S.-Y. Lan and D.-Q.

Dai
Nagoya Math. J.
Vol. 194 (2009), 149167
C

-CONVERGENCE OF CIRCLE PATTERNS TO

MINIMAL SURFACES
SHI-YI LAN and DAO-QING DAI
Abstract. Given a smooth minimal surface F : R
3
dened on a simply
connected region in the complex plane C, there is a regular SG circle pattern
Q

. By the Weierstrass representation of F and the existence theorem of SG

circle patterns, there exists an associated SG circle pattern P

in C with the
combinatoric of Q

. Based on the relationship between the circle pattern P

and the corresponding discrete minimal surface F

: V

R
3
dened on the
vertex set V

of the graph of Q

, we show that there exists a family of discrete

minimal surface

: V

R
3
, which converges in C

() to the minimal
surface F : R
3
as 0.
1. Introduction
The theory of discrete dierential geometry is presently emerging on the
border of dierential and discrete geometry, which studies geometric shapes
with a nite number of elements (polyhedra) and aims at a development of
discrete equivalents of the geometric notions and methods of surface theory
(see [1], [2], [3], [4], [8], [12], etc.). A smooth geometric shape (such as
surface) appears then as a limit of the renement of the discretization. One
of the central problems of discrete dierential geometry is to nd proper
discrete analogues of special classes of surfaces, such as minimal, constant
mean curvature, isothermic, etc. In [2], a new discrete model was intro-
duced to investigate conformal discretizations of minimal surface, i.e., the
analogous discrete minimal surfaces consisting of touching spheres, and of
circles which intersect the spheres orthogonally in their points of touch. It is
proved that the discrete minimal surfaces converge to the smooth ones. The
advantages of the discretizations are that they respect conformal properties
of surfaces, possess a maximum principle, etc. Here, we are concerned with
the C

-convergence of discrete minimal surfaces given in terms of circles

Received June 2, 2006.
Revised March 19, 2008.
Accepted September 7, 2008.
2000 Mathematics Subject Classication: 52C26, 53A10, 53C42.
150 S.-Y. LAN AND D.-Q. DAI
and spheres, that is, the convergence of discrete minimal surface discussed
in [2] is extended to C

-convergence.
For each > 0, let SG

denote the square grid with mesh > 0. The

vertices of SG

form the square lattice V

= {n + mi : (n, m) Z Z},
and an edge connects any two vertices of SG

at distance . The 1-skeleton

of SG

is the graph of regular SG circle pattern Q

each circle of which has

radius equal to /

2. Aussume that is a simply-connected domain in C

with = C. Let Q

be the largest sub-pattern of Q

that is contained
in , and let SG

be the sub-complex of SG

whose 1-skeleton is equal to

the graph of Q

. The vertex set of SG

is denoted by V

. Suppose that
F : R
3
is a minimal immersion without umbilic points in conformal
curvature line coordinates. First, by the Weierstrass representation of F and
the local theory of SG circle patterns (see [12, 6]), there is an associated
SG circle pattern P

in C with the combinatoric of Q

. In the meantime,
one gets a discrete minimal surface F

: V

R
3
corresponding to F, which
consists of spheres and circles. Secondly, in terms of Mobius invariants of
circle pattern P

, we dene the discrete Schwarzians of P

and show that

they are uniformly bounded in C

(). Thirdly, we construct a Mobius

transformation T

through circle pattern P

such that they can be expressed

by the discrete Schwarzians. By the C

-boundedness of the Schwarzians,

we will prove that T

converges in C

() to some Mobius transformation

T as 0. Lastly, using the relation between the discrete minimal surface
F

and circle pattern P

, and combining with the denition and the C

-
convergence of T

, we will show that there exists a family of discrete surface

: V

R
3
, obtained by scaling appropriately the centers of spheres and
circles in F

, which converges in C

() to the minimal surface F : R

3
as 0.
This paper is organized as follows. For a given smooth minimal surface
F : R
3
, we will give the associated SG circle patterns P

, the cor-
responding discrete minimal surfaces F

and the relation between them in

Section 2. In Section 3, we rst give the denitions of C

-convergence and
C

-boundedness for discrete functions. Then by using Mobius invariants

of P

, we dene the discrete Schwarzians of P

and show that they are

uniformly bounded in C

(). In Section 4, we construct the Mobius trans-

formations T

through circle pattern P

such that they can be expressed

by the discrete Schwarzians. Then it is proved that T

converges in C

()
to some Mobius transformation T as 0. In Section 5, by the relation
between the discrete minimal surface F

and the circle pattern P

, we will
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 151
show that there exists a family of discrete minimal surface

