4 Multidimensional Geometry

The notion of a curve in Rn is readily extended to a notion of an n-

dimensional surface in Rn+l , l ≥ 0.
A regular n-dimensional surface of class C m (m ≥ 1) is the point set in
real (n + l)-dimensional space Rn+l locally represented by some parametric
n-dimensional domain S n and a parametrization

r(s) : S n → Rn+l , r(s) = [r1 (s), . . . , rn+l (s)] , s = (s1 , . . . , sn ) , (4.1)

such that all partial derivatives of r(s) of order m are continuous in S n and
the rank of the matrix (∂ri /∂sj ), i = 1, . . . , n + l, j = 1, . . . , n, equals n at
each point of S n . The vector equation (4.1) is called a parametric equation
of the n-dimensional surface while the variables si , i = 1, . . . , n, are referred
to as curvilinear coordinates on the surface. We shall use the designation S rn
for the surface represented by (4.1). In accordance with the definition a curve
is meant as a one-dimensional surface.
In grid generation methods regular n-dimensional surfaces are typical ob-
jects as boundaries of the domains under consideration, coordinate hyper-
surfaces, and monitor surfaces specified to generate adaptive meshes. The
advanced grid technology also requires the application of the theories of
more sophisticated geometries, namely, Riemannian manifolds which general-
ize regular surfaces. These geometries have real potential to provide efficient
means to control the qualitative properties of grids and develop advanced
grid technologies.
This chapter gives an introduction to the theory of multidimensional sur-
faces and Riemannian manifolds.

4.1 Tangent and Normal Vectors and Tangent Plane

Let s(t) : [a, b] → S n be a representation of a curve in S n . Then the
parametrization

r[s(t)] : [a, b] → Rn+l (4.2)

represents a curve in the n-dimensional surface S rn specified by (4.1). The

tangent vector to this curve forms a tangent vector to the surface S rn . The
62 4 Multidimensional Geometry

Fig. 4.1. Base tangent and normal vectors to coordinate lines

set of all tangent vectors at a point on the surface S rn forms the tangent
n-dimensional plane to the surface at this point.
Analogously to the definition of the coordinate line in space (section 2.2)
there is defined the si th coordinate line in the surface S rn as a curve repre-
sented by the following vector-valued function dependent upon a variable ϕ
in the capacity of si :
r[si (ϕ)] : [a, b] → Rn+l ,
(4.3)
si (ϕ) = (s10 , . . . , si−1 i+1 n
0 , ϕ, s0 , . . . , s0 ) , i fixed ,

here r(s) is the function from (4.1), si (ϕ) ∈ S n , ϕ ∈ [a, b], the constants sj0 ,
j = i are fixed.
Each si th coordinate line defines one basic tangent vector along this curve
∂r
rsi = , i = 1, . . . , n ,
∂si
assuming in (4.3) ϕ = si . The transformation r(s) is of rank n hence the
basic tangent vectors rsi , i = 1, . . . , n, at a point P are independent and
therefore form the tangent plane at this point (Fig. 4.1 for n = 2).
Similarly to the coordinate hypersurface in space (section 2.3) we define a
coordinate hypersurface in S rn as an (n − 1)-dimensional surface lying in S rn
along which all of the coordinates s1 , . . . , sn except one, say si , are varied.
Thus the si th coordinate hypersurface is specified by the parametrization
r[si (s1 , . . . , si−1 , si+1 , . . . , sn ) : S n−1 → Rn+l ,
(4.4)
si (s1 , . . . , si−1 , si+1 , . . . , sn ) = (s1 , . . . , si−1 , si0 , si+1 , . . . , sn ) ,

where si0 fixed, while r(s) is the function from (4.1). We personify this coor-
dinate hypersurface with the equation si = si0 . Equation (4.4) readily yields
that the basic tangent vectors to the hypersurface si = si0 are the vectors rsj ,
j = i.
 4.2 First Groundform 63

A vector lying in the tangent plane to S rn and orthogonal to the coordi-

nate hypersurface si = si0 in S rn (consequently to its basic tangent vectors
rsj , j = i) is called a vector orthogonal or a normal vector to this hyper-
surface in S rn . Let us introduce, analogously to the normal vectors in space,
as a basic normal vector to the coordinate hypersurface si = si0 in S rn the
vector designated by ∇si for which

∇si · rsj = δji , i, j = 1, . . . , n , i fixed , (4.5)

(Fig. 4.1).
The basic normal vectors to the coordinate hypersurface in S rn are de-
scribed inambiguously in the following section through the basic tangent
vectors and elements of the metric tensors of S rn .

4.2 First Groundform

All the properties of an n-dimensional surface S rn which can be described
without referring to the surrounding space are called intrinsic properties of
the surface and their description constitutes the intrinsic geometry of the
surface. Any characteristic of the intrinsic geometry is defined by the surface
metric tensor whose elements are specified through the operation of the dot
product on the basic tangent vectors.

4.2.1 Covariant Metric Tensor

Definition

The covariant metric tensor of any regular n-dimensional surface S rn rep-

resented in the coordinates s1 , . . . , sn by (4.1) is the matrix (gij
rs
), i, j =
1, . . . , n, where
rs
gij = rsi · rsj , i, j = 1, . . . , n . (4.6)

In particular, when the surface is a monitor surface i.e. it is identified with

the graph of the values of some vector-valued function u(s) over a domain
S n then this surface is parametrized by the equation r(s) = [s, r(s)] and
consequently
∂u ∂u
rs
gij = δij + i · j , i, j = 1, . . . , n .
∂s ∂s

Quadratic Form

The differential quadratic form

rs
gij dsi dsj , i, j = 1, . . . , n ,
64 4 Multidimensional Geometry

relating to the line elements in space, is called the first groundform or funda-
mental form of the surface. It represents the value of the square of the length
of an elementary displacement dr (see Fig. 2.4) on the surface. Therefore the
length of the curve (4.2) in the surface S rn is computed by the formula
 b
dsi dsj
l= gij dt , i, j = 1, . . . , n ,
a dt dt

which is similar to (2.16).

rs
Let the Jacobian of (gij ) be designated by g rs . Then, analogously to the
length of the line (4.2), the n-dimensional area of the surface S rn is computed
from the formula 
√ rs
S= g ds .
Sn

Basic Parallelepiped

The basic parallelepiped in S rn with respect to the coordinates s1 , . . . , sn

is an n-dimensional parallelepipedwhose edges are the basic tangent vectors
rs for a fixed index i has the geometrical
rsi , i = 1, . . . , n. So the quantity gii
meaning of the length of the ith edge of the basic parallelepiped (see Fig.
4.2 in the case n = 2). Note the uniformly contracted basic parallelepiped
represents to a high order of accuracy the cell of the coordinate grid in S rn
in the case when the parametric domain S n is a logical domain (see also
Sect. 2.2).

Fig. 4.2. Geometric meaning of the metric elements

 4.2 First Groundform 65

4.2.2 Contravariant Metric Tensor

Definition

The contravariant metric tensor of the surface S rn in the coordinates s1 , . . . , sn

ij rs
is the matrix (gsr ), i, j = 1, . . . , n, inverse to (gij ), i.e. the following relations
are held
rs jk
gij gsr = δki , i, j, k = 1, . . . , n . (4.7)
Thus analogously to (2.21) we obtain, in the case n = 2,
ij
gsr = (−1)i+j g3−i
rs
3−j /g
rs
,
rs 3−i 3−j
gij = (−1)i+j g rs gsr , i, j = 1, 2 , (4.8)
with fixed indices i and j. Similarly to (2.22) we find, in the case n = 3,
1 rs
ij
gsr = (g rs
gi+2 − gi+1
rs rs
gi+2 j+1 ) ,
g rs i+1 j+1 j+2 j+2

rs
gij = g rs (gsr
i+1 j+1 i+2
gsr j+2
− gsr
i+1 j+2 i+2
gsr j+1
), (4.9)

i, j = 1, 2, 3 , i, j fixed ,
with the convention that any index, say l, is identified with l ± 3.

Computation of Basic Normal Vectors

Using the elements of the contravariant metric tensor we can readily find the
expression for the basic normal vector ∇si to the coordinate hypersurface
si = si0 , satisfying (4.5), through the basic tangent vectors rsj , j = 1, . . . , n.
Namely
∇si = gsr
ij
rsj , i, j = 1, . . . , n . (4.10)
Indeed, the condition (4.5) is observed since
∇si · rsk = gsr
ij
rsj · rsk = gsr
ij rs
gjk = δki , i, j, k = 1, . . . , n ,

i.e. the vector ∇si is orthogonal to the coordinate hypersurface si = const in

S rn at the point of consideration. As ∇si · rsi = 1, i fixed, we conclude that
the both basic vectors ∇si and rsi have the same direction with respect to
the hypersurface si = const in S rn .
From (4.10) we readily find that
∇si · ∇sj = gsr
ij
, i, j = 1, . . . , n . (4.11)
Note also that for the length of ∇si we obtain the following expression
√ 
|∇si | = ∇si · ∇si = gsr ii , i = 1, . . . , n , i fixed .
66 4 Multidimensional Geometry

Since the normal vector ∇si is orthogonal to the vectors rsj , j = i, we found
that the distance di between the ith (n − 1)-dimensional faces of the basic
parallelepiped formed by the tangent vectors rsj , j = 1, . . . , n, is computed
as follows:
∇si 1
di = rsi · = , i = 1, . . . , n , i fixed .
∇si ii
gsr
 rs
Thus with respect to the basic parallelepiped the quantity gii (i fixed) is

ii
the length of its ith edge, while the quantity 1/ gsr (i fixed) is the distance
between the parallel (n − 1)-dimensional faces of the parallelepiped, which
are formed by the vectors rsj , j = i (see Fig. 4.2 for n = 2).

Normal Vector to a Hypersurface

Formula (4.10) is readily extended to the case of the hypersurface in S rn

defined by an equation ϕ(s) = 0. Namely, a normal n to this hypersurface in
S rn is specified by the following formula
ik
n = ϕsi gsr rsk , i, k = 1, . . . , n , (4.12)

that is a generalization of (4.10) obtained for the equation ϕ(s) ≡ si − const.

The validity of (4.12) will be proved if we show that

n·t=0,

where t is an arbitrary tangent vector to the hypersurface ϕ(s) = 0. Without

loss of generality we can assume that ϕsn = 0 at a point s ∈ S n under
consideration. Then the equation ϕ(s) = 0 is resolved with respect to sn , i.e.
there exists a function sn (s1 , . . . , sn−1 ), (s1 , . . . , sn−1 ) ∈ S n−1 such that

ϕ[s1 , . . . , sn−1 , sn (s1 , . . . , sn−1 )] ≡ 0 , (s1 , . . . , sn−1 ) ∈ S n−1 .

Therefore, using (4.1), we can locally specify the hypersurface ϕ(s) = 0 in

the coordinates s1 , . . . , sn−1 by the following parametrization

rϕ (s1 , . . . , sn−1 ) = r[s1 , . . . , sn−1 , sn (s1 , . . . , sn−1 )] : S n−1 → Rn+l .

Consequently the basic tangent vectors to this hypersurface with respect to

the coordinates s1 , . . . , sn−1 are computed by
∂sn ϕi
rϕ
si = rs +
i
i
rsn = rsi − s rsn , i = 1, . . . , n − 1 ,
∂s ϕsn

since ∂sn /∂si = −ϕsi /ϕsn , i = 1, . . . , n − 1. Using these relations and (4.12)
gives
 4.3 Generalization to Riemannian Manifolds 67
 ϕi 
n · rϕ
si
mk
= ϕsm gsr rsk rsi − s rsn
ϕsn
 ϕsi rs 
= ϕsm gsr gki −
mk rs
g
ϕsn kn
ϕ i ϕsn
= ϕsi − s = 0 , i = 1, . . . , n − 1 .
ϕsn
As an arbitrary tangent vector t to the hypersurface is expanded by rϕ si ,
i = 1, . . . , n − 1, we obtain that n · t = 0, i.e. formula (4.12) gives a real
expression for the vector n normal to the hypersurface ϕ(s) = 0.

We have, by virtue of (4.12), the following formula for the length of n
√ 
|n| = n · n = (ϕsi gsr ik r k ) · (ϕ m g ml r l )
s s sr s

ik g ml g rs
= ϕsi ϕsm gsr (4.13)
sr kl

ml ,
= ϕsm ϕsl gsr i, k, l, m = 1, . . . , n .

