ a2 eiµ j2
Clearly, the net force is exerted by the vortex on the cylinder is directed to-
ward the vortex. Its precise form is obtained
H by evaluating the pressure gradient
¡2
px of (22) along the circle f = 2(4¼)2 C px ds.
15