: V

R
3
which converges in C

() to F : R
3
as 0.
2. Circle patterns and discrete minimal surfaces
In this section, for any smooth minimal surface, we apply its Weierstrass
representation to yield associated circle patterns with the combinatorics of
square grid and dene corresponding discrete minimal surfaces. Moreover,
we give the relationship between the circle patterns and the discrete minimal
surfaces (also see [12], [2] for more details).
For each positive number > 0, let SG

be the cell complex whose

vertices form the square lattice V

= Z + iZ = {v = n + im : (n, m)
ZZ}, whose edges are the pair [v, v

] such that |v v

| = and v, v

V

,
and whose 2-cells are the squares {v +x +iy : x, y [0, ]}, v V

.
An indexed collection P

= {P

(v) : v V

} of oriented circles in the

Riemann sphere

C is said to be a circle pattern with the combinatorics of
square grid SG

(or a SG circle pattern) if the following three conditions

hold.
(a) Whenever v, v

are neighbors in SG

, the corresponding circles

P

(v), P

(v

) intersect orthogonally.
(b) If v
1
, v
2
are neighbors of a vertex v in SG

, and they belong to

the same square of SG

, then the circles P

(v
1
), P

(v
2
) are distinct and
tangent.
(c) Whenever the situation is as in (b) and v
2
is neighbor of v, which is
one step counterclockwise from v
1
, the circular order of the triplet of points
P

(v) P

(v
1
) P

(v
2
), P

(v
1
) P

(v
2
), P

(v) P

(v
2
) P

(v
1
) agrees
with the orientation of P

(v).
Clearly, the 1-skeleton of SG

is the graph of regular SG circle pattern

Q

each circle of which has radius equal to /

2.
Suppose that is a simply connected domain in the complex plane C.
Without loss of generality, we may assume 0 . Let

Q

be the largest sub-

pattern of Q

that is contained in . Let Q

be the connected component

of

Q

that contains 0, and let SG

be the cell complex whose 1-skeleton is

equal to the graph of Q

. The set of vertices of SG

is denoted by V

, and
the set of centers of squares of SG

by

V

.
A smooth immersed surface in R
3
is said to be isothermic if it admits a
conformal curvature line parametrization in a neighborhood of every non-
umbilic. An isothermic immersion is a minimal surface if and only if the
dual immersion is contained in a sphere. In that case the dual immersion is
152 S.-Y. LAN AND D.-Q. DAI
in fact the Gauss map of the minimal surface, up to scale and translation.
We rst give the following lemma about the Weierstrass representation of
minimal surfaces, which follows from [2].
Lemma 1. Suppose that F : R
3
is a minimal immersion without
umbilic points in conformal curvature line coordinates. Then
(1) F =
_
Re
_
1 f(z)
2
f

(z)
dz, Re
_
i(1 +f(z)
2
)
f

(z)
dz, Re
_
2f(z)
f

(z)
dz
_
,
where f : C is a locally injective meromorphic function.
For the locally injective meromorphic function f in Lemma 1, set
(2)
(1)

(v) = 1 +
2
Re S
f
(v)
for each boundary vertex v V

, where S
f
denotes the Schwarzian deriva-
tive of f. Then SG

Dirichlet principle [12, Theorem 6.2] implies that

(1)

is extended to a solution of the SG

-Dirichlet problem on SG

. Let
(2)

be
the companion of
(1)

in the solution of the SG

CR equation [12, 5],

i.e.,
(3)

(2)

(v +
0
)

(2)

(v +
1
)
=
_

(1)

(v +i)
1
+ 1

(1)

(v)
1
+ 1
_
2
and
(4)

(2)

(v +
0
)

(2)

(v +
3
)
=
_

(1)

(v +) + 1

(1)

(v) + 1
_
2
for any v V

, such that
(5)
(2)

(
0
) = 1 +
2
ImS
f
(
0
),
where
j
= i
j
(/2 +i/2) (j = 0, 1, 2, 3).
Based on the local theory of SG circle patterns [12, Theorem 6.1], there
exists an SG circle pattern P

= {P

(v) : v V

} for SG

in the com-
plex C that has
(1)

and
(2)

as its Mobius invariants. That is,

(1)

(v) =
cr[q
0
(v), q
2
(v); q
3
(v), q
1
(v)] for any v V

and
(2)

(u) = cr[q
j
(v +
), q
j
(v ); q
j
(v i), q
j
(v + i) for any u = v +
j


V

, where q
j
(v)
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 153
(j = 0, 1, 2, 3) denotes the point of contact with circles P

(v) and P

(v+2
j
)
and cr[q
1
, q
2
, q
3
, q
4
] denotes the cross ratio of the four points q
1
, q
2
, q
3
, q
4


C.
On the other hand, we further suppose that there are two kinds of
vertices v
s
and v
c
in V

such that each edge of SG

has vertices of dierent

kinds. Then a discrete isothermic surface is a mapping
F

: V

R
3
such that images F

(v
s
) and F

(v
c
) of v
s
and v
c
are spheres and circles
respectively. Spheres F

(v
s
) and circles F

(v
c
) intersect orthogonally if v
s
and v
c
belong to the same edge of SG

, and spheres F

(v
(1)
s
) and F

(v
(2)
s
)
(respectively, the circles F

(v
(1)
c
) and F

(v
(2)
c
)) are distinct and tangent if
v
(1)
s
and v
(2)
s
(respectively, v
(1)
c
and v
(2)
c
) belong to the same face of SG