4.3 Generalization to Riemannian Manifolds

4.3.1 Definition of the Manifolds

Formula (4.6) readily yields the result that the elements of the covariant
rs rv
metric tensor (gij ) and (gij ) of the regular surface S rn in arbitrary coordi-
1 1
nates s , . . . , s and v , . . . , v n , respectively, are connected by the following
n

relations

rs rv ∂v k ∂v l
gij = gkl , i, j, k, l = 1, . . . , n . (4.14)
∂si ∂sj
Indeed,

∂v k ∂v m
rs
gij = rsi · rsj = rvk i
· rvm j
∂s ∂s
∂v k ∂v m k
rv ∂v ∂v
m
= rvk · rvm = gkm , i, j, k, m = 1, . . . , n ,
∂si ∂sj ∂si ∂sj
i.e. equations (4.14) are held.
ij
Analogously, the elements of the contravariant metric tensor (gsr ) and
1 1
(gvr ) of the regular surface S in the coordinates s , . . . , s and v , . . . , v n ,
ij rn n

respectively, are connected by

ij kl ∂v i ∂v j
gvr = gsr , i, j, k, l = 1, . . . , n . (4.15)
∂sk ∂sl
ij
For showing that the components gvr are subject to (4.15), it is sufficient to
demonstrate that the matrix
68 4 Multidimensional Geometry

Fig. 4.3. Illustration for Riemannian manifold maps

 ∂v i ∂v j 
kl
gsr , i, j, k, l = 1, . . . , n ,
∂sk ∂sl
ij rv ij
coincides with (gvr ), i.e. it is the inverse of (gij ) provided the matrix (gsr ) is
rs
the inverse of (gij ). Since (4.14)

rv kl ∂v j ∂v t m p
rs ∂s ∂s ∂v
j
kl ∂v
t
gij gsr = gmp g = δit , i, j, k, l, m, p, t = 1, . . . , n ,
∂sk ∂sl ∂v i ∂v j ∂sk sr ∂sl
i.e. equations (4.15) are valid.
The relations (4.14) and (4.15) valid for the metrics of arbitrary regular
n-dimensional surfaces give rise to the definition of the geometrical objects
called Riemannian manifolds which generalize the regular surfaces.
Namely, a point set M n of Rn+l , l ≥ 1 is called a C m -differential Rie-
mannian manifold (Fig. 4.3) of dimension n if there is a collection (atlas) Φ
of local parametrizations of M (local maps)

rϕ (ϕ) : Sϕn → M , ϕ∈Φ, ϕ ∈ Sϕn ,

(4.16)
ϕ = (ϕ1 , . . . , ϕn ) , rϕ = (rϕ1 , . . . , rϕn+l ) ,

where Sϕn ⊂ Rn is an n-dimensional parametric domain, and the same col-

lection of local variable matrices (local covariant metric tensors)
rϕ
(gij (ϕ)) , i, j = 1, . . . , n , ϕ ∈ Sϕn , ϕ∈Φ, (4.17)

such that
1) ∪ϕ∈Φ rϕ (Sϕn ) = M n , where rϕ (Sϕn ) is the image of Sϕn in M n built by rϕ ,
2) for each ϕ ∈ Φ , rϕ (Sϕn ) is an open space of M n and

rϕ : Sϕn → rϕ (Sϕn ) , ϕ∈Φ,

is a one-to-one continuous mapping,

 4.3 Generalization to Riemannian Manifolds 69

3) for each ϕ, ψ ∈ Φ with Bϕψ = ∅, where

Bϕψ = rϕ (Sϕn ) ∩ rψ (Sψn ) ,

the map
ϕ(ψ) : Sψn ∩ r−1
ψ (Bϕψ ) → R ,
n

where ϕ(ψ) = r−1ϕ [rψ (ψ)], is a C

m
– map,
rϕ
4) for each ϕ ∈ Φ, ϕ ∈ Sϕ the matrix (gij
n
(ϕ)) is positive, symmetric, and
nondegenerate and the functions
rϕ
gij (ϕ) : Sϕn → R, i, j = 1, . . . , n,

are C m -functions,
5) for each ϕ, ψ ∈ Φ, Bϕψ = ∅,

rψ rϕ ∂ϕk ∂ϕl
gij (ψ) = gkl [ϕ(ψ)] (ψ) (ψ) , i, j, k, l = 1, . . . , n , (4.18)
∂ψ i ∂ψ j
where
ϕ ∈ Sϕn ∩ r−1
ϕ (Bϕψ ) , ψ ∈ Sψn ∩ r−1
ψ (Bϕψ ) ,

ϕi (ψ), i = 1, . . . , n, is the ith component of the function ϕ(ψ) = r−1

ϕ (rψ ),
6) Φ is maximal relative to (2-5) that is if there is one more local parametriza-
tion
rθ (θ) : Sθn → X , Sθn ⊂ Rn , θ ∈ Sθn ,
and a nondegenerate, positive, symmetric C m – matrix function
rθ
(gij (θ)) , θ ∈ Sθn ,

and their inclusion into local parametrizations and local matrices, respec-
tively, does not violate the requirements (2-5) – they are in the corre-
sponding collections of Φ.
Here m may be 1, 2, . . . , ∞. C m for m finite means all partial derivatives
of order less than or equal to m exist and are continuous.
The variables ϕ1 , . . . , ϕn of the parametric domain Sϕn , i.e.

ϕ = (ϕ1 , . . . , ϕn ) ∈ Sϕn , ϕ∈Φ,

together with the corresponding transformation

rϕ (ϕ) : Sϕn → M n

are called a local coordinate system or local coordinates of M n .

rϕ
Each local covariant metric tensor (gij (ϕ)) derives the local contravariant
ij rϕ
metric tensor (gϕr (ϕ)) as the inverse matrix of (gij (ϕ)), i.e.
70 4 Multidimensional Geometry
ij
gϕr (ϕ) = Gϕ rϕ
ij /det(gij (ϕ)) , i, j = 1, . . . , n ,

where Gϕ rϕ
ij is the (ij)th cofactor of (gij (ϕ)). Analogously to (4.15) it is readily
verified that the equations (4.18) yield

ij kl ∂ψ i ∂ψ j
gψr (ψ) = gϕr [ϕ(ψ)] , i, j, k, l = 1, . . . , n , (4.19)
∂ϕk ∂ϕl

where ψ k (ϕ), k = 1, . . . , n, is the kth component of the function ψ(ϕ) =

r−1
ψ [rϕ (ϕ)].
The need of multiple charts is essential for geometries that are not diffeo-
morphic to an n-dimensional cube. Such geometries appear in various appli-
cations, the most complicated and difficult for gridding of them are human
organs.

4.3.2 Example of a Riemannian Manifold

In the grid generation theory the Riemannian manifolds appear as tools

to control the grids derived by the generalized Laplace equations. Typi-
cally these manifolds are formulated as a generalization of monitor surfaces.
Namely, let S xn be a regular n-dimensional surface lying in Rn+k and repre-
sented locally by a mapping
x(s) : S n → Rn+k , s = (s1 , . . . , sn ) , x = (x1 , . . . , xn+k ) ,
xs
with a covariant metric tensor (gij ) in the coordinates s1 , . . . , sn
xs
gij = xsi · xsj , i, j = 1, . . . , n . (4.20)
By a monitor surface over S xn there is meant a regular surface whose
points are
[x, f (x)] ∈ Rn+l+k ,
where x ∈ S xn , while
f (x) : S xn → Rl , f = (f 1 , . . . , f l ) ,
is some vector-valued function called a monitor function. The monitor sur-
face is represented in the coordinates s1 , . . . , sn by the following local
parametrization
r(s) : S n → Rn+k+l , r(s) = {x(s), f [x(s)]} .
An extension of the monitor surface over S xn is produced, for example,
by two scalar-valued weight functions z(s) > 0 and v(s) ≥ 0 and one vector-
valued smooth function f (s) = [f 1 (s), . . . , f l (s)] which form a Riemannian
manifold M n whose points and parametrizations are the same as of S xn , while
the elements of the covariant metric tensor in the coordinates s1 , . . . , sn ,
s
designated as gij , are defined as follows:
 4.3 Generalization to Riemannian Manifolds 71

s ∂f ∂f
gij xs
= z(s)gij + v(s) (s) · (s) , i, j = 1, . . . , n . (4.21)
∂si ∂sj

We shall call the Riemannian manifold with the metric (4.21) imposed by
the functions z(s), v(s), and f (s) as a monitor manifold. The function f (s)
will be referred to as a monitor function, while z(s) and v(s) will be called
weight functions.

4.3.3 Christoffel Symbols of Manifolds

Definition of the Symbols

Any covariant metric tensor of a Riemannian manifold M n , designated by

s
(gij ) in the local coordinates si , i = 1, . . . , n, derives the quantities called
the Christoffel symbols of the first and second kinds, which are also referred
to as the three-index symbols. The symbols of the first kind, designated in
the coordinates si , i = 1, . . . , n, by [ij, k]s , are defined as follows:

1  ∂gjk
s
∂g s s 
∂gij
[ij, k]s = i
+ ikj
− k , i, j, k = 1, . . . , n . (4.22)
2 ∂s ∂s ∂s
Equations (4.22) easily yield the following relations for the first derivatives
of the elements of the covariant metric tensor
s
∂gik
= [ij, k]s + [kj, i]s , i, j, k = 1, . . . , n . (4.23)
∂sj
The Christoffel symbols of the second kind, designated in the coordinates
si , i = 1, . . . , n, by s Υijl , are defined by the equations
s
Υijl = gslm [ij, m]s , i, j, l, m = 1, . . . , n . (4.24)

It is seen at once from (4.22) and (4.24) that the Christoffel symbols are
symmetrical in i, j.
Further, when a coordinate system in a formula is fixed we, for simplicity,
shall omit the superscript personifying a coordinate system in the Christoffel
symbols of the first and second kinds thus designating them merely by [ij, k]
and Υijl , respectively.
Analogously to (4.23) we find an expression for the first derivatives of the
elements of the contravariant metric tensor through the Christoffel symbols
of the second kind
s
∂gsij im
lj s ∂gs lj im ∂gml
= gs gml = −gs gs
∂sk ∂sk ∂sk
j
= −gslj gsim ([mk, l] + [lk, m]) = −gsim Υkm − gslj Υlk
i
, (4.25)

i, j, k, l, m = 1, . . . , n .
72 4 Multidimensional Geometry

The Christoffel symbols of the second kind also have some relation to the
formula of the differentiation of the Jacobian g s of the metric tensor (gij
s
).
s
Indeed the rule of the differentiation of the Jacobian g gives
s
∂g s s jm
∂gjm
= g gs , i, j, m = 1, . . . , n ,
∂si ∂si
and the application of (4.23) to this formula yields

∂g s
= g s gsjm ([ji, m] + [mi, j]) = g s [Υji
j m
+ Υmi ] = 2g s Υji
j
, i, j, m = 1, . . . , n .
∂si
Remind repeated indices in a single term mean a summation over them so
we imply in the above equations

∂g s
n
= 2g s Υji
j
= 2g s j
Υji , i, j = 1, . . . , n . (4.26)
∂si j=1

Symbols for Regular Surfaces

In the case of the regular surface S rn represented by (4.1) whose intrinsic

metric is expressed as
s
gij = rsi · rsj , i, j = 1, . . . , n ,

it is readily found, from (4.22), that

[ij, k] = rsi sj · rsk , i, j, k = 1, . . . , n . (4.27)

Using (4.27), (4.24), and (4.10) yields the following formula for the Christoffel
symbols of the second kind of the regular surface S rn in the coordinates
s1 , . . . , sn :

Υijl = gslm (rsi sj · rsm ) = rsi sj · ∇sl , i, j, l, m = 1, . . . , n , (4.28)

where ∇sl is the basic normal vector to the coordinate hypersurface sl = c0

in S rn (see (4.10)).

Transformation of the Christoffel Symbols

Now we shall establish how the Christoffel symbols of two coordinate systems
s v
are related. Let us designate by gij and gij the elements of the covariant
metric tensor of a manifold M in the coordinates s1 , . . . , sn and v 1 , . . . , v n ,
n

respectively. Then, according to (4.18), we have

v s ∂sk ∂sl
gij = gkl , i, j, k, l = 1, . . . , n .
∂v i ∂v j
 4.3 Generalization to Riemannian Manifolds 73

Differentiating these equations with respect to v p gives

v s 2 k
∂gij ∂gkl ∂sm ∂sk ∂sl s ∂ s ∂sl
= + 2gkl , i, j, k, l, m, p = 1, . . . , n ,
∂v p ∂sm ∂v p ∂v i ∂v j ∂v i ∂v p ∂v j
and consequently, applying (4.23),
v
∂gij
= [ip, j]v + [jp, i]v
∂v p
2 k
∂sm ∂sk ∂sl s ∂ s ∂sl
= ([km, l]s + [lm, k]s ) + 2gkl ,
∂v p ∂v i ∂v j ∂v i ∂v p ∂v j
i, j, k, l, m, p = 1, . . . , n .

Therefore, after computing by these formulas the following expression

v v v
∂gij ∂gpi ∂gpj
− + + , i, j, p = 1, . . . , n ,
∂v p ∂v j ∂v i
we readily obtain, using first (4.22) and then (4.24),
2 k
∂sl ∂sk ∂sm s ∂ s ∂sl
[ij, p]v = [km, l]s p i j
+ gkl ,
∂v ∂v ∂v ∂v ∂v ∂v p
i j

v ∂sk ∂sl ∂v p ∂ 2 sk ∂v p (4.29)

Υijp = s Υkl
m
+ ,
∂v i ∂v j ∂sm ∂v i ∂v j ∂sk
i, j, k, l, m, p = 1, . . . , n .

Geometric Meanings of the Symbols

Let us consider the Christoffel symbols of a regular n-dimensional surface

S rn represented by (4.1) and whose covariant metric tensor is defined by
(4.6). We know that the first partial derivatives of the parametrization r(s)
are the tangent vectors forming the tangent plane to the surface and the
rs
elements of the covariant metric tensor (gij ). It appears that the second
partial derivatives of r(s) are connected with the Christoffel symbols.
Let us designate by P the operator which projects the vectors from Rn+l
on the tangent plane to the regular surface S rn ⊂ Rn+l at a point P . Now
considering the vector-valued function

∂2r
rsm sp = , m, p = 1, . . . , n ,
∂sm ∂sp
we can expand the vector P[rsm sp ] (lying in the tangent n-dimensional plane)
in both the base tangential rsi , i = 1, . . . , n, and normal ∇si , i = 1, . . . , n,
vectors. Applying the formula (2.6) in the case of the tangential vectors, i.e.
assuming in (2.6) ai = rsi , i = 1, . . . , n, we find
74 4 Multidimensional Geometry

P[rsm sp ] = aij (P[rsm sp ] · rsj )rsi , i, j, m, p = 1, . . . , n , (4.30)

where aij are the elements of the matrix which is inverse to the matrix (aij )

aij = rsi · rsj , i, j = 1, . . . , n .

