Professional Documents
Culture Documents
Manjunath Krishnapur
Indian Institute of Science
A BSTRACT. Lecture notes from SLE seminar in Summer and Fall of 2010 at IISc,
Bangalore. Mainly, we follow (some chapters of) Wendelin Werners St. Flour lecture
notes on SLE, but filling in most proofs. Much of this latter is in Greg Lawlers book
from which we borrow a lot of things. Some proofs are a little different. However, we
want to keep it streamlined to a minimal introduction to SLE. In particular, we stick
to chordal versions of all theorems.
Contents
Chapter 1. Complex analysis background
1.1. Boundary behaviour of conformal maps
1.2. The standard setting
1.3. Brownian motion
1.4. Harmonic measure
1.5. Beurlings theorem
1.6. Schwarzs lemma and its many variants
1.7. Half-plane capacity
1.8. A heuristic derivation of Lwners equation
1.9. Closeness of conformal maps
1.10. Chordal version of Lwners differential equation
1.11. Continuously growing hulls
1.12. Lwner evolution in reverse
1
1
2
3
4
5
7
8
9
11
13
14
15
CHAPTER 1
2
3
(Im z0 ) |0 ( z0 )| 1 + a y0 + O ( y0 )
=
Im ( z0 )
1 a y02 + O ( y03 )
where in the end we chose z0 = i y0 . Letting y0 , we see that a must be nonnegative. Further, a = 0 if and only if K is empty.
Remark 1.2.4. Schwarzs lemma (or Schwarz-Pick lemma) has the following meaning. Consider the disk with the hyperbolic metric (1 | z|2 )2 ( dx2 + d y2 ). If f :
| f 0 ( z )| 2
( dx2 + d y2 ).
(1| f ( z)|2 )2
Schwarz-Pick
we can see that K ( z) = z + rz . A not so trivial exercise is to find K for the oblique
slit [0, te i ] for some 0 < < and t > 0.
1.3. Brownian motion
What is Brownian motion? We give a quick definition and a convenient way to
visualize it. It can be shown that Brownian motion does exist.
Definition 1.3.1. Brownian motion in the plane is W = (Wt ) t0 is a collection of
complex-valued random variables on a probability space such that (a) W0 = 0 w.p.1.
(b) For any t 1 < t 2 < . . . < t k the random variables Wt i Wt i1 are independent, and
for any s < t the real and imaginary parts of Wt Ws are independent N (0, t s)
distributed. (c) The function t Wt is a continuous function w.p.1.
A convenient way to visualize Brownian motion is topfix h > 0, imagine a particle
which picks a random direction and moves a distance 2 h p
in a time duration of h,
then picks a random direction again and moves a distance 2 h etc. As h 0, this
random trajectory converges to Brownian motion in a precise sense.
What we really use in the sequel are the following basic properties of Brownian
motion.
d
d p
(1) Symmetries: e i W () = W () (rotation invariance). W ( r ) = rW () (scale ind
(3) Assume that for every , there is a set K as in (2). Then, for any
f C (), the function u is the unique solution to the
Dirichlet problem: Find v C () such that v = 0 in , and v = f on .
P ROOF. (1) We leave the part about P z ( < ) = 1 to Exercise 1.3.3. Fix z and
r such that z + r . Define the stopping
y
dt
( x t )2 + y2
1 1| z|2
2 | z e it |2
is the Poisson
These examples are atypical. Usually we cannot find z exactly but only estimate it. The next section will introduce one such estimate which will be of great use
to us.
1.5. Beurlings theorem
If K is a compact subset of the plane we let K := {| z| : z K } be its radial projection.
Theorem 1.5.1 (Beurlings projection theorem and estimate). Let K be compact
and let V = \K and let V = \K .
p
| z |.
P ROOF.
(1) By rotating we may assume that z = | z|. Let L = { xe i : x
R}. Apply Lemma 1.5.2 with the lines L /2 , L /4 , L /8 , . . . successively to
get sets K 1 = R /2 (K ), K 2 = R /4 R /4 (K 1 ), K 3 = R /8 R /8 (K 2 ) etc. with
Vj
z (K j ) increasing in j . Then, K j is confined to the sector {2 j arg z
Vj
Lemma 1.5.2. Let z and let K be a compact subset of . Let L be a line passing
through 0 and let R L denote reflection across L. Let K + be the part of K on the same
S
side as z and let K be the part of K on the other side. Let K := K R L (K + ). Set
V = \K and V = \K . Then, V (K ) V (K ).
z
V
z (K ) = P z (K ) + P z ( AK ) + P z ( IK ) + P z ( AIK ) + P z ( I AK ) + P z ( AI AK ) + . . .