.
Let

F

(v) be the center of sphere (or circle) F

(v) for each v V

,
p
j
(v
s
) be the intersection points of spheres F

(v
s
) and F

(v
s
+
j
) (j =
0, 1, 2, 3), and let p
j
(v
s
) =

F

(w
s
) + b
j
. Then a discrete isothermic surface
F

: V

R
3
is called a discrete minimal surface if it satises any one of
the equivalent conditions below.
(a) The points

F

(v
s
) + (1)
j
b
j
lie on a circle.
(b) There is a d R
3
such that (1)
j
(b
j
, d) is the same for j = 0, 1, 2, 3.
(c) There is a plane through

F

(v
s
) such that the points {p
j
(v
s
) | j =
0, 2} and the points {p
j
(v
s
) | j = 1, 3} lie in planes which are parallel to it
at the same distance on opposite sides.
From the denition above, it is easy to see that a discrete isothermic
surface is a discrete minimal surface, if and only if the dual discrete isother-
mic surface corresponds to a Koebe polyhedron (see [2, 4]). Let A

(v)
denote the center of circle P

(v) in the circle pattern P

for each v V

.
Then the following lemma, which can follow from [2, 5], gives the relation
between the discrete minimal surfaces F

and the circle patterns P

.
Lemma 2. For any vertices v
(1)
s
and v
(2)
s
that belong to the same square
of SG

, let

F

(v
(1)
s
) and

F

(v
(2)
s
) denote the centers of spheres F

(v
(1)
s
) and
F

(v
(2)
s
) in the discrete minimal surface F

, respectively. Then

(v
(1)
s
)

F

(v
(2)
s
) (6)
= Re
_
R(v
(1)
s
) +R(v
(2)
s
)
1 +|q|
2
A(v
(1)
s
) A(v
(2)
s
)
|A(v
(1)
s
) A(v
(2)
s
)|
(1 q
2
, i(1 +q
2
), 2q)
_
,
154 S.-Y. LAN AND D.-Q. DAI
where q denotes the point of contact with circles P

(v
(1)
s
) and P

(v
(2)
s
) in
the circle pattern P

and the radii R(v

(j)
s
) of the sphere F

(v
(j)
s
) (j = 1, 2)
are
R(v
(j)
s
) =

1 +|A(v
(j)
s
)|
2
|A(v
(j)
s
q|
2
2|A(v
(j)
s
q|

. (7)
For the discrete minimal surface F

, we dene a discrete surface

F

:
V

R
3
by

F

(v) =

F

(v) for each v V

. Then

F is called a dis-
crete minimal surface comprised of points and F

is called one consisting

of spheres and circles for distinction. Next, we will show that after scaling
appropriately,

F

converges in C

() to F : R
3
.
3. The C

-boundedness of discrete Schwarzians

In this section we rst give some denitions and notations related to
discrete dierential operators (also see [8]). Then using the Mobius invari-
ants
(1)

and
(2)

given in Section 2, we dene the discrete Schwarzians of

circle pattern P

and show that they are uniformly bounded in C

().
Let W be a subset of V

. A vertex v W is said to be an interior

vertex of W if all its neighboring vertices are contained in W. Let int W
denote the set of the interior vertices of W. Given a function : W R,
we dene the discrete directional derivative
j
: int W R by

j
(v) = ((v +i
j
) (v))/,
for each j Z
4
. The discrete Laplacian of a function : W R is a
function in int W dened by the formula

(v) = 1/(4
2
)
3

j=0
((v +i
j
) (v)).
For any dierentiable function H : R, let
j
H denote the direc-
tional derivate

j
H(z) = lim
s0
H(z +i
j
s) H(z)
s
,
where j = 0, 1, 2, 3. Let f : C
d
be a function dened in . For each
> 0, let f

be a function dened on some set of vertices

V

, with
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 155
values in C
d
. Assume that for each z , there are some
1
,
2
> 0 such
that {v V

: |v z| <
2
}

V

whenever (0,
1
).
If for every z and every > 0, there are some
1
,
2
> 0 such that
|f(z) f

(v)| < , for every (0,

1
) and every v V

with |v z| <
2
,
then we say f

converges to f, locally uniformly in .

Let n N, and suppose that f is C
n
-smooth. If for every sequence
j
1
, j
2
, . . . , j
k
Z
4
with k n we have

j
k

j
k1

j
1
f

j
k

j
k1

j
1
f
locally uniformly in , then we say that f

converges to f in C
n
(). If that
holds for all n N, then the convergence is C

. The functions f

are said
to be uniformly bounded in C
n
() provided that for every compact K
there is some constant C(K, n) such that

j
k

j
k1

j
1
f

KV

< C(K, n)
whenever k n, and is suciently small, where denotes the L

-norm.
The functions f

are uniformly bounded in C

(), if they are uniformly

bounded in C
n
() for every n N.
For a smooth vector function F : R
3
and a discrete vector function
F

: V

R
3
, it is said that F

converges in C

() to F if every component
of F

converges in C

() to the corresponding one of F, and that F

are
uniformly bounded in C

() if all components of F

are uniformly bounded

in C

().
For
(1)

and
(2)

given in Section 2, we dene the two discrete Schwarz-

ians of P

as follows: let
(8) h
(1)

(v) =
2
(
(1)

(v) 1)
for each vertex v V

and let
(9) h
(2)

(u) =
2
(
(2)

(u) 1)
for every u

V

. It is easy to see that h

(1)

= h
(2)

0 if P

is a regular SG
circle pattern, because
(1)

=
(2)

1 for regular SG circle patterns.

