a

r
X
i
v
:
0
8
0
9
.
3
6
0
8
v
2


[
m
a
t
h
.
D
G
]


1
3

J
a
n

2
0
0
9
ISOTHERMIC HYPERSURFACES IN R
n+1
NEIL DONALDSON

AND CHUU-LIAN TERNG

Abstract. A diagonal metric

P
n
i=1
giidx
2
i
is termed Guichard
k
if
P
nk
i=1
gii
P
n
i=nk+1
gii = 0. A hypersurface in R
n+1
is isothermic
k
if it admits
line of curvature co-ordinates such that its induced metric is Guichard
k
.
Isothermic1 surfaces in R
3
are the classical isothermic surfaces in R
3
.
Both isothermic
k
hypersurfaces in R
n+1
and Guichard
k
orthogonal co-
ordinate systems on R
n
are invariant under conformal transformations.
A sequence of n isothermic
k
hypersurfaces in R
n+1
(Guichard
k
orthog-
onal co-ordinate systems on R
n
resp.) is called a Combescure sequence
if the consecutive hypersurfaces (orthogonal co-ordinate systems resp.)
are related by Combescure transformations. We give a correspondence
between Combescure sequences of Guichard
k
orthogonal co-ordinate sys-
tems on R
n
and solutions of the
O(2nk,k)
O(n)O(nk,k)
-system, and a corre-
spondence between Combescure sequences of isothermic
k
hypersurfaces
in R
n+1
and solutions of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system, both being inte-
grable systems. Methods from soliton theory can therefore be used to
construct Christoel, Ribaucour, and Lie transforms, and to describe
the moduli spaces of these geometric objects and their loop group sym-
metries.
1. Introduction
A parameterised surface f(x
1
, x
2
) in R
3
is isothermic if (x
1
, x
2
) is a con-
formal line of curvature co-ordinate system. For example, constant mean
curvature surfaces in R
3
are isothermic away from umbilic points. Classi-
cal geometers constructed various geometric transforms for these surfaces
that gave methods to generate new isothermic surfaces from a given one.
The Christoel transform associates to each isothermic surface f
1
(x
1
, x
2
)
a second isothermic surface f
2
(x
1
, x
2
) (a Christoel dual of f
1
) such that
the principal curvature directions of f
1
and f
2
are parallel and the map
f
1
(x) f
2
(x) is orientation reversing. A Ribaucour transform is a dif-
feomorphism : M M

between two surfaces in R

3
satisfying: (i) the
normal line of M at p intersects the normal line of M

at p

= (p) at equal
distance for all p M, (ii) maps principal directions of M to those of
M

. Given an isothermic surface in R

3
, one can solve a system of compat-
ible ordinary dierential equations to construct a two-parameter family of
Ribaucour transforms so that the target surfaces are also isothermic; these
transforms are known classically as Darboux transforms. These geometric

Research supported in part by NSF Advance Grant.

Research supported in part by NSF Grant DMS-0707132.

1
2 NEIL DONALDSON

AND CHUU-LIAN TERNG

transforms provide a rich class of isothermic surfaces. The connection of

isothermic surfaces to soliton theory was rst noted in [3], that the Gauss
Codazzi equation for isothermic surfaces in R
3
has a Lax pair and is a soliton
equation. Techniques from soliton theory have been used to identify classical
transformations of isothermic surfaces in [4, 6].
A notion of isothermic surfaces in R
n
was introduced in [1, 2]: An immer-
sion f(x
1
, x
2
) in R
n
is isothermic if the normal bundle is at and (x
1
, x
2
)
are conformal line of curvature co-ordinates. It was proved in [1, 2] that the
GaussCodazzi equation for isothermic surfaces in R
n
is the soliton equa-
tion associated to
O(n+1,1)
O(n)O(1,1)
, and analogues of classical transformations
were constructed for these higher co-dimension isothermic surfaces.
Burstall asked in [2]: Is there any interesting theory of isothermic sub-
manifolds of R
n
of dimension greater than two? An attempt was made by
Tojeiro in [11]: A parameterised submanifold f(x) in Euclidean space is k-
isothermic if the normal bundle is at, x is line of curvature co-ordinates, the
induced metric is conformal to a Riemannian product, and there are exactly
k distributions E
1
, . . . , E
k
such that TM =
k
i=1
E
i
and each E
i
(x) is con-
tained in the common eigenspace of the shape operators A
v
[v (M)
x
.
However, the class of k-isothermic submanifolds is not as rich and does not
have all the geometric transforms that isothermic surfaces have.
In this paper we give a positive answer to Burstalls question. We rst
dene Guichard diagonal metrics. Fix 1 k n 1. A diagonal metric

n
i=1
g
ii
dx
2
i
is Guichard
k
if
nk

i=1
g
ii
=
n

i=nk+1
g
ii
.
Let I
nk,k
= diag(
1
, . . . ,
n
) with
i
= 1 for i n k and
i
= 1 for
n k < i n, and ( , )
k
the bilinear form on C
n
dened by
(x, y)
k
= x
T
I
nk,k
y.
If ds
2
=

n
i=1
u
2
i
dx
2
i
is a diagonal metric, then ds
2
is Guichard
k
if and only
if (u, u)
k
= 0, where u = (u
1
, . . . , u
n
)
T
.We call u a metric eld for ds
2
.
An orthogonal co-ordinate system (x) on R
n
is Guichard
k
if

(ds
2
0
)
is Guichard
k
, where ds
2
0
is the standard Euclidean metric on R
n
. Let
(x), (x) be orthogonal co-ordinate systems on R
n
. Classically, a map
(x) (x) is said to be a Combescure transform if
x
i
(x) is parallel to

x
i
(x) for all 1 i n.
A hypersurface in R
n+1
is isothermic
k
if it admits line of curvature co-
ordinates such that its induced metric is Guichard
k
. It follows from the
denition that isothermic
k
hypersurfaces in R
n+1
and Guichard
k
orthogonal
co-ordinate systems on R
n
are invariant under conformal transformations.
Note that isothermic
1
surfaces in R
3
are the classical isothermic surfaces in
R
3
.
ISOTHERMIC HYPERSURFACES 3
Let f
1
, f
2
be two isothermic hypersurfaces in R
n+1
. The map f
1
(x)
f
2
(x) is called a Combescure transform if (f
1
)
x
i
is parallel to (f
2
)
x
i
for all
1 i n.
Denition 1.1. A sequence
1
, . . . ,
n
of Guichard
k
orthogonal co-
ordinate systems on R
n
is called a Combescure sequence if (i)
i
(x)

i+1
(x) is a Combescure transform for all 1 i n 1, (ii) u
1
, . . . , u
n
are linearly independent at every point, where u
i
= (u
i1
, . . . , u
in
)
T
are met-
ric elds for

i
(ds
2
0
) (i.e.,

i
(ds
2
0
) =

n
j=1
u
2
ij
dx
2
j
) chosen in such a way
that
(1.1)
(
i
)
x
j
u
ij
=
(
1
)
x
j
u
1j
, 1 i, j n.
Condition (i) implies that given a choice of metric eld u
1
for