The sum is over all strings in which A and I alternate a number of times and is
V
z (K ) = P z ( I K ) + P z (BI K ) + P z ( IBI K ) + P z (BIBI K ) + . . .
(1.5.2)
We show that each term here is equal to one of the terms in (1.5.1). More precisely,
Claim :
A I
and similarly
P z (BIBI K ) =
Z Z Z Z
B I B I
U
V
U
V
U
z ( dw1 )w1 ( dt 1 ) t 1 ( dw2 )w2 ( dt 2 ) t 2 (K ).
V
By considering reflection as before, V
t 2 (K ) = t 2 (K ) for any t 2 I . Further, as B =
R ( A ), the integrals are also equal (make a change of variables from w i to R (w i )).
Beurlings estimate is very useful to us later. Here is one example that we use
later.
Corollary 1.5.3. Let be a s.c. domain (other than the whole plane). Then, for any
z , if d = dist( z, c ), then for any x > 0 we have
c0
P z (W travels at least distance xd away from z before exiting ) p .
x
P ROOF. Let with | z | = d . Consider B(, xd ) with K = . By assumption that is s.c., it follows that K is connected and connects to B(, xd ) (if xd is
small enough, which is all we need to consider). Apply Beurlings estimate.
A standard application of harmonic measure in complex analysis is to PhragmenLindelf type theorems. This may be omitted and is included only for general interest.
Corollary 1.5.4. Let f be an analytic function on a neighbourhood of = {0 Re z
1} and suppose that | f ( z)| M R := exp{exp{CR 1 }} for some > 0. If | f ( z)| M on
{Re z = 0 or 1} and Then, | f ( z)| M for all z .
P ROOF. For R > 0, let S R := [0, 1] [R, R ] and let u( z) be a harmonic function
on S R whose boundary values are equal to log | f |. Since log | f | is subharmonic, we
get log | f ( z)| u( z) for any z S R . Fix z and let z (R ) denote the harmonic measure of the top and bottom edges of S R , as seen from z. Thus, log | f ( z)| u( z)
z (R ) log M R + log M , by the Brownian motion solution to the Dirichlet problem. We
need an estimate for z (R ). Divide S R into b2R c squares of side 1 and argue that
z (R ) e cR for some c > 0. Letting R we get | f ( z)| M .
| f ( z) f (w)|
|1 f ( z) f (w)|
| f 0 ( z)|
| z w|
1
and
for any z, w
|1 zw|
1 | f ( z)|2 1 | z|2
with strict inequalities everywhere unless f is one-one and onto the disk. These
inequalities follow from Schwarzs statement by pre-composing (and post-composing)
f with linear fractional transformation that takes 0 to z (respectively, f ( z) to 0).
In the form given by Pick, Schwarzs lemma attains a deep geometric meaning.
On the disk , define a Riemannian metric (1| z|2 )2 ( dx2 + d y2 ) called the hyperbolic
R1
metric. This means that an curve : [0, 1] has length `() = 0 (1 | t |2 )1 | t | dt.
The length minimizing curves are called geodesics and they may be seen to be exactly
arcs of circles that intersect at right angles (this includes diameters of which
are arcs of circles of infinite radius).
If we call each z a point and each geodesic a line, then this system of points
and lines satisfy the axioms of the hyperbolic geometry discovered by Bolyai, Gauss
and Lobachevsky. The axioms are the same as that of Euclidean geometry, except for
the parallel postulate, which is replaced by a statement that given a line and a point
not on it, there exist at least two lines through the point that do not intersect the given
line. We say that the disk with the hyperbolic metric forms a model for hyperbolic
geometry.
Exercise 1.6.1. Show that with the metric y2 ( dx2 + d y2 ) is isometric to the
hyperbolic metric on . What are the geodesics?
Returning to Schwarz-Pick lemma, the inequalities 1.6.1 state precisely that any
holomorphic map from the disk to itself is a contraction of the hyperbolic metric. And
if it is not strictly contracting even at one point, then it must be an isometry of the
disk, in other words a linear fractional transformation that maps the disk injectively
onto itself. Here are some variants of Schwarzs lemma that we shall need.
Lemma 1.6.2.
f ( z ) f ( w) | z w|
| f 0 ( z)|
1
and
f ( z ) f ( w) | z w|
Im f ( z) Im z
(1) Let ( z) =
z i
z+ i
(2) Derive this one from part (3) using a linear fractional transformation that
maps a and b to 0 and . We have not omitted absolute values by mistake.
