Académique Documents
Professionnel Documents
Culture Documents
Characterization of Navier-Stokes
Twistor theory and uids p. 3/41
Plan of Talk
Topics to look at:
Characterization of Navier-Stokes
Characterization of Navier-Stokes
Laplaces Equation
Twistor theory and uids p. 3/41
Plan of Talk
Topics to look at:
Characterization of Navier-Stokes
Laplaces Equation
Biharmonic Equation
Twistor theory and uids p. 3/41
Plan of Talk
Topics to look at:
Characterization of Navier-Stokes
Laplaces Equation
Biharmonic Equation
Characterization of Navier-Stokes
Laplaces Equation
Biharmonic Equation
t
+.(v) = 0 (1)
and the conservation of momentum is expressed by
the Navier-Stokes equations
(
v
t
+ v.v) = p +
2
v (2)
Twistor theory and uids p. 4/41
Incompressibility
If the uid is incompressible in the sense that is a
constant in both time and space, we have the
condition:
.v = 0 (3)
To analyse matters further, we introduce the vorticity
vector
= v (4)
We demand incompressibility but allow for non-zero
vorticity. We let
H = p +
1
2
v
2
(5)
Twistor theory and uids p. 5/41
Recasting of Navier-Stokes
If .v = 0 then
2
v = . Use the identity
(
1
2
v.v) = v.v + v (6)
Recast Navier-Stokes equations as
(
v
t
v ) +H = . (7)
Taking curl, we get vorticity equation
t
+ v. .v =
2
(8)
where kinematic viscosity = /.
Twistor theory and uids p. 6/41
Vector potential
Since the velocity eld is divergence-free, we may
introduce a vector potential such that
v = (9)
and furthermore we may choose it so that it is
divergence free (a gauge condition):
. = 0 (10)
It is perhaps a tacet assumption of uid dynamics that
this object is of no use except when it can be reduced
to a single function. For example, planar 2D ow is
obtained by setting (gauge condition guaranteed)
= (x, y)e
z
(11)
Twistor theory and uids p. 7/41
Good idea in 3D too
We note that under the assumption that satises
. = 0
=
2
(12)
and the vorticity equation becomes, denoting
t
by :
4
=
1
_
(( )
2
) +
2
_
(13)
Well-known in the 2D planar case where it reduces to:
4
=
1
_
(,
2
)
(x, y)
+
2
_
(14)
Twistor theory and uids p. 8/41
Difcult (impossible?) Goal
Can we:
Understand
Twistor theory and uids p. 9/41
Difcult (impossible?) Goal
Can we:
Understand
Understand
Understand
i
4
w
2
w
2
=
1
2
_
w w
2
w
w
2
w
_
(18)
Twistor theory and uids p. 9/41
Difcult (impossible?) Goal
Can we:
Understand
i
4
w
2
w
2
=
1
2
_
w w
2
w
w
2
w
_
(19)
(20)
and is harmonic conjugate of in 2D case.
2
= 0 ,
2
= 0 (21)
Twistor theory and uids p. 10/41
The biharmonic limit
When viscosity , ignore non-linearities.
Time-independent Navier-Stokes equations reduce to
4
= 0 (22)
which is the biharmonic limit, also known as Stokes
ow. We want to understand the holomorphic
structure for the 3D vector version of:
w
2
w
2
= 0 (23)
Key issues: what do we do about w = x +iy when we
also have a z? How can we make it all holomorphic
when we have all these w littered around?
Twistor theory and uids p. 11/41
Holomorphic methods for 3D
We need a new picture to proceed. It is very well
known that the Laplace equation can be solved in
terms of holomorphic functions in two dimensions.
Among devotees of twistor methods, and students of
Bateman, Whittaker, it is known that this can be
carried out in three dimensions. Want to extend to the
biharmonic case. This can be done. This is at least an
opportunity to explain how to use complex methods in
3 dimensions. In 2D we let w = x + iy. What if we
also have a z? (Never put z = x + iy!) But what
complex structure do we use? The key is twistor
space for 3D, a la N. Hitchin.