Since rsi ·rsj = gij

rs
, i, j = 1, . . . , n, the functions aij in (4.30) are the elements
ij
of the contravariant metric tensor (gsr ), i.e.

aij = gsr
ij
, i, j = 1, . . . , n .

Further, as the operator P projects the vector rsm sp on the plane formed by
the tangent vectors rsi , i = 1, . . . , n, we conclude that

P[rsm sp ] · rsj = rsm sp · rsj , j, m, p = 1, . . . , n .

Therefore equation (4.30) has the following form

ij
P[rsm sp ] = gsr (rsm sp · rsj )rsi , i, j, m, p = 1, . . . , n . (4.31)

Now, applying (4.28) to this equation, we find

i
P[rsm sp ] = Υmp rsi , i, m, p = 1, . . . , n . (4.32)
i
Thus the Christoffel symbol Υmp of the second kind is the ith component
of the vector P[rsm sp ] expanded in the base tangent vectors rsi (see Fig. 4.4
where the vector rsm sp is identified with rmp ).
Similarly we obtain

P[rsm sp ] = [mp, i]∇si , i, m, p = 1, . . . , n . (4.33)

This formula can also be inferred from (4.32). Indeed multiplying (4.10) by
rs
gik gives
rs
rsk = gik ∇si , i, k = 1, . . . , n ,
and substituting this equation for rsi in (4.32), we readily come to (4.33).
So the Christoffel symbols of the first kind represent the components of
the vector P[rsm sp ] expanded in the base normal vectors to the coordinate
hypersurfaces in S rn (see Fig. 4.4 for n = 2).

4.4 Tensors
The theory of multidimensional geometry operates largely with the quantities
called tensors. This section gives an introduction to such geometric objects.
 4.4 Tensors 75

Fig. 4.4. Expension of the vector rsm sp in the base vectors

4.4.1 Definition

Let M n be some n-dimensional Riemannian manifold, in particular, a regular

surface S rn . A tensor of rank k ≥ 0 at a point P of M n is a set of values
defined for each local coordinate system s1 , . . . , sn around this point and
indices i = (i1 , . . . , ik ) ∈ Rk , ij = 1, . . . , n, j = 1, . . . , k, such that the
values obey certain transformation laws when the coordinates are changed.
The number k is called a tensor order or its rank, while the values with the
indices are referred to as tensor components.
The tensors are also distinguished by their types. There are two basic
tapes for the tensors of order k > 0: covariant and contravariant and the
third mixed type having the features of the basic types.

Tensors of Order Zero

A quantity which has the same fixed value at the point P in an arbitrary
coordinate system is called a scalar, or an invariant, or a tensor of order zero.

Covariant Tensors

In the pure covariant case the components of a tensor f of order k, whose des-
ignation is distinguished by indices being subscripts, for example, by fis1 ...ik
for the indices i1 , . . . , ik and coordinates s1 , . . . , sn , are subject to the fol-
lowing relations with respect to arbitrary coordinate systems s1 , . . . , sn and
v1 , . . . , vn :
∂v j1 ∂v jk
fis1 ...ik = fjv1 ...jk · · · , il , jl = 1, . . . , n , l = 1, . . . , k . (4.34)
∂si1 ∂sik
76 4 Multidimensional Geometry

Thus it is sufficient to know the values of the tensor components for some one
fixed coordinate system since its values for other systems can be computed
by (4.34).

Contravariant Tensors

The components of the pure contravariant tensor f of rank k > 0, desig-

nated with the help of indices being superscripts: by fsi1 ...ik in the coordi-
nates s1 , . . . , sn , obey the following law with respect to arbitrary coordinate
systems s1 , . . . , sn and v 1 , . . . , v n :

∂si1 ∂sik
fsi1 ...ik = fvj1 ...jk · · · , il , jl = 1, . . . , n , l = 1, . . . , k . (4.35)
∂v j1 ∂v jk

Mixed Tensors

The notion of the pure covariant and contravariant tensors gives rise to the
concept of a mixed tensor if it is covariant in some indices and contravariant
in the rest of them. Consequently there are used in designations superscripts
for contravariant indices and subscripts for covariant indices. Namely, a mixed
tensor f k times covariant and l times contravariant is a set of values fij11...i
...jl
k
(s)
1 n
dependent on the coordinate system s , . . . , s at the point of consideration
such that for an arbitrary another coordinate system v 1 , . . . , v n

∂v p1 ∂v pk ∂sj1 ∂sjl
fij11...i
...jl
k
(s) = fpm1 ...p
1 ...ml
k
(v) · · · · · · ,
∂si1 ∂sik ∂v m1 ∂v ml (4.36)
ia , pa , jb , mb = 1, . . . , n , a = 1, . . . , k , b = 1, . . . , l .

4.4.2 Examples of Tensors

Covariant Tensors

A typical covariant tensor of the first order is the vector gradϕ where ϕ is a
tensor of order zero. The components of this vector designated as (gradϕ)si
in the coordinates s1 , . . . , sn are computed as

∂ϕ(s)
(gradϕ)si = = ϕsi , i = 1, . . . , n . (4.37)
∂si
It is obvious that the relations (4.34), for k = 1, are held for these values.
Note the second derivatives of ϕ, i.e. the set

ϕsij = ϕsi sj (s) , i, j = 1, . . . , n ,

do not form a tensor of the second rank since

 4.4 Tensors 77

∂v l ∂v m ∂ 2 vl
ϕsi sj (s) = ϕvl vm i j
+ ϕvl i j , i, j, l, m = 1, . . . , n , (4.38)
∂s ∂s ∂s ∂s
and the second term in the right-hand part of these equations impedes sat-
isfaction of (4.34) for k = 2. However using the Christoffel symbols of the
second kind which are not tensors as well, since (4.29), produces the fol-
lowing covariant tensor ∇ij (ϕ) of the second kind whose (ij)th component
designated in the coordinates s1 , . . . , sn as ∇sij (ϕ) is computed as follows:

∇sij (ϕ) = ϕsi sj − ϕsk Υijk , i, j, k = 1, . . . , n . (4.39)

Formulas (4.29) and (4.38) readily yield that these components satisfy the
condition (4.34) for k = 2. Indeed
∇vij (ϕ) = ϕvi vj − ϕvk v Υijk

∂sk ∂sm ∂ 2 sk
= ϕsk sm i j
+ ϕsk i j
∂v ∂v ∂v ∂v
 p l k
∂ 2 sp ∂v k  ∂st
m ∂s ∂s ∂v
−ϕst s Υpl + (4.40)
∂v i ∂v j ∂sm ∂v i ∂v j ∂sp ∂v k
∂sk ∂sm ∂sk ∂sm
= (ϕsk sm − ϕsp s Υkm
p
) i j
= ∇skm (ϕ) i ,
∂v ∂v ∂v ∂v j
i, j, k, l, m, p, t = 1, . . . , n ,
i.e. the quantities ∇skm (ϕ) form a covariant tensor of the second kind. This
tensor is called a tensor of mixed covariant derivatives of the invariant ϕ.
Analogous construction over an arbitrary covariant vector f = (fis ) de-
fines a covariant tensor of the second kind called a covariant derivative of
this vector. Its (ij)th component, designated by (∇f )sij in the coordinates
s1 , . . . , sn , is computed by the following formula
∂ s
(∇f )sij = f − fks Υijk , i, j, k = 1, . . . , n . (4.41)
∂sj i
Tensor relations (4.34), for n = 2, are verified for these components similarly
as in (4.40).
It is obvious, comparing (4.39) and (4.41), that
∇sij (ϕ) = (∇gradϕ)sij , i, j = 1, . . . , n .

In the case of a regular surface S rn ⊂ Rn+l represented by (4.1) another

example of a covariant vector is formed through an arbitrary fixed vector
P ∈ Rn+l by the following formula for its components in the coordinates
s1 , . . . , sn :
fis = P · rsi , i = 1, . . . , n , (4.42)
rn 1
where rsi is the ith basic tangent vector of S in the coordinates s , . . . , sn .
78 4 Multidimensional Geometry

If P ∈ Rn+l is a vector orthogonal to S rn at a point P then the quantities

bsij = rsi sj · P , i, j = 1, . . . , n , (4.43)

form a covariant tensor of the second order. Indeed

 ∂sk ∂sm ∂ 2 sk 
bvij = rvi vj · P = r k m + r k ·P
∂v i ∂v j s s ∂v i ∂v j s
∂sk ∂sm k
s ∂s ∂s
m
= rsk sm · P = bkm , i, j, k, m = 1, . . . , n .
∂v i ∂v j ∂v i ∂v j
Since (4.14) or (4.18), a typical example of a symmetric covariant tensor
of the second order also gives the metric tensor of any regular m-dimensional
surface S rn defined by (4.1) or a manifold M n .
There is an evident rule for forming a new covariant tensor from two
original ones. Namely, let f and v be two covariant tensors of the rank k and
l, respectively. Then the new tensor f ⊗ v is the covariant tensor of the rank
k + l whose components in the coordinates s1 , . . . , sn are computed as

(f ⊗ v)si1 ...ik+l = fis1 ...ik visk+1 ...ik+l , ij = 1, . . . , n . (4.44)

In particular, two smooth functions f and ϕ specified in the vicinity of a

point P ∈ S rn produce a covariant tensor of the second rank, through the
covariant vectors formulated by (4.37):

(f ⊗ ϕ)sij = fsi ϕsj , i, j = 1, . . . , n . (4.45)

Note, generally, this tensor is not symmetric.

Contravariant Tensors

Since (4.15) and (4.19), an example of the contravariant tensor of the sec-
ond rank is represented by the contravariant metric tensor of S rn and M n ,
respectively.
By virtue of (4.15) we can readily conclude that for a fixed vector P ∈
Rn+l a set of values defined in the coordinates s1 , . . . , sn as

P · ∇si , i = 1, . . . , n , (4.46)

where ∇si is the ith normal vector to the ith coordinate hypersurface in a
regular surface S rn , is a contravariant tensor of the first rank. Indeed, by
(4.10) and (4.15)

∂si ∂sj  ∂v l  ∂v i
P · ∇si = gsr
ij
(P · rsj ) = gvr
km
k m
P · rvl j = gvr
km
(P · rvm ) k
∂v ∂v ∂s ∂s
∂si
= P · ∇v k , i, j, k, l, m = 1, . . . , n .
∂v k
 4.4 Tensors 79

Similarly to the case of the covariant tensors considered above two con-
travariant tensors f and v of rank k and l, respectively, form a contravariant
tensor f ⊗v of the order k+l, whose components in the coordinates s1 , . . . , sn
are computed by
i ...ik+l i ...ik+l
(f ⊗ v)s1 = fsi1 ...ik vsk+1 , ij = 1, . . . , n . (4.47)

Mixed Tensors

Examples of the mixed tensors are readily constructed by the product of two
tensors one of which is covariant and the order is contravariant. For instance,
two tensors of the covariant and contravariant types formed by (4.42) and
(4.46), respectively, produce, through a vector P ∈ Rn+l , the following mixed
tensor
Pji (s) = Psi Pjs = (P · ∇si )(P · rsj ) = gsr
il
(P · rsl )(P · rsj )
(4.48)
i, j, l = 1, . . . , n .

There also is an operation of covariant differentiation of a contravariant

vector f = (fsi ) which results in a mixed tensor by the following formula for
its components designated by (∇f )ij (s) in the coordinates s1 , . . . , sn :

∂ i
(∇f )ij (s) = f + fsk Υjk
i
, i, j, k = 1, . . . , n . (4.49)
∂sj s
Using the relations (4.29), and (4.35) we easily find that the quantities
(∇f )ij (s) comprise a mixed tensor.

4.4.3 Tensor Operations

The operations over tensors defined at the same point of M n are addition,
multiplication and contraction.

Operation of Addition

The addition operation is carried out over tensors of the same order and
type merely by adding the values of their components. A particular case of
addition is the operation of subtraction.

Operation of Multiplication

The operation of multiplication is carried out over tensors of arbitrary or-

der and type by multiplying each component of one tensor, say f1 by every
component of another tensor, say f2 , in particular, as in (4.44) and (4.47).
As a result the order of the product equals the sum of the orders of the two
80 4 Multidimensional Geometry

original tensors f1 and f2 . The same rule of summation is valid for the type
of the product, namely, it is k1 + k2 times covariant and l1 + l2 times con-
travariant if the original tensor fi , i = 1, 2, is ki times covariant and li times
contravariant.
The operations of multiplication and addition allow one to formulate a
monitor manifold over a physical geometry S xn presented by a parametriza-
tion
x(s) : S n → Rn+k
with the use of covariant vectors B1 , . . . , Bl . The covariant metric elements
s
gij of the manifold are computed by the following formula
s xs
gij = (s)gij + Bim Bjm , i, j = 1, . . . , n, m = 1, . . . , l, (4.50)

where (s) > 0 is an arbitrary function. Analogously a set of contravariant

vectors D1 , . . . , Dl determines the contravariant metric elements gsij of a
corresponding monitor manifold over S xn :

gsij = (s)gsx
ij i
+ Dm j
Dm , i, j = 1, . . . , n, m = 1, . . . , l.

These very metrics prove to be crucial tools in controlling numerical grid

properties.

Operation of Contraction

The operation of contraction is carried out over mixed tensors only. Let us
take a mixed tensor, say one time covariant and two times contravariant
whose components in the coordinates s1 , . . . , sn are designated, correspon-
dently, as fkij (s). Assume the indices j and k the same then summation over
them gives quantities designated by fjij (s) which are dependent on one index
i only. It is readily shown that these quantities form a contravariant tensor
of order 1. Indeed
∂v k ∂sj ∂si ∂si
fjij (s) = fklm (v) j l m
= fllm (v) m , i, k, l, m = 1, . . . , n ,
∂s ∂v ∂v ∂v

i.e. the system bis = fjij (s) is a contravariant vector.