1
(ds
2
0
),
there exist unique metric elds u
i
for

i
(ds
2
0
) satisfying (1.1). Note that the
condition that u
1
, . . . , u
n
are independent is independent of choices of u
1
.
Denition 1.2. A sequence of isothermic hypersurfaces f
1
, . . . , f
n
in
R
n+1
is called a Combescure sequence if (i) f
i
(x) f
i+1
(x) is a Combescure
transform for all 1 i n 1, (ii) u
1
, . . . , u
n
are linearly independent at
every point, where u
i
= (u
i1
, . . . , u
in
)
T
are chosen such that the induced
metrics I
f
i
=

n
j=1
u
2
ij
dx
2
j
satisfy
(f
i
)x
j
u
ij
=
(f
1
)x
j
u
1j
for all 1 i, j n.
Next we explain the relation between Combescure sequences and soliton
equations. To each symmetric space U/K is associated a soliton hierarchy.
The ows in the hierarchy are parameterised by (, j) with 1 rk(U/K)
(the rank of U/K) and j a positive integer. Each ow is equivalent to the
condition that a family of u
C
-valued connection 1-forms

[ C is at:
this family is the Lax pair of the ow. In fact, the Lax pair

of the (, j)-th
ow is a degree j polynomial in . For example, the SU(2),
SU(2)
SO(2)
- and
SU(3)
SO(3)
hierarchies are hierarchies for the non-linear Schr odinger equation, modied
KdV equation, and the reduced 3-wave equation respectively. We put all
(, 1)-th ows with 1 rk(U/K) together to construct a rst order
non-linear system, the so-called U/K-system (cf. [7]). These integrable
systems often arise in submanifold geometry.
We call a basis

C = c
1
, . . . , c
n
of R
nk,k
a null basis if each c
i
is a null
vector. One main result of this paper is to give a construction of a Combes-
cure sequence of isothermic
k
hypersurfaces f
,

C
1
, . . . , f
,

C
n
from a solution
of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system and a null basis

C of R
nk,k
. Moreover, we
show that:
(1) All Combescure sequences of isothermic
k
hypersurfaces arise from
this construction.
(2) If

B,

C are two null bases for R
nk,k
, then f
,

C
i
(x) f
,

B
i
(x) is again
a Combescure transform for all 1 i n.
4 NEIL DONALDSON

AND CHUU-LIAN TERNG

(3) Let f
1
, . . . , f
n1
be isothermic
k
hypersurfaces in R
n+1
such that (i)
f
i
(x) f
i+1
(x) is Combescure for all 1 i n2, (ii) u
1
, . . . , u
n1
are linearly independent at every point, where u

= (u
1
, . . . , u
n
) is
the metric eld for the induced metric I
f

chosen so that
(f

)x
j
u
j
=
(f
1
)x
j
u
1j
for all 2 n 1. Then there exists an isothermic
k
hypersurface f
n
so that f
1
, . . . , f
n
is a Combescure sequence of
isothermic
k
hypersurfaces in R
n+1
.
(4) If (f
1
, f
2
) is a Combescure sequence of isothermic
1
surfaces in R
3
,
then either (f
1
, f
2
) is a classical Christoel pair, or (f
1
, g
2
) is, where
g
2
(x
1
, x
2
) = f(x
1
, x
2
) is oppositely oriented to f
1
.
We have similar results for Combescure sequences of Guichard
k
orthogonal
co-ordinate systems and solutions of the
O(2nk,k)
O(n)O(nk,k)
-system.
We apply techniques from soliton theory to construct loop group actions,
Ribaucour transforms, and Lie transforms for Combescure sequences of
isothermic
k
hypersurfaces. We prove that Christoel and Ribaucour trans-
forms commute, and that the conjugation of a Ribaucour transform and a
Lie transform is again Ribaucour. Moreover, our results generalise those for
isothermic surfaces.
Results in [7] and [9] imply that Combescure sequences of isothermic
k
hypersurfaces are determined by n
2
functions of one variable and a null
basis of R
nk,k
.
An n-dimensional submanifold in R
n+m
is isothermic
k
if it has at nor-
mal bundle and line of curvature co-ordinates such that the rst fundamental
form is Guichard
k
. Isothermic
k
submanifolds in R
n+m
are invariant under
conformal transformations. All results for the co-dimension one case gener-
alise easily to higher co-dimension.
All Guichard
k
co-ordinate systems and isothermic
k
hypersurfaces in this
paper are dened on simply connected open subsets of R
n
. Although the
Cauchy problems for the
O(2nk,k)
O(n)O(nk,k)
- and
O(2n+1k,k)
O(n+1)O(nk,k)
- systems with
rapidly decaying initial data on a non-characteristic line have global solu-
tions, the corresponding co-ordinate systems and hypersurfaces may have
singularities.
The paper is organised as follows: We write down the
O(2nk,k)
O(n)O(nk,k)
- and
O(2n+mk,k)
O(n+1)O(n+mk,k)
-systems and their Lax pairs in section 2, give the corre-
spondence between solutions of the
O(2nk,k)
O(n)O(nk,k)
-system and Combescure
sequences of Guichard
k
orthogonal co-ordinate systems on R
n
in section
3. In section 4 we describe the correspondence between solutions of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system and Combescure sequences of isothermic
k
hypersur-
faces in R
n+1
, and construct Christoel transforms of these. In the nal
section we (a) use the dressing action of a rational loop with two poles on
ISOTHERMIC HYPERSURFACES 5
the space of solutions of integrable systems to construct geometric Ribau-
cour transforms of Combescure sequences of isothermic
k
hypersurfaces and
prove that Christoel and Ribaucour transforms commute, and (b) con-
struct Lie transforms of Combescure sequences of isothermic
k
hypersurfaces
and show that the conjugation of a Ribaucour transform by a Lie transform
is again a Ribaucour transform. These geometric transforms can easily be
generalised to Combescure sequences of Guichard
k
orthogonal co-ordinate
systems on R
n
and Combescure sequences of isothermic
k
submanifolds in
Euclidean space.
2. The
O(2nk,k)
O(n)O(nk,k)
- and
O(2n+1k,k)
O(n+1)O(nk,k)
-systems
Let U be a real simple Lie group, an involution of U (we also use to
denote d
e
on the Lie algebra u of U), K the xed point set of , and k, p
the 1 eigenspaces of on u. Let a be a maximal abelian
1
subalgebra in
p, a
1
, . . . , a
n
a basis of a, and a

the orthogonal complement of a with

respect to the Killing form. The U/K-system [7] is the following PDE for
v : R
n
a

p,
[a
i
, v
x
j
] [a
j
, v
x
i
] [[a
i
, v], [a
j
, v]] = 0, i ,= j.
Set
(2.1)

=
n

i=1
(a
i
+ [a
i
, v])dx
i
.
The following Proposition is well-known and the proof follows easily from a
direct computation:
Proposition 2.1. The following statements are equivalent for a smooth map
v : R
n
a

p and

dened by (2.1):
(1) v is a solution of the U/K-system,
(2)

is at for all C,
(3)
r
is at for some r R iR.
We call

the Lax pair for the solution v of the U/K-system. Note that

satises the U/K-reality condition:

, (

) =

.
Denition 2.2. Given a solution v of the U/K-system, an extended frame
E

for v is a parallel frame of

(i.e., E
1

dE

) that satises the

U/K-reality condition:
(2.2) E

= E

, (E

) = E

.
1
Throughout this paper maximal abelian will mean maximal semisimple abelian: when
U/K is non-Riemannian there will generally be non-semisimple abelian subalgebras of
dimension greater than the rank of U/K. E.g. p stab() where is an isotropic line in
R
1,1
is 3-dimensional abelian for the rank 2 space
O(4,1)
O(3)O(1,1)
.
6 NEIL DONALDSON

AND CHUU-LIAN TERNG

An extended frame E is called the normalized extended frame for the solution
v if E(0, ) = I.
The following Proposition is well-known (cf. [1, 2]).
Proposition 2.3. Let v be a solution of the U/K-system,

n
i=1
(a
i
+
[a
i
, v]) dx
i
the Lax pair of v, and E

an extended frame. Then:

(1) E
0
K,
(2) The gauge transformation
E
0

= E
0

E
1
0
dE
0
E
1
0
=
n

i=1
E
0
a
i
E
1
0
dx
i
is a p-valued closed 1-form,
(3) Z :=
E

E
1

=0
is a p-valued map and dZ =

n
i=1
E
0
a
i
E
1
0
dx
i
.
Proof. A simple computation implies that E
0

= . Since is at and
= 0, d = 0. The third statement follows from a direct computation.

Example 2.4 (The
O(2nk,k)
O(n)O(nk,k)
-system). Here the involution on O(2n
k, k) is (g) = I
n,n
gI
1
n,n
, and
p =
__
0
T
J
0
_

/
nn
_
, J = I
nk,k
,
where /
nn
is the space of real n n matrices. Set
a
i
=
_
0 e
ii
J
e
ii
0
_
, 1 i n.
Then a
1
, . . . , a
n
is a basis of a maximal abelian subalgebra a in p, and
moreover
a

p =
__
0 F
T
J
F 0
_

F = (f
ij
), f
ii
= 0 1 i n
_
.
The Lax pair for the
O(2nk,k)
O(n)O(nk,k)
-system is
(2.3)

=
_
J

_
, = JF + F
T
J, = F
T
J + FJ,
where = diag(dx
1
, . . . , dx
n
).
The
O(2nk,k)
O(n)O(nk,k)
-system is given by the condition that
0
is at, i.e.,
d = , d = .
In other words, the
O(2nk,k)
O(n)O(nk,k)
-system for F = (f
ij
) is
_

i
(f
ij
)
x
j
+
j
(f
ji
)
x
i

i

k
f
ik
f
jk
= 0, i ,== j,
(f
ij
)
x
i
+ (f
ji
)
x
j

k

k
f
ki
f
kj
= 0, i ,= j,
(f
ij
)
x
k
+
k
f
ik
f
kj
= 0, i, j, k distinct.
ISOTHERMIC HYPERSURFACES 7
Example 2.5 (The
O(2n+1k,k)
O(n+1)O(nk,k)
-system). Here (g) = I
n+1,n
gI
1
n+1,n
,
p =
__
0
T
J
0
_

/
n,n+1
_
,
where J = I
nk,k
. Set
a
i
=
_
0 A
T
i
J
A
i
0
_
, A
i
= (e
ii
, 0), 1 i n.
Then a
1
, . . . , a
n
forms a basis of a maximal abelian subalgebra a in p, and
a

p =
__
0
T
J
0
_

= (F, ), F = (f
ij
), f
ii
= 0, 1 i n
_
.
The
O(2n+1k,k)
O(n+1)O(nk,k)
-system is the PDE for = (F, ) with F = (f
ij
) and
= (
1
, . . . ,
n
)
T
such that f
ii
= 0 for all 1 i n. The Lax pair for the
O(2n+1k,k)
O(n+1)O(nk,k)
-system is
(2.4)

=
_
_

_

0
_
J
(, 0)
_
_
,
where J = I
nk,k
, = diag(dx
1
, . . . , dx
n
), and
(2.5) =
_
JF + F
T
J J

T
J 0
_
, = (F F
T
)J,
Note that and are o(n + 1)-valued and o(n k, k)-valued 1-forms re-
spectively, and that the
O(2n+1k,k)
O(n+1)O(nk,k)
-system is given by the atness of
and . In other words, the
O(2n+1k,k)
O(n+1)O(nk,k)
-system is the following PDE for
F = (f
ij
) and = (
1
, . . . ,
n
)
T
:
(2.6)
_

i
(f
ij
)
x
j
+
j
(f
ji
)
x
i

i

k
f
ik
f
jk
= 0, i ,= j,
(f
ij
)
x
i
+ (f
ji
)
x
j

k

k
f
ki
f
kj
= 0, i ,= j,
(f
ij
)
x
k
+
k
f
ik
f
kj
= 0, i, j, k distinct,
(
i
)
x
j
+
j
f
ij

j
= 0, i ,= j.
Example 2.6 (The
O(2n+mk,k)
O(n+m)O(nk,k)
-system). The
O(2n+mk,k)
O(n+m)O(nk,k)
-system
is the PDE for = (F, ) dened by the atness of and , where and
are dened in terms of F, the same way as in (2.5), F = (f
ij
) is an n n
map with f
ii
= 0 for all 1 i n, and a n m-valued map. The Lax
pair has the same form as (2.4).
3. Guichard orthogonal co-ordinate systems
We exhibit a relation between Combescure sequences of Guichard
k
orthog-
onal co-ordinate systems on R
n
and solutions of the
O(2nk,k)
O(n)O(nk,k)
-system.
First recall a simple and well-known Lemma (the proof follows from a
direct computation):
8 NEIL DONALDSON

AND CHUU-LIAN TERNG

Lemma 3.1. Let (x) be an orthogonal system on R

n
, and

(ds
2
0
) =

n
i=1
u
2
i
dx
2
i
, where ds
2
0
is the standard Euclidean metric on R
n
. Set
e
i
=

x
i
u
i
, g = (e
1
, . . . , e
n
),
i
= u
i
dx
i
,
and F = (f
ij
) with
f
ij
=
_

i
(u
i
)x
j
u
j
, i ,= j,
0, i = j.
Then := (
ij
) = g
1
dg is the at Levi-Civita connection 1-form for

(ds
2
0
) and

ij
=
(u
i
)
x
j
u
j
dx
i

(u
j
)
x
i
u
i
dx
j
=
i
f
ij
dx
i
+
j
f
ji
dx
j
,
or equivalently, = (
ij
) = JF + F
T
J, where = diag(dx
1
, . . . , dx
n
).
Theorem 3.2. Let F = (f
ij
) be a solution of the
O(2nk,k)
O(n)O(nk,k)
-system, and
E

an extended frame of F. Then:

(a) E
0
=
_
g
1
0
0 g
2
_
O(n) O(n k, k).
(b)
E

E
1

=0
is of the form
_
0 Y
JY
T
0
_
for some /
nn
-valued map
Y and dY =

n
i=0
g
1
e
ii
Jg
1
2
dx
i
.
(c) Let c R
nk,k
be a constant null vector. Then Yc is a Guichard
k
orthogonal co-ordinate system on R
n
with ds
2
=

n
i=1
u
2
i
dx
2
i
dened
on the open subset x R
n
[

n
i=1
u
i
(x) ,= 0, where (u
1
, . . . , u
n
)
T
:=
Jg
1
2
c.
(d) If

C = c
1
, . . . , c
n
is a null basis for R
nk,k
, then Y

C = Y c
1
, . . . , Yc
n

is a Combescure sequence of Guichard

k
orthogonal co-ordinate sys-
tems on R
n
.
(e) If

b
1
, . . . ,

b
n
is another null basis for R
nk,k
, then Y (x)c
i
Y (x)

b
j
is a Combescure transform for all 1 i, j n.
Proof. (a) Since E

is an extended frame, it satises the U/K-reality con-

dition (2.2), E
0
K. Hence we can write E
0
=
_
g
1
0
0 g
2
_
.
(b) By Proposition 2.3, we have dY = g
1
Jg
1
2
. Set g
1
= (e
1
, . . . , e
n
)
and u = (u
1
, . . . , u
n
)
T
:= g
1
2
c. Then
(3.1) dYc = (e
1
, . . . , e
n
)(
1
u
1
, . . . ,
n
u
n
)
T
=
n

i=1

i
u
i
dx
i
e
i
.
(c) The Euclidean metric is therefore ds
2
=

n
i=1
u
2
i
dx
2
i
. But c being a
null vector in R
nk,k
implies that (u, u)
k
= 0. Hence Yc is a Guichard
k
orthogonal co-ordinate system.
Statements (d) and (e) follow.
ISOTHERMIC HYPERSURFACES 9
Theorem 3.3. Let
1
, . . . ,
n
be a Combescure sequence of Guichard
k
orthogonal co-ordinate systems on R
n
, and
i
(0) = 0. Then there is a
solution F of the
O(2nk,k)
O(n)O(nk,k)
-system and a null basis

C = c
1
, . . . , c
n

of R
nk,k
so that
1
, . . . ,
n
is the Combescure sequence of Guichard
k
orthogonal co-ordinate systems constructed from F and

C as in Theorem
3.2.
Proof. Let u

= (u
1
, . . . , u
n
)
T
be the metric eld for

(ds
2
0
) chosen as
in Denition 1.1, so
(

)x
j
u
j
=
(
1
)x
j
u
1j
. Then the vector eld e
j
:=
(
i
)x
j
u
ij
is
independent of i. Set
h = (e
1
, . . . , e
n
), = h
1
dh.
Since h does not depend on , we have, by Lemma 3.1, that
f
ij
:=
i
(u
i
)
x
j
u
j
is independent of for i ,= j. Let F = (f
ij
) with f
ii
= 0 for all i. Then
= JF + F
T
J = h
1
dh
is a at o(n)-valued connection 1-form, where = diag(dx
1
, . . . , dx
n
).
Claim 1. F is a solution of the
O(2nk,k)
O(n)O(nk,k)
-system.
The Lax pair

for the
O(2nk,k)
O(n)O(nk,k)
-system is (2.3). By Proposition 2.1,
F is a solution of the
O(2nk,k)
O(n)O(nk,k)
-system if and only if both and are
at, where = JF +F
T
J and is the o(k, nk)-valued 1-form dened
by
= (
ij
) = (F
T
+ F)J, i.e.,
ij
= f
ji

j
dx
i
+ f
ij

j
dx
j
.
To prove Claim 1 it thus suces to prove that is at. We will show that
(3.2) du
T

= u
T

, 1 n,
i.e., u
T
1
, . . . , u
T
n
are linearly independent parallel sections for . It then
follows that is at. By the denition of f
ij
,
(u
i
)
x
j
=
i
f
ij
u
j
, i ,= j.
Since

n
i=1

i
u
2
i
= 0,

i
u
i
(u
i
)
x
i
=

j=i

j
u
j
(u
j
)
x
i
= u
i

j=i
f
ji
u
j
.
So we have
(3.3) (u
i
)
x
i
=
i

j=i
f
ji
u
j
.
It is a short calculation using (3.3) to establish (3.2). This proves Claim 1.
10 NEIL DONALDSON

AND CHUU-LIAN TERNG

Since
T
J + J = 0 and (3.2), we have
(3.4) du

=
T
u

= JJu

, 1 n.
Let

denote the Lax pair for F, and E

(x) the solution of

E
1

dE

=
_
J

_
, E

(0) =
_
h(0) 0
0 I
_
,
where h(0) = (e
1
(0), . . . , e
n
(0)) O(n). Since E

(0) O(n) O(n

k, k) and

satises the
O(2nk,k)
O(n)O(nk,k)
-reality condition, E

satises the
O(2nk,k)
O(n)O(nk,k)
-reality condition. In particular, E
0
(x) O(n) O(n k) so
we may write E
0
=
_
g
1
0
0 g
2
_
. Then
g
1
1
dg
1
= , g
1
2
dg
2
= .
But h
1
dh = and g
1
(0) = h(0). So g
1
= h.
Claim 2. u

= Jg
1
2
Ju

(0).
Since g
1
2
dg
2
= ,
d(Jg
1
2
) = Jg
1
2
= JJ(Jg
1
2
).
By (3.4), h
2
= (u
1
, . . . , u
n
) also satises dh
2
= JJh
2
. There is thus
a constant null vector c

such that u

= Jg
1
2
c

. Since g
2
(0) = I, c

=
Ju

(0). This proves Claim 2.

By Theorem 3.2,
E

E
1

=0
=
_
0 Y
JY
T
0
_
,
for some n n valued map Y . Since E

(0) is independent of , Y (0) is the

zero matrix.
It remains to prove that Y

, where Y

= Yc

.
Use u

= Jg
1
2
c

, g
1
= h = (e
1
, . . . , e
n
), and Theorem 3.2 (b) to see that
dY

i
u
i
e
i
dx
i
. But d

i
u
i
e

dx
i
and Y

(0) =

(0) = 0. Hence
Y

.
Since an o(n k, k)-valued connection 1-form with n 1 linearly inde-
pendent parallel sections is at, we have:
Corollary 3.4. Suppose
1
, . . . ,
n1
are Guichard
k
orthogonal coordinate
systems on R
n
such that (i)
i
(x)
i+1
(x) is a Combescure transform
for 1 i n 2, (ii) u
1
, . . . , u
n1
are linearly independent at every point,
where u

= (u
1
, . . . , u
n
) is the metric eld of

i
(ds
2
0
) chosen such that
(

)x
j
u
j
=
(
1
)x
j
u
1j
for all 2 n 1 and 1 j n. There then exists a
Guichard
k
orthogonal co-ordinate system
n
on R
n
such that
1
, . . . ,
n

is a Combescure sequence of Guichard

k
orthogonal co-ordinate systems.
As a consequence of Theorem 3.2(e) and Theorem 3.3 we have:
ISOTHERMIC HYPERSURFACES 11
Corollary 3.5. Let =
1
, . . . ,
n
be a Combescure sequence of orthogo-
nal co-ordinate systems on R
n
, and

B =

b
1
, . . . ,

b
n
a null basis of R
nk,k
.