Let SG

h
denote (1/2)SG

+ (1 +i)/4, and the set of vertices of SG

h
is denoted by V

h
. For any w V

h
, let v V

be the unique vertex of SG

that is closest to w. Let M

w
be the Mobius transformation that takes ,
0, 1 to q
0
(v), q
1
(v), q
3
(v), respectively. Set
M
[w
1
,w
2
]
= M
1
w
1
M
w
2
for each directed edge [w
1
, w
2
] of SG

h
. Write e
j
(v) = [v +
j
/2, v +
j+1
/2]
for any v V

and for any j Z

4
, then we have
156 S.-Y. LAN AND D.-Q. DAI
Lemma 3. There hold the following equalities
M
e
0
(u)
(z) = M
e
2
(u)
(z) = 1 i(h
(2)

(u) + 1)
1/2
z, (10)
M
e
1
(u)
(z) = M
e
3
(u)
(z) = 1 i(1/)(h
(2)

(u) + 1)
1/2
z (11)
for each u

V

and
M
e
0
(v)
(z) = M
e
2
(v)
(z) = (
2
(h
(1)

(v) + 1)
1
+ 1)
1
/(1 z), (12)
M
e
1
(v)
(z) = M
e
3
(v)
(z) = (
2
(h
(1)

(v) + 1) + 1)
1
/(1 z) (13)
for every v V

.
Proof. For any u

V

, we rst consider the directed edge e

2
(u) =
[u +
2
/2, u +
3
/2] SG

h
. By the denition of M
[w
1
,w
2
]
, we get that
M
[w
1
,w
2
]
does not change if we apply a Mobius transformation to P

. So we
may assume that M
u+
2
/2
is the identity. It follows from the denition of
M
w
that
q
1
(v) = , q
1
(v ) = 0, q
1
(v i) = 1,
where v V

is the unique vertex of SG

that is closest to u+
3
/2. Hence
we deduce that the four points q
1
(v ), q
1
(v i) form the vertices of a
rectangle. From the denition of
(2)

in Section 2, we get

(2)

(u) =
_
q
1
(v +) q
1
(v i)
q
1
(v i) q
1
(v )
_
2
,
which implies
q
1
(v +) = q
1
(v i) i(
(2)

)
1/2
(q
1
(v i) q
1
(v )) = 1 i(
(2)

)
1/2
.
Since M
e
2
(u)
takes , 0, 1 to q
1
(v), q
1
(v i), q
1
(v + i), respectively, we
conclude from (9) that
M
e
2
(u)
(z) = 1 i(
(2)

)
1/2
z = 1 i(h
(2)

(u) + 1)
1/2
z.
Similarly, we deduce that M
e
0
(u)
(z) = 1i(h
(2)

(u) +1)
1/2
z. So (10) holds.
Identical to the above arguments, we conclude that (11) holds.
Next, we consider the directed edge e
0
(v) = [v +
0
/2, v +
1
/2] SG

h
for any v V

. We assume with no loss of generality that M

v+
0
/2
is the
identity. Then we obtain
q
0
(v) = , q
1
(v) = 0, q
1
(v) = 1.
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 157
By the denition of
(1)

, we get q
2
(v) = 1/(
(1)

(v) + 1). Note that M

e
0
(v)
takes , 0, 1 to q
1
(v), q
1
(v), q
1
(v), respectively, combining with (8), we
deduce
M
e
0
(v)
= ((
(1)

(v))
1
+ 1)
1
/(1 z)
= (
2
(h
(1)

(v) + 1)
1
+ 1)
1
/(1 z).
Similarly, we get
M
e
2
(v)
(z) = (
2
(h
(1)

(v) + 1)
1
+ 1)
1
/(1 z).
Thus (12) holds. With the same arguments as above, we get that (13) holds.
Lemma 4. (i) The equality
(14) h
(1)

(v) = Re S
f
(v) +
2
O(1)
holds for each v V

.
(ii) The equality
(15) h
(2)

(u) = ImS
f
(u) +

(u)
2
O(1)
holds for each u

V

, where

(u) denotes the combinatorial distance in

SG

from u to
0
, i.e., the least l such that there is a sequence {u
1
, u
2
, . . . ,
u
l
= u}

V

such that u
1
=
0
and |u
j+1
u
j
| = for j = 1, 2, . . . , l 1.
Proof. First, for any v int V

, expending S
f
in power series about v
and noting that

3
k=0
i
jk
= 0 for j = 1, 2, 3, we obtain

Re S
f
= Re
_
1/(4
2
)
3

j=0
[S
f
(v +i
j
) S
f
(v)]
_
= O(
2
).
Consider the function
g
1
(v) = h
(1)

