See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341266769

Riemann Spaces and Pfaff Differential Forms

Article · May 2020

CITATIONS READS

0 13

1 author:

Nikos Bagis
Aristotle University of Thessaloniki
68 PUBLICATIONS   133 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Elliptic functions & integrals View project

All content following this page was uploaded by Nikos Bagis on 09 May 2020.

The user has requested enhancement of the downloaded file.

 Riemann Spaces and Pfaff Differential Forms

Nikos D. Bagis
Aristotele University of Thessaloniki-AUTH-Greece
nikosbagis@hotmail.gr

Abstract
In this work we study differential geometry in N dimensional Riemann curved
spaces using Pfaff derivatives. Avoiding the classical partial derivative the Pfaff
derivatives are constructed in a more sophisticated way and make evaluations
become easier. In this way Christofell symbols Γikj of classical Riemann geom-
etry as also the elements of the metric tensor gij are replaced with one symbol
(the qikj ). Actually to describe the space we need no usage of the metric tensor
gij at all. We also don’t use Einstein’s notation and this simplifies also things
a lot. For example we don’t have to use upper and lower indexes, which in
eyes of a beginner, is quite messy. Also we don’t use the concept of tensor. All
quantities of the surface or curve or space which form a tensor field are called in-
variants or curvatures of the space. Several new ideas are developed in this basis.

Keywords: Riemann geometry; Curved spaces; Pfaff derivatives; Differential

Operators; Invariant theory

1 Introduction and Development of the Theory

Here we assume a N − 1 dimensional space S, which is embedded in the N
dimensional Euclidean space E. The space S will be described by the vector
N
X
x= xi (u1 , u2 , . . . , uN )i , (1)
i=1

where i is usual orthonormal base of E = RN .

We assume that in every point of the space S correspond N orthonormal vectors
{e1 , e2 , . . . , eN } with the together assumption that vector eN is vertical to every
affine plane of S. Hence the first N − 1 vectors {e1 , e2 , . . . , eN −1 } together span
the tangent space of S and all N span the Euclidean space E. We will use
Pfaff derivatives to write our equations. We also study some properties of S
which will need us for the construction of these equations. The Pfaff derivatives
related with the structure of the space S which produces differential forms ωk ,
k = 1, 2, . . . , N . These are defined as below.

1
It holds that
N
X
∂j x = ∂j xi i . (2)
i=1

The linear element of S is

N
X N
X X
2 2
(ds) = (dx) = h∂i x, ∂j xi dui duj = gii du2i + 2 gij dui duj . (3)
i,j=1 i=1 i<j

Hence  
∂x ∂x
gij = , , (4)
∂ui ∂uj
are the structure functions of the first linear form.
The Pfaff differential forms ωk are defined with the help of {ek }, k = 1, 2, . . . , N
as
XN
dx = ωk e k . (5)
k=1

Then the Pfaff derivatives of the function f are ∇k f , k = 1, 2, . . . , N and holds

N
X N
X
df = (∂k f )duk = (∇k f )ωk , (6)
k=1 k=1

From (6) we get

N
X
dx = (∂k x)duk . (7)
k=1

Also
N
X
dx = ωk ek ⇒ ∇m (x) = em . (8)
k=1

Derivating the vectors el we can write them as linear composition of their self,
since they form a basis of E:
N
X
dei = ωik ek . (9)
k=1

Then we define the connections qijm and bkl as

N
X N
X
ωij = qijm ωm , ωk = bkl dul . (10)
m=1 l=1

Hence from (10),(7) and hek , dxi = ωk , we get that

N
X
hek , ∂l xi = bkl and ∂l x = bkl ek (11)
k=1

2
and from (4)
N
X
gij = bki bkj . (12)
k=1

Also
N
X ∂xm
bsl = cos (φms ) , where cos (φms ) = hm , es i (13)
m=1
∂ul
and φms is the angle formed by m and es . Also it holds
N N
!
X X
ωij = qijm bml dul (14)
l=1 m=1

By this way we get the Christofell symbols

N
X
Γijl = qijm bml . (15)
m=1

Thus in view of (19) below

N
X
ωij = Γijl dul , Γijl + Γjil = 0 (16)
l=1

and easy (see and Proposition 1 below)

N
X N
X
∂k em = Γmjk ej and ∇k em = qmjk ej . (17)
j=1 j=1

From the orthonormality of ek we have

δij = hei , ej i . (18)

Derivating the above relation, we get

ωij + ωji = 0 (19)

and hence
qijm + qjim = 0. (20)

Theorem 1.
The structure equations of S are (19),
N
X N
X
dωj = ωm ∧ ωmj , dωij = ωim ∧ ωmj . (21)
m=1 m=1

3
Proof.
We have
N
! N
X X
d (dx) = 0 ⇒ d ei ωi =0⇒ (dei ∧ ωi + ei dωi ) = 0.
i=1 i=1

Hence
N
X N
X N
X
dωi ei + ωik ∧ ωi ek = 0 ⇒ dωi = ωl ∧ ωli .
i=1 i,k=1 l=1

The same arguments hold and for the second relation of (20). qed

Definition 1.
We write
rotij (Ak...i...j...l ) = Ak...i...j...l − Ak...j...i...l (22)

Definition 2.
We define the Kronecker−δ symbol as follows:
If {i1 , i2 , . . . , iM }, {j1 , j2 , . . . , jM } are two set of indexes, then

δi1 i2 ...iM ,j1 j2 ...jM = 1,

if {i1 , i2 , . . . , iM } is even permutation of {j1 , j2 , . . . , jM }.

δi1 i2 ...iM ,j1 j2 ...jM = −1,

if {i1 , i2 , . . . , iM } is odd permutation of {j1 , j2 , . . . jM } and

δi1 i2 ...iM ,j1 j2 ...jM = 0

in any other case.

PN PN
Let a = i=1 ai ωi and b = j,k=1 bjk ωj ∧ ωk , then
N
1 1 X
(a ∧ b)123 = δijk,123 aj bjk =
1! 2!
i,j,k=1

1
= [a1 b23 δ123,123 + a1 b32 δ132,123 + a2 b13 δ213,123 + a2 b31 δ231,123 +
2
+a3 b12 δ312,123 + a3 b21 δ321,123 ] =
1
(a1 b23 − a1 b32 − a2 b13 + a2 b31 + a3 b12 − a3 b21 ) .
=
2
We remark here that we don’t use Einstein’s index notation.

4
 By this way equations (21) give
N
X N
X
dωj = ωm ∧ ωmj = qmjs ωm ∧ ωs .
m=1 m,s=1

Hence X
dωj = Qmjs ωm ∧ ωs , (23)
m<s

with
Qmjs := rotms (qmjs ) = qmjs − qsjm (24)
Hence with the above notation holds Qikl + Qlki = 0. Also if we set
N
X
Rijkl := − rotkl (qimk qjml ) , (25)
m=1

then we have X
dωij = Rijkl ωk ∧ ωl . (26)
k<l

Using the structure equations of the surface S we have the next

Proposition 1.
N
X
∇k ∇m (x) = ∇k (em ) = qmjk ej . (27)
j=1

Proof.
N
X N
X N
X
dei = ∇k (ei )ωk ⇔ ωik ek = ∇k (ei )ωk ⇒
k=1 k=1 k=1
N N
! N
X X X
qikm ek ωm = ∇m (ei )ωm .
m=1 k=1 m=1

Theorem 2.
For every function f hold the following relations
N
X
∇l ∇m f − ∇ m ∇ l f + (∇k f )Qlkm = 0, ∀ l, m ∈ {1, 2, . . . , N } (28)
k=1

or equivalently !
N
X
rotlm ∇ l ∇m f + (∇k f )qlkm = 0. (29)
k=1

5
Proof.
N
X N
X N
X
df = (∇k f )ωk ⇒ d(df ) = d(∇k f ) ∧ ωk + (∇k f )dωk = 0
k=1 k=1 k=1

or
N N
! N
!
X X X X
(∇s ∇k f )ωs ∧ ωk + (∇k f ) rotlm (qlkm )ωl ∧ ωm =0
k=1 s=1 k=1 l<m

or !
X N
X
∇l ∇m f − ∇ m ∇ l f + (∇k f )Qlkm ωl ∧ ωm = 0.
l<m k=1

Corollary 1.
If λij = −λji , i, j ∈ {1, 2, . . . , N } is any antisymetric field, then
N
X N
X
λij ∇i ∇j f + λij qikj ∇k f = 0. (30)
i,j=1 i,j,k=1

Theorem 3.
N
X N
X
Riklm = ∇l (qikm ) − ∇m (qikl ) + qiks Qlsm = − rotlm (qisl qksm ) (31)
s=1 s=1

and it holds also

Riklm = −Rkilm , Riklm = −Rikml , (32)

Proof.
We describe the proof.
N
X N
X
dωij = d(qijk ) ∧ ωk + qijk dωk =
k=1 k=1

N N
!
X X
= ∇m (qijl ) − ∇l (qijm ) + qijs Qmsl ωm ∧ ωl
m<l s=1

and !
N
X N
X N
X
dωij = ωik ∧ ωkj = rotml qism qsjl ωm ∧ ωl
k=1 m<l s=1

Having in mind the above two relations we get the result.

Note 1.