Both f 0 (a) and f 0 ( b) must be positive as f maps (a, ) and ( b, )
into .
(3) must be positive as f maps into . Without loss of generality take
x = 0 = f ( x) and write f ( z) = z + O ( z2 ) as z 0 so that f 0 (0) = which is
also positive as f maps into near 0. Now take z = i y and w = iv and
apply part (1) to get
i y i v + O (1) | i y iv + O (1)|
.
i y + i v + O (1) | i y + iv + O (1)|
y
The left hand side is 1 2 v + O ( y2 /v2 ) while the right hand side is 1 2 v +
O ( y2 /v2 ) from which it follows that as claimed.
S
(3) Let K := inf{ t : Wt K R}. Then, a(K ) = lim yE i y Im WK .
y
P ROOF.
(1) We saw that a(K ) 0 with equality if and only if K = ;. Now,
let K 3 := K 1 (K 2 \K 1 ). Then, K 2 = K 3 K 1 from which it follows that
a(K 2 ) = a(K 1 ) + a(K 3 ) a(K 1 ) with equality if and only if K 3 = ; which is
equivalent to K 1 = K 2 .
2
)
2
(2) True, since rK + t ( z) = r K zr t + t = ( z t) + r za(K
t + O (( z t) ) + t = z +
r 2 a( K )
+ O ( z2 ).
z
Im{K ( z) z} is
R
Exercise 1.7.4. If K is a hull with r (K ) < 1, then a(K ) = 2 E e it [Im B ] sin t dt.
(3)
With the probabilistic interpretation, we can estimate a(K ) using Brownian motion
hitting probabilities which in turn are estimated by Beurlings estimate. Thus we
have the following fact.
Theorem 1.7.5. Let K 1 , K 2 be two hulls.
S
(1) If K 1 K 2 and every point of (\K 2 ) (which may be smallerpthan K 2 R)
S
is within distance of K 1 R. Then a(K 1 ) a(K 2 ) a(K 1 ) + c 0 r (K 2 )3/2 .
(2) K 1 , K 2 general with r (K i ) r . If every point of (p\K 2 ) is within distance
S
of K 1 R and vice versa, then |a(K 1 ) a(K 2 )| c 0 r 3/2 .
P ROOF.
(1) We know that a(K 1 ) a(K 2 ). To show the second inequality,
start Brownian motion W at i y and let 2 1 be the hitting times of
S
S
S
(2 r 2 ) R, of K 2 R and of K 1 R, respectively.
Let A := {|W | = 2 r 2 }. If A c occurs, then W R and 1 = 2 = and
hence Im W2 Im W1 = 0. On the event A := {|W | = 2 r 2 }, we may deduce
p
yE i y Im W2 yE i y Im W1
= yP i y ( A ) E i y Im W2 Im W1 A
Z r2 p
c0 r 2
p dx = c 0 1/2 r 3/2
2 .
x
0
S
(2) Immediate upon applying the first part to the pairs K 1 K 1 K 2 and K 2
S
K1 K2.
g t ( z) =
a t
g t ( z) U t
where a t = a([0, t]), U t = g t ( t ) and a t , g t denote time ( t) derivatives. Among several issues raised by this statement, we have addressed a few. For example U t is
well-defined because t corresponds to only one prime end in \[0, t]. What about
10
differentiability of a t ? Theorem 1.7.3 shows that a t is strictly increasing and Theorem 1.7.5 shows that a t is continuous. Therefore, can always be reparameterized
so that a t is smooth. In particular, if we set s = a1 (2s) , defined for s < a(), then
a([0, s]) = 2 s. is called the natural parameterization of .
Assume now that (0) is in its natural parameterization, that is a t = 2 t. Fix t
and let s = g t ( t+s ). If G s = [0,s] , then for any h > 0we have g t+h = G h g t . From
this, it is clear that is also a simple curve in its natural parametrization (see the
proof of the firstppart of Lemma 1.7.3). If we heuristically replace [0, h] by a vertical
h] having the same half-plane capacity 2 h, then by Example 1.2.5,
slit [U t ,U t + i 2 p
we get g t+h ( z) ( g t ( z) U t )2 + 4 h. Differentiate by h and set h = 0 to get
2
g t ( z) =
g t ( z) U t
which is exactly Lwners equation! If you have been able to calculate the conformal
maps for oblique slits asked for in Example 1.2.5, then check that replacing [0, h]
by any oblique slit with the same half-plane capacity 2 h will also result in the same
differential equation.