Twistor theory and uids p. 12/41
Twistor space for 3D
The twistor space associated with R
3
is rst, as a real
space, the set of oriented straight lines in R
3
. Relative
to some origin O, let r denote the position vector of
the point on a given line nearest to O. Then r is
orthogonal to the the direction of the line, which we
denote by u with u.u = 1. So the set of oriented
straight lines is the set
TS
2
=
_
(r, u) R
3
S
2
| r.u = 0
_
(24)
This tangent bundle for the unit sphere is also the
tangent bundle to a complex manifold, S
2
is the
Riemann sphere CP
1
. TCP
1
is our twistor space.
Twistor theory and uids p. 13/41
Dening Points I
How do we dene a point in ordinary space in terms
of structure on TCP
1
. A point may be regarded as the
intersection of all straight lines through it. So a point
is necessarily some vector eld in TCP
1
that is
dened globally.
Twistor theory and uids p. 14/41
Dening Points II
To see the implications of this we introduce two open
sets that cover CP
1
. We can take coordinates for the
sphere as on one patch (covering everything except
innity), and
= 1/ on another patch, covering
everything except = 0. Over each of these
respective patches we can dene coordinates for the
tangent bundle as (, ) and ( ,
), where the relevant
vector elds are, respectively
(25)
Twistor theory and uids p. 15/41
Dening Points III
Consider now a holomorphic vector eld. On the
patch it can be written as
f
0
()
(26)
for some f
0
, and on the
patch, it can be written as
f
1
(
(27)
for some f
1
.
Twistor theory and uids p. 16/41
Dening Points IV
On the intersection of the two patches we have:
f
1
(
1
)(
2
)
= f
0
()
(28)
Taylor series expansion of both functions,
f
i
() =
n=0
a
i
n
n
, we deduce that the coefcients
a
i
n
vanish if n > 2. Global vector elds must be of the
form, for example on the patch:
() = a + b + c
2
(29)
Quadratics are only global holomorphic vector elds,
correspond to points of C
3
.
Twistor theory and uids p. 17/41
Summary of reality and metric
Further analysis allows the identication of real points
in R
3
, and the construction of a natural metric. The
points are real if and only if
c = a AND b = b (30)
Metric discriminant of the quadratic. Normalize as
ds
2
= dx
2
+ dy
2
+ dz
2
=
1
4
db
2
dadc (31)
The metric for real points:
ds
2
= dx
2
+ dy
2
+ dz
2
=
1
4
db
2
+ dada (32)
Twistor theory and uids p. 18/41
The nal point correspondence
If we pick our coordinate system such that the real
part of a is x, we see that we can take the imaginary
part of a to be y and set b = 2z. The convention is
to set:
r
() = (x + iy) + 2z (x iy)
2
(33)
This gives us the correspondence between real points
in 3D and global holomorphic vector elds
Twistor theory and uids p. 19/41
The scalar Laplace equation
We consider a function f(, ) dened on twistor
space. This can then be thought of as restricted to the
special global sections of twistor space represented by
r
(), and the -dependence integrated out by
integration over a contour C. We set:
(r) =
_
C
f(
r
(), )d (34)
Note that what really matters is f modulo other
functions that are holomorphic inside or outside C.
Cauchys Theorem ensures that these do not
contribute.
Twistor theory and uids p. 20/41
Laplace II
It is easy to check that satises the scalar Laplaces
equation. Observe that
k
f(
r
(), )
x
k
= (1
2
)
k
k
f
k
|
=
r
(35)
k
f(
r
(), )
y
k
= i
k
(1 +
2
)
k
k
f
k
|
=
r
(36)
k
f(
r
(), )
z
k
= (2)
k
k
f
k
|
=
r
(37)
and that adding these three expressions with k = 2
gives zero identically for any choice of f.