6
When exist a field Φij such that ∇k Φij = qijk : (a), then from Theorem 2 we
have d(ωij ) = 0 and using Theorem 3:
N
X
Riklm = − rotlm (qisl qksm ) = 0 (33)
s=1

and in view of (26),(16):

∂l Γijk − ∂k Γijl = 0. (34)

Proposition 2.
N N
!
X X
∇l ∇m (ei ) = ∇l (qikm ) − qism qksl ek (35)
k=1 s=1

Hence
N
X
∇2k (em ), em = − 2


qmsk = invariant (36)
s=1

Tijlm := h∇l ∇m (ei ) − ∇m ∇l (ei ), ej i = invariant, (37)

and !
N
X
Tijlm = rotlm ∇l (qijm ) + qisl qjsm (38)
s=1

for all i, j, l, m ∈ {1, 2, . . . , N }.

Theorem 4.
N
X
Tijlm = − qijs Qlsm (39)
s=1

and
∇l (qijm ) − ∇m (qijl ) (40)
are invariants.

Proof.
Use Theorem 3 along with Proposition 2.

Definition 3.
(1)
We construct the differential operator Θlm , such that for a vector field Y =
Y1 e1 + Y2 e2 + . . . + YN eN it is
N
(1)
X
Θlm (Y ) := ∇l (Ym ) − ∇m (Yl ) + Yk Qlkm . (41)
k=1

7
Theorem 5.
The derivative of a vector Y = Y1 e1 + Y2 e2 + . . . + YN eN is
N N
!
X X
dY = Yj;l ωl ej (42)
j=1 l=1

where
N
X
Yj;l = ∇l Yj − qjkl Yk = invariant. (43)
k=1

Remark 1.
(1)

Θlm Y = Ym;l − Yl;m (44)

Lemma 1.
For every vector field Y we have


∇k Y , el = Yl;k = invariant (45)

Proof. * ! +
N
X


∇k Y , e l = ∇k Ym em , el =
m=1
N
X
= h∇k (Ym ) em + Ym ∇k (em ) , el i =
m=1
N
X N
X N
X
= ∇k Yl + Ym qmsk hes , el i = ∇k Yl + Ym qmlk =
m=1 s=1 m=1
N
X
= ∇k Yl − Ym qlmk = Yl;k = invariant.
m=1

Proposition 3.
The connections qijk are invariants.

Proof. *N +
X
h∇k em , el i = qmjk ej , el = qmlk .
j=1

Definition
PN 4.
If ω = k=1 ak ωk is a Pfaff form, then we define
N
(2)
X
Θlm (ω) := ∇l am − ∇m al + ak Qlkm (46)
k=1

8
Hence
N
(2)
X
dω = Θlm (ω) ωl ∧ ωm , (47)
l<m
(1) (2)
The above derivative operators Xm;l , Θlm (ω) and Θlm (ω) are invariant un-
der all acceptablePchange of variables. As application of such differentiation are
N
the forms ωij = k=1 qijk ωk , which we can write
(2)
Θlm (ωij ) = Rijlm (48)

and lead us concluding that Rijlm are invariants of S. Actually

Theorem 6.
Rijlm is the curvature tensor of S.

Theorem
PN 7.
If ω = k=1 ak ωk , then
X  ∇l f ∇m f  
 + Θ(2) (ω)f ωl ∧ ωm .

d(f ω) = 
 al am  lm (49)
l<m

In particular
X  ∇l f ∇m f  

d(f ωij ) =   + Rijlm f ωl ∧ ωm . (50)
 qijl qijm 
l<m

Also  
(2)  ∇ f ∇m f
=  l  + f Θ(2) (ω).

Θlm (f ω) lm (51)
al am 
If we set
R := R1 ω1 + R2 ω2 + . . . + RN ωN , (52)
then
N
(2)
X
Θlm (R) = ∇l Rm − ∇m Rl + Rk Qlkm (53)
k=1

and
N
(2)
X X X
Θlm (ω) = (∇l am − ∇m al ) + ak Ak . (54)
l<m l<m k=1

Theorem 8.
We set
R := R1 e1 + R2 e2 + . . . + RN eN , (55)
where the Rk are as in Definition 6 below, then
X
(∇l Rm − ∇m Rl ) = invariant. (56)
l<m

9
Proof. PN
From Definition 6 below and k=1 Rk Ak = 0 (relation (76) below), we get
N
!
X (1)
X X
Θlm R = ∇l Rm − ∇m Rl + Rk Qlkm (57)
l<m l<m k=1

Note 2.
i)
(2)
Θlm (ωj ) = Qljm = invariant (58)
Hence
(2)
X
Θlm (ωj ) = Aj = invariant (59)
l<m
PN
ii) If we assume that ω = dg = k=1 (∇k g)ωk and use (27) we get
(2)
Θlm (dg) = 0 (60)

and hence for all multivariable functions f, g we have the next

Theorem 9.
Z XZ Z
f dg = (∇l f ∇m g − ∇m f ∇l g) ωl ∧ ωm (61)
∂A l<m A

Proposition 4.
If exists multivariable function f = f (u1 , u2 , . . . , uN ) ∈ R such that
     
 ∇1 f ∇ 2 f   ∇2 f ∇3 f 
 = . . . =  ∇N −1 f ∇N f
 
 =  = 0, (62)
 qij1 qij2   qij2 qij3   qijN −1 qijN 

then exists function µij such that

Z XZ Z
µij df = Rijlm ωl ∧ ωm . (63)
∂A l<m A

Proof.
Obviously we can write
∇1 f ∇2 f ∇N f 1
= = ... = = .
qij1 qij2 qijN µij

From Theorem 3 we have

N
X
∇l (qijm ) − ∇m (qijl ) + qijk Qlkm = Rijlm .
k=1

10
Hence
N
X
∇l (µij ∇m f ) − ∇m (µij ∇l f ) + (∇k f )µij Qlkm = Rijlm ,
k=1

or equivalently using Theorem 2

∇m f · ∇l µij − ∇l f · ∇m µij = Rijlm .

Hence from relation (61) (Stokes formula) we get the result.

Note 3.
Condition (62) is equivalent to say that exist functions µij and f such that

ωij = µij df. (64)

Theorem 10.
If exists function field Fij such that
(1)
X X
Fij Θlm (xωij ) = 0, (65)
l<m i,j∈I

then !
N P
X i,j∈I Fij rijk
x= P P ek . (66)
l<m i,j∈I Fij Rijlm
k=1

Proof.
From (50) and (47) we have
 
X X  ∇l (x) ∇m (x)  X
Θlm (xωij ) = 
 qijl qijm
+x
 Rijlm =
l<m l<m l<m

X  el em  X N
X X

= 
 qijl qijm
+x
 Rijlm = − ks qijs ek + x Rijlm =
l<m l<m k,s=1 l<m

N
X X
=− rijk ek + x Rijlm , (67)
k=1 l<m

where the rijk are defined in (68) and the ks are that of Definition 6 below.
Hence, if exists function field Fij such that (65) holds, then we get the validity
of (66).

Above we have set

N
X
rijk = ks qijs . (68)
s=1

11
Also * +
X X
Θlm (xωij ) , ek = −rijk + wk Rijlm , (69)
l<m l<m

where wk = hx, ek i is called support function of the hypersurface.

Setting the symbols

fk = fk (x1 , x2 , . . . , xN ), k = 1, 2, . . . , N

be such that
N
X
x= fk e k , (70)
k=1

then
N
X N
X N
X N
X N
X
d(x) = dfk ek + fk d(ek ) = (dfk )ek + fk ωkj ej =
k=1 k=1 k=1 k=1 j=1
 
N
X N
X N
X N
X N
X
= (dfk )ek + fj ωjk ek = dfk + fj ωjk  ek .
k=1 j=1 k=1 k=1 j=1

Hence from (5) and the above relation we get

N
X
ωk = dfk + fj ωjk . (71)
j=1

If we use the Pfaf

N
X N
X N
X
ωk = (∇l fk )ωl + fj qjkm ωm
l=1 j=1 m=1

Or equivalent
N
X N X
X N
ωk = (∇l fk )ωl + fj qjkl ωl
l=1 l=1 j=1

Hence
N
X
∇l f k − fj qkjl = δlk
j=1

Or equivalently we conclude that:

The necessary conditions such that fk be the coordinates of the vector x
(who generates the space S), in the moving frame ek , are

fk;l = δkl .

12
Hence we get the next

Theorem 11.
If
fk;l = δkl , where k, l ∈ {1, 2, . . . , N }, (72)
then we have
N
X
x= fk ek + h, where dh = 0
k=1
and the opposite.

Finally having in mind of (66) we get

P
i,j∈I Fij rijk
fk = P P (73)
l<m i,j∈I Fij Rijlm
For two generalized hyper-vectors F = Fij...k and G = Gij...k we define the
generalized inner product as
N
X
(F, G) := Fij...k Gij...k . (74)
i,j,...,k=1

Hence relation (73) can be written as

N
X Fij
fk = rijk ,
i,j=1
(F, U )

where X
U := Uij := Rijlm
l<m
Hence
N   N
X Fij X Fij
fk;l = ∇l rijk − rijm qkml =
i,j=1
(F, U ) i,j,m=1
(F, U)
N   N N
X Fij X Fij X Fij
= ∇l rijk + ∇l (rijk ) − rijm qkml =
i,j=1
(F, U ) i,j=1
(F, U ) i,j,m=1
(F, U )
N   N
X Fij X Fij
= ∇l rijk + rij{k};l
i,j=1
(F, U ) i,j=1
(F, U)

Theorem 12.
If Fij is such that
N   N
X Fij X Fij
∇l rijk + rij{k};l = δkl , (75)
i,j=1
(F, U ) i,j=1
(F, U)

13
then exists constant vector h such that
N
X Fij
x= rijk ek + h, (76)
(F, U )
k,i,j=1

where dh = 0.