Analogously, the operation of contraction is defined for arbitrary mixed
tensors by identifying some upper and lower indices in the components and
producing summation over them. In particular, if f is a tensor k times co-
variant and k times contravariant then the operation of contraction over all
indices produces an invariant (tensor of order zero).
For example the operation of contraction over the tensor (4.48) produces
the following invariant

Pii (s) = gsr

ij
(P · rsi )(P · rsj ) , i, j = 1, . . . , n .
 4.5 Basic Invariants 81

With the operation of contraction one can define an invariant for arbi-
trary two tensors of the same order k provided one of them is covariant and
another contravariant. This invariant is obtained by the composition of two
operations: the first is multiplication of the tensors and the second is contrac-
tion of the obtained mixed tensor with respect to all indices. For example,
the covariant and contravariant metric tensors yield the invariant
s
gsij gij =n, i, j = 1, . . . , n ,
which equals n at all points of M n .

4.5 Basic Invariants

This section reviews the most important invariants indispensable in the anal-
ysis of grid properties. They are formed by the operation of contraction over
the contravariant metric tensor and some covariant tensors of order 2.

4.5.1 Beltrami’s Differential Parameters

Mixed Differential Parameters
The successive operations of multiplication and contraction over the covari-
ant tensor (4.45) and the contravariant metric tensor gsij of a manifold M n
produce the invariant
∇(f, ϕ) = fsi ϕsj gsij , i, j = 1, . . . , n , (4.51)
called Beltrami’s mixed differential parameter of f and ϕ. In accordance with
this designation formula (4.12) is also read as
ji
n = ϕsj gsr rsi = ∇(ϕ, r) , i, j = 1 . . . , n . (4.52)
By putting in (4.51) f equal to ϕ the following invariant is formulated:
∇(f ) = ∇(f, f ) = fsi fsj gsij , i, j = 1, . . . , n , (4.53)
which is referred to as Beltrami’s first differential parameter of f . Thus for-
mula (4.13) with this parameter is read as

n = ∇(ϕ) , (4.54)
i.e. the length of the normal n defined by (4.12) is an invariant.
Let ϕ1 = const and ϕ2 = const be two hypersurfaces in a regular surface
S rn . The angle θ between these hypersurfaces is defined as the angle between
the corresponding normals n1 and n2 to them. In accordance with (4.52) and
(4.54) the cosine of this angle is computed through the Beltrami’s differential
parameters:
∇(ϕ1 , r) · ∇(ϕ2 , r) ∇(ϕ1 , ϕ2 )
cos θ =   =  . (4.55)
∇(ϕ1 ) ∇(ϕ2 ) ∇(ϕ1 ) ∇(ϕ2 )
82 4 Multidimensional Geometry

Second Differential Parameter

Another important Beltrami’s differential parameter of a scalar ϕ is obtained

by contracting the mixed tensor formed through the multiplication operation
of the tensor of mixed derivatives (4.39) and the contravariant metric ten-
sor. This invariant, designated by ∆B [ϕ], is referred to as Beltrami’s second
differential parameter of ϕ. Namely

∆B [ϕ] = gsij ∇sij (ϕ) , i, j = 1, . . . , n . (4.56)

There exists one more important form of the invariant ∆B [ϕ] helpful for
its computing. To deduce it we note that, from (4.25) and (4.26),

1 ∂ √
√ s j ( g s gsij ) = gsij Υkj
k j
− gsim Υjm − gslj Υlji =
g ∂s (4.57)
= −gslj Υlji , i, j, k, l, m = 1, . . . , n ,

and consequently, taking advantage of these relations and (4.39) in (4.56),

we obtain the following expression for ∆B [ϕ]

∆B [ϕ] = gsij ϕsi sj − ϕsk gsij Υijk

ϕk ∂ √
= gsij ϕsi sj + √s s j ( g s gsjk ) (4.58)
g ∂s
1 ∂ √
= √ s j ( g s gsjk ϕsk ) , i, j, k = 1, . . . , n .
g ∂s

This very expression is used in Chap. 5 for formulating comprehensive grid

generation equations.

4.5.2 Measure of Relative Spacing

In grid technology there is often a need in estimating grid spacing near some
hypersurface in S rn . Typically the hypersurface is specified by the equation
ϕ(s) = 0. This equation describes one more hypersurface in the parametric
domain S n as well. The parametric mapping r(s) : S n → S rn transforms
a band of the thickness h around the hypersurface in S n to a band in S rn
whose thickness l is computed by the following formula
 ∂r 
l= · n2 h + O(h2 ) , (4.59)
∂n1
where n1 is a unit normal to the hypersurface in S n while n2 is a unit normal
to the hypersurface in S rn (Fig. 4.5 for n = 2). Since (4.52) and (4.54)
 4.5 Basic Invariants 83

Fig. 4.5. Illustration for the measure of relative spacing


n1 = (ϕs1 , . . . , ϕsn )/ ∇E (ϕ) ,
 (4.60)
ki
n2 = ϕsk gsr rsi / ∇(ϕ) , i, k = 1, . . . , n ,

where

∇E (ϕ) = ϕsi ϕsj δji = (ϕs1 )2 + · · · + (ϕsn )2 , i, j = 1, . . . , n ,

is Beltrami’s first differential parameter of ϕ(s) in the Euclidean metric δji of

S n . Substituting (4.60) and the relation
∂r 1
= ϕsj rsj , j = 1, . . . , n ,
∂n1 ∇E (ϕ)

in (4.59) yields

ϕsj rsj · (ϕsk gsr

ki
rsi )
l=   h + O(h2 )
∇E (ϕ) ∇(ϕ)

∇E (ϕ)
= h + O(h2 ) , i, j, k = 1, . . . , n .
∇(ϕ)

So the invariant
 
s(ϕ) = ∇E (ϕ)/ ∇(ϕ) (4.61)

defined through Beltrami’s first differential parameters of ϕ can be considered

as a measure of relative spacing produced by the parametric transformation
r(s) near the hypersurface ϕ(s) = 0 in S rn .
When a logical domain Ξ n introduced for generating grids is considered as
one more parametric domain then the quantity (4.61) with the identification
84 4 Multidimensional Geometry

Fig. 4.6. Illustration for the measure of relative clustering

S n = Ξ n (i.e. ∇E (ϕ) = (ϕξ1 )2 + · · · + (ϕξn )2 ) is referred to as a measure

of relative grid spacing near the hypersurface ϕ(s) = 0 in S rn . In particular,
let the hypersurface ϕ(s) = 0 be the grid hypersurface ξ i (s) − c = 0. As
∇E (ξ i ) = 1, ∇(ξ i ) = gξr
ii
for i fixed, hence (4.61) has the following form

s(ξ i ) = 1/ gξr
ii , i fixed . (4.62)

4.5.3 Measure of Relative Clustering

The rate of change of relative spacing in the direction n2 normal to the

hypersurface ϕ(s) = 0 in S rn is called a measure of relative clustering near
this hypersurface (see Fig. 4.6). Designating this measure by v we find, since
(4.60) and (4.61),

d 1 ji ∂
v(ϕ) = s(ϕ) =  ϕsj gsr s(ϕ)
dn2 ∇(ϕ) ∂si
(4.63)
1
=  ∇(ϕ, s(ϕ)) , i, j, k = 1, . . . , n .
∇(ϕ)

i.e. it is defined through the Beltrami’s first and mixed differential parameters.
In particular, if ϕ(s) ≡ ξ i (s) − c, where ξ i is the ith grid coordinate, using
(4.62) and (4.63) yields
1  1 
v(ξ i ) =  ∇ ξ i ,  , i fixed . (4.64)
ii
gξr ii
gξr

It will be shown in Chap. 6 that the measure (4.64) can also be expressed
through Beltrami’s second differential parameters and the so called mean
curvature of the grid hypersurface ξ i = const.
 4.6 Geometry of Hypersurfaces 85

4.5.4 Mean Curvature

One more invariant of a regular surface S rn , important in grid technology, is

obtained from the covariant tensor (4.43) and contravariant metric tensor of
S rn
ij
σ = gsr rsi sj · P , i, j = 1, . . . , n . (4.65)

When the surface S rn lies in a surface S r(n+1) and the vector P being orthog-
onal to S rn belongs also to the tangent plane to S r(n+1) then the invariant
σ from (4.65), scaled by the factor 1/(n P ) and designated as Km , i.e.
1
Km = g ij r i j · P , i, j = 1, . . . , n , (4.66)
n P sr s s

is called the mean curvature of S rn in S r(n+1) with the respect to the normal
P.
In particular for n = 1 (S r1 is a curve) the invariant Km is referred to as
the geodesic curvature of the curve S r1 in the surface S r2 .
In the following section a formula for this invariant will be established for
an arbitrary hypersurface lying in some regular surface. If this hypersurface
is found from the equation ϕ(s) = 0 then the mean curvature is defined by
Beltrami’s first and second differential parameters of ϕ.
An amusing role played by the mean curvature in grid technology is
demonstrated in Chap. 6.

4.6 Geometry of Hypersurfaces

We consider in this section the geometric characteristics which appear when
a regular n-dimensional hypersurface S xn represented by a set of local
parametrizations
x(s) : S n → Rn+l , S n ⊂ Rn ,
is a subset of an (n + 1)-dimensional surface S r(n+1) specified by local
parametrizations

r(s) : S n+1 → Rn+l , S n+1 ⊂ Rn+1 .

This situation occurs in the grid technology when a scalar-valued monitor

function is considered or grid hypersurfaces are analyzed.

4.6.1 Normal Vector to a Hypersurface

A unit vector n at a point x ∈ S xn which lies in the tangent plane to S r(n+1)

at this point and orthogonal to the tangent plane to S xn at the same point
is called the unit normal vector to the surface S xn in S rn .
86 4 Multidimensional Geometry

General Case

A normal vector is readily computed by the formula analogous to (2.25).

Namely, let P be a point of the surface S xn and let vectors x1 , . . . , xn and
r1 , . . . , rn+1 be the basic tangent vectors at this point of S xn and S rn re-
spectively. Now we form, analogously to (2.25), the following (n + 1) × (n + 1)
matrix
⎛ ⎞
r1 . . . rn+1
⎜ ⎟
⎜ x1 · r1 . . . x1 · rn+1 ⎟
⎜ ⎟
A=⎜ ⎟ , (4.67)
⎜ ... ... . . . ⎟
⎝ ⎠
xn · r1 . . . xn · rn+1
which derives the vector
b = det(A) . (4.68)
The rank of the n × (n + 1) matrix obtained from A by eliminating its top
row equals n hence b = 0. As
⎛ ⎞
xi · r1 . . . xi · rn+1
⎜ ⎟
⎜ x1 · r1 . . . x1 · rn+1 ⎟
⎜ ⎟
b · xi = ⎜ ⎟=0,
⎜ ... ... ... ⎟
⎝ ⎠
xn · r1 . . . xn · rn+1
The vector b is orthogonal to x1 , . . . , xn and consequently to the surface
S xn at the point of consideration. Thus for the unit normal vector n to the
surface S xn in S r(n+1) we find
n = b/|b| . (4.69)
In a one-dimensional case, i.e. when n = 1, we find from (4.67) and (4.68)
b = (x1 · r2 )r1 − (x1 · r1 )r2
(Fig. 4.7), while for n = 2 we obtain, similar to (2.26),
b = [(x1 · rk+1 )(x2 · rk+2 ) − (x1 · rk+2 )(x2 · rk+1 )]rk , k = 1, 2, 3 ,
where any index, say l, is identified with l ± 3. Thus the vector b obtained
by the formula (4.68) can be considered as the tensor product of the vectors
x1 , . . . , xn .
When the surface S xn lies in Rn+1 and is determined by the equation
ξ(x1 , . . . , xn+1 ) = c then there also is the well-known formula for the unit
normal n:

n = gradξ/|gradξ| = (ξx1 , . . . , ξxn+1 )/ (ξx1 )2 + . . . + (ξxn+1 )2 . (4.70)
 4.6 Geometry of Hypersurfaces 87

Fig. 4.7. Normal vector to the curve on the surface

Note if the surface S xn is a coordinate hypersurface of S r(n+1) in some

coordinates s1 , . . . , sn+1 , for example, it is defined by the equation s1 = s10
then S xn is represented in the coordinates s2 , . . . , sn+1 by

x(s2 , . . . , sn+1 ) = r(s10 , s2 , . . . , sn+1 ) .

Therefore
xi = rsi (s10 , s2 , . . . , sn+1 ) , i = 2, . . . , n + 1 ,
and consequently (4.67) and (4.68) result in
⎛ ⎞
rs1 . . . rsn+1
⎜ rs ⎟
⎜ g21 . . . g2n+1
rs
⎟
⎜ ⎟
A=⎜ ⎟ ,
⎜ ... ... ... ⎟
⎝ ⎠
rs rs
gn+11 . . . gn+1n+1

b1 = G1i rsi , i = 1, . . . , n + 1 ,
where G1i is the (1i)th cofactor of the matrix (gij
rs
), i, j = 1, . . . , n + 1. Since

G1i = g rs gsr
1i
, i = 1, . . . , n + 1 ,

we obtain that
b1 = g rs gsr
1i
rsi , i = 1, . . . , n + 1 ,
and comparing this expression with (4.10) gives, in this case,

b1 = g rs ∇s1 .

Analogously one readily shows that the vector bi expressed as follows:

88 4 Multidimensional Geometry

Fig. 4.8. Base vectors of the two-dimensional surface in R3

bi = g rs gsr
ij
rsj , i, j = 1, . . . , n + 1 , (4.71)

or using (4.10)

bi = g rs ∇si , i = 1, . . . , n + 1 , (4.72)

is orthogonal to the coordinate hypersurface si = si0 .