B then denes a Combescure sequence of Guichard

k
orthogonal co-ordinate
systems =
1
, . . . ,
n
on R
n
such that
i
(x)
i
(x) is a Combescure
transform for each i.
4. Isothermic hypersurfaces in R
n+1
In this section we describe the correspondence between solutions of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system and isothermic
k
hypersurfaces in R
n+1
.
The following result can be proved in a similar manner to Theorem 3.2:
Theorem 4.1. Let (F, ) be a solution of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system, and
E

an extended frame for (F, ). Then:

(a) E
0
O(n + 1) O(n k, k), so E
0
=
_
g
1
0
0 g
2
_
.
(b)
E

E
1

=0
is of the form
_
0 Y
JY
T
0
_
for some (n+1) n valued
map Y , and dY =

n
i=1
g
1
e
ii
Jg
1
2
dx
i
, where J = I
nk,k
.
(c) Let

b be a constant null vector in R
nk,k
, then Y

b is an immersed
isothermic
k
hypersurface in R
n+1
with
I

b
=
n

i=1
u
2
i
dx
2
i
, II

b
=
n

i=1

i
u
i
dx
2
i
,
where (u
1
, . . . , u
n
)
T
= Jg
1
2

b.
(d) If

C = c
1
, . . . , c
n
is a null basis of R
nk,k
, then Y

C = Yc
1
, . . . , Yc
n

is a Combescure sequence of isothermic

k
hypersurfaces in R
n+1
.
(e) If

b
1
, . . . ,

b
n
is another null basis of R
nk,k
, then Yc
i
(x) Y

b
i
(x)
is a Combescure transform for 1 i n.
We need the following Lemma to prove an analogue of Theorem 3.3 for
Combescure sequences of isothermic
k
hypersurfaces in R
n+1
:
Lemma 4.2. Suppose that f(x
1
, . . . , x
n
) is an immersed isothermic
k
hyper-
surface in R
n+1
with fundamental forms
I =
n

i=1
u
2
i
dx
2
i
, II =
n

i=1
u
i
h
i
dx
2
i
,
where (u, u)
k
= 0 and u = (u
1
, . . . , u
n
). Set F = (f
ij
) and = (
1
, . . . ,
n
)
T
by
(4.1) f
ij
=
_

i
(u
i
)x
j
u
j
, i ,= j,
0, i = j,
= (
1
, . . . ,
n
)
T
,
i
=
i
h
i
,
where J = I
nk,k
= diag(
1
, . . . ,
n
). Then
12 NEIL DONALDSON

AND CHUU-LIAN TERNG

(a)
(4.2) =
_
JF + F
T
J J

T
J 0
_
,
is at,
(b) du
T
= u
T
, where
(4.3) = F
T
J + FJ.
Proof. Set
e
i
=
f
x
i
u
i
,
i
= u
i
dx
i
, 1 i n,
and e
n+1
the unit normal eld. Then
1
, . . . ,
n
is the dual frame and
df =
n

i=1

i
e
i
=
n

i=1
u
i
e
i
dx
i
,
Write de
i
=

n+1
j=1

ji
e
j
for 1 i n + 1. Then
(4.4)
ij
= (de
j
, e
i
)
o
, 1 i, j n + 1.
It follows from elementary hypersurface theory that (
ij
)
1i,jn
is the Levi-
Civita connection 1-form for the induced metric I
f
and
(4.5)
ij
=
(u
i
)
x
j
u
j
dx
i

(u
j
)
x
i
u
i
dx
j
,
n+1,i
= h
i
dx
i
, 1 i, j n.
The GaussCodazzi equation for isothermic
k
hypersurfaces in R
n+1
is given
by the atness of the o(n + 1)-valued 1-form = (
ij
)
1i,jn+1
, which
written in terms of F, is (4.2). This proves statement (a).
Use the condition

n
i=1

i
u
2
i
= 0 and the same computation as for the
Guichard
k
orthogonal co-ordinate system case to conclude du
T
= u
T
.

We have an analogue of Theorem 3.3 for Combescure sequences of

isothermic
k
hypersurfaces in R
n+1
:
Theorem 4.3. Let f
1
, . . . , f
n
be a Combescure sequence of isothermic
k
hypersurfaces in R
n+1
, and f
i
(0) = 0. Then there is a solution (F, )
of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system and a null basis

C = c
1
, . . . , c
n
so that

1
, . . . ,
n
is the Combescure sequence of isothermic
k
hypersurfaces of
R
n+1
constructed from (F, ) and

C as in Theorem 4.1.
Proof. Let u

= (u
1
, . . . , u
n
) be the metric elds for I
f

(i.e., I
f

n
i=1
u
2
i
dx
2
i
) such that
(f

)
x
i
u
i
=
(f
1
)
x
i
u
1i
for all 1 n. We denote these vector elds by e
1
, . . . , e
n
(independent
of ). This means that (e
1
, . . . , e
n
) is the orthonormal principal curvature
ISOTHERMIC HYPERSURFACES 13
frame for each f

. So the unit normal e

n+1
for f
1
is also the unit normal for
each f

. By (4.4) the at o(n + 1) connection 1-form (

ij
())
(
ij
)() = (de
j
, e
i
)
0
,
is independent of , denoted it by = (
ij
). But
ij
is also given by (4.5). So
(u
i
)x
j
u
j
is independent of . Let F

and

denote the matrix maps dened for

f

as in (4.1). So we have proved that F

= F and

= are independent of
, and is given by (4.2). Let be the o(nk, k) connection 1-form dened
by (4.3). By Lemma 4.2 (b), du
T

= u
T

for 1 n, i.e., u
T

is a parallel
frame for . By Denition 1.2 of a Combescure sequence of isothermic
k
hypersurfaces, u
1
, . . . , u
n
are linearly independent. This proves that is
at. In Example 2.5, we see that the Lax pair

for (F, ) at = 0 is
_
0
0
_
. By Proposition 2.1 (3), (F, ) is a solution of the
O(2n+1k,k)
O(n+1)O(nk,k)
-
system.
The rest of the proof can be carried out the same way as that of Theorem
3.3.
Since an o(nk, k)-connection with n1 linearly independent parallel
sections is at, we have
Corollary 4.4. Suppose f
1
, . . . , f
n1
are isothermic
k
of R
n+1
such that (i)
f
i
(x) f
i+1
(x) is Combescure for each 1 i n 2, (ii) u
1
, . . . , u
n1
are
linearly independent at every point, where u

= (u
1
, . . . , u
n
) is the metric
eld for the induced metric I
f

chosen in such a way that

(f

)x
j
u
j
=
(f
1
)x
j
u
1j
for all 2 n 1 and 1 j n. There then exists an isothermic
k
hypersurface f
n
in R
n+1
such that f
1
, . . . , f
n
is a Combescure sequence of
isothermic
k
hypersurfaces.
Classical Christoel transforms
When n = 2, we may assume that c = (1, 1)
T
,

b = (1, 1)
T
, f = Yc,
g
1
= (e
1
, e
2
, e
3
), I
f
= e
2u
(dx
2
1
+ dx
2
2
), and g
1
2
=
_
cosh u sinhu
sinhu cosh u
_
. Then
f(x) = Y (x)c

f(x) = Y (x)

b given in Theorem 4.1 is a classical Christoel

transform of isothermic surfaces in R
3
. Since g
1
2
c = e
u
(1, 1)
T
and g
1
2

b =
e
u
(1, 1)
T
, we have df = e
u
(e
1
dx
1
+ e
2
dx
2
), d

f = e
u
(e
1
dx
1
e
2
dx
2
),
e
1
, e
2
are principal directions, and f

f is orientation reversing.
Isothermic
k
submanifolds
An n-dimensional submanifold in R
n+m
is isothermic
k
if it has at nor-
mal bundle and line of curvature co-ordinates so that the induced metric is
Guichard
k
. Theorems 4.1 and 4.3 for isothermic
k
hypersurfaces can be gen-
eralised easily to isothermic
k
submanifolds (with essentially the same state-
ments except replacing
O(2n+1k,k)
O(n+1)O(nk,k)
by
O(2n+mk,k)
O(n+m)O(nk,k)
). In particular,
14 NEIL DONALDSON