Theorem 13.
There holds
qik{m};l − qik{l};m = Riklm . (77)

Proof.
Easy

Note 4.
In general in tij...{k}...m;l the brackets indicate where the differential acts. Hence
N
X
tij...{k}...m;l = ∇l tij...k...m − qkνl tij...ν...m (78)
ν=1

Also we can use more brackets {} in the vector.

tij...{k1 }...{k2 }...m;l =
N
X N
X
= ∇l tij...k1 ...k2 ...m − qk1 ν1 l tij...ν1 ...k2 ...m − qk2 ν2 l tij...k1 ...ν2 ...m (79)
ν1 =1 ν2 =1

Theorem 14.
If eN is orthogonal to S, then ωN = 0 and the following forms are invariants of
the hypersurface S:
i) The linear element is
N
X
I = (ds)2 = ωk2 (80)
k=1
ii) The area element is
E = ω1 ∧ ω2 ∧ . . . ∧ ωN −1 (81)
iii) The second invariant form is
N
X
II = hdx, deN i = ωk ωN k (82)
k=1

iv) The linear element of n = eN is

N
X
III = (dn)2 = hdeN , deN i = 2
ωN k (83)
k=1

14
v) The Gauss curvature K is given from
ωN 1 ∧ ωN 2 ∧ . . . ∧ ωN N −1
K = det(κij ) = , (84)
E
where
κkm := qkN m . (85)
The above forms remain unchanged, in every change of position of x and rota-
tion of {ej }j=1,2,...,N , except for possible change of sign.

Definition 5.
We define the Beltrami differential operator
 
N N  
∇2l f + ∇2m f + 1
X X  ∇i f ∇j f 
∆2 f :=   . (86)
N − 1 i<j  Qlim Qljm 
l<m

We also call f harmonic if ∆2 f = 0.

Remark 2.
In the case of N = 3 where the space S is a two dimensional surface and we
have
∆2 A = ∇1 ∇1 A + ∇2 ∇2 A + q2 ∇1 A − q1 ∇2 A, (87)
where
dx = ω1 e1 + ω2 e2 (88)
and ω3 = 0, dω1 = q1 ω1 ∧ ω2 and dω2 = q2 ω1 ∧ ω2 , there exists functions f, f ∗
such that ∇1 f = −∇2 f ∗ and ∇2 f = ∇1 f ∗ and ∆2 f = ∆2 f ∗ = 0, (f, f ∗ har-
monics). In higher dimensions is not such easy to construct harmonic functions.
However we will give here one way of such construction. Before going to study
this operator, we simplify the expansion (86) of the definition of Beltrami dif-
ferential operator ∆2 . We also generalize it as we show below in Definition 8.
First we give a definition.

Definition 6.
Set
X N
X
As := Qlsm , Rk := ks As (89)
l<m s=1
and more generally if
N
X N
X
as = tlsm and rs := sk ak , (90)
l<m k=1

PN1 ks := −1 if k < s, 1 if k > s and 0 if k = s. Then from identity

where
k,s=1 ks fk fs = 0, for every choice of fk , we get

N
X N
X
Rk Ak = 0, rs αs = 0 (91)
k=1 s=1

15
 From the above definition we have the next

Theorem 15.
N  
X 1
∆2 f = (N − 1)∇2k f − (∇k f ) Rk . (92)
N −1
k=1

Also
N
X 1 D E
∆2 f = (N − 1) ∇2k f − grad(f ), R . (93)
N −1
k=1

Proof.
 
N
X
∇2l f + ∇2m f − 1 X
∆2 f = (∇k f ) ks Qlsm  =
N −1
l<m k,s=1

N N N
X 1 X X
= (N − 1) ∇2k f − (∇k f ) ks As =
N −1 s=1
k=1 k=1
N N
X 1 X
= (N − 1) ∇2k f − (∇k f ) Rk .
N −1
k=1 k=1

Note 5. PN
1) If exists f such k=1 Ak ωk = df , then ∇k f = Ak and
N
X
∆2 f = (N − 1) ∇ k Ak .
k=1
PN
2) If hk is any vector such that k=1 hk = 0 and fk = ∇k F is such that

1
(N − 1)∇k fk − fk Rk = hk , ∀k = 1, 2, . . . , N, (94)
N −1
then
∆2 F = 0.
Hence solving first the PDE’s
1
(N − 1)∇k Ψ − ΨRk = hk , ∀k = 1, 2, . . . , N, (eq1)
N −1
N
X
hk = 0, (eq2)
k=1

we find N solutions Ψ = fk . Then we solve fk = ∇k F and thus we get a solution

F of ∆2 F = 0.

16
3) This lead us to define the derivative Dk , k = 1, 2, . . . , N , which acts in every
function f as
1
Dk (f ) := Dk f := (N − 1)∇k f − Rk f. (95)
N −1
Obviously Dk is linear. But
N
X N
X N
X
d(f g) = gdf + f dg = g ∇k f ωk + f ∇k gωk = (g∇k f + f ∇k g) ωk ,
k=1 k=1 k=1

or equivalently
∇k (f g) = g∇k f + f ∇k g. (96)
Hence the differential operator Dk acts in the product of f and g as
1
Dk (f g) = (N − 1)∇k (f g) − f gRk =
N −1
1
= (N − 1)f ∇k g + (N − 1)g∇k f − f gRk = f Dk g + (N − 1)g∇k f =
N −1
= gDk f + (N − 1)f ∇k g
and finally we have
1
Dk (f g) = f Dk g + gDk f + f gRk , (97)
N −1
f Dk g − gDk f = (N − 1)(f ∇k g − g∇k f ). (98)
The commutator of Dk and ∇k acting to a scalar field is

[Dk , ∇k ] f = Dk ∇k f − ∇k Dk f =
 
1 1
= (N − 1)∇2k f − Rk ∇k f − (N − 1)∇2k f − ∇k (Rk f ) .
N −1 N −1
Hence finally after simplifications
∇k Rk
[Dk , ∇k ] f = f. (99)
N −1
i) If f = const, then
1
Dk f = − Rk f.
N −1
ii) The derivative Dk is such that if exists function f with

Dk f = 0, ∀k = 1, 2, . . . , N,

then equivalently
1
∇k (log f ) = Rk , ∀k = 1, 2, . . . , N.
(N − 1)2

17
But then
Dk f = 0 ⇔ ∃g = (N − 1)2 log f : Rk = ∇k g ⇔ R = dg.


(100)
By these arguments we conclude to the result, that only in specific spaces S ex-
ists f such Dk f = 0, for all k = 1, 2, . . . , N . Hence we have a definition-theorem:

Definition 7.
A space is called S-R iff exists function g such that R = dg.

Theorem 16.
A space S is S-R iff exists g such that Dk g = 0, for all k = 1, 2, . . . , N .

Corollary 2.
In a S-R space
Θlm (R) = 0, ∀l, m ∈ {1, 2, . . . , N }. (101)
Proof. P
N
It is R = k=1 Rk ωk and S is S-R. Hence
N
X
Θlm (R) = ∇l Rm − ∇m Rl + Rk Qlkm .
k=1

But exists g such that Rk = ∇k g, for all k = 1, 2, . . . , N . Hence

N
X
Θlm (R) = ∇l ∇m g − ∇m ∇l g + (∇k g)Qlkm = 0.
k=1

The last equality is due to Theorem 2. Hence the result follows.

Corollary 3.
In a S-R space holds
X
d(f R) = (Rm ∇l f − Rl ∇m f ) ωl ∧ ωm . (101.1)
l<m

Proof.
The result is application of Theorem 7.

Corollary 4.
In every space S holds
N
X N
X
Ak (Dk f ) = (N − 1) Ak (∇k f ). (101.2)
k=1 k=1

Theorem 17.
Set
Df := (D1 f )ω1 + (D2 f )ω2 + . . . + (DN f )ωN , (101.3)

18
then
1
Df = (N − 1)df − f R. (101.4)
N −1
Also



 R

D V 1 , V 2 = DV 1 , V 2 + V 1 , DV 2 + V 1, V 2 . (101.5)
N −1

Theorem 18.
If a space S is S-R, then the PDE ∆2 Ψ = µΨ have non trivial solution.