Twistor theory and uids p. 21/41
Not just a modal trick
It would be wrong to think that this is just a trick for
building modes that solve Laplaces equation and
then adding them up by integration. There is much
more to it than that. See NH Monopoles paper -
consider curves in twistor space and eld of
osculating quadratics. The real part of this solves the
minimal surface problem. This is another story - but
the point is that the geometric structure is very
important.
Twistor theory and uids p. 22/41
The scalar biharmonic equation
How do we modify the integrand f(
r
(), ), say to
some holomorphic function g, to arrange that
4
g = 0 but
2
g = 0? We try to build h from f by
multiplying by some prefactor h(r, ), so that
g = h(r, )f(
r
(), ) (38)
Now
2
g = f
2
h + h
2
f + 2h.f (39)
That is, as f satises Laplace equation:
2
g = f
2
h + 2h.f (40)
Twistor theory and uids p. 23/41
Biharmonics II
If we furthermore choose h to be linear in r matters
simplify further and we have
2
g = 2h.f (41)
Let us set, w.l.o.g., h = u().r, so that h = u().
We also note that
f =
f
=
f
(1
2
, i(1 +
2
), 2). (42)
Twistor theory and uids p. 24/41
Biharmonics III
Putting this all together, we arrive at
2
g = 2u().(1
2
, i(1+
2
), 2)
f
= 2
u()
()
f
(43)
We can now see that
2
of this last expression
vanishes identically, while this expression does not
itself vanish unless
u()
() 0. (44)
Twistor theory and uids p. 25/41
Biharmonics IV
To see what is happening, we can now make matters
more explicit. We let u() = (u
1
(), u
2
(), u
3
()),
then
u.r = u
1
()x + u
3
()y + u
3
()z (45)
and
u()
() = (u
1
()+iu
2
())+2u
3
()(u
1
()iu
2
())
2
(46)
Twistor theory and uids p. 26/41
Biharmonics V
In terms of these variables the proposed integral
representation for solutions of the 3D scalar
biharmonic equation is just
=
_
C
d
_
xu
1
() + yu
2
() + zu
3
()
_
f(
r
(), )
(47)
or indeed, with w = x + iy:
=
1
2
_
C
d
_
wg
()+ wg
+
()+2zu
3
()
_
f(
r
(), )
(48)
where g
() = u
1
() iu
2
())
Twistor theory and uids p. 27/41
Relationship to 2D
Suppose we want no z-dependence. Set u
3
= 0 and
w = x + iy:
=
_
C
d
_
wg
() + wg
+
()
_
f(
r
(), ) (49)
= w
_
C
df
1
(
r
(), ) + w
_
C
df
2
(
r
(), ) (50)
Consider second term. This is w(x, y, z), where
(x, y, z) =
_
C
df
2
(
r
(), ) (51)
is a solution of Laplaces equation.
Twistor theory and uids p. 28/41
Relationship to 2D II
(x, y, z) =
_
C
df
2
(
r
(), ) (52)
We want this not to depend on z either. But this looks
awkward given
r
() = (x + iy) + 2z (x iy)
2
.
Actually it is not!
(x, y, z + h/2) =
_
C
df
2
(
r
() + h, ) (53)
The equation we need is
(x, y, z + h/2) = (x, y, z) (54)
Twistor theory and uids p. 29/41
Relationship to 2D III
This does not require that
f
2
(
r
() + h, ) = f
2
(
r
(), ) (55)
Instead we need
f
2
(
r
()+h, ) = f
2
(
r
(), )+g
0
(, , h)g
1
(, , h)
(56)
where g
0
is holomorphic on and inside C and g
1
is
likewise outside. (Cauchy!) Lets take C to be unit
circle. Now differentiate w.r.t h then set h = 0.
f
2
= G
0
(, ) G
1
(, ) (57)
Twistor theory and uids p. 30/41
Relationship to 2D IV
Integrated w.r.t. and divide by .
f
2
=
H
0
(, )
H
1
(, )
(58)
Now evaluate the integral using calculus of residues.