The gaps in this heuristic are that we only calculated the right derivative of g t
above and that we did not justify how to replace [0, h] by a slit. Both these issues
are addressed by proving several estimates in the next section about how conformal
maps K differ if the hulls K differ only a little. Then we put together the proof in
section 1.10.
11
x K ( x) x +
1
1
for x 1 and x K ( x) x + for x 1.
x
x
Z
2
=
E e it [Im W ] p( z, e it ) sin t dt
1
| z|
Z2
2
Im( z)
E e it [Im W ] sin tdt c 0 a(K )
.
| z |3
0
12
Now, consider |w| 2. Then the disk (w, |w|/2) is contained in {| z| > 1} and
hence, we can use Poisson integral formula to write for |w0 w| < R := |w|/2,
Z2
R 2 |w0 |2 dt
v(w ) = v(w + R e it ) 0
|w R e it |2 2
0
from which we differentiate with respect to x (or y) and set w0 = w and use the bounds
on v (that holds for all t since |w + R e it | |w|/2 1). We get
1 Im w
1
max |v x (w)|, |v y (w)| c 0 a(K )
c0
.
3
R | w|
| w |3
Since v = Im g, we get | g0 (w)| = |v y (w) + iv x (w)| c 0 a(K ) |w1|3 . Note that g() = 0.
Therefore, integrating g0 along a path from z to we get
Z
Z
a( K )
| z + it|3
.
t c0
| g( z)| | g0 ( z + it)| dt c 0 a(K )
d
| z |2
0
P ROOF. Let | z| > 1. For any arc I , the harmonic measure z ( I ) is the value at
S
S
z of the harmonic function which has boundary values 1 on I and 0 on R {} (\ I ).
By conformal invariance of harmonic functions, we can map it forward by z z + 1/ z
which takes to [2, 2]. By knowledge of the Poisson kernel on , the hitting density
w
of W , started from w = z + 1/ z is given by p( s) = |Im
for s R. Pulling it back to
w s |2
the original problem, we find that
Im( z + z1 )
2 sin t
p( z, e it ) =
z + z1 2 cos t 2
which can easily be seen to be equal to Im(1/ z) 2 sin t (1 + O (1/| z|)).
13
Exercise 1.9.7. Let K be a connected set in the plane connecting 0 to . Then for
p
any , we have P (W hits before hitting K ) c 0 .
1.10. Chordal version of Lwners differential equation
Let be a simple curve in with 0 = 0 and t for t > 0. Let g t : \[0, t]
be the unique conformal map such that g t ( z) = z + azt + O ( z2 ) as z . Here
3 r (K t,s U t )
(by Theorem 1.9.2 applied to K t,s U t )
p
C ,T dia([ t, s])
(by Theorem 1.9.5).
g s ( z) g t ( z) 2( s t) = h t,s (w) w 2( s t)
(where w = g t ( z))
g t ( z) U t
w Ut
( s t)dia(K t,s )
(1.10.1)
C ,T
| g t ( z ) U t |2
where the last inequality is implied by Theorem 1.9.2 if | g t ( z) U t | > 2 r (K t,s ).
If we fix and z \K s and let t s, then the last condition holds for t close
enough to s. Divide (1.10.1) by t s and let t s to get g s ( z) = g s ( z2)Us . Here we used
14
g t ( z) =
2
g t ( z)U t .
Exercise 1.11.2. Let be a curve in and let K t be the hull generated by [0, t],
that is \K t is the unique unbounded component of \[0, t]. Assume that K t , K s
for any t < s.3 Then K t is strictly increasing and right continuous.
The left continuity of U and the left derivative of g t dont follow automatically.
For example, if as in the above exercise crosses itself, the U jumps. To be specific,
3If K is an increasing family of hulls, it can be reparameterized so that it is strictly increasing. But
t
if K is also generated b a curve , then it may not be possible to reparameterize so that K t is strictly
increasing in t.
15
All this will be done locally. Inductively it is clear that each X n is analytic in x.
Picards iterates converge locally uniformly to the solution of the ODE. Therefore
the solution is also holomorphic.
Thus we have shown that g t are conformal maps from \K t onto . For fixed t
and large z, we can argue (How?) that g t ( z) = 2/ g t ( z) + O ( g t ( z)2 ) and hence g t ( z) =
z + 2zt + O ( z2 ). Thus K is in its natural parameterization.
16