Proof.
If S is S-R, then from (100) we get that exists f, g such g = (N − 1)2 log f and
R = dg, Dj f = 0, ∀j. Hence
1
(N − 1)∇j f − Rj f = 0 ⇒
N −1
1 ∇j Rj
(N − 1)∇j ∇j f − Rj ∇ j f − f =0⇒
N −1 N −1
N
f X
∆2 f = ∇j R j .
N − 1 j=1

Theorem 19.
In every space S we have
N
X 1
Df := (Dk f )ek = (N − 1)gradf − Rf (102)
i=1
N −1

Note 6.
1) If C is a curve, then
     
Df df 1 dR
= (N − 1) − f
ds C ds C N − 1 ds C

and if (dR)C 6= 0, then

   
Df df 1
= (N − 1) − f
dR C dR C N − 1
 
and Df dR = 0 iff
C  
df 1
= f.
dR C (N − 1)2
Hence  
RC
fC = const · exp .
(N − 1)2

19
  
Hence if f = f (x1 , x2 , . . . , xN ) and C : xi = xi (s), s ∈ [a, b], such Df ds = 0,
C
then
 
R (x1 (s), x2 (s), . . . , xN (s))
f (x1 (s), x2 (s), . . . , xN (s)) = const · exp .
(N − 1)2

2) Also assuming C : xi = xi (s) is

a curve
 in S and ν is the tangent vector of
C, which is such that hν, νi = 1 ⇒ ν, dνds = 0, then

Dν dν 1 dR
= (N − 1) − ν.
ds ds N − 1 ds
Consequently we get  
Dν dν
, = (N − 1)k(s)
ds ds
and  
Dν 1 dR
,ν =− .
ds N − 1 ds
But
dR dR
D hν, νi = D1 ⇔ 2 hDν, νi + =− .
N −1 N −1
Hence
dR
hDν, νi = − . (103)
N −1
Hence differentiating we get
d hDν, νi = 0.

Theorem 20.
For every curve C : xi = xi (s) of a general space S, with s being its normal
parameter, the unitary tangent vector ν of C has the property
    
Dν 1 dR
,ν = − .
ds C N − 1 ds C

Moreover we define
    
∗ ∗ 1  Dν 
k = (k )C := := 
 ,
ρR C ds C 

then s   2
∗ 1 dR
k (s) = (N − 1)2 k 2 (s) + .
(N − 1)2 ds C

If the space is S-R, then

(k ∗ )C = (N − 1)k(s).

20
Corollary 5.
Let C : xi = xi (s) be a curve of a S-R space. If s is the normal parameter of C
and ν is the unitary tangent vector of C, then
 
Dν dν  Dν 
= (N − 1) and   = (N − 1)k(s),
ds ds ds 

where k(s) is the curvature of C.

Next we generalize the deffinition of ∆2 operator:

Definition 8.
N
X
∆(λ) (f ) = λij Di ∇j f. (104)
i,j=1

Hence if λij = δij , then

N
X N
X
∆2 f = δij Di ∇j f = Dk ∇k f. (105)
i,j=1 k=1

Theorem 21.
If λij = ij , then
N N
!
()
X 1 X
∆ f= (N − 1)Ak + ks Rs ∇k f. (106)
N − 1 s=1
k=1

Proof.
For λij = ∗ij = −ij we have

N
∗ X X X
∆( ) f = ∗ij Di ∇j f = ∗ij Di ∇j f + ∗ij Di ∇j f =
i,j=1 i<j i>j

X
= (Di ∇j f − Dj ∇i f ) =
i<j
X 1 X
= (N − 1) (∇i ∇j f − ∇j ∇i f ) − (Ri ∇j f − Rj ∇i f ) =
i<j
N − 1 i<j

N N
XX 1 X
= −(N − 1) (∇k f )Qikj − ks Rs ∇k f =
i<j k=1
N −1
k,s=1

21
 N N
X 1 X
= −(N − 1) Ak ∇ k f − ks Rs ∇k f =
N −1
k=1 k,s=1

N N
!
X 1 X
=− (N − 1)Ak + ks Rs ∇k f.
N − 1 s=1
k=1

qed

Example 1.
If λij = ij and S is S-R, then

∆() g = 0, (107)

where dg = R. That is because

N N
X 1 X
∆() g = (N − 1) Ak Rk + ks Rk Rs = 0,
N −1
k=1 k,s=1

PN PN
since k=1 Ak Rk = 0 and k,s=1 ks Rk Rs = 0. Also easily we get
D E
∆() x, R = 0. (108)

Example 2.
In a S-R space we have R = dg, for a certain g and thus ∇k g = Rk . Hence we
can write
N N
X 1 X 2
∆2 g = (N − 1) ∇k Rk − Rk . (109)
N −1
k=1 k=1

Theorem 22.
In the general case when λij is any field we can write
N N
(λ)
X 1 X
∆ f = (N − 1) λij ∇i ∇j f − λij Ri ∇j f. (110)
i,j=1
N − 1 i,j=1

and
N N
D E X 1 X
∆(λ) x, ek = (N − 1) λij qjki − λik Ri . (111)
i,j=1
N − 1 i=1

Proof.
The first identity is easy. For the second we have: Setting f → x, we get
N
X N
X
∆(λ) x = λij Di (∇j x) = λij Di (ej ) =
i,j=1 i,j=1

22
 N N
X X λij
= (N − 1)λij ∇i ej − Ri ej =
i,j=1 i,j=1
N −1

N  
X 1
= λij (N − 1)∇i ej − Ri ej =
i,j=1
N −1

N N
X λik X
= (N − 1)λij qjki ek − Ri ek
N −1
i,j,k=1 i,k=1
 
N N N
X
(N − 1)
X 1 X
= λij qjki − λik Ri  ek .
i,j=1
N − 1 i=1
k=1

We need some notation to proceed further

Definition 9.
For any field λij we define
(λ) (λ)∗
X X
Ak := λij Qikj , Ak := λij Q∗ikj ,
i<j i<j

where
Qikj := qikj − qjki , Q∗ikj := qikj + qjki
and
N N
(λ) (λ)∗
X X
Rk := ks A(λ)
s , Rk := ks A(λ)∗
s .
s=1 s=1

Proposition 5.
N
X N
X
A(λ)
s Rs
(λ)
= 0, A(λ)∗
s Rs(λ)∗ = 0.
s=1 s=1

Theorem 23.
Let λij be antisymmetric i.e. λij = −λji , then we have in general:
1) !
N N
(λ)
X (λ) 1 X
∆ f =− (N − 1)Ak + λik Ri ∇k f.
N − 1 i=1
k=1

2)
N
D E
(λ) 1 X
∆(λ) x, ek = −(N − 1)Ak − λik Ri .
N − 1 i=1

23
3)
D E N
(λ)
X
∆(λ) x, R = −(N − 1) Ak R k
k=1

and
N
D (λ)
E 1 X (λ)
∆(λ) x, R =− λik Ri Rk .
N −1
i,k=1

4) In case the space is S-R, with ∇k g = Rk , then

N
(λ)
X
∆(λ) g = −(N − 1) Ak Rk .
k=1

Proof.
The case of (1) follows with straight forward evaluation. Since λij is antisym-
metric, we have
X X
∆(λ) f = λij Di ∇j f − λij Dj ∇i f =
i<j i<j
  
X 1 1
= λij (N − 1)∇i ∇j f − Ri ∇j f − (N − 1)∇j ∇i f − Rj ∇i f
i<j
N −1 N −1
X 1 X
= (N − 1) λij (∇i ∇j f − ∇j ∇i f ) − λij (Ri ∇j f − Rj ∇i f ) =
i<j
N − 1 i<j

N
!
X X 1 X
= (N − 1) λij − ∇k f Qikj − λij rotij (Ri ∇j f ) =
i<j
N − 1 i<j
k=1

N
X (λ) 1 X
= −(N − 1) Ak ∇ k f − λij rotij (Ri ∇j f ) =
N − 1 i<j
k=1

N N N
!
X 1 X X
(λ)
= −(N − 1) − Ak ∇ k f λik Ri ∇k f =
N −1
k=1 k=1 i=1
N N
!
X (λ) 1 X
=− (N − 1)Ak + λik Ri ∇k f.
N − 1 i=1
k=1

In the second we use ∇k x = ek .

The third relation can be proved if we consider the formula
N
X X X
λik Ri Rk = λik Ri Rk − λik Rk Ri = 0.
i,k=1 i<k i<k

24
The fourth case follows easily from the first with ∇k g = Rk . Note that in the
most general case where λij is arbitrary we still have
N
(λ) (λ)
X
Ak Rk = 0.
k=1

Theorem 24.
In case λij is any antisymmetric field, we have
N D
X E D E
∆(λ) f = ∆(λ) x, ek ∇k f = ∆(λ) x, grad(f ) , (112)
k=1

where
N
X
grad(f ) := (∇k f )ek . (113)
k=1

Proof.
If λij is any antisymmetric field, then using Theorem 22, Theorem 2 and Corol-
lary 1, we get
 
N D N N N
X E X X 1 X
∆(λ) x, ek ∇k f = (N − 1) λij qjki − λik Ri  ∇k f =
i,j=1
N − 1 i=1
k=1 k=1

N N
X 1 X
= −(N − 1) λij qikj ∇k f − λik Ri ∇k f = ∆(λ) f.
N −1
i,j,k=1 i,k=1

But
N
X X
λij qikj = λij (qikj − qjki ) = Aλk
i,j=1 i<j

and the result follows.