First term is easy, get
2iH
0
(
r
(0), 0) = K(w) (59)
and a contribution to the eld of wK(w). Calculation
of second piece gives another function of w, to be
multiplied by w. Other terms; we get
= wK(w) + wH( w) + w
K( w) + w
H(w) (60)
Twistor theory and uids p. 31/41
Relationship to 2D V
A fully holomorphic picture in 3D projects to 2D
picture and generates the familiar yet supercially
non-holomorphic 2D representation of solutions to
biharmonic and Laplace equations. In 3D our
functions are contour integrals.
Twistor theory and uids p. 32/41
Axis-symmetric Stokes Flow
This is traditionally modelled in terms of the Stokes
stream function
S
(r, ). The components of the
velocity eld are given by
u
r
=
1
r
2
sin()
, u
=
1
r sin()
S
r
(61)
What this representation is really telling us, as is
made clear in modern uid theory, is that the vector
potential for the ow is given by
=
S
r sin()
e
(62)
Twistor theory and uids p. 33/41
The PDE
For an axis-symmetric function f(r, )
(
f
r sin()
e
) =
1
r sin()
(D
2
f)e
(63)
where the operator D
2
is given by
D
2
f =
2
f
r
2
+
sin()
r
2
_
1
sin()
f
_
(64)
The biharmonic condition may be expressed as the
scalar PDE
D
4
S
= 0 (65)
Twistor theory and uids p. 34/41
Going to Cartesian Basis
e
=
ye
x
+ xe
y
_
x
2
+ y
2
=
ye
x
+ xe
y
r sin()
=
1
r sin()
{w[ie
x
+e
y
]}
(66)
where w = x + iy as before.
=
S
r sin()
e
(67)
and if
S
is real we can write the vector potential as
=
_
S
w
r
2
sin
2
()
[ie
x
+ e
y
]
_
(68)
Twistor theory and uids p. 35/41
Cartesian Analysis
Cartesian components must satisfy the scalar
biharmonic equation
S
= r
2
sin
2
()g(r, ) = w w
1
2i
_
d
1
f
_
_
(69)
for some complex function f. f can be expanded as a
Laurent series:
S
= w w
1
2i
_
d
n=
a
n
n+1
(w + 2z w
2
)
n
(70)
The terms in the series can be evaluated in terms of
Legendre functions.
Twistor theory and uids p. 36/41
Adding Harmonic Components
A corresponding analysis of the purely harmonic
pieces gives their contribution as
SH
=
w
2i
_
d
_
_
+
w
2i
_
d
1
_
(71)
for some choice of complex functions and . Can
be worked out in spatial terms.
Twistor theory and uids p. 37/41
The Stokes stream function
The contour integral solution for the Stokes stream
function for axis-symmetric biharmonic ow:
S
= w w
1
2i
_
d
1
f
_
_
+
w
2i
_
d
_
_
+
w
2i
_
d
1
_
(72)
where is written in terms of x, y, z and where f, ,
have Laurent series expansions that generate
expansions in terms of powers of r and regular (f)
and modied (, ) functions.
Twistor theory and uids p. 38/41
Simple Twistor Functions
f constant gives a contribution to
S
proportional to
w w = r
2
sin
2
() (73)
f(z) = 1/z gives a Coulomb eld and a contribution
to
S
proportional to
w w
r
= r sin
2
() (74)
(z) = 1/z
2
, a contribution to
S
of the form
w w
r
3
=
1
r
sin
2
() (75)
Twistor theory and uids p. 39/41
Does it all add up?
The particular combination
S
=
U
2
sin
2
()
_
r
2
3ar
2
+
a
3
2r
_
(76)
gives the well-known stream function for very viscous
ow around a sphere of radius a and uniform ow at
at rate U at innity. In general we have a novel
technique for solving the PDE
D
4
S
= 0 (77)
Twistor theory and uids p. 40/41
Where next?