Note by the vectors x1 , . . . , xn and r1 , . . . , rn+1 in (4.67) for defining the
unit normal vector n to S xn in S r(n+1) by (4.69) one can use arbitrary sets
of vectors if only they are independent and lie in the corresponding tangent
planes.

Hypersurface in a Domain

Let S r(n+1) = Dn+1 ⊂ Rn+1 be an (n + 1)-dimensional domain containing

the n-dimensional surface S xn represented by

x(s) : S n → Rn+1 .

Since a unit normal vector n to S xn in Dn+1 is orthogonal to the tangential

vectors xsi , i = 1, . . . , n, the vectors

xs1 , . . . , xsn , n ,

constitute the basis of Rn+1 . These vectors are called the base vectors of S xn
in Rn+1 (Fig. 4.8 for n = 2). The vector xsm sp , m, p = 1, . . . , n, at a point
P ∈ S xn is expanded in these vectors as

xsm sp = P[xsm sp ] + (xsm sp · n)n , m, p = 1, . . . , n , (4.73)

Where P is the operator which projects vectors in Rn+1 on the tangent plane
to S xn . Taking advantage of (4.32) yields
 4.6 Geometry of Hypersurfaces 89
i
xsm sp = Υmp xsi + (xsm sp · n)n , i, m, p = 1, . . . , n , (4.74)
(see Fig 4.4 for n = 2 with the identification x = r, xsm sp = rmp , xsm sp · n =
bmp ). Similarly, using (4.33) in (4.73),
xsm sp = [mp, i]∇si + (xsm sp · n)n , i, m, p = 1, . . . , n , (4.75)
where
∇si = gsx
ij
xsj , i, j = 1, . . . , n .
For example, let S be a monitor surface over S n ⊂ Rn with a scalar-
xn

valued monitor function f (s), i.e. S xn is represented in the parametric coor-

dinates s1 , . . . , sn by
x(s) : S n → Rn+1 , x(s) = [s, f (s)] ,
It is readily verified that for the elements of the covariant and contravariant
metric tensors of S xn in the coordinates s1 , . . . , sn we have
xs
gij = xsi · xsj = δji + fsi fsj , i, j = 1, . . . , n ,
1 ∂f ∂f (4.76)
ij
gsx = δji − , i, j = 1, . . . , n ,
g xs ∂si ∂sj
where
g xs = det(gij
xs
) = 1 + (fs1 )2 + . . . + (fsn )2 = 1 + ∇E (f ) .
Since
xsi = (0, . . . , 0, 1, 0, . . . , 0, fsi ) , i = 1, . . . , n ,
     
i−1 n−i

it is obvious that one unit normal vector n to S xn in Dn+1 can be computed

as follows:
1
n = √ xs (fs1 , . . . , fsn , −1) . (4.77)
g
Therefore in this case the expansion (4.74) has the form
1
i
xsm sp = Υmp xsi − √ xs fsm sp n , i, m, p = 1, . . . , n . (4.78)
g
Since (4.76)
 1  1
il
gmp fsl = δli − xs fsi fsl fsl = xs fsi , i, l, m, p = 1, . . . , n ,
g g
therefore from (4.24) and (4.27) we find
i li li 1
Υmp = gsx [mp, l] = gsx fsm sp fsl = fsm sp fsi , i, l, m, p = 1, . . . , n .
g xs
So (4.78) results in
1 √
xsm sp = xs
fsm sp (fsi xsi − g xs n) , i, m, p = 1, . . . , n .
g
90 4 Multidimensional Geometry

4.6.2 Second Fundamental Form

Assuming in the coordinates s1 , . . . , sn , similarly to (4.43),

bij = xsi sj · n , i, j = 1, . . . , n , (4.79)

where n is the unit normal to the surface S xn in S r(n+1) (Figs. 4.7 and 4.8
for n = 2) we define the so called second fundamental form of the surface
S xn in S r(n+1) by
bij dsi dsj , i, j = 1, . . . , n .
The covariant tensor (bij ) reflects the local warping of the surface S xn in
S r(n+1) .

4.6.3 Surface Curvatures

Multidimensional Case
ij
The covariant tensor (bij ) and the contravariant tensor (gsx ) of the surface
S xn in S r(n+1) define the mixed tensor (Kji ), where

Kji = gsx
ik
bkj , i, k, j = 1, . . . , n . (4.80)

An nth part of the trace of (Kji ), namely, the quantity

1 1 ij
Km = tr(Kji ) = gsx bij , i, j = 1, . . . , n , (4.81)
n n
is called the mean curvature of the surface S xn in S r(n+1) . From (4.79 – 4.81)
we readily conclude that the quantity Km is invariant of parametrizations of
S xn and S r(n+1) .

Two-Dimensional Case

When n = 2 then the determinant of (Kji ), i.e.

KG = det(Kji ) , i, j = 1, 2 ,

is called the Gaussian curvature of S x2 in S r3 . It is obvious that

ij 1
KG = det(bij )det(gsx )= det(bij ) , (4.82)
g xs

where g xs = det(gij
xs
) = det(xsi · xsj ).
 4.6 Geometry of Hypersurfaces 91

4.6.4 Formulas of the Mean Curvature

Formula for the Mean Curvature of a Monitor Surface over a

Domain

Let S xn be the monitor surface over the domain S n parametrized by

x(s) : S n → Rn+1 , x(s) = [s, f (s)] .

Then, taking into account (4.77), we find

1
bij = xsi sj · n = − √ s fsi sj , i, j = 1, . . . , n .
g

where g s = det(xsi · xsj ). Using (4.76), (4.77), and (4.81) gives

1 1 
Km = √ s s fsi fsj fsi sj − ∇2 [f ] , i, j = 1, . . . , n . (4.83)
n g g

Formula for the Mean Curvature of the Hypersurface Specified by

the Equation ϕ(s) = 0

Multidimensional Case

Let the surface S r(n+1) be represented by a parametrization

r(s) : S n+1 → Rn+l , s = (s1 , . . . , sn+1 ) , l≥1,

while the n-dimensional surface S xn in S r(n+1) is the image through r(s)

of the n-dimensional surface in S n+1 defined from the equation ϕ(s) = 0,
i.e. the hypersurface S xn consists of the points r(s), s ∈ S n+1 , such that
ϕ(s) = 0. In this paragraph we will find a formula for the mean curvature of
this hypersurface S xn in S r(n+1) .
Let the equation ϕ(s) = 0 be locally resolved with respect to sn+1 , i.e.
there is a function sn+1 (s1 , . . . , sn ) such that

ϕ[s1 , . . . , sn , sn+1 (s1 , . . . , sn )] ≡ 0 , (s1 , . . . , sn ) ∈ S n ,

for some n-dimensional domain S n . This is possible if ϕsn+1 = 0 at some

point s ∈ S n+1 . To make things definite we assume that ϕsn+1 < 0. Then the
hypersurface S xn is represented locally by the coordinates s1 , . . . , sn from
the domain S n as

x(s1 , . . . , sn ) : S n → Rn+l , (4.84)

where
x(s1 , . . . , sn ) = r(s1 , . . . , sn , sn+1 (s1 , . . . , sn )) .
92 4 Multidimensional Geometry

Since
∂sn+1 ϕi
i
=− s , i = 1, . . . , n ,
∂s ϕsn+1
the basic tangent vectors to S xn in the coordinates s1 , . . . , sn are subject to
the equations

∂sn+1 1
xsi = rsi + rsn+1 = (ϕ n+1 rsi − ϕsi rsn+1 ) ,
∂si ϕsn+1 s (4.85)
i = 1, . . . , n ,
xs
therefore the values for the elements gij , i, j = 1, . . . , n, of the covariant
1
metric tensor of S in the coordinates s , . . . , sn can be computed by the
xn

following formula
xs
gij = xsi · xsj
1
= [(ϕ n+1 )2 gij
rs
− ϕsn+1 (ϕsj gn+1i
rs rs
+ ϕsi gn+1j ) (4.86)
(ϕsn+1 )2 s
rs
+ϕsi ϕsj gn+1n+1 ], i, j = 1, . . . , n ,
rs
where the quantities gkl , k, l = 1, . . . , n + 1, are the elements of the covariant
metric tensor of S r(n+1)
in the coordinates s1 , . . . , sn+1 computed by the
formula
rs
gkl = rsk · rsl , k, l = 1, . . . , n + 1 .
For the unit normal vector n to S xn in S r(n+1) we have from (4.12)

n = b/|b| , (4.87)

where
lk
b = ϕsl gsr rsk , l, k = 1, . . . , n + 1 ,
  (4.88)
lj
|b| = ϕsl ϕsj gsr = ∇(ϕ) , j, l = 1, . . . , n + 1 .

Since our assumption that ϕsn+1 < 0, equations (4.87) and (4.88) result in
1 1
n · rsn+1 = √ lk
ϕsl gsr rsk · rsn+1 = √ ϕsn+1 < 0 ,
∇ϕ ∇ϕ
i.e. the vectors n and rsn+1 have opposite directions with respect to the
tangent plane to S xn in S r(n+1) .
ij
Now we compute the elements gsx of the contravariant metric tensor of
the hypersurface S in the coordinates s1 , . . . , sn .
xn

First note that the following n + 1 vectors

x1 , x2 , . . . , xn , n ,
 4.6 Geometry of Hypersurfaces 93

where n is defined by (4.87), comprise a basis of Rn+1 . As n is orthogonal

to xi , i = 1, . . . , n, so, in accordance with (2.6), any vector v ∈ Rn+1 is
represented through the vectors of this basis by
ij
v = gsx (v · xj )xi + (v · n)n , i, j = 1, . . . , n . (4.89)

Note every basic normal vector ∇si , i = 1, . . . , n, to the coordinate sur-

face si = c in S r(n+1) defined, in accordance with (4.10), by

∇si = gsr
il
rsl , l = 1, . . . , n + 1 , (4.90)

is orthogonal to xk , k = 1, . . . , n, and k = i. Indeed, from (4.85) and (4.89),

1
xk · ∇si = (ϕ n+1 rsk − ϕsk rsn+1 ) · gsil rsl
ϕsn+1 s
ϕk (4.91)
= δik − s δin+1 = 0 ,
ϕsn+1
l = 1, . . . , n + 1 , i, k = 1, . . . , n , i = k .

Therefore, from (4.87 – 4.91), we obtain

∇si = gsx
ij
(∇si · xsi )xsj + (∇si · n)n
1
ij
= gsx xsj +  il
ϕsl gsr n, (4.92)
∇(ϕ)
l = 1, . . . , n + 1 , i, j = 1, . . . , n , i fixed .

Thus by expanding the basic tangent vectors rsl , l = 1, . . . , n, through

x1 , . . . , xn , and n and availing us of this expansion in (4.90) we can com-
ij
pute the contravariant elements gsx , i, j = 1, . . . , n, of S xn in the coordinates
1 n
s ,... ,s .
In order to find this expansion we first note that, in accordance with
(4.85),
1
rsk = (ϕsn+1 xsk + ϕsk rsn+1 ) , k = 1, . . . , n . (4.93)
ϕsn+1
Multiplying this equation by

lk
ϕsl gsr / ∇(ϕ) , l = 1, . . . , n + 1 , k = 1, . . . , n ,

and using (4.87) and (4.88) we find

ln+1
ϕsl gsr ϕ l g lk ∇(ϕ) − ϕsl ϕsn+1 gsr
ln+1
n−  rsn+1 = s sr xsk +  rsn+1 ,
∇(ϕ) ∇(ϕ) ∇(ϕ)ϕsn+1
l = 1, . . . , n + 1 , k = 1, . . . , n .
94 4 Multidimensional Geometry

Therefore
ϕ n+1  ϕ l g lk 
rsn+1 =  s n − s sr xsk ,
∇(ϕ) ∇(ϕ) (4.94)
l = 1, . . . , n + 1 , k = 1, . . . , n ,

and using this equation in (4.93) gives

1  lk
ϕsj ϕsl gsr 
rsj = xsj +  ϕsj n −  xsk ,
∇(ϕ) ∇(ϕ) (4.95)
l = 1, . . . , n + 1 , j, k = 1, . . . , n .

Substituting (4.94) and (4.95) in (4.90) we obtain

1
∇si = gsr
ij in+1
rsj + gsr rsn+1 = gsrij
xsj +  ji
ϕsj gsr n
∇(ϕ)
1 1
− ϕsj ϕsl gsr gsr xsk + 
ij lk in+1
ϕsn+1 gsr n
∇(ϕ) ∇(ϕ)
1
− ϕ n+1 ϕsl gsr lk in+1
gsr xsk
∇(ϕ) s
 1  1
= gsr ij
− ϕsp ϕsl gsr gsr xsj + 
ip lj il
ϕsl gsr n,
∇(ϕ) ∇(ϕ)
i, j, k = 1, . . . , n , l, p = 1, . . . , n + 1 .

Comparing this expansion with (4.92) we find

1
ij
gsx ij
= gsr − ip jl
ϕsp ϕsl gsr gsr ,
∇(ϕ) (4.96)
i, j = 1, . . . , n , l, p = 1, . . . , n + 1 .

This very matrix is the contravariant metric tensor of the subsurface S xn

in the coordinates s1 , . . . , sn . In order to check this statement we note first
that for arbitrary symmetric matrices (akl ) and (bkl ), k, l = 1, . . . , n + 1, the
following relations are held:

bij ajk = bil alk − bin+1 an+1k ,

bij ϕsj akn+1 = bil ϕsl akn+1 − bin+1 ϕsn+1 akn+1 ,
(4.97)
bij ϕsk an+1j = bil ϕsk an+1l − bin+1 ϕsk an+1k+1 ,
i, j, k = 1, . . . , n , l = 1, . . . , n + 1 .