Corollary 6.
If λij is any antisymmetric field and

∆(λ) (x) = 0. (114)

Then for every f we have
∆(λ) f = 0. (115)

Remark 3. The above corollary is very strange. The Beltrami operator over
an antisymmetric field is zero for every function f when (114) holds. This force
us to conclude that in any space the vector
N
(λ)
X
σ (λ) := σk ek = ∆(λ) (x) (116)
k=1

25
must play a very prominent role in the geometry of S. Then (in any space):
N N
(λ)
X 1 X
σk = (N − 1) λij qjki − λik Ri . (117)
i,j=1
N − 1 i=1

In case λij is antisymmetric, then we get

N
(λ) (λ) 1 X
σk = −(N − 1)Ak − λik Ri =
N − 1 i=1

N
(λ) 1 X (λ)
= −(N − 1)Ak − kl Al ,
N −1
l=1

where
N
(λ)
X
kl := λik il . (118)
i=1

Hence when λij is antisymmetric, then

N
(λ) (λ) 1 X (λ)
σk = −(N − 1)Ak − kl Al .
N −1
l=1

Hence " #
N N
(λ)
X 1 X (λ)
σk =− (N − 1)λij Qikj + kl Qilj (119)
i<j
N −1
l=1

Definition 10.
We define
N
(λ) 1 X (λ)
Λkij := (N − 1)λij Qikj + kl Qilj (120)
N −1
l=1

and
N
(λ) 1 X (λ)
Mkij := (N − 1)λij Q∗ikj − kl Qilj . (121)
N −1
l=1

Also we define the mean curvature with respect to the λij as

N
(λ)
X
Hk := (N − 1) λij qjki . (122)
i,j=1

Also we define
N
(λ± )
X
Hk := (N − 1) λ±
ij Qikj (123)
i,j=1

26
and
N
(λ± )∗
X
Hk := (N − 1) λ± ∗
ij Qikj . (124)
i,j=1

Theorem 25.
If λij is antisymmetric we have
N
(λ) (λ)
X
σk =− Λkij . (125)
i<j

− −
If λij = λ+ +
ij + λij , where λij is the symmetric part of λij and λij is the antisym-
metric part of λij , then
∗
(λ)
X + X −
λ
σk = Mkij − Λλkij =
i≤j i<j

N
X + X − X
λ
= Mkij − Λλkij + (N − 1) λii qiki , (126)
i<j i<j i=1

where the asterisk on the summation means that when i = j we must multiply
the summands with 21 .

Corollary 7.
We have
∆(λ) x = 0, (127)
iff for all k = 1, 2, . . . , N we have
(λ+ )∗ (λ− ) N
Hk − Hk 1 X
− λik Ri = 0. (128)
2 N − 1 i=1

Remark 4. Actually then we have

N
(λ) (λ) 1 X
σk = Hk − λik Ri . (129)
N − 1 i=1

where
N
(λ)
X N − 1  (λ+ )∗ (λ− )

Hk := (N − 1) λij qjki = Hk − Hk . (130)
i,j=1
2

In case λij = gij is the metric tensor, then we write

N
(g)
X
Hk = Hk = (N − 1) gij qikj (131)
i,j=1

27
and call Hk mean curvature of the surface S.

Proof.
Using Theorem 22 and Definition 10, we get
N
(λ+ )∗ (λ− ) 1 X  (λ)+ (λ)−
X
(N − 1)Ak − (N − 1)Ak − kl − kl Qilj = 0 ⇔
N −1 i<j
l=1

N N
X X 1 X
(N − 1) rot∗ij (λij qjki ) + (N − 1) λii qiki − λik Ri = 0 ⇔
i<j i=1
N − 1 i=1

(λ+ )∗ (λ− ) N
Hk − Hk 1 X
= λik Ri
2 N − 1 i=1

Theorem 26.
If λij = gij is the metric tensor, then ∆(g) x = 0 iff
N
1 X
Hk − gik Ri = 0. (132)
N − 1 i=1

Corollary 8.
If λij is antisymmetric, then

∆(λ) (f g) = f ∆(λ) g + g∆(λ) f

Proof.
Use Theorem 24 with

grad(f g) = f grad(g) + ggrad(f ).

Theorem 27.
Assume that λij = ij −antisymmetric. Then
N
1 X (2)
∆() x = 0 ⇔ (N − 1)Ak + kl Al = 0, (133)
N −1
l=1

where
N
(2)
X
kl = ik il . (134)
i=1

For every space S with Ai as above and for every function f , we have

∆() f = 0. (135)

28
Proof.
Use Theorem 23 with λij = ij and then Definitions 9,6.

Remark 5. From the above propositions we conclude that in every space

S we have at least one antisymmetric field (the λij = ij ) that under condition
N N
1 X (2) X ()
(N − 1)Ak + kl Al = 0 ⇔ Λkij = 0, (136)
N −1
l=1 k=1

we have
∆() (f ) = 0, ∀f. (137)
Hence in every space S the quantity
N
() 1 X (2)
σk := (N − 1)Ak + kl Al , k = 1, 2, . . . , N (138)
N −1
l=1
(λ)
is important. More general the quantities σk of (119), for λij antisymmetric
are of extreme interest.

Corollary 9.
If λij is antisymmetric, then
1 D E 1 D E
∆(λ) f = ∆ (λ)
x, R f + Df , ∆ (λ)
x . (139)
(N − 1)2 N −1

Proof.
It holds
1
Df = (N − 1)gradf − Rf
N −1
Hence
D E D E 1 D E
Df , ∆(λ) x = (N − 1) gradf , ∆(λ) x − f ∆(λ) x, R f.
N −1
Now since λij is antisymmetric we have from Theorem (24) the result.

Corollary 10.
If λij is antisymmetric and Df = 0, then
1
∆(λ) f =


gradf , R f. (140)
(N − 1)2

Lemma 2.
If λij is symmetric, then
∗ N N
!
(λ)
X X (λ) 1 X
∆ f = (N − 1) λij ∇i ∇j f + (N − 1)Ak − λik Ri (∇k f ),
N − 1 i=1
i≤j k=1
(141)

29
where the asterisk on the summation means that when i < j the summands are
multiplied with 2 and when i = j with 1.

Proof.
Assume that λij is symmetric, then we can write
N
X
∆(λ) f = λij Di ∇j f =
i,j=1

N
X X
= λkk Dk ∇k f + λij (Di ∇j f + Dj ∇i f ) .
k=1 i<j

But
Ri ∇j f Rj ∇i f
Di ∇j f + Dj ∇i f = (N − 1)∇i ∇j f − + (N − 1)∇j ∇i f − =
N −1 N −1
1
= (N − 1) (∇i ∇j f + ∇j ∇i f ) − (Ri ∇j f + Rj ∇i f ) =
N −1
N
!
X 1
(N − 1) 2∇i ∇j f + ∇k f Qikj − (Ri ∇j f + Rj ∇i f ) .
N −1
k=1

Hence X
λij (Di ∇j f + Dj ∇i f ) =
i<j

X N
X X
= 2(N − 1) λij ∇i ∇j f + (N − 1) (∇k f ) λij Qikj −
i<j k=1 i<j

N N
1 X 1 X
− λik Ri ∇k f + λkk Rk ∇k f =
N −1 N −1
i,k=1 k=1

N N
X X (λ) 1 X
= 2(N − 1) λij ∇i ∇j f + (N − 1) Ak (∇k f ) − λik Ri ∇k f +
i<j
N −1
k=1 i,k=1

N
1 X
+ λkk Rk ∇k f.
N −1
k=1

Also
N N N
X X 1 X
λkk Dk ∇k f = (N − 1) λkk ∇2k f − λkk Rk (∇k f ).
N −1
k=1 k=1 k=1

Hence combining the above we get the first result.

30
Theorem 28.
If λij is symmetric, then
∗
X D E
∆(λ) f = (N − 1) λij (∇j f );i + ∆(λ) x, gradf , (142)
i≤j

where the asterisk in the sum means that if i < j, then the summands are
multiplied with 2. In case i = j with 1.

Proof.
From Lemma we have that if λij is symmetric, then
∗ N N
!
(λ)
X X (λ) 1 X
∆ f = (N − 1) λij ∇i ∇j f + (N − 1)Ak − λik Ri (∇k f ).
N − 1 i=1
i≤j k=1

But also from Theorem 22 we have

N N
1 X X D E
λik Ri = (N − 1) λij qikj − ∆(λ) x, ek
N − 1 i=1 i,j=1

Hence !
N N
X (λ) 1 X
(N − 1)Ak − λik Ri (∇k f ) =
N − 1 i=1
k=1
N X
X N
X N D
X E
= (N −1) λij Qikj ∇k f −(N −1) λij qikj ∇k f + ∆(λ) x, ek ∇k f =
k=1 i<j i,j,k=1 k=1
 
N
X X X X X
= (N − 1) ∇k f  λij qikj − λij qjki − λij qikj − λij qikj  −
k=1 i<j i<j i<j i>j

N
X D E
− λii qiki ∇k f + ∆(λ) x, gradf =
k,i=1

N X
X N
X D E
= −2(N − 1) λij qjki ∇k f − λii qiki ∇k f + ∆(λ) x, gradf .
k=1 i<j k,i=1

Hence
X N
X X
∆(λ) f = 2(N −1) λij ∇i ∇j f +(N −1) λii ∇2i f −2(N −1) λij qjki ∇k f −
i<j i=1 i<j

N
X D E
− λii qiki ∇k f + ∆(λ) x, gradf =
k,i=1

31
 ∗
X D E
= (N − 1) λij (∇j f );i + ∆(λ) x, gradf .
i≤j

Definition 11.
Assume that gij is the metric tensor of the surface S. Then

∆(g) x = 0 (143)

Iff
N
1 X (g)
(N − 1)gij Q∗ikj − kl Qilj = 0, (144)
N −1
l=1

where
N
(g)
X
kl := gik il . (145)
i=1

We call a space S, G−space iff ∆(g) (x) = 0.