AND CHUU-LIAN TERNG

2-submanifolds of R
2+m
are isothermic
1
if and only if they are isothermic in
the sense of [1] and [2].
5. Loop Groups, Ribaucour and Lie transformations
Classically, geometric transforms for isothermic surfaces were constructed
using dierential geometric techniques. The GaussCodazzi equations for
surfaces in R
3
admitting a huge class of geometric transforms often turn out
to be soliton equations, and their geometric transforms can be constructed
from the dressing action on the space of solutions. Although the classical
geometric constructions are beautiful, they often seem mysterious. But if
the GaussCodazzi equation is a soliton equation, then the techniques from
soliton theory give a unied method to construct these geometric trans-
forms. In this section, we rst review the dressing action of a loop group
on the space of solutions of the U/K-system, then use the dressing action
of a rational loop with two poles and the scaling transformation of solutions
of the U/K-system to, respectively, construct geometric Ribaucour and Lie
transforms for Combescure sequences isothermic
k
hypersurfaces in R
n+1
.
We also show that the conjugation of a Ribaucour transform by a Lie trans-
form is Ribaucour and that Christoel transforms commute with Ribaucour
transforms.
The Dressing action
Fix > 0. Let L(U
C
) denote the group of holomorphic maps from

1
< [[ < to U
C
satisfying the U/K-reality condition (2.2), L
+
(U
C
)
the subgroup of g L(U
C
) that can be extended holomorphically to C, and
L

(U
C
) the subgroup of g L(U
C
) that can be extended holomorphically
to = with g() = I. The dressing action of L

(U
C
) on L
+
(U
C
) is
dened as follows: Given g

(U
C
), use the Birkho Factorization The-
orem to see that for generic g

we can factor g

g
+
= g
+
g

uniquely with
g

(U
C
), then the dressing action of L

(U
C
) on L
+
(U
C
) is dened by
g

g
+
= g
+
.
It is proved in [8] that the L

(U
C
) action induces an action on the space
of solutions of the U/K-system and on the space of normalized extended
frames: If E is the normalized frame of a solution v of the U/K-system, then
E(x) L
+
(U
C
). Given g

(U
C
), let

E(x) = g

E(x). Recall that

E(x) is obtained by factoring g

E(x) =

E(x) g

(x) with

E(x) L
+
(U
C
)
and g

(x) L

(U
C
). Then

E is the normalized frame for a new solution
v = v + (m
1
) of the U/K-system, where m
1
is the coecient of
1
for
g

, and () is the projection from p to a

p. Note that Theorem 5.1 is

a special case of this general Theorem when we take g

= p
,L
dened by
(5.2). In this case, the factorization is carried out explicitly.
The Cauchy problem
The U/K-system as an exterior dierential system is involutive and, by
CartanKahler Theory, local analytic solutions are determined by a germ of
ISOTHERMIC HYPERSURFACES 15
an a

p-valued analytic function on a regular line (cf. [9]). Recall that

a a is regular if ad(a) : a

p k is injective. Solutions in the orbit of

L

(U
C
) at the vacuum solution are local analytic.
We can also apply the theory of inverse scattering to nd global solutions
to the Cauchy problem (cf. [7]). Explicitly: x a regular b a, and let
0
:
R a

p be rapidly decreasing and have L

1
-norm less than 1; then there
exists a unique global solution to the U/K-system such that (tb) =
0
(t).
The Dressing action of a simple element
Sections 3 and 4 explain how to construct Combescure sequences of
Guichard
k
orthogonal co-ordinate systems on R
n
and isothermic
k
hypersur-
faces in R
n+1
from solutions of the
O(2nk,k)
O(n)O(nk,k)
-systems and
O(2n+1k,k)
O(n+1)O(nk,k)
-
systems respectively. The dressing action of the loop group on the space of
solutions of the U/K-system gives rise to various transformations of the
corresponding geometric objects. However only the action of the simplest
kind of rational loops in L

(U
C
) are known to give geometrically interesting
transforms.
The dressing action of a rational germ on the space of solutions of the
U/K-system can be computed by residue calculus. We can thus use The-
orems 3.2 and 4.1 to construct geometric transformations of Guichard
k
orthogonal co-ordinate systems on R
n
and isothermic
k
hypersurfaces in
R
n+1
. We give the explicit formulae for the dressing actions of the simplest
type of rational loops and show that the corresponding transformations of
isothermic
k
hypersurfaces are Ribaucour transformations.
Simple elements p
,L
Let be the conjugation by = I
n+1,n
as in Example 2.5 that gives the
symmetric space
U
K
=
O(2n+1k,k)
O(n+1)O(nk,k)
. Choose a scalar R

iR

, and
an isotropic line such that either
(5.1) R
2n+1k,k
and R

, or R
n+1
iR
nk,k
and iR

,
where R

= R 0. Let L =
C
and suppose in addition that L ,= L
(equivalently L , L). Let
L
denote the projection onto L away from
L

. In fact, if = v) with (v, v)

k
= 0 and (v, v)
k
= 1, then

L
= vv
T
,
L
= vv
T
.
Dene the simple element p
,L
by
(5.2) p
,L
() =

+

L
+
(LL)
+
+

L
.
It is easily checked that p
,L
satises the U/K-reality condition.
The following is known ([1, 5]) for the
O(2n1,1)
O(n)O(n1,1)
-system, and ex-
actly the same proof works for the
O(2n+1k,k)
O(n+1)O(nk,k)
-system and for the
O(2nk,k)
O(n)O(nk,k)
-system.
16 NEIL DONALDSON

AND CHUU-LIAN TERNG

Theorem 5.1. Let

U
K
=
O(2nk,k)
O(n)O(nk,k)
, E

(x) the normalized extended

frame for a solution (F, ) of the
U
K
-system, C, L =
C
an isotropic
line in C
2n+1k,k
such that satises (5.1), and p
,L
the simple element
dened by (5.2). Then:
(1) There is an open subset B of the origin in R
n
such that E
1

(x)L ,=
(E
1

(x)L) for all x B, where (g) = I

n+1,n
gI
1
n+1,n
,
(2)

L(x) = E
1

(x)L and satisfy (5.1), so there exist

Q(x) C
n+1
and

Z(x) C
n
such that

L(x) =
__

Q(x)

Z(x)
__
with (

Q,

Q)
0
= (

Z,

Z)
k
=
2 and

Q =

Q and

Z = sgn(
2
)

Z,
(3)

E(x) := E(x)p
1
,E
1

(x)L
is an extended frame for a new solution
(

F, ) of the U/K-system, where
(

F, ) = (F, ) +

Z

Q
T
.
Ribaucour transforms
Let M and

M be two hypersurfaces in R
n+1
. A dieomorphism : M

M is a Ribaucour transform (cf. [1]) if:

(1) For each p M there is an n-sphere S
p
containing p and (p) such
that M and

M are tangent to S
p
at p and (p) respectively.
(2) d
p
maps eigenvectors of the shape operator of M at p to eigenvec-
tors of the shape operator of

M at (p).
(3) The tangent line through p in a principal direction v meets the tan-
gent line through (p) in the direction of d
p
(v) at equal distance.
Remarks
(i) The equidistance claim of (3) follows automatically from (1) and (2)
provided that the principal directions through p, (p) in fact meet.
(ii) In classical dierential geometry, a sphere congruence in R
n+1
is an
n parameter family of n-spheres o = S
x
[ x O, where O is an
open subset of R
n
. It is easy to see that there are two hypersurfaces
f(x) and

f(x) in R
n+1
so that f(x
0
),

f(x
0
) S
x
0
and S
x
0
is tangent
to f and

f at f(x
0
) and

f(x
0
) respectively. These are the called
envelopes of o. We also call the resulting map f(x)

f(x) a sphere
congruence. A Ribaucour transform is thus a sphere congruence that
preserves lines of principal curvature.
(iii) Let M be a hypersurface in R
n+1
, and (e
1
, . . . , e
n+1
) an orthonor-
mal frame on M such that e
1
, . . . , e
n
are principal directions (i.e.,
unit eigenvectors for the shape operator of M). Let

M be another
hypersurface, : M

M a dieomorphism,

e
n+1
the unit normal
eld on

M, and

e
i
is the direction of d(e
i
) for 1 i n. Then
is a Ribaucour transform if:
(a)

e
i
is a principal direction for

M for 1 i n.
ISOTHERMIC HYPERSURFACES 17
(b) There exist functions h
1
, . . . , h
n+1
on M such that
(p) + h
i
(p)

e
i
(p) = p + h
i
(p)e
i
(p), 1 i n + 1
for all p M.
Theorem 5.2. Let (F, ) be a solution of the
O(2n1k,k)
O(n+1)O(nk,k)
-system, E
the normalized extended frame for (F, ), Y dened by
E

E
1

=0
=
_
0 Y
JY
T
0
_
,
where c a constant null vector in R
nk,k
, and f
c
= Yc (the resulting
isothermic
k
hypersurface in R
n+1
). Let p
,L
be the simple element dened
by (5.2), and

Q,

Z and

E as in Theorem 5.1. Then:
(1) E
0
=
_
g
1
0
0 g
2
_
and

E
0
=
_
g
1
0
0 g
2
_
for some g
1
, g
1
O(n + 1) and
g
2
, g
2
O(n k, k).
(2) Write g
1
= (e
1
, . . . , e
n+1
), g
1
= (

e
1
, . . . ,

e
n+1
), u = Jg
1
2
c =
(u
1
, . . . , u
n
)
T
,

Q = (q
1
, . . . , q
n+1
)
T
, and

Z = (z
1
, . . . , z
n
)
T
. Set

f
c
:= f
c

Z
T
u

n+1

j=1
q
j
e
j
.
Then
(a)

f
c
is isothermic
k
,
(b) e
i

n
i=1
and

e
i

n
i=1
are principal curvature directions of f
c
and

f
c
respectively,
(c) R
,L
: f
c


f
c
is a Ribaucour transform, in fact,

f
c

(

Z, u)
k
q
i

e
i
= f
c

Z, u)
k
q
i
e
i
, 1 i n + 1.
Proof. Note that
(5.3) p
1
,E
1

L
() =
_
I +

2

Q
T

Z
T
J

Z

Q
T
I

2

Z
T
J
_
, J = I
nk,k
,
from which we write

= E

_
I +

2

Q
T

Z
T
J

Z

Q
T
I

2

Q
T
J
_
.
Evaluate at = 0 to get
(5.4)
_
g
1
0
0 g
2
_
=
_
g
1
0
0 g
2
_ _
I

Q

Q
T
0
0 I +

Z

Z
T
J
_
.
18 NEIL DONALDSON

AND CHUU-LIAN TERNG

Let

e
i
and e
i
denote the ith column of g
1
and g
1
respectively. By (5.4) we
have

e
i
= e
i
q
i
n+1

j=1
q
j
e
j
, 1 i n + 1. (5.5)
Recall that u = Jg
1
2
c. Notice that

E
1

=0
=
_
0

Y
J

Y
T
0
_
,
E

E
1

=0
=
_
0 Y
JY
T
0
_
,
where J = I
nk,k
. By (5.3),
(5.6)

Y = Y
1
g
1

Z
T
Jg
1
2
.
Set

f
c
=

Yc. Then since f = Yc and (5.6),

f
c
= f
c

1
g
1

Z
T
Jg
1
2
c.
Or equivalently,

f
c
f
c
=
1
g
1

Q(

Z, g
1
2
c)
k
=

Z
T
u

n+1

i=1
q
i
e
i
.
By (5.5),

e
i
e
i
= q
i
n+1

j=1
q
j
e
j
, 1 i n + 1.
Hence

f
c
+ f
i

e
i
= f
c
+ f
i
e
i
for 1 i n + 1, where f
i
=

Z
T
u
q
i
.
As a consequence of Theorems 4.3 and 5.1 we have:
Corollary 5.3. Let f
1
, . . . , f
n
be a Combescure sequence of isothermic
k
hypersurfaces in R
n+1
, and p
,L
a simple element dened by (5.2) so that
, L satisfying condition (5.1). Then there is a new Combescure sequence of
isothermic
k
hypersurfaces

f
1
, . . . ,

f
n
so that f
i
(x)

f
i
(x) is a Ribaucour
transform for each i. Moreover, the radius of S
i
(x) depends only on x but
not on i, where S
i
(x
0
) is the sphere tangent to f
i
and

f
i
at f
i
(x
0
) and

f
i
(x
0
)
respectively.
Ribaucour transforms of a Combescure sequence of isothermic
k
hypersur-
faces in R
n+1
can also be constructed by solving a system of compatible
ODEs:
Theorem5.4. Let f = f
1
, . . . , f
n
be a Combescure sequence of isothermic
k
hypersurfaces in R
n+1
with I
f
1
=

n
i=1
u
2
i
dx
2
i
, f
i
(0) = 0, and f
x
i
(0) is par-
allel to v
i
for 1 i n, where v
1
, . . . , v
n+1
is the standard basis of R
n+1
.
Let (F, ) denote the solution of the
O(2n+1k,k)
O(n+1)O(nk,k)
-system constructed from
f as in Theorem 4.3, i.e., F, are dened by (4.1). Let

the Lax pair (2.4)

ISOTHERMIC HYPERSURFACES 19
for (F, ). Given R

iR

and = v) satisfying (5.1), then the fol-

lowing rst order system for y : R
n
C
2n
is solvable and has a unique
solution:
dy =

y, y(0) = v.
Moreover,
(1) y(x)) satises condition (5.1), so there exists

Q(x) R
n+1
and

Z(x) R
n
if R and in iR
n
if iR

such that y) =
__

Z
__
and (

Q,

Q)
0
= (

Z,

Z)
k
= 2,
(2)

f
i
:= f
i


Z
T
u

n+1
j=1
q
j
e
j
denes a Ribaucour transform R
,L
,
where e
i
=
fx
i
u
i
for 1 i n, e
n+1
is the unit normal for f,

Q = (q
1
, . . . , q
n+1
)
T
, and u = (u
1
, . . . , u
n
)
T
,
(3)

f
1
, . . . ,

f
n
is a Combescure sequence of isothermic
k
hypersurfaces.
Proof. Note that Theorem 5.1(2) says that

L = E

L, where E is an extended
frame of (F, ). Then
d

L = E
1

dE

E
1

L =

L.
This gives a system of ODE dy =

y. This system has n independent

solutions if and only if

is at. The rest follows.