(v) Re S
f
+|v|
2
,
where (0,
2
) is some function of . Similar to the proof of [12, Lemma
9.2], we deduce from the Taylors formula and the properties of Mobius
invariant h
(1)

that g
1
has no maxima in int V

if is chosen as = C
2
with C > 0 a suciently large constant. By the assumption (2), it follows
158 S.-Y. LAN AND D.-Q. DAI
that g
1
(v) = |v|
2
= O(
2
) on V

. Hence we obtain g
1
(z) O(
2
) in V

,
which implies
h
(1)

(v) Re S
f
(v) +O(
2
).
On the other hand, if we let g
2
(v) = h
(1)

(v) Re S
f
(v) |v|
2
, then similar
to the above arguments, we conclude that
h
(1)

(v) Re S
f
(v) +O(
2
).
So it follows that
h
(1)

(v) = Re S
f
(v) +O(
2
),
which implies (i).
Next, by (3), the relationship between
(1)

and h
(1)

and Taylors for-

mula, we get
log
(2)

(v +
0
) log
(2)

(v +
1
)
= 2 log(
(1)

(v +i)
1
+ 1) 2 log(
(1)

(v)
1
+ 1)
= 2 log((1 +
2
h
(1)

(v +i))
1
+ 1) 2 log((1 +
2
h
(1)

(v))
1
+ 1)
= 2 log(2
2
h
(1)

(v +i)) 2 log(2
2
h
(1)

(v)) +O(
4
)
=
2
h
(1)

(v)
2
h
(1)

(v +i) +O(
4
).
Hence it follows from (14) and Taylors formula
log
(2)

(v +
0
) log
(2)

(v +
1
) (16)
=
2
Re(S
f
(v) S
f
(v +i)) +O(
4
)
=
2
Re(iS

f
(v)) +O(
4
)
=
3
ImS

f
(v) +O(
4
)
=
2
ImS
f
(v +i)
2
ImS
f
(v) +O(
4
).
Similarly, we conclude from (4) that
log
(2)

(v +
0
) log
(2)

(v +
3
) (17)
=
2
ImS
f
(v +)
2
ImS
f
(v) +O(
4
).
By (5) and Taylors formula, we deduce that
log
(2)

(/2 +i/2) =
2
ImS
f
(/2 +i/2) +O(
4
). (18)
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 159
So we conclude from (16), (17) and (18) that
log
(2)

(u) =
2
ImS
f
(u) +

(u)O(
2
),
By the relationship between
(2)

and h
(2)

and Taylors formula, we obtain

h
(2)

(u) = ImS
f
(u) +

(u)
2
O(1),
which implies (ii). This completes the proof of the lemma.
Lemma 5. Let v V

, u

V

, and suppose that the distance from v

(respectively, u) to is greater than 2. Then
h
(1)

(v) C, h
(2)

(u) C
for some constant C = C(, f) which depends only on and f.
Proof. Note that Re S
f
and ImS
f
are bounded on compact subset
K and

= O(1/) on K. By Lemma 4, we conclude that the lemma

holds.
Similar to the case of regular hexagonal lattice (see [8, 7]), for regular
square lattice we have
Lemma 6. Suppose that (i) W is a subset of V

(or

V

); (ii) u int W;
(iii) is the distance from u to V

W (or

V

W). If : W R is any
function, then the inequality
(19) |

j
(u)| < 5 + (1/2)
2

int W
holds for any j = 0, 1, 2, 3.
Proof. The proof of the Lemma is similar to that of [8, Lemma 7.1]
where regular hexagonal lattices were investigated.
Theorem 1. Let n be an integer, and let V

(respectively,

V

) be the
set of vertices of V

(respectively,

V

) whose distance to is at least for

each > 0. Then there are constants C = C(n, ) and = (n, ) > 0 such
that

j
n

j
n1

j
1
h
(1)

< C,

j
n

j
n1

j
1
h
(2)

< C (20)
hold for each < , and j
0
, j
1
, . . . , j
n
Z
4
.
160 S.-Y. LAN AND D.-Q. DAI
Proof. The proof proceeds by induction on n. In the case n = 0, by
Lemma 5, we get that (20) holds. So we suppose that n > 0, and that (20)
holds for n = 0, 1, 2, . . . , n 1. Let

1
=
j
n1

j
1
h
(1)

,
2
=
j
n1

j
1
h
(2)

.
Note that the operators

and
j
can commute with each other, so we
have

1
=

j
n1

j
1
h
(1)

=
j
n1

j
1

h
(1)

,
and

2
=

j
n1

j
1
h
(2)

=
j
n1

j
1

h
(2)

.
By Lemma 4, it follows that

h
(1)

Re S
f
+

(
2
O(1)),

h
(2)

ImS
f
+

2
O(1)).
Since

S
f
= O(
2
) and

= O(1/) on compact subset K , we have

h
(1)

= O(
2
) +O(1),

h
(2)

= O(
2
) +

(O(1))
on compact subset K . Note that O(1) C

(), we deduce that there

exists a constant C
1
= C
1
(, n) such that

1
(v)
V

C
1
,

2
(v)