Availing us of these relations we obtain

 4.6 Geometry of Hypersurfaces 95
 1 
ij xs
gsx gjk ij
= gsr − ip jl
ϕsp ϕsl gsr gsr
∇(ϕ)
 1 ϕ j ϕ k rs 
× gjk rs
− rs
(ϕsj gn+1k rs
+ ϕsk gn+1j ) + s s gn+1n+1
ϕsn+1 ϕsn+1
1
= gsr glk − gsr
il rs
gn+1k −
in+1 rs
g rs (g il ϕ l − gsrin+1
ϕsn+1 )
ϕsn+1 n+1k sr s
ϕk
− s (gsr gln+1 − gsr
il rs in+1 rs
gn+1n+1 )
ϕsn+1
ϕsk
+ g rs (g il ϕ l − gsrin+1
ϕsn+1 )
(ϕsn+1 )2 n+1n+1 sr s
ϕsp ϕsm ip  ml rs
− g g g − gsr mn+1 rs
gn+1k
∇(ϕ) sr sr lk
1
− g rs (g ml ϕ l − gsr mn+1
ϕsn+1 )
ϕsn+1 kn+1 sr s
ϕk
− s (gsr gln+1 − gsr
ml rs mn+1 rs
gn+1n+1 )
ϕsn+1
ϕsk 
− g rs
)(g ml
ϕ sl − gsr
mn+1
ϕ sn+1 )
(ϕsn+1 )2 n+1n+1 sr

ϕsl il rs ϕ k ϕ l il rs
= δki − gsr gkn+1 + s s 2 gsr gn+1n+1
ϕsn+1 (ϕsn+1 )
ϕsp ip rs ϕsp ϕsk ip rs
+ g g − g g = δki ,
ϕsn+1 sr kn+1 (ϕsn+1 )2 sr n+1n+1
i, j, k = 1, . . . , n , m, l, p = 1, . . . , n + 1 ,
ij
i.e. the matrix (gsx ), i, j = 1, . . . , n, whose elements are defined by (4.96) is
xs
the inverse to the covariant metric tensor (gij ), i, j = 1, . . . , n, of S xn , ex-
ij
pressed by (4.86). Consequently the elements gsx specified by (4.96) comprise
the contravariant metric tensor of S xn in the coordinates s1 , . . . , sn .
Now we proceed to the computation of the quantity xsi sj · n. From (4.85)
we find
1 1
xsi sj = Lij [r] − Lij [ϕ]rsn+1 , i, j = 1, . . . , n , (4.98)
(ϕsn+1 )2 (ϕsn+1 )3

where Lij is the operator defined at a function v(s1 , . . . , sn+1 ) as

Lij [v] = (ϕsn+1 )2 vsi sj − ϕsn+1 (ϕsj vsi sn+1 + ϕsi vsj sn+1 )
(4.99)
+ϕsi ϕsj vsn+1 sn+1 , i, j = 1, . . . , n .

Therefore, using (4.87) and (4.88),

96 4 Multidimensional Geometry

1
Lij [r] · n =  [(ϕsn+1 )2 ϕsl gsr
lk
rsi sj · rsk
∇(ϕ)
−ϕsn+1 ϕsl gsrlk
(ϕsj rsi sn+1 · rsk + ϕsi rsj sn+1 · rsk )
lr n+1 n+1
+ϕsi ϕsj ϕsl gsr rs s · rsk ] (4.100)
ϕl
=  s [(ϕsn+1 )2 Υijl − ϕsn+1 (ϕsj Υin+1
l l
+ ϕsi Υjn+1 )
∇(ϕ)
l
+ϕsi ϕsj Υn+1n+1 ], i, j = 1, . . . , n , l, k = 1, . . . , n + 1 .

where Υijl , k, l, p = 1, . . . , n + 1, are the Christoffel symbols of the second

kind of S r(n+1) in the coordinates s1 , . . . , sn+1 . Analogously
ϕ n+1
Lij [ϕ]rsn+1 · n =  s Lij [ϕ] , i, j = 1, . . . , n . (4.101)
∇(ϕ)

Thus for the mean curvature of the hypersurface S xn in S r(n+1) we have,

from (4.81) and (4.96),
1 ij
Km = g x i j ·n
n sx s s
1  1 
=  ij
gsr − ip jl
ϕsp ϕsl gsr gsr
2(ϕsn+1 )2 ∇(ϕ) ∇(ϕ)
×{(ϕsn+1 )2 (ϕsl Υijl − ϕsi sj ) − ϕsn+1 ϕsj (ϕsl Υin+1
l
− ϕsi sn+1 )

−ϕsn+1 ϕsi (ϕsl Υjn+1

l
− ϕsj sn+1 ) (4.102)
l
+ϕsi ϕsj (ϕsl Υn+1n+1 − ϕsn+1 sn+1 )}
1  1 
=−  ij
gsr − ip jl
ϕsp ϕsl gsr gsr dij ,
n(ϕsn+1 )2 ∇(ϕ) ∇(ϕ)
i, j = 1, . . . , n , l, p = 1, . . . , n + 1 ,

where
dkl = (ϕsn+1 )2 ∇kl (ϕ) − ϕsn+1 ϕsk ∇n+1l (ϕ) − ϕsn+1 ϕsl ∇n+1k (ϕ)

+ϕsk ϕsl ∇n+1n+1 (ϕ) , k, l, = 1, n + 1 ,

while
p
∇kl (ϕ) = ϕsk sl − ϕsp Υkl , k, l, = 1, n + 1 ,
is the mixed covariant derivative of ϕ with respect to sk and sl in the metric
of S r(n+1) .
In order to compute (4.102) we use an analog of the formulas (4.97) which
states that the following combinations of the same matrices (akl ) and (bkl ),
as in (4.97), are subject to the relations
 4.6 Geometry of Hypersurfaces 97

bij aij = bkl akl − 2bkn+1 akn+1 + bn+1n+1 an+1n+1 ,

bij ϕsi ajn+1 = bij ϕsj ain+1 = aln+1 (bkl ϕsk − bn+1l ϕsn+1 )
(4.103)
−an+1n+1 (bln+1 ϕsl − bn+1n+1 ϕsn+1 ) ,

i, j = 1, . . . , n , k, l = 1, . . . , n + 1 .
Therefore
1 ϕ iϕ j
bij (aij − (ϕ i an+1j + ϕsj an+1i ) − s s2 an+1n+1 )
ϕsn+1 s (ϕsn )
= bkl akl − 2bkn+1 akn+1 + bn+1n+1 an+1n+1

2
− [aln+1 (bkl ϕsk − bn+1l ϕsn+1 )
ϕsn+1
−an+1n+1 (bln+1 ϕsl − bn+1n+1 ϕsn+1 )] (4.104)

an+1n+1 kl
+ (b ϕsk ϕsl − 2bkn+1 ϕsk ϕsn+1 + bn+1n+1 ϕsn+1 ϕsn+1 )
(ϕsn+1 )2
2 an+1n+1 kl
= bkl akl − bkl ϕsk aln+1 + b ϕsk ϕsl ,
ϕsn+1 (ϕsn+1 )2
i, j = 1, . . . , n , k, l = 1, . . . , n + 1 .
Assuming now
1
bkl = gsr
kl
− pk tl
ϕsp ϕst gsr g ,
∇(ϕ)
akl = ∇kl (ϕ) , k, l, p, t = 1, . . . , n + 1 ,
we obtain from (4.102) and (4.104)
1 ij
Km = g x i j ·n
n sx s s
1 1
=−  kl
[(gsr − pk tl
ϕsp ϕst gsr g )∇kl (ϕ)
2 ∇(ϕ) ∇(ϕ)
2  kl 1 
+ gsr ϕsk − pk
ϕsp ϕsk gsr ϕst g lt ∇ln+1 (ϕ)
ϕsn+1 ∇(ϕ) (4.105)
1  1 
− g kl
ϕs k ϕsl − ϕ spϕ kg
s
pk
ϕ sl ϕst g
lt
∇n+1n+1 (ϕ)
(ϕsn+1 )2 sr
∇(ϕ) sr

1
= [ϕsp ϕst gsr gsr − ∇(ϕ)gsr
pk tl kl
]∇kl (ϕ) ,
n(∇(ϕ))3/2
i, j = 1, . . . , n , k, l, p, t = 1, . . . , n + 1 .
It is obvious that the same formula for Km is obtained if ϕsi (s) = 0 for some
i, 1 ≤ i ≤ n, and s ∈ S n+1 .

98 4 Multidimensional Geometry

Hypersurface in a Domain

In the case S r(n+1) is an (n + 1)-dimensional domain with the Euclidean

rs
metric gkl = δlk the equation (4.105) results in

1
Km = [ϕ k ϕ l ϕ k l − |grad ϕ|2 ϕsp sp ] ,
n|grad ϕ|3 s s s s (4.106)
k, l, p = 1, . . . , n + 1 .

In particular, let S xn be an n-dimensional sphere of a radius ρ, i.e.

n+1
ϕ(s) ≡ ρ2 − (si )2 .
i=1

Then, at the points of the sphere,

|gradϕ|2 = 4ρ2 ,
ϕsk sk = −2(n + 1) , k = 1, . . . , n + 1 ,
2
ϕsk ϕsl ϕsk sl = −8ρ , k, l = 1, . . . , n + 1 .

Thus, from (4.106),

1 1
Km = 3
[−8ρ2 + 8(n + 1)ρ2 ] = .
8nρ ρ
If S xn is a monitor surface over a domain S n defined by the values of a
scalar-valued function f (s), then this surface in Rn+1 is also specified by the
equation
ϕ(s1 , . . . , sn+1 ) ≡ f (s1 , . . . , sn ) − sn+1 = 0 .
Consequently
√ √
|gradϕ| = g s = 1 + fsi fsi , i = 1, . . . , n ,

0, k = n + 1 or l = n + 1 ,
ϕsk sl =
fsk sl , k = n + 1 and l = n + 1 ,

therefore, from (4.106), we have the following formula for Km

1
Km = (fsi fsj fsi sj − g s fsi si ) , i = 1, . . . , n ,
n(g s )3/2

which coincides with (4.83).

 4.6 Geometry of Hypersurfaces 99

Expression Through Beltrami‘s Differential Parameters

Note that similarly to the formula (4.25)

∂ ij im j
g = −gsr Υmk − gsr
lj i
Υlk , i, j, l, m = 1, . . . , n + 1 .
∂sk sr
Taking advantage of these relations yields
∂ ∂
∇(ϕ) = i (ϕsk ϕsl gsr
kl kl
) = 2gsr ϕsl (ϕsk si − ϕsm Υki
m
)
∂si ∂s (4.107)
kl
= 2gsr ϕsl ∇ki (ϕ) , i, k, l, m = 1, . . . , n + 1 ,

and consequently
1 pk ∂
gsr ∇kl (ϕ) =
pk tl
ϕsp ϕst gsr ϕsp gsr ∇(ϕ)
2 ∂sk
(4.108)
1
= ∇(ϕ, ∇(ϕ)) , k, l, p, t = 1, . . . , n + 1 .
2
Further, in accordance with (4.56),
kl
gsr ∇kl (ϕ) = ∆B [ϕ] , k, l = 1, . . . , n + 1 . (4.109)

Therefore (4.105) is transformed, with the help of (4.108) and (4.109), to

1  ∆B [ϕ]  1 
Km = −  + ∇ ϕ,  . (4.110)
n ∇(ϕ) ∇(ϕ)

This formula does not require the knowledge of a normal to the hypersurface
ϕ(s) = 0 in S rn so it is used for determining the mean curvature of such a
hypersurface in an arbitrary Riemannian manifold.

Another Form

One more formula for the mean curvature is found from the following relation
  1  1 ∂ √
∆B [ϕ] + ∇(ϕ)∇ ϕ,  = √ rs j ( g rs gsr ij
ϕsi )
∇(ϕ) g ∂s
  rs
 ∂ 1 ∇(ϕ) ∂  g 
+ ∇(ϕ)gsr ϕsi j 
ij
= g ij
ϕ i ,
∂s ∇(ϕ) g rs ∂sj ∇(ϕ) sr s
i, j = 1, . . . , n + 1 .

Hence equation (4.110) also becomes

 rs
1 ∂  g 
Km = − √ rs j ij
gsr ϕsi , i, j = 1, . . . , n + 1 . (4.111)
n g ∂s ∇(ϕ)
100 4 Multidimensional Geometry

In particular, for the n-dimensional coordinate hypersurface ϕ ≡ si − c0 = 0,

we have
∇(ϕ) = gsr
ii
, i fixed ,
therefore, in accordance with (4.111), the mean curvature of the coordinate
hypersurface si = c0 is expressed as follows:

1 ∂  g rs ij 
Km = − √ rs j g
ii sr
, i, j = 1, . . . , n + 1 , i fixed . (4.112)
n g ∂s gsr

One-Dimensional Case

When n = 1, i.e. S xn is a curve, while S r(n+1) is a two-dimensional surface

rs
then (gij ) is a 2 × 2 matrix and its elements satisfy the relations
rs 3−i3−j
gij = (−1)i+j g rs gsr , i, j = 1, 2 ,

where g rs = det(gij
rs
). Substituting these relations in (4.86) for n = 1 we find
xs
the following expression for the metric element g11 of the curve S x1 in S r2
represented by the equation ϕ(s) = 0, s = (s1 , s2 ),
1 1
xs
g11 = g rs ϕsl ϕsk gsr
lk
= g rs ∇(ϕ) , k, l = 1, 2 .
(ϕs2 )2 (ϕs2 )2

Therefore the contravariant metric element of S x1 is expressed as

11
gsx = (ϕs2 )2 /[g rs ∇(ϕ)] .