Theorem 29.
If S is a G−space, then for all functions f , we have
∗
X
∆(g) f = (N − 1) gij (∇j f );i (146)
i≤j

Theorem 30.
If λij is any antisymmetric field, then
N N N
X 1 X (λ) f X (2)
Ak Dk f 2 =


λij [Di , ∇j ]f + λij Θij (R). (147)
i,j=1
f 2(N − 1) i,j=1
k=1

In case that S is S-R, then

N N
X 1 X (λ)
Ak D k f 2 .


λij [Di , ∇j ]f = − (148)
i,j=1
f
k=1

Proof.
We first evaluate the bracket.
1
[Di , ∇j ]f = Di ∇j f − ∇j Di f = (N − 1)∇i ∇j f − Ri ∇j f −
N −1
 
1
− (N − 1)∇j ∇i f − ∇j (Ri f ) =
N −1
1 1
(N − 1) (∇i ∇j f − ∇j ∇i f ) − Ri ∇j f + (f ∇j Ri + Ri ∇j f ) =
N −1 N −1

32
 N
X f
= −(N − 1) (∇k f )Qikj + ∇j Ri .
N −1
k=1

Hence if we multiply with λij and sum with respect to i, j, we get

N N N
X X f X
P = λij [Di , ∇j ]f = −(N − 1) λij ∇k f Qikj + λij ∇j Ri .
i,j=1
N − 1 i,j=1
k,i,j=1

Also we have from Theorem 7 and the antisymmetric property of λij :

N N N
X 1 X 1 X
λij ∇j Ri = λij (∇j Ri − ∇i Rj ) = − λij Rk Qjki +
i,j=1
2 i,j=1 2
i,j,k=1

N N N
1 X (2) 1 X 1 X (2)
+ λij Θij (R) = λij Rk Qikj + λij Θij (R).
2 i,j=1 2 2 i,j=1
i,j,k=1

Hence combining the above two results, we get

N N
X f X
P = −(N − 1) λij ∇k f Qikj + λij Rk Qikj +
2(N − 1)
k,i,j=1 i,j,k=1

N
f X (2)
+ λij Θij (R) ⇒
2(N − 1) i,j=1
N
Rk f 2
 
1 X
λij Qikj (N − 1)∇k f 2 −


fP = − +
2 N −1
k,i,j=1

N
f2 X (2)
+ λij Θij (R)
2(N − 1) i,j=1

and the results follow easily.

Corollary 11.
If λij is antisymmetric, then
N
X X (λ) f X
λij [Di , ∇j ]f = −(N − 1) Ak ∇ k f + λij ∇j Ri . (149)
i<j
N − 1 i<j
k=1

Theorem 31.
−
If λij = λ+
ij + λij , is any field, then

N N
X 1 X (λ− )
Π(λ) f := Hk Dk f 2 +


λij [Di , ∇j ] f = −
i,j=1
2f (N − 1)
k=1

33
 N N
f X (2) f X
+ λ− Θ (R) + λ + ∇j R i . (150)
2(N − 1) i,j=1 ij ij N − 1 i,j=1 ij

Corollary 12.
If λij is any field and D (f 2 ) = 0, then exists µ independed of f such

Π(λ) f = µf. (151)

In particular
N N
1 X (2) 1 X
µ= λ− Θ (R) + λ+ ∇j Ri (152)
2(N − 1) i,j=1 ij ij N − 1 i,j=1 ij

Theorem 32.
If λij is symmetric, then Π(λ) is simplified considerably
N
f X
Π(λ) f = λij ∇j Ri . (153)
N − 1 i,j=1

Theorem 33.
If λij is any symmetric field, then for every function f we have

Π(λ) f = 0, ∀f (154)

iff
N
X
λij ∇j Ri = 0. (155)
i,j=1

Remark 6.
1) In any space S we define the quantity
N
1 X
η (λ) := λij ∇j Ri . (156)
N − 1 i,j=1

2) If λij is symmetric, then

Π(λ) f = η (λ) f. (157)
(g) (g)
3) If λij = gij (the metric tensor), then η := η, Π := Π and
N
X
Πf := gij [Di , ∇j ]f = ηf, (158)
i,j=1

where
N
1 X
η= gij ∇j Ri . (159)
N − 1 i,j=1

34
Theorem 34.
(S) 1
If λij is any field and λij = 2 (λij + λji ), then
∗ D E
(S)
X
∆(λ) f = (N − 1) λij (∇j f );i + ∆(λ) x, gradf . (160)
i≤j

Proof.
(S) (A) (S)
Write λij = λij + λij , where λij = 12 (λij + λji ) is the symmetric part of λij
(A)
and λij = 12 (λij − λji ) is the antisymmetric part of λij . Then we have

N N
(S) (A)
X X
(λ)
∆ f= λij Di ∇j f + λij Di ∇j f =
i,j=1 i,j=1

N N X
(S) (S) (S)
X X X
= 2(N −1) λij ∇i ∇j f +(N −1) λii (∇i f );i −2(N −1) λij qjki ∇k f +
i<j i=1 k=1 i<j

N N
X (S) 1 X (S)
+(N − 1) λij qikj ∇k f − λik Ri ∇k f −
N −1
k,i,j=1 i,k=1

N N
X (A) 1 X (A)
−(N − 1) Ak ∇ k f − λik Ri ∇k f =
N −1
k=1 i,k=1

N
(S) (S)
X X
= 2(N − 1) λij ∇i ∇j f + (N − 1) λii (∇i f );i −
i<j i=1

N X
X N
X
−(N − 1) λij (qikj − qjki ) ∇k f + (N − 1) λii qiki ∇k f −
k=1 i>j k,i=1

N
1 X
− λik Ri ∇k f =
N −1
k,i=1

N
(S) (S)
X X
= 2(N − 1) λij ∇i ∇j f + (N − 1) λii (∇i f );i −
i<j i=1

N X
X N
X
−(N − 1) λij (qikj − qjki ) ∇k f + (N − 1) λii qiki ∇k f +
k=1 i>j k,i=1

D E N
X
+ ∆(λ) x, gradf − (N − 1) λij qjki ∇k f =
k,i,j=1

35
 N
(S) (S)
X X
= 2(N − 1) λij ∇i ∇j f + (N − 1) λii (∇i f );i −
i<j i=1

N X
X D E
−(N − 1) λij (qikj − qjki ) ∇k f + ∆(λ) x, gradf −
k=1 i>j

N X
X N X
X
−(N − 1) λij qjki ∇k f − (N − 1) λij qjki ∇k f =
k=1 i<j k=1 i>j

N
(S) (S)
X X
= 2(N − 1) λij ∇i ∇j f + (N − 1) λii (∇i f );i −
i<j i=1

N X
X D E
−(N − 1) λji qjki ∇k f + ∆(λ) x, gradf −
k=1 i<j

N X
X
−(N − 1) λij qjki ∇k f =
k=1 i<j

N
(S) (S)
X X
= 2(N − 1) λij ∇i ∇j f + (N − 1) λii (∇i f );i −
i<j i=1

N X D E
(S)
X
−2(N − 1) λij qjki ∇k f + ∆(λ) x, gradf =
k=1 i<j
∗ D E
(S)
X
= (N − 1) λij (∇j f );i + ∆(λ) x, gradf .
i≤j

Definition 12.
We call mean curvature vector of the surface S the vector
N
X
H= Hk ek , (161)
k=1

where
N
X
Hk = (N − 1) gij qikj . (162)
i,j=1

Theorem 35.
In case λij = gij , then we have
N
X D E
∆(g) f = (N − 1) gij ∇i ∇j f + ∆(g) x − H, gradf . (163)
i,j=1

36
Proof.
We know that (∇j f );i = (∇i f );j and λij = gij is symmetric. Hence from
Theorem 32 we have
X N
X D E
∆(g) f = 2(N − 1) gij (∇j f );i + (N − 1) gii (∇i f );i + ∆(g) x, gradf =
i<j i=1

N
X N
X N
X
= (N − 1) gij (∇j f );i + (N − 1) gij (∇j f );i + (N − 1) gii (∇i f );i +
i<j i>j i=1
D E
+ ∆(g) x, gradf =
N
X D E
= (N − 1) gij (∇j f );i + ∆(g) x, gradf =
i,j=1

N
X N
X N
X D E
= (N − 1) gij ∇i ∇j f − (N − 1) gij qjki ∇k f + ∆(g) x, gradf =
i,j=1 i,j=1 k=1

N
X N
X D E
= (N − 1) gij ∇i ∇j f − Hk ∇k f + ∆(g) x, gradf =
i,j=1 k=1

N
X D E
= (N − 1) gij ∇i ∇j f + ∆(g) x − H, gradf .
i,j=1

and the theorem is proved.

Note 7.
Actually the above theorem can be generalized for any λij as
N D E
X (S) (S)
∆(λ) f = (N − 1) λij ∇i ∇j f + ∆(λ) x − H , gradf , (164)
i,j=1

(S) λij +λji

where λij = 2 ,

N
(S) (S)
X
Hk = (N − 1) λij qjki (165)
i,j=1

and
N
(S) X (S)
H = Hk e k . (166)
k=1

Definition 13.