C
1
.
Since |
1
| and |
2
| are bounded on V

and

V

, respectively, it follows from

Lemma 4 that |
1
| and |
2
| are bounded on V

and

V

, respectively,
which completes the induction step. So (20) holds for any integer n and
any j
0
, j
1
, . . . , j
n
Z
4
.
4. The Mobius transformations of circle patterns
In this section, we construct the Mobius transformations T

through cir-
cle pattern P

such that they can be expressed by the discrete Schwarzians

h
(1)

and h
(2)

. By the boundedness of h
(1)

and h
(2)

, we will prove that T

converges in C

() to some Mobius transformation T as 0 and obtain

the relation between T and f.
Note that 0 V

, so we may suppose that P

is normalized by Mobius
transformations so that
(21) q
0
(0) = f(
0
), q
1
(0) = f(
1
), q
3
(0) = f(
3
),
where
j
= i
j
(1 +i)/2 (j = 0, 1, 2, 3). Then we have
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 161
Theorem 2. For each v int V

, let T

= T

(v) be the M obius trans-

formation that takes the three points
0
,
1
,
3
to the points q
0
(v), q
1
(v),
q
3
(v), respectively. Then
(i) the limit
(22) T(z) = lim
0
T

(v)
exists for any 0, and the convergence is in C

().
(ii) T(z)(0) = f(z).
Proof. (i) Let B = B

be the Mobius transformation that takes , 0,

1 to
0
,
1
,
3
, respectively. Then we have
(23) B(z) =
z (1 +i)/2
(1 i)z +i
.
Recall the denition of M
w
and M
[w
1
,w
2
]
as in Section 3, we deduce that
T

(v) B = M
v+(1+i)/4
.
and
T

(v)
1
T

(v +) = B M
[v+
0
/2,v+
0
/2+/2]
M
[v+
0
+/2,v+
0
+]
B
1
.
Using the usual matrix notation for Mobius transformations and the fact
that M
[w
1
,w
2
]
= M
1
[w
2
,w
1
]
, we obtain from (10), (12) and (23) that
T

(v)
1
T

(v +)
=
_
(1 +i)/2
1 i i
_

_
i(h
(2)

(v +
0
) + 1)
1/2
1
0 1
_

_
0 (
2
(h
(1)

(v +) + 1)
1
+ 1)
1
1 1
_
1

_
(1 +i)/2
1 i 1
_
1
.
By Lemma 4 and noting that S
f
is Lipschitz, we deduce that
T

(v)
1
T

(v +) =
_
1
S
f
(v)/2 1
_
+O(
2
) = I +C(v) +O(
2
),
where I denotes the identity matrix and
C(v) =
_
0 1
S
f
(v)/2 0
_
.
162 S.-Y. LAN AND D.-Q. DAI
This implies
(24) T

(v +) = T

(v) +T

(v)C(v) +T

(v)O(
2
).
Similarly, we have
(25) T

(v +i
j
) = T

(v) +i
j
T

(v)C(v) +T

(v)O(
2
)
for j = 1, 2, 3. We assume, without loss of generality, that
f(0) = 0, f

(0) = 1, f

(0) = 0,
because the statement of Theorem 2 is Mobius invariant. Thus we get from
Taylors formula that
f(i
j
(1 +i)/2) = i
j
(1 +i)/2 +O(
3
)
for j = 0, 1, 2, 3. From (21) and the denition of T

(v), it follows that

T

(0)(i
j
(1 +i)/2) = i
j
(1 +i)/2 +O(
3
)
for j = 0, 1, 3, which implies
T

(0) = I +O().
By (24) and (25), we deduce that the matrices T

(v) (v int V

) are
bounded in compact subsets of , independently of . On the other hand,
(24) and (25) imply
(26)

j
T

(v) = i
j
T

(v) C(v) +T

(v) O() = T

(v) O(1)
for j = 0, 1, 2, 3, where O(1) = i
j
C(v) + O(). It follows from Theorem 1
that h
(1)

and h
(2)

are C

-bounded, so O(1) is bounded in C

(). By re-
peating dierentiation of (26), we conclude that T

(v) is bounded in C

()
uniformly in . By the properties of C

-convergence of functions (see [8,

Lemma 2.1]), we obtain that (22) holds for some subsequence of 0, and
the convergence is C

().
In the following we will show that (22) is also valid for every sequence of
0. Indeed, let D(v) be the matrix solution of the dierential equation
(27) D

(v) = D(v)C(v)
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 163
with initial condition D(0) = I, then we have
D(v +i
j
) = D(v) +i
j
D(v)C(v) +O(
2
)
for j = 0, 1, 2, 3. From (24) and (25), we obtain
|T

(v +i
j
) D(v +i
j
)|
|T

(v) D(v)| +|(T

(v) D(v))C(v)| + (1 +|T

(v)|)O(
2
)
|T

(v) D(v)|(1 +O()) + (1 +|T

(v)|)O(
2
).
Note that T

(0) = I +O() = C(0) +O(), we deduce

|T

(v) D(v)| = ((v)O(

2
) +O())(1 +O())
(v)
,
where (v) is the combinatorial distance from v to 0 in SG