Further note that the formula (4.99) with n = 1 has the following form

L11 [v] = |gradϕ|2 vsl sl − ϕsk ϕsl vsk sl , k, l = 1, 2 ,

hence equation (4.105) for n = 1 is transformed to

1
σ= [|gradϕ|2 (ϕsl Υkk
l
− ϕsk sk )
g rs (∇(ϕ))3/2
−ϕsk ϕsp (ϕsl Υkp
l
− ϕsk sp )]
1 (4.113)
= [ϕ k ϕ l ∇kl (ϕ) − |gradϕ|2 ∇kk (ϕ)]
g (∇(ϕ))3/2 s s
rs

1
−(−1)k+l rs ϕ 3−k ϕs3−l ∇kl (ϕ) , k, l, p = 1, 2 ,
g (∇(ϕ))3/2 s

where σ = Km is the geodesic curvature of S x1 in S r2 .

The geodesic curvature of the coordinate line si = c0 is also computed
from (4.112) by the formula
 4.6 Geometry of Hypersurfaces 101

1 ∂  1 
σi = −(−1)i+j √ rs j  rs rs
g3−i3−j ,
g ∂s g3−i3−i (4.114)
i, j = 1, 2 , i fixed ,
i.e.
1  ∂  1  ∂  1 
σ1 = − √ rs 1
 rs g22
rs
− 2
 rs g12
rs
,
g ∂s g22 ∂s g22
1  ∂  1  ∂  1 
σ2 = − √ rs 2
 rs g11
rs
− 1
 rs g12
rs
,
g ∂s g11 ∂s g11
where σi , i = 1, 2, is the geodesic curvature of the coordinate curve si = c0 .

Computation of the Mean Curvature in the Case of a Parametric

Representation

Multidimensional Case

Let the coordinates s1 , . . . , sn+1 of the points of the n-dimensional surface

S xn in S r(n+1) be represented locally by the following transformation

s(t) : T n → S n+1 , t = (t1 , . . . , tn ) , (4.115)

where T n ⊂ Rn is some n-dimensional domain of Rn . Then S xn is parametrized

in the coordinates t1 , . . . , tn by

x(t) = r[s(t)] : T n → Rn+l , (4.116)

where
r(s) : S n+1 → Rn+l , l≥1,
r(n+1)
is the parametrization of S . So the basic tangent vectors xi , i =
1, . . . , n, to the hypersurface S xn in S r(n+1) are expressed as follows:
dsk
xi = rsk , i = 1, . . . , n , k = 1, . . . , n + 1 . (4.117)
dti
For finding a normal to S xn in S r(n+1) we use formulas (4.68) and (4.69)
which require the computation of the determinant of the matrix (4.67) which
has, in accordance with (4.117), the following form
⎛ ⎞
rs1 · · · rsn+1
⎜ ⎟
⎜ ⎟
⎜ rs dsl dsl ⎟
⎜ g1l · · · g rs ⎟
⎜ dt1 n+1l
dt1 ⎟
A=⎜ ⎜ ⎟ , l = 1, . . . , n + 1 . (4.118)
⎟
⎜ · ··· · ⎟
⎜ ⎟
⎜ ⎟
⎝ l l ⎠
ds ds
rs
g1l · · · gn+1l
rs
dtn dtn
102 4 Multidimensional Geometry

We readily see that the matrix A is composed as the product of the covariant
metric tensor and the following matrix B
⎛ j1 ⎞
gsr rsj · · · gsr
jn+1
rsj
⎜ ⎟
⎜ ⎟
⎜ ds1 dsn+1 ⎟
⎜ · · · ⎟
⎜ dt1 dt1 ⎟
B=⎜ ⎜
⎟ , j = 1, . . . , n + 1 ,
⎟ (4.119)
⎜ · ··· · ⎟
⎜ ⎟
⎜ ⎟
⎝ 1 n+1 ⎠
ds ds
···
dtn dtn
namely,
rs
A = B(gij ).
Therefore, analogously to formula (4.68), a normal vector b to S xn in S r(n+1)
is also computed by
b = det B .
In accordance with the rule of the computation of the determinant of a matrix
we find from (4.119)

b = −(−1)i gsr
ji
(det Di )rsj , i, j = 1, . . . , n + 1 , (4.120)

where Di is the n × n matrix obtained by deleting the ith column of the

n × (n + 1) matrix (dsi /dtj ), i = 1, . . . , n + 1, j = 1, . . . , n, i.e.
⎛ ⎞
ds1 dsi−1 dsi+1 dsn+1
⎜ dt1 · · · · · · ⎟
⎜ dt1 dt1 dt1 ⎟
⎜ ⎟
Di = ⎜
⎜
· ··· · · ··· · ⎟ .
⎟
⎜ ⎟
⎝ 1 ⎠
ds dsi−1 dsi+1 dsn+1
· · · · · ·
dtn dtn dtn dtn
Hence

|b|2 = b · b = (−1)i+k gsr

ki
det Di det Dk , i, k = 1, . . . , n + 1 . (4.121)

As for
∂ 2 x(t)
xti tj = , i, j = 1, . . . , n ,
∂ti ∂tj
we have from (4.116)

dsm dsl d2 sm
xti tj = rsm sl i j
+ rsm i j , i, j = 1, . . . , n , l, m = 1, . . . , n + 1 ,
dt dt dt dt
therefore
 4.6 Geometry of Hypersurfaces 103

b
bij = xti tj · n = xti tj ·
|b|
1 pk  dsm dsl d2 sm 
= −(−1)p gsr det Dp rsm sl · rsk + i j rsm · rsk
|b| i
dt dtj dt dt
1  m l 2 p 
p ds ds d s
= −(−1)p det Dp Υml + i j ,
|b| i
dt dtj dt dt
i, j = 1, . . . , n , k, l, m, p = 1, . . . , n + 1 ,
and consequently
1 ij 1
Km = g bij = det C , i, j = 1, . . . , n , (4.122)
2 tx 2|b|
ij
where gtx is the (ij)th element of the contravariant metric tensor of S xn in
the coordinates t1 , . . . , tn , while
⎛  ⎞
ij
m
1 ∂s ∂s
l
∂ 2 s1  ij
 m
n+1 ∂s ∂s
l
∂ 2 sn+1 
g Υ
⎜ tx ml ∂ti ∂tj + . . . g Υ +
⎜ ∂ti ∂tj tx ml
∂ti ∂tj ∂ti ∂tj ⎟ ⎟
⎜ ⎟
⎜ ∂s1 ∂sn+1 ⎟
⎜ ... ⎟
⎜ ∂t 1 ∂t 1 ⎟
C=⎜ ⎟ .
⎜ ⎟
⎜ · ... · ⎟
⎜ ⎟
⎜ ⎟
⎝ 1 n+1 ⎠
∂s ∂s
. . .
∂tn ∂tn
Now we establish a relation between the elements of the contravariant
metric tensor (gtx ij
), i, j = 1, . . . , n, of S xn in the coordinates t1 , . . . , tn and
the metric elements of S r(n+1) in the coordinates s1 , . . . , sn+1 . For this pur-
pose we notice that the vectors
x1 , . . . , xn , n ,
where n = b/|b|, constitute a basis for the tangent plane to S r(n+1) . The
vectors rsk , k = 1, . . . , n + 1, are expanded in these basis by the formula
(2.6), namely,
ij
rsk = gtx (rsk xi ) · xj + (rsk · n)n , i, j = 1, . . . , n , k = 1, . . . , n + 1 ,
therefore, using (4.117) and (4.120),
ij kp
rs
gkl = rsk · rsl = [gtx (rsk · xi )xj + (rsk · n)n] · [gtx (rsl · xp )xk + (rsl · n)n]
ij kp xt
= gtx gtx gkj (rsk · xi )(rsl · xp ) + (rsk · n)(rsl · n)

ij 1
= gtx (rsk · xi )(rsl · xj ) + (r k · b)(rsl · b)
|b|2 s
m p rs 2
ij rs ∂s rs ∂s k+l (g )
= gtx gkm i glp + (−1) det Dk det Dl ,
∂t ∂tj |b|2
i, j = 1, . . . , n , k, l = 1, . . . , n + 1 , k, l fixed .
104 4 Multidimensional Geometry

Fig. 4.9. Illustration for a curve on the surface

lh
Multiplying these relations by gsr yields

ij rs ∂sm ∂sh (g rs )2
δhk = gtx gkm + (−1)k+l lh
det Dk det Dl gsr ,
i
∂t ∂t j |b|2 (4.123)
i, j = 1, . . . , n , k, l, h = 1, . . . , n + 1 , k, fixed .
kp
Further multiplication of (4.123) by gsr gives

hp ij ∂sp ∂sh (g rs )2 hl mp
gsr = gtx + (−1)m+l g g det Dm det Dl ,
i
∂t ∂t j |b|2 sr sr (4.124)
i, j = 1, . . . , n , h, l, m, p = 1, . . . , n + 1 .
Therefore
ij p ∂sm ∂sl ml p p
gtx Υml = gsr Υml − dm dl Υml , i, j = 1, . . . , n , l, m, p = 1, . . . , n + 1 ,
∂ti ∂tj
where
g rs mk
dk = (−1)m g det Dm , k, m = 1, . . . , n + 1 .
|b| sr
One-Dimensional Subsurface

In the case n = 1 (Fig. 4.8), when the surface S x1 is a curvilinear line in S r2 ,

the twice mean curvature is called the geodesic curvature of the curve S x1 in
S r2 . Typically the geodesic curvature is designated by σ. It is obvious that
in this case KG = σ. Besides this it is evident that the geodesic curvature of
S x1 in S r2 is its curvature determined by (3.8) if S r2 is a two-dimensional
domain.
The geodesic curvature of S x1 in S r2 can be computed through the el-
ements of the first groundforms of S x1 and S r2 . In order to prove this we
assume that the coordinates s1 and s2 of S x1 are specified by the function
 4.6 Geometry of Hypersurfaces 105

s(t) : [a, b] → S 2 ,

i.e. S x1 is parametrized as

r[s(t)] : [a, b] → Rl+2 ,

where
r(s) : S 2 → R2+l , l≥0
is the parametrization of S r2 . The geodesic curvature of S x1 in S r2 is defined
through the scalar product of

d2 r[s(t)] dsi dsj d2 si

rtt = = rsi sj + rsi , i, j = 1, 2 ,
dt2 dt dt dt2
and a normal to this curve in S r2 . Namely,
1
σ= rtt · n ,
g rt
where
i j
dr[s(t)] dr[s(t)] rs ds ds
g rt = · = gij , i, j = 1, 2 ,
dt dt dt dt
while the normal n may be computed by (4.69). This formula yields

n = b/|b|

where
 dsi   dsi 
b= rsi · rs2 rs1 − rsi · rs1 rs2
dt dt
3−i
dsi ik ds
rs
= (g2i rs1 − g1i
rs
rs2 ) = −(−1)i g rs gsr rk, i, k = 1, 2 .
dt dt s
From this equation we readily find

ds3−i ds3−j
|b|2 = b · b = (−1)i+j (g rs )2 gsr
ik rs jl
gkl gsr
dt dt
ds3−i ds3−j
= (−1)i+j (g rs )2 gsr
ij
= g rs g rt , i, j, k, l = 1, 2 ,
dt dt
and consequently

g rs ik ds3−i
n = −(−1) i
g rk, i, k = 1, 2 .
g rt sr dt s

Thus using the above formulas gives the following expression for the geodesic
curvature of S x1 in S r2
106 4 Multidimensional Geometry
√ rs 
g dsi dsj d2 si   lk ds3−k 
σ = −(−1)k rsi sj + rsi 2 · gsr rl
rt
(g ) 3/2 dt dt dt dt s
√ rs
g ds3−k  d2 sk i
k ds ds
j
= −(−1)k
+ Υ , i, j, k, l = 1, 2 ,
(g rt )3/2 dt dt2 ij
dt dt

or using the matrix notation

√ rs
g
σ = rt 3/2 det(A) (4.125)
(g )

where ⎛ j ⎞
d2 s1 i j 2 2
1 ds ds d s
i
2 ds ds
⎜ dt2 + Υ + Υ
ij
dt dt dt2 ij
dt dt ⎟
A=⎜ ⎝
⎟ .
⎠
1 2
ds ds
dt dt
Note formula (4.125) is determined through the elements of the first
groundform only, therefore it is used to formulated the geodesic curvature
of curves in arbitrary two-dimensional Riemannian manifolds too.

4.7 Relations to the Principal Curvatures of

Two-Dimensional Surfaces
4.7.1 Second Fundamental Form

The coefficients of the second fundamental form

bij dsi dsj , i, j = 1, 2 ,

of the surface S x2 in R3 represented by

x(s) : S 2 → R3 , x = (x1 , x2 , x3 ) , s = (s1 , s2 ) , (4.126)

are defined by the dot products of the second derivatives of the vector func-
tion x(s) and the unit normal vector n to the surface at the point s under
consideration:
bij = xsi sj · n , i, j = 1, 2 .
Thus, from (2.26) we obtain for bij , i, j = 1, 2,

1  ∂ 2 xl  ∂xl+1 ∂xl+2 ∂xl+2 ∂xl+1 

bij = √ rs − , l = 1, 2, 3 , (4.127)
g ∂si ∂sj ∂s1 ∂s2 ∂s1 ∂s2

with the identification convention for the superscripts that k is equivalent to

k±3. Correspondingly, for the monitor surface with the scalar-valued monitor
function f (s), we obtain
 4.7 Relations to the Principal Curvatures of Two-Dimensional Surfaces 107

1
bij =  fsi sj , i, j = 1, 2 . (4.128)
1 + (fs1 )2 + (fs2 )2

The tensor (bij ) reflects the local warping of the surface, namely its devi-
ation from the tangent plane at the point under consideration. In particular,
if (bij ) ≡ 0 at all points of S 2 then the surface is a plane.