37
We can also expand the definition of Beltrami operator acting to vectors. This
can be done as follows:
If A = A1 1 + A2 2 + . . . + AN N , where {i }i=1,2,...,N is ”constant” orthonormal
base of E = RN , then

∆2 (A) = ∆2 (A1 )1 + ∆2 (A2 )2 + . . . + ∆2 (AN )N (167)

Theorem 36.
If V 1 and V 2 are vector fields, then
N
X





∆2 V 1 , V 2 = ∆2 V 1 , V 2 + V 1 , ∆2 V 2 + 2(N − 1) ∇k V 1 , ∇k V 2 (168)
k=1

Proof.
The proof follows by direct use of Theorem 7 and the identity




∇k V 1 , V 2 = ∇k V 1 , V 2 + V 1 , ∇k V 2 . (169)

Corollary 13.
N
X
2
h∆2 (ek ) , ek i = −(N − 1) qkjl = invariant. (170)
j,l=1

Proof.
From Theorem 36 and
hek , ek i = 1,
we have
N
X 2
2 h∆2 (ek ) , ek i + 2(N − 1) |∇l (ek )| = 0. (171)
l=1

Hence we get the result.

From Theorem 15 and Proposition 1 one can easily see that

N N
!
X X 1
∆2 (x) = (N − 1) qlkl − Rk ek . (172)
N −1
k=1 l=1

Hence if we define as ”2-mean curvature vector” the (not to confused with hk

of (eq1),(eq2) in pg.16):
XN
h= hk , (173)
k=1

38
where
N
X
hk := (N − 1) qlkl , (174)
l=1
then
1
h∗k = h∆2 (x), ek i = hk − Rk (175)
N −1
and
N
X
∆2 (x) = h∗k ek . (176)
k=1
Also we can write
1
∆2 x = h − R
N −1
and
D E D E N
X
∆2 x, grad(f ) = H, grad(f ) + ∆2 f − (N − 1) ∇2k f,
k=1

Proposition 6. PN PN 2
The quantities hk = (N − 1) i=1 qiki and hij = k=1 qikj are invariants.

Proof.

It follows from invariant property of ∇l Y , ek for all Y .

Proposition 7.
For the mean curvature holds the following relation
1
hk = Rk + h∆2 (x), ek i = invariant. (177)
N −1

Definition 14.
We call S 2-minimal if hk = 0, ∀k = 1, 2, . . . , N .

Theorem 37. PN
The invariants Rj∗ = (N − 1) i=1 κ2ij have interesting properties. The sum
PN
R∗ = j=1 Rj∗ is also invariant and

2 Rk∗
|∇k (eN )| = (178)
N −1
and
h∆2 (eN ), eN i = −R∗ . (179)

Proof.
We have
N
X
2
h∇l ek , ∇l ek i = qksl ,
s=1

39
which gives (178).
From relations (170) and (178) we get (179).

Definition 15.
We call the R∗ as R∗ −curvature.

The extremely case R∗ = 0 happens if and only if κij = 0, for all i, j =

1, 2, . . . , N . This leads from relation (10) to ωiN = 0, for all i = 1, 2, . . . , N
which means that III = 0 and hence eN is constant vector. We call such space
flat.

Application 1.
In the case x = eN , then S is a hypersphere and we have:

∇
e k (eN ) = ek , k = 1, 2, . . . , N. (180)

Also
N
(Σ)
X
∇k (eN ) = ek = qN jk ej .
j=1

Hence
(Σ)
qmN k = −δmk (181)
(Σ)
HN = −N (N − 1) (182)
From Theorem 36 we have

0 = ∆2 heN , eN i = 2 h∆2 eN , eN i + 2(N − 1)N

Hence
(Σ) ∗ 1 (Σ)
HN = h∆2 eN , eN i = −N (N − 1) = −N (N − 1) − R (183)
N −1 N
Hence
(Σ)
RN = 0, (184)
(Σ) ∗
R = N (N − 1) (185)
and
(Σ) ∗
Rk = N − 1, k = 1, 2, . . . , N (186)
(Σ) N
K = (−1) (187)
From
N
X N
X
RiN ml = − rotlm (qisl qN sm ) = rotlm (qisl qsN m ) =
s=1 s=1
N
X N
X N
X
=− rotlm (qisl δsm − qism δsl ) = − qisl δsm + qism δsl =
s=1 s=1 s=1

40
 = −qiml + qilm = qmil − qlim = Qmil = −Qlim . (188)

Theorem 38.
In general
N
X 1

∇2k f −

∆2 f = (N − 1) R, gradf .
N −1
k=1
1) If S is 2-minimal, then
1
∆2 x = − R.
N −1
2) If ti is any vector field, then easily in general
N
X N
X N
X
(ti );i = ∇k t k − tk hk
i=1 k=1 k=1

and if S is 2-minimal, then

N
X N
X
(tk );k = ∇k t k .
k=1 k=1

Note 8.
Assume that (with Einstein’s notation)
∂2Φ
 
a ∂Φ
∆2 Φ := g lm − Γ lm . (a)
∂xl ∂xm ∂xa
Then
hk = h∆2 (x), ek i = g lm blk;m . (b)
Hence hk is invariant. We call hk mean curvature tensor.

Proof.
We know that  
∂x
blk = , ek and ∂l ek = Γalk ea
∂xl
We differentiate the above first of two identities with respect to xm and we
have
∂blk
2
2
x, ek + Γamk h∂l x, ea i .
 
m
= ∂lm x, ek + h∂l x, ∂m ek i = ∂lm
∂x
Hence
2
∂lm x, ek = ∂m blk − Γamk bla


But
hk = h∆2 (x), ek i = g lm ∂lm

2
x, ek − g lm Γalm h∂a x, ek i =


g lm (∂m blk − Γamk bla ) − g lm Γalm bak =

= g lm (∂m blk − Γamk bla − Γalm bak ) = g lm blk;m .

41
2 The Spherical Form
Consider also the Pfaf derivatives of a function f with respect to the form ωN m .
It holds
XN
df = (∇e k f )ωN k . (189)
k=1
1
Assume the connections qmN j such that

N
X
1
ωm = qmN j ωN j . (190)
j=1

Then from (10) we have

N
X
1
ωm = qmN j qN js ωs . (191)
s,j=1

Hence
N
X
1
qmN j qN js = δms , (192)
j=1

where δij is the usual Kronecker symbol. Using this (189) becomes
N N N N
!
X X X X
df = ∇k f
e qN ks ωs = (∇k f )qN ks ωs .
e (193)
k=1 s=1 s=1 k=1

Hence
N
X
∇s f = (∇
e k f )qN ks (194)
k=1

and using (192)

N
X
1
∇
e lf = ∇s f qsN l. (195)
s=1

We set qemjl to be the connection

N
X
∇
e l (em ) = qemjl ej . (196)
j=1

One can easily see that

N
X
1
qemjl = qmjs qsN l. (197)
s=1
PN D E
From deN = ωN k ek we get ∇
e k (eN ) = ek and ∇e 2 (eN ), eN = qekN k
k=1 k

42
 Also if w = hx, eN i, then
* N
+
X
dw = hdx, eN i + hx, deN i = x, ∇
e k (eN )ωN k =
k=1

N D
X E N
X
= x, ∇
e k (eN ) ωN k = hx, ek i ωN k .
k=1 k=1

Hence
e k w = hx, ek i .
∇ (198)
Also
N N N N
!
X X X X
1 1
dx = (∇k x)ωk = ek qkN j ωN j = qkN j ek ωN j .
k=1 k,j=1 j=1 k=1

From this we get

N
X
1 1
∇
e jx = qkN j ek , qkN j is invariant (199)
k=1

and D E D E
∇
e l∇ e m (w) = ∇e l x, em + x, ∇ e l em =
*N + * N + N
X X X
1 1
= qkN l ek , em + x, qemjl ej = qmN l + hx, ej i qemjl
k=1 j=1 j=1

But
N N
X 1 X e
∆III
2 w = (N − 1)
e 2j w −
∇ (∇j w)R
ej =
j=1
N − 1 j=1

N N N
X
1
X 1 X
= (N − 1) qlN l + (N − 1) hx, ej i qeljl − hx, ej i R
ej (200)
N − 1 j=1
l=1 l,j=1

Also
N N
X 1 X e
∆III
2 eN = (N − 1)
e 2j eN −
∇ (∇j eN )R
ej =
j=1
N − 1 j=1

N N
X 1 Xe
= (N − 1) qeljl ej − Rj ej (201)
N − 1 j=1
l,j=1

Hence
N N

III  X 1 X
∆2 eN , x = (N − 1) hx, ej i qeljl − hx, ej i R
ej (202)
N − 1 j=1
l,j=1

43
From (200) and (202) we get
N
X
∆III III 1


2 hx, e N i − x, ∆ 2 e N = (N − 1) qlN l. (203)
l=1

From Theorem 36 and formula (197) we get

N D
X E
∆III

III 
III

2 hx, eN i = ∆2 x, eN + x, ∆2 eN + 2(N − 1) ∇e k x, ∇
e k eN
k=1

Or
N N
*N N
+
X X X X
1
∆III 1


(N − 1) qlN l = 2 x, eN + 2(N − 1) qjN k ej , qeN jk ej
l=1 k=1 j=1 j=1

Or
N
X N
X
1

III  1
(N − 1) qlN l = ∆2 x, eN + 2(N − 1) qjN kq
eN jk
l=1 k,j=1

But
 
N
X N
X N
X N
X XN
1 1 1 1 1
qkN jq
eN jk = qkN j qN js qsN k = qsN k
 qkN j qN js
=
k,j=1 j,k=1 s=1 k,s=1 j=1

N
X N
X
1 1
= qsN k δks = qkN k.
k,s=1 k=1

Hence

Theorem 39.
In S holds
N
X
∆III 1


2 x, e N = −(N − 1) qlN l (204)
l=1

3 General Forms and Invariants

Let C be a curve (one dimensional object) on S. Let also s be its physical
parameter of C. Then  
dx
t= (205)
ds C
is the tangent vector of C in P ∈ S. If n = eN = eN (P ) is the vertical vector of
S in every P ∈ S, then

t, n = 0 (206)

44
Definition 16.
We call vertical curvature of C the quantity
   
1 dt
= ,n (207)
ρ∗ C ds

Theorem 40.  
The vertical curvature ρ1∗ is an invariant (by the word invariant we mean
C
that remains unchanged in every acceptable transformation of the coordinates
ui ).