. Hence we
deduce
|T

(v) D(v)| O()e

O(1)
= O()
on a compact subset K , because (v) = O(1/) on K. So T

(v) =
D(v) +O(), which implies that (22) holds for every 0.
In equation (26), taking a limit as 0, we obtain
(28)
j
T(z) = i
j
T(z)C(z).
Hence, we get

j
T(z) =
j+2
T(z),
which implies that T(z) is a matrix-valued analytic function of z. In addi-
tion, it follows from (28) that the determinant of T(z) is constant in . Note
that at z = 0 this determinant is 1. So T(z) is a Mobius transformation for
each z .
(ii) Let
T(z) =
_
a(v) b(v)
c(v) d(v)
_
.
Then T(z) satises the dierential equation (27). By the denitions of
Schwarzian derivative and C(v), we deduce that
S
b/d
= S
f
.
Note that b/d(0) = f(0) = 0, (b/d)

(0) = f

(0) = 1, (b/d)

(0) = f

(0) = 0.
So we conclude that T(z)(0) = b(z)/d(z) = f(z).
164 S.-Y. LAN AND D.-Q. DAI
5. C

-convergence to minimal surfaces

We will show that there exists a family of discrete minimal surface

: V

R
3
which converges in C

() to the smooth minimal surface

F : R
3
in this section.
We rst give some properties of C

-convergence of discrete functions,

that is
Lemma 7. Suppose that f

, g

converge in C

() to functions f, g :
C, respectively. Then the following statements hold in C

(): (i)
f

+ g

f + g; (ii) f

fg; (iii) 1/f

1/f if f = 0 in ; (iv)

f if f

> 0; (v) |f

| |f|.
Proof. Using the denition of discrete directional derivative as well as
the denition of C

-convergence of discrete function, we deduce easily that

the lemma holds.
For each v V

, let f

(v) and r

(v) denote the center and radius of

circle P

(v) in the circle pattern P

, respectively, and let g

(v) denote the

intersection point q
0
(v) of circles P

(v) and P

(v + 2
0
) in P

. Then we
have
Lemma 8. The discrete functions f

(v) and g

(v) converge in C

()
to the locally injective meromorphic function f : C, and r

(v)/ con-
verges in C

() to |f

| as 0.
Proof. First, we write T

(v) and T(z) as matrices

T

(v) =
_
a

(v) b

(v)
c

(v) d

(v)
_
, T(z) =
_
a(z) b(z)
c(z) d(z)
_
.
Then it follows from Theorem 2(i) that
0
c

+d

and
0
a

+b

converge to d
and b in C

(), respectively. By Theorem 2(ii), we obtain that b(z)/d(z) =

f(z). Since the determinant of T(z) is nonzero, d is nonzero in . By
Lemma 7, we get that (
0
a

+ b

)/(
0
c

+d

) b/d in C

(). Note that

g

(v) = T

(v)(
0
), so we conclude that the discrete function g

f in
C

() as 0.
Next, let c be the circle that contains the three points
0
,
1
,
3
. Since
T

(v) maps the three points

0
,
1
,
3
to q
0
(v), q
1
(v), q
3
(v), respectively,
and since the three points q
0
(v), q
1
(v), and q
3
(v) lie in the circle P

(v),
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 165
it follows that T

(v) maps the circle c onto P

(v). Hence it follows that

T

(v)(0) = f

(v). Similar to the above arguments, we conclude that f

(v) =
T

(v)(0) converges in C

() to f as 0.
Lastly, note that
_
r

(v) +r

(v +) = |f

(v) f

(v + )| = |

0
f

|,
r

(v) + r

(v + 2
0
) = |f

(v) f

(v + 2
0
)| = |

0
f

(v) +
1
f

(v + )|, and
_
r

(v +) +r

(v + 2
0
) = |f

(v +) f

(v +2
0
)| = |

1
f

(v +)|. So we
deduce
2r

(v)/ = |

0
f

(v) +
1
f

(v +)| +(|

0
f

|
2
|

1
f

(v +)|
2
).
Because
j
f

converges in C

() to f

, it follows from Lemma 7 that

|

j
f| converges in C

() to |f

|. Hence we obtain that r

/ converges
C

() to |f

|.
Now we give the following C

-convergence theorem for smooth minimal

surfaces.
Theorem 3. Suppose that F : R
3
is a minimal immersion with-
out umbilic points in conformal curvature line coordinates. Then there exists
a family of discrete surface

: V

R
3
that converges in C

() to F as
0.
Proof. It follows from Lemma 1 that the smooth minimal surface F :
R
3
is expressed by (1). Thus, to prove that there is a discrete minimal
surface

that converges in C

() to F, it suces to show that each

component of

converges in C

() to the corresponding one of F.