4.7.2 Principal Curvatures

Let a curve on the surface S x2 ⊂ R3 be defined by the intersection of a

plane containing the normal n with the surface. Taking into account (3.8),
we obtain for the curvature of this curve
bij dsi dsj
k= rs dsi dsj , i, j = 1, 2 . (4.129)
gij

Here (ds1 , ds2 ) is the direction of the curve, i.e. dsi = c(dsi /dϕ), where s(ϕ)
is a curve parametrization. The two extreme quantities KI and KII of the
values of k are called the principal curvatures of the surface at the point under
consideration. In order to compute the principal curvatures, we consider the
following relation for the value of the curvature:

(bij − kgij
rs
)dsi dsj = 0 , i, j = 1, 2 , (4.130)

which follows from (4.129). In order to find the maximum and minimum
values of k, the usual method of equating to zero the derivative with respect
to dsi is applied. Thus the components of the (ds1 , ds2 ) direction giving an
extreme value of k are subject to the restriction

(bij − kgij
rs
)dsj = 0 , i, j = 1, 2 ,

which, in fact, is the eigenvalue problem for curvature. One finds the eigen-
values k by setting the determinant of this equation equal to zero, obtaining
thereby the secular equation for k:

det(bij − kgij
rs
)=0, i, j = 1, 2 .

This equation, written out in full, is a quadratic equation

k2 − gsr
ij
bij k + [b11 b22 − (b12 )2 ]/g rs = 0 ,

with two roots, which are the maximum and minimum values KI and KII of
the curvature k:

1 ij 1 ij 1
KI,II = gsr bij ± (gsr bij )2 − rs [b11 b22 − (b12 )2 ] . (4.131)
2 4 g
108 4 Multidimensional Geometry

Mean Curvature

One half of the sum of the principal curvatures is, in fact, the mean surface
curvature:
1 ij 1
Km = gsr bij = (KI + KII ) , i, j = 1, 2 . (4.132)
2 2
In the case of the monitor surface represented by the scalar-valued function
f (s1 , s2 ), we obtain using (4.128)
fs1 s1 [1 + (fs2 )2 ] + fs2 s2 [1 + (fs1 )2 ] − 2fs1 fs2 fs1 s2
Km = .
2[1 + (fs1 )2 + (fs2 )2 ]3/2
A surface whose mean curvature is zero, i.e. KI = −KII , possesses the
following unique property. Namely, if a surface bounded by a specified contour
has a minimum area then its mean curvature is zero. Conversely, of all the
surfaces bounded by a curve whose length is sufficiently small, the minimum
area is possessed by the surface whose mean curvature is zero. So the surface
whose mean curvature is equal zero at all its points is referred to as a minimal
surface.

Gaussian Curvature

Taking into account (4.82), we readily see that the Gaussian curvature is the
product of the two principal curvatures KI and KII , i.e.
1
KG = KI KII = rs [b11 b22 − (b12 )2 ] . (4.133)
g
In terms of the derivatives of a scalar-valued function f (s) representing the
monitor surface S r2 we have, from (4.128) and (4.133),
f 1 1 f 2 2 − (fs1 s2 )2
KG = s s s s 2 . (4.134)
[1 + (fs1 ) + (fs2 )2 ]2

Formula Dependent on Christoffel Symbols

In the case of a general two-dimensional surface represented by (4.126) an

expression for the quantity b11 b22 − (b12 )2 can be obtained through deriva-
tives of the elements of the metric tensor and the coefficients of the second
fundamental form of the surface. This is accomplished by using the expansion
(4.74) which yields the following relation:
xs1 s1 · xs2 s2 − xs1 s2 · xs1 s2 = gij
rs i
(Υ11 j
Υ22 − Υ12
i j
Υ12 ) + b11 b22 − (b12 )2 ,

i, j = 1, 2 . (4.135)
The left-hand part of (4.135) equals
 4.7 Relations to the Principal Curvatures of Two-Dimensional Surfaces 109

∂ ∂
xs1 s1 · xs2 s2 − xs1 s2 · xs1 s2 = 1
(xs2 s2 · xs1 ) − 2 (xs1 s2 · xs1 ) . (4.136)
∂s ∂s
Since
1  ∂g12
rs
∂g rs 
xs2 s2 · xs1 = [22, 1] = 2 2 − 22 ,
2 ∂s ∂s1
rs
1 ∂g11
xs1 s2 · xs1 = [12, 1] = ,
2 ∂s2
we obtain from (4.136)
1  ∂ 2 g12
rs
∂ 2 g rs ∂ 2 g rs 
xs1 s1 · xs2 s2 − xs1 s2 · xs1 s2 = 2 1 2 − 2 112 − 1 221 .
2 ∂s ∂s ∂s ∂s ∂s ∂s
Therefore (4.133) results in
1  1  ∂ 2 g12
rs
∂ 2 g rs ∂ 2 g rs  
KG = rs 2 1 2 − 2 112 − 1 221 − gij rs i
(Υ11 j
Υ22 − Υ12
i j
Υ12 ) ,
g 2 ∂s ∂s ∂s ∂s ∂s ∂s
i, j = 1, 2 .
Applying (4.8) transforms this equation to
1  1 ∂2 j j

KG = − rs (g rs ij
gsr ) + g rs
ij (Υ i
Υ
11 22 − Υ i
Υ
12 12 ) ,
g 2 ∂si ∂sj (4.137)
i, j = 1, 2 .

This equation depends on the elements of the first surface groundform
only. Therefore it can be applied to compute the Gauss curvature of two-
dimensional Riemannian manifolds with arbitrary metric tensors.
Since (4.24)
rs i j ij
gij (Υkl Υmp ) = gsr [kl, i][mp, j] , i, j, k, l, m, p = 1, 2 ,
therefore (4.137) has also the following form
1  1 ∂2
KG = − rs (g rs gsr
ij rs
) + gij ([11, i][22, j]
g 2 ∂si ∂sj (4.138)

−[12, i][12, j]) , i, j = 1, 2 .
In particular, in the case of the spherical metric
s
gij = v(s)δji , i, j = 1, 2 ,
we readily find from (4.138)
1 1 2 1 
KG = − ∇ [v] + ([11, i][22, i] − [12, i][12, i])
[v(s)]2 2 v(s)
1  2 1 2 2

=− ∇ [v] − [(vs1 ) + (v s2 ) ] (4.139)
2[v(s)]2 v(s)
1
=− ∇2 [ln v] , i = 1, 2 .
2v(s)
110 4 Multidimensional Geometry

Formula Dependent on Metric Elements

Now we establish one more important expression for KG . For this purpose
we compute the quantity
d = det A
where ⎛ rs rs rs
⎞
g11 g12 g22
⎜ ⎟
⎜ rs ⎟
⎜ ∂g11 rs
∂g12 rs
∂g22 ⎟
A=⎜
⎜ ∂s1
⎟ .
⎟
⎜ ∂s1 ∂s1 ⎟
⎝ ∂g rs rs
∂g12 rs
∂g22 ⎠
11
∂s2 ∂s2 ∂s2
Using (4.22 – 4.24) gives
 ∂g rs ∂g rs rs
∂g12 rs 
∂g22  rs rs rs rs 
12 22 rs ∂g11 ∂g22 ∂g11 ∂g22
rs
d = g11 − − g12 −
∂s1 ∂s2∂s2 ∂s1 ∂s1 ∂s2 ∂s2 ∂s1
 ∂g rs ∂g rs rs
∂g ∂g12 rs 
11 12
rs
+g22 1 2
− 11
∂s ∂s ∂s2 ∂s1
rs
= 2g11 {([11, 2] + [21, 1])[22, 2] − ([12, 2] + [22, 1])[21, 2]}

−4g12 ([11, 1][22, 2] − [12, 1][21, 2])

rs
+2g22 {[11, 1]([12, 2] + [22, 1]) − [12, 1]([11, 2] + [21, 1])}
rs
= 2g11 k rs
[(Υ11 k rs
gk2 + Υ21 gk1 )Υ22 gl2 − (Υ12
l rs k rs k rs
gk2 + Υ22 l rs
gk1 )Υ21 gl2 ]

−4g12
rs
gk1 Υ22 gl2 − Υ12
k rs l rs
(Υ11 k rs l rs
gk1 Υ21 gl2 )
(4.140)
rs k rs
+2g22 l rs
[Υ11 gk1 (Υ12 gl2 + Υ22 gl1 ) − Υ12
l rs k rs l rs
gk1 (Υ11 l rs
gl2 + Υ21 gl1 )]
k l
= 2(Υ11 Υ22 − Υ12
k l
Υ12 )(g11 gk2 gl2 − 2g12
rs rs rs rs rs rs rs rs rs
gk1 gl2 + g22 gk1 gl1 )
rs k l
+2(g11 rs k l
Υ21 Υ22 + g22 Υ11 Υ12 )(gk1 gl2 − gl1
rs rs rs rs
gk2 )
1 1 1 1 2 2 2 2
= 2g rs [g11
rs
(Υ11 Υ22 − Υ12 rs
Υ12 ) + g22 (Υ11 Υ22 − Υ12 Υ12 )
2 1 2 1 1 2 2 1
rs
+2g12 (Υ11 Υ22 − Υ12 rs
Υ12 ) + g11 (Υ21 Υ22 − Υ21 Υ22 )
1 2 2 1
rs
+g22 (Υ11 Υ12 − Υ11 Υ12 )]
j j 2 1 1 2
= 2g rs [gij
rs i
(Υ11 Υ22 − Υ12
i
Υ12 rs
) + g12 (Υ11 Υ22 − Υ11 Υ22 )
1 2 2 1 1 2 2 1
rs
+g11 (Υ21 Υ22 − Υ21 rs
Υ22 ) + g22 (Υ11 Υ12 − Υ11 Υ12 )] , k, l = 1, 2 .
 4.7 Relations to the Principal Curvatures of Two-Dimensional Surfaces 111

On the other hand

∂  ∂ ∂ rs 
k
Υki (g rs ij
g ) = (−1)i+1 k
Υ g rs
23−i − g
∂sj sr ki
∂s1 ∂s2 13−i
k
= Υk1 ([21, 2] − [22, 1]) + Υk2
k
([12, 1] − [11, 2])
k
= Υk1 (Υ21 gl2 − Υ22
l rs l rs k
gl1 ) + Υk2 gl1 − Υ11
l rs
(Υ12 l rs
gl2 )
rs
= gl2 k l
(Υk1 Υ21 − Υk2
k l rs
Υ11 ) + gl1 k l
(Υk2 Υ12 − Υk1
k l
Υ22 )
rs 1 1 2 1 1 2
= g12 [Υ21 (Υ11 + Υ21 ) − Υ11 (Υ12 + Υ22 )]
(4.141)
rs 2 1 2 2 1 2
+g21 [Υ12 (Υ12 + Υ22 ) − Υ22 (Υ11 + Υ21 )]
1 2 1 2 rs 2 1 2 2 1 2
rs
= 2g12 (Υ21 Υ21 − Υ11 Υ22 ) + g22 [Υ21 (Υ11 + Υ21 ) − Υ11 (Υ12 + Υ22 )]
rs 1 1 2 1 1 2
+g11 [Υ12 (Υ12 + Υ22 ) − Υ22 (Υ11 + Υ21 )]
j j 1 2 1 2
rs
= gij i
(Υ12 Υ12 − Υ11
i
Υ22 rs
) + g11 (Υ12 Υ22 − Υ22 Υ21 )
2 1 2 1 2 1 1 2
rs
+g22 (Υ21 Υ11 − Υ11 rs
Υ12 ) + g12 (Υ11 Υ22 − Υ11 Υ22 ) .

Availing us of (4.140) and (4.141) we find

∂ j j
d − 2g rs Υki
k
(g rs gsr
ij
) = 4g rs gij
rs i
(Υ11 Υ22 − Υ12
i
Υ12 ), i, j = 1, 2 . (4.142)
∂sj
Since (4.25)
√ ∂  1 
k
Υki = − g rs i √ rs , i, k = 1, 2 ,
∂s g
therefore, using (4.137), the equation (4.142) becomes
1  1 ∂2 d
KG = − (g rs gsr
ij
) + rs
g rs 2 ∂si ∂sj 4g
√ rs
g ∂  1  ∂ 
+ √ (g rs ij
g sr ) (4.143)
2 ∂si g rs ∂sj
d 1 ∂  1 ∂ 
=− − √ √ (g rs ij
g ) , i, j = 1, 2 .

4(g rs )2 2 g rs ∂si g rs ∂sj sr

In particular, when the coordinate system s1 , s2 is orthogonal, then d = 0,

12
gsr = 0 and (4.143) yields in this case
1  ∂  1 ∂ rs ∂  1 ∂ rs 
KG = − √ rs √ g + √ g . (4.144)
2 g ∂s1 g rs ∂s1 22 ∂s2 g rs ∂s2 11
Classification of Points of Two-Dimensional Surfaces

A surface point is called elliptic if KG > 0, i.e. both KI and KII are both neg-
ative or both positive at the point of consideration. A saddle or hyperbolic
112 4 Multidimensional Geometry

point has principal curvatures of opposite sign, and therefore has negative
Gaussian curvature. A parabolic point has one principal curvature vanish-
ing and, consequently, a vanishing Gaussian curvature. This classification of
points is prompted by the form of the curve which is obtained by the in-
tersection of the surface with a slightly offset tangent plane. For an elliptic
point the curve is an ellipse; for a saddle point it is a hyperbola. It is a pair
of lines (degenerate conic) at a parabolic point, and it vanishes at a planar
point, where both principal curvatures are zero.