Let R
,L
and C

B
denote the Ribaucour and Combescure transforms for
Combescure sequences of isothermic
k
hypersurfaces in R
n+1
. It follows from
the constructions of these transforms commute, i.e.,
C

B
R
,L
= R
,L
C

B
.
Scaling transforms
We review the scaling transform on the space of solutions of the U/K-
system (cf. [8]). Let Aut(L) be the group of automorphisms of the group
L dened in the beginning of this section, and : R
+
Aut(L) the group
homomorphism dened by
((r)(g))() = g(r).
It is easy to see that (r)(L

) L

. The multiplication for the semi-direct

product R
+

is
(r
1
, g
1
)(r
2
, g
2
) = (r
1
r
2
, g
1
((r
1
)(g
2
))).
The R
+
-action on the solutions of the U/K-system is dened as follows: If
r R
+
and v is a solution of the U/K-system, then v(x) = r
1
v(rx) is
again a solution and the normalized frames are related by

E(x, ) = E(r
1
x, r).
Then r v = v and r E =

E dene an R
+
-action on the space of solutions
and the space of normalized extended frames of the U/K-system.
20 NEIL DONALDSON

AND CHUU-LIAN TERNG

It is proved in [8] that the R

+
-action and the L

-action on the spaces

of solutions and normalized extended frames extend to an action of the
semi-direct product R
+

on these spaces. Since (r, 1)(1, g)(r

1
, 1) =
(1, (r)(g)), we have
(5.7) r (g (r
1
v)) = ((r)(g)) v, r (g (r
1
E)) = ((r)(g)) E,
where v is a solution of the U/K-system and E is its normalized extended
frame.
Lie Transforms
As a consequence of the R
+
-action and Theorem 4.1 we get,
Theorem 5.5. If f = f
1
, . . . , f
n
is a Combescure sequence of isothermic
k
hypersurfaces in R
n+1
and r R
+
, then

f =

f
1
, . . . ,

f
n
is a Combescure
sequence of isothermic
k
hypersurfaces in R
n+1
, where

f
i
(x) = rf
i
(r
1
x).
We denote this transform by S
r
.
The scaling transform exists for the Sine-Gordon equation q
xt
= sinq:
q(x, t) = q(rx, r
1
t) is again a solution of the SGE if q is. The Gauss
Codazzi equation for K = 1 surfaces in R
3
is the SGE, and the scaling
transformation induces a transformation of surfaces in R
3
with K = 1.
Such transformations were constructed by Lie. We therefore call the trans-
formation given in Theorem 5.5 the Lie transform for isothermic
k
hypersur-
faces.
It follows from (r)(p
,L
) = p
r
1
,L
and (5.7) that we have:
Corollary 5.6. Let S
r
and R
,L
denote Lie transform and the Ribaucour
transform for Combescure sequences of isothermic
k
hypersurfacesin R
n+1
respectively. Then S
r
R
,L
S
r
1 = R
r
1
,L
.
Ribaucour transforms are dened for submanifolds with at normal bun-
dle (cf. [1]): Let M,

M be n-dimensional submanifolds in R
n+m
with at nor-
mal bundle, and (e
1
, . . . , e
n
) an orthonormal frame on M such that e
1
, . . . , e
n
are common eigenvectors for the shape operators of M, and e
n+1
, . . . , e
n+m
are a parallel normal eld to M. A Ribaucour transformation from M to

M is a bundle morphism P : (M) (

M) which covers a dieomorphism
: M

M satisfying the following conditions:
(1) For 1 i n, d
p
(e
i
(p)) is a common eigenvector of the shape
operators of

M at (p),
(2) (

e
n+1
, . . . ,

e
n+m
) := (P(e
n+1
), . . . , P(e
n+m
)) is an orthonormal par-
allel normal frame on

M.
(3) There exist smooth functions r
i
on M such that
p + r
i
(p)e
i
(p) = (p) + r
i
(p)

e
i
(p), 1 i n + m,
where

e
i
is the direction of d(e
i
) for 1 i n.
ISOTHERMIC HYPERSURFACES 21
It follows that if is Ribaucour, then M and

M envelop m congruences
of n-spheres: x n + 1 i n + m, the ith congruence is the family of
n-spheres S
p
in R
n+m
with centres p + r
i
(p)e
i
(p) and radii [r
i
(p)[.
Ribaucour and Lie transforms can similarly be constructed for Combes-
cure sequences of isothermic
k
n-submanifolds in R
n+m
as for Combescure
sequences of isothermic
k
hypersurfaces in R
n+1
.
The rst author was supported in part by an NSF Advance Grant. The
second author was supported in part by NSF Grant DMS-0707132.
References
[1] Br uck, M., Du, X., Park, J., Terng, C-L.: The submanifold geometries associated to
Grassmannian systems. Mem. Amer. Math. Soc. 735, viii+95 (2002)
[2] Burstall, F.E.: Isothermic surfaces, conformal geometry, Cliord algebras and inte-
grable systems; in Integrable systems, geometry, and topology, AMS/IP Stud. Adv.
Math. 36, 182 (2006)
[3] Ciesli nski, J., Goldstein, P., Sym, A.: Isothermic surfaces in E
3
as soliton surfaces.
Phys. Lett. A 205, 3743 (1995)
[4] Ciesli nski,J.: The Darboux-Bianchi transformation for isothermic surfaces. Di.
Geom. Appl. 7, 1-28 (1997)
[5] Donaldson, N.M., Terng, C-L.: Conformally at submanifolds in spheres and inte-
grable systems. arXiv:math-dg/0803.2754v2 (2008)
[6] Hertrich-Jeromin, U., Pedit, F.: Remarks on the Darboux transform of isothermic
surfaces. Doc. Math. 2, 313333 (1997)
[7] Terng, C-L.: Soliton equations and dierential geometry. J. Di. Geom. 45, 407445
(1997)
[8] Terng, C-L., Uhlenbeck, K.: Backlund transformations and loop group actions.
Comm. Pure Appl. Math. 53, 175 (2000)
[9] Terng, C-L., Wang, E.: Curved ats, exterior dierential systems, and conservation
laws; in Complex, contact and symmetric manifolds, Progr. Math. 234, 235254
(2005)
[10] Terng, C-L., Wang, E.: Transformations of at Lagrangian immersions and Egoro
nets, Asian J. Math. 12, 99119 (2008)
[11] Tojeiro, R.: Isothermic submanifolds of Euclidean space. J. reine angew. Math. 598,
124 (2006)
Department of Mathematics, University of California at Irvine, Irvine, CA
92697-3875
E-mail address: ndonalds@math.uci.edu
Department of Mathematics, University of California at Irvine, Irvine, CA
92697-3875
E-mail address: cterng@math.uci.edu