Proof.
This can be shown as follows. Derivate (206) to get
   
dt dn
, n + t, = 0.
ds ds
Hence   PN
1 dx dn II k=1 ωk ωN k
=− , =− = = invariant.
ρ∗
PN
ds ds I k=1 ωk2

Using relation (10), we get

N
1 X ωk ωm
∗
= κkm (208)
ρ ds ds
k,m=1

Assume in general that exist N curves Ci , i = 1, 2 . . . , N − 1 passing through

every point P of S and are vertical to each other in all P , then
*    +
dx dx
, = δij (209)
ds Ci ds Cj

Hence for these curves hold

  N  N
dx X ωk  X
= ek = λik ek , (210)
ds Ci ds Ci
k=1 k=1

where we have set ω 

k
= λik . (211)
ds Ci

Clearly from the orthogonality of Ci we have that

N
X
λik λjk = δij .
k=1

45
Hence the matrix M = (λik ) is orthogonal. Assuming these facts in (208), we
get
N   N N N N
X 1 X X X X
= κkm λ ik λ im = κkm δ km = κkk = invariant.
i=1
ρ∗ Ci i=1 k,m=1 k,m=1 k=1

From the above we get the next

Theorem 41.
In every space S there are N − 1 vertical curves Ci , i = 1, 2, . . . , N − 1, which
their directions satisfy
N −1  
X 1
= hN . (212)
i=1
ρ∗ Ci

Note 9.
1) We mention here that with the word invariant we mean any quantity that
remains unchanged in every acceptable choose of parameters {ui }i=1,2,...,N as
also change of position and rotation of S.  
2) Here eN is always vertical to S. That is why ρ1∗ = 0.
CN

Set
N
X N
X N
X N
X
T = Ai ωi2 + 2
Bij ωij + Cijk ωi ωjk + Eijf m ωij ωf m (213)
i=1 i,j=1 i,j,k=1 i,j,f,m=1

and assume the N orthogonal curves Ci . If the direction of Ci is

 
T
di = , (214)
(ds)2 Ci

then summing in all five directions we get, after using the orthogonality and
making simplifications
N
X N
X N
X N
X
2
di = Ai + Bij qijm + Cijk qjki +
i=1 i=1 i,j,m=1 i,j,k=1

N
X
+ Eijf mn qijn qf mn (215)
i,j,f,m,n=1

As application we get that

N
X N
X N
X
T0 = AI +BII +CIII = A ωi2 +B 2
ωN i +C ωi ωN i = invariant. (216)
i=1 i=1 i=1

46
Then summing the vertical directions we get
N
X −1 N
X N
X
2
di = A + B qiN j +C qiN i = invariant (217)
i=1 i,j=1 i=1

For every choice of costants A, B, C.

Definition 17.
Let now C be a surface curve and ti , i = 1, 2, . . . , N is a family of orthonormal
vectors of C. We define
    
1 dti
:= , tk , ∀i, k ∈ {1, 2, . . . , N } (218)
ρik C ds C
 
and call ρ1ik as ”ik−curvature” of C and
C

N  
dti X 1
= tk , ∀i = 1, 2, . . . , N. (219)
ds ρik C
k=1

Theorem
  42.
The ρ1ij −curvatures are semi-invariants of the space. Moreover
C
   
1 1
=− (220)
ρij C ρji C

and if
N
X
ti = Aki ek ,
k=1

then
  N X
N
1 X ω 
l
= A{m}i;l Amj , (221)
ρij C ds C
l=1 m=1
PN
where A{m}i;l = ∇l Ami − k=1 qmkl Aki .

Proof.
We express ti in the base ek and differentiate with respect to the canonical
parameter s of C, hence
N
X
ti = Aki ek ,
k=1

where Aki is orthogonal matrix i.e.

N
X
Aki Ali = δkl .
i=1

47
Hence
N
X
ej = ATkj tk , (222)
k=1

where AT is the symmetric of A.

Differentiating with respect to s of ti , we get
N N
dti X dAki X dek
= ek + Aki =
ds ds ds
k=1 k=1

N N N
X dAki X X ωl
= ek + Aki ∇ l ek =
ds ds
k=1 k=1 l=1
N N
X dAki X ωl
= ek + Aki qkml em =
ds ds
k=1 k,l,m=1

N N
X (∇l Aki )ωl X ωl
= ek + Aki qkml em =
ds ds
k,l=1 k,l,m=1

N N
X (∇l Ami )ωl X ωl
= em + Aki qkml em .
ds ds
m,l=1 k,l,m=1

Hence we can write

  XN N
! N
dti X ωl X ωl
cim := , em = ∇l Ami − qmkl Aki = A{m}i;l . (223)
ds ds ds
l=1 k=1 l=1

Hence
N ω  N ω 
l l
X X
cim = ∇l Ami − Aki qlkm . (224)
ds C ds C
l=1 l,k=1

Hence
  N
dAmi dti X ω 
l
= , em + Aki qlkm (225)
ds ds ds C
l,k=1

Also    

 dti dtj
ti , tj = δij ⇒ , tj + ti , = 0.
ds ds
Hence    
1 1
=− . (226)
ρij C ρji C

From the orthonormality of Aij and ATij = Aji we have

N
X N
X
Ami Aji = δmj ⇔ Aim Ajm = δij . (227)
i=1 m=1

48
Differentiating we get
N
X N
X
A{m}i;l Aji + Ami A{j}i;l = 0 (228)
i=1 i=1

Moreover
N N XN
dti X X ω 
l
= cim em = A{m}i;l Amj tj (229)
ds m=1 j=1
ds C
l,m=1

and the curvatures will be

  N X
N
1 X ω 
l
= A{m}i;l Amj . (230)
ρij C ds C
l=1 m=1

Hence
N   N N
X 1 X X ω 
l
Apj = A{m}i;l Amj Apj =
j=1
ρij C ds C
l=1 m,j=1

N X
N ω  N ω 
l l
X X
= A{m}i;l δmp = A{p}i;l = cip .
ds C ds C
l=1 m=1 l=1

This lead us to write

N   N N
X 1 X X ω 
l
Ani Apj = Ani cip = A{p}i;l Ani =
i,j=1
ρij C i=1
ds C
l,i=1

N ω  N ω 
l l
X X
= AT{i}p;l ATin = AT{m}p;l ATmn =
ds C ds C
l,i=1 l,m=1
 T
N ω   T
X l 1
= A{m}n;l Amp  = =
ds C ρnp C
l,m=1

N N
!T
ω 
l
X X
= ∇l A{m}n Amp − qmkl Akn Amp  =
ds C
l,m=1 k=1

N N
!
ω 
l
X X
= ∇l A{m}p Amn − qpkl Akp Amn =
ds C
l,m=1 k=1

N ω 
l
X
= A{m}p;l Amn
=
ds C
l,m=1
   
1 1
= =− .
ρpn C ρnp C

49
 Since we have
  N
X
AT{i}j = ∇l Aji − qjkl Aki
;l
k=1

and !T
N N

T X X
A{i}j;l = ∇l Aij − qikl Akj = ∇l Aji − qlkj Aki
k=1 k=1

and  
T
AT{i}j = A{i}j;l . (231)
;l

References

[1]: Nirmala Prakash. ”Differential Geometry An Integrated Approach”.

Tata McGraw-Hill Publishing Company Limited. New Delhi. 1981.

[2]: Bo-Yu Hou, Bo-Yuan Hou. ”Differential Geometry for Physicists”.

World Scientific. Singapore, New Jersey, London, Hong Kong. 1997.

[3]: N.K. Stephanidis. ”Differential Geometry”. Vol. I. Zitis Pub. Thessa-

loniki, Greece. 1995.

[4]: N.K. Stephanidis. ”Differential Geometry”. Vol. II. Zitis Pub. Thessa-
loniki, Greece. 1987.

[5]: D. Dimitropoulou-Psomopoulou. ”Calculus of Differential Forms”. 2nd

edition. Zitis Pub. Thessaloniki, Greece. 1993.

[6]: E.A. Iliopoulou, F. Gouli-Andreou. ”Introduction to Riemann Geome-

try”. Zitis Pub. Thessaloniki, Greece. 1985.

[7]: N.K. Spyrou. ”Introduction to the General Theory of Ralativity”.

Gartaganis Pub. Thessaloniki, Greece. 1989.

50

View publication stats