For any > 0, let F

: V

R
3
be the discrete minimal surface
corresponding to F, which consists of spheres and circles, and

F

: V

R
3
be the discrete minimal surface which is comprised of the centers of spheres
and circles in F

. Then it follows from (6) that

(v
(1)
s
)

F

(v
(2)
s
)
(29)
= Re
_
R(v
(1)
s
) +R(v
(2)
s
)
1 +|g

|
2
|f

(v
(1)
s
) f

(v
(2)
s
)|
f

(v
(1)
s
) f

(v
(2)
s
)
(1 (g

)
2
, i(1 + (g

)
2
), 2g

)
_
Re(

1
,

2
,

3
)
166 S.-Y. LAN AND D.-Q. DAI
for any v
(1)
s
, v
(2)
s
lying in the same square of SG

, where

1
=
R(v
(1)
s
) +R(v
(2)
s
)
1 +|g

|
2
|f

(v
(1)
s
) f

(v
(2)
s
)|
f

(v
(1)
s
) f

(v
(2)
s
)
(1 (g

)
2
),

2
=
R(v
(1)
s
) +R(v
(2)
s
)
1 +|g

|
2
|f

(v
(1)
s
) f

(v
(2)
s
)|
f

(v
(1)
s
) f

(v
(2)
s
)
i(1 + (g

)
2
),

3
=
R(v
(1)
s
) +R(v
(2)
s
)
1 +|g

|
2
|f

(v
(1)
s
) f

(v
(2)
s
)|
f

(v
(1)
s
) f

(v
(2)
s
)
(2g

).
By Lemma 8, we get that f

and g

converge in C

() to f, and that
r

(v)/ converges in C

() to |f

| as 0. Moreover, it is easy to see that

(f

(v
(1)
s
) f

(v
(2)
s
))/ converges in C

() to 2f

. On the other hand, we

get from (7) that
R(v
(k)
s
) =

1 +|f

(v
(k)
s
)|
2
|f

(v
(k)
s
) g

|
2
2|f

(v
(k)
s
) g

for k = 1, 2. So we deduce from the denitions of

l
(l = 1, 2, 3) and Lemma
7 that

1
,

2
and

3
converge in C

() to (1f
2
)/f

, i(1+f
2
)/f

and
2f/f

, respectively.
Set

(v) = 2
2

F

(v) for each v V

, then we conclude from (29) that

the discrete minimal surface

converges in C

() to the smooth minimal

surface F : R
3
. This completes the proof of Theorem 3.
Acknowledgments. This work is supported in part by NSF of China
(60575004, 10771220), the Ministry of Education of China (SRFDP-
20070558043), NSF of Guangxi (0991081), the Department of Education
of Guangxi (200707MS043), and the Grant for Guangxi University for Na-
tionalities (2008ZD009).
References
[1] S. I. Agafonov, Imbedded circle patterns with the combinatorics of the square grid and
discrete Painleve equations, Discrete Comput. Geom., 29 (2003), no. 2, 305319.
[2] A. I. Bobenko, T. Homann, and B. A. Springborn, Minimal surfaces from circle
patterns: Geometry from combinatorics, Ann. of Math., 164 (2006), 231264.
[3] A. I. Bobenko and U. Pinkall, Disrete isothermic surfaces, J. Reine Angew. Math.,
475 (1996), 187208.
CONVERGENCE OF CIRCLE PATTERNS TO MINIMAL SURFACE 167
[4] A. I. Bokenko and Y. B. Suris, Discrete Diernetial Geometry. Consistency as In-
tergrabilty, arXiv:math.DG/0504358 v1, 18 Apr. 2005.
[5] P. G. Doyle, Z.-X. He and B. Rodin, Second derivatives of circle packings and con-
formal mappings, Disrete Comput. Geom., 11 (1994), 3549.
[6] U. Dierkes, S. Hildebrandt, A. K uster and O. Wohlrab, Minimal Surfaces I, Springer-
Verlag, Berlin, 1992.
[7] Z.-X. He and O. Schramm, On the convergence of circle packings to the Riemann
map, Invent. Math., 125 (1996), 285305.
[8] Z.-X. He and O. Schramm, The C

-convergence of hexagonal disk packings to the

Riemann map, Acta Math., 180 (1998), 219245.
[9] N. Kapouleas, Complete embedded minimal surfaces of nite total curvature, J. Dif-
ferential Geom., 47 (1997), 95169.
[10] J. C. Nitsche, Lectures on Minimal Surfaces, Vol. 1, Cambridge Univ. Press, 1989.
[11] B. Rodin and D. Sullivan, The convergence of circle packings to the Riemann map-
ping, J. Dierential Geom., 26 (1987), 349360.
[12] O. Schramm, Circle patterns with the combinatorics of the square grid, Duke Math.
J., 86 (1997), no. 2, 347389.
[13] W. Thurston, The geometry and topology of 3-manifolds, Chapter 13, Princeton
University Notes, Princeton, N.J., 1980.
Shi-Yi Lan
Department of Mathematics
Guangxi University for Nationalities
Nanning 530006
P. R. China
Dao-Qing Dai
Department of Mathematics
Sun Yat-Sen (Zhongshan) University
Guangzhou 510275
P. R. China
stsddq@mail.sysu.edu.cn