746 MA THEMA TICS: A. D. MICHAL PROC. N. A. S.

THE GROUP MANIFOLD OF FINITE CONTINUOUS POINT

AND FUNCTIONAL TRANSFORMATION GROUPS
By ARIsTomII D. MIcHAiL
D1PARTM1NT OF MATHZMATICS, THji OHIo STAT}3 UNIVERSITY
Communicated August 8, 1928
1. Introduction.-The differential geometry of the manifold of the r
parameters a', a2, ... ., a' of an r-parameter continuous group of trans-
formations
si=si(X1p X2, . .., X"; a', a2,. .., at) (i= 1,2 .*.,n
has recently been studied by Cartan' with the aid of ingenious geometric
methods peculiar to his way of looking at an affinely connected manifold.2
From the analytic and invariant-theoretic standpoint, however, Cartan's
geometric methods are by no means the last word on the subject. Further-
more, the proof of many of his theorems bring into play the variables
xl, x2, .. ., x". It is the purpose of this paper to develop some of the
essential results of Cartan's theory analytically in such a way as to make
the theory applicable to finite continuous groups of functional trans-
formations in function space as well as to point transformations in n-
dimensional space, and to exhibit various fundamental tensor invariants
of the manifold in an arbitrary co6rdinate system.
A simple example of an r-parameter group of functional transformations
is given by

y(x) = y(x) + / K(x,s; a', a2, . . a)y(s)ds

where
K(x,s; al, a2,.. ,ad) = ca'H (x,s)
i=1
+ d1af H (xu)H (ups)du + . +
and

f Hs(x,u)Hj(u,s)du Hj(x- )Hj(u,s)du

8 E C*jHk(x,s) (ij= l,2,. r)
k=1

What we have to say will apply equally well to finite continuous groups
G in any abstract space for which a continuous group theory is possible.
The central feature of all these group theories is found in their possession
of an r-dimensional manifold: the manifold of the r-essential parameters
VOL. 14, 1928 MA THEMA TICS: A.D. MICHAL 747

of the group. We shall call this manifold the group manifold of the group
G.
2. The Three Coefficients of Connection.-The allowable point trans-
formations in the group manifold are given by either of the two parameter-
groups R, and R2 of G
Ri: a' = f'(al, a2,. . .,a; bl, b2,. . .,bT) (i = 1, 2,. . .,r) (1)
R2: ai = f (b, b2,. . .,by; a', a2. . . a) (i = 1, 2,. . . ,r). (2)
The two parameter-groups R, and R2 essentially translate the fact that
the O r transformations in G actually form a group.
Let"
Uaf = X(a) a (3)
be the r infinitesimal generators4 of the parameter-group Ri.
The r vectors e(1), e(2), . . ., (r) are linearly independent with respect to
function multipliers since R, is a simple transitive group.5 Thus the
determinant
A\ = |a)I|i 0. (4)
If we let Al-) denote the cofactor of a in A divided by A, we shall
have obviously
(a) As -at(j)a - a (5)
where 6,i is zero or unity according as a 0 j3 or a =
Since Ai( is a covariant vector for each j3, it follows that the set of func-
tions Pt s defined by

Pia#aa (6)
are the components of an asymmetric linear connection.6 It is well
known that the infinitesimal generators UJ satisfy the identities4
(Ua, U )f = Ca U7f (7)
In other words the contravariant vectors ea) satisfy the relations
ra) - COay
a)a= Ca . (8)

The quantities C', in (8) are the structural constants of the group R,
and thus satisfy the identities
{ CO + C"O'i= 0
cw,c + C;acMcp + c#'W GOO = 0.(9
748 MA THEMA TICS: A. D. MICHAL IPRiOC. N. A. S.

Let
Vaf = 7(a) a (10)

be the infinitesimal generators of the second parameter-group R2. The

r vectors (a) are connected with the vectors ea) of R1 by the reciprocity
relations7
(Ua, VO)f = 0. (11)
An equivalent form of these conditions is given by

{)7(8) k
at(a) -
= (12)
If we let 4I stand for the determinant of the m (4a) for the co-
and
factor of t in 4 divided by 1', then the set of functions Q'# defined by
= 11 -y
1i a ( $_00W (13)
are the components of an asymmetric connection. Thus in the group
manifold there exist two asymmetric coefficients of connection:8 the first
corresponds to the parameter-group R1 and the second to the parameter-group
R2.
If we multiply both sides of (12) by dI4(A) and sum on a and j3, we
arrive at the relation
-Qa + PS= .
We are thus led to the following fundamental theorem.
Theorem 1. If Ps , is the coefficient of connection formed from the first
parameter-group R1, then Q =-POa is the coefficient of connection formed
from the second parameter-group R2.
The geometry of paths9 determined by
a= /2(poSO + P,@a) (14)
is thus the same for both parameter-groups R1 and R2.
3. Curvature Tensors of Asymmetric Connections and the Torsion
Tensor.-Let
Qt = 1/2(Piaf - Ppa). (15)
Thus it is clear that the torsion tensor'0 based on the asymmetric con-
nection Qi is equal to Co.
The torsion tensor l.m can be expressed in terms of the group struc-
tural constants C$, and the vectors (. In fact, on multiplying both
sides of (8) by 1/2A(a) &A) and summing on a and 3, we obtain the result
Ulm = 1/2Cap ey) Ala) -'.(6 (16)
Voi,. 14 1928 MA THEMA TICS: A. D. MICHAL '7A9

By a judicious choice of coordinate systems a number of theorems and

formulas can be established quite simply. For the purposes of this paper
it is convenient to have a coordinate system V' such that
==s and (P'O(d))d'0 = 0. (17)
A coordinate system for which the relations (17) are satisfied will be called
a canonical co6rdinate system.'"
Such a canonical coordinate system d' can be defined by the coordinate
transformation
a = q$ + (2ira))qaa - e(r1 )r,)daa (18)
For the sake of clarity we note that the point with coordinates a =
is the origin of the special coordinate system. Since the parameter-
group R, is simply transitive it follows that the Jacobian of the trans-
formation (18) does not vanish at the point a' = q' and hence (18) is a
reversible transformation in the neighborhood of that point.
It is a consequence of (17) that
(Ra=)o - (P.ia)o ; (Ak )o = ik- (19)
The symbol ( )o in (19) is an abbreviation for the operation whi'ch evalu-
ates the quantity within the parenthesis at the origin of a canonical co-
ordinate system. The omission of the bars above the quantities in the
parenthesis need not cause any confusion.
Theorem 2. The curvature tensors Piki and Q, of the parameter-groups.
R, and R2, respectively, vanish.
To prove the first part of this theorem we proceed as follows. By
means of (19) and (16) we see that
(Pia)o = 1/2CIP (20)
and hence

~_6790 2 up = ;2 (a)
\Ia 00I
-
2 CI- (21)
But from the form of the curvature tensor Pjk, we have
-p (aP;k ZbPJA
Evaluating this relation with the aid of (6) and (21) we get
(P. i)o= -14C:l CCk + 1/4 CaikjG. (22)
Consequently, =
(Pjskl)o = °-
750 MA THE MA TICS: A. D. MICHAL PROC. N. A. S.

Since Pj,k is a tensor, we obtain by a familiar reasoning the result

Pj;k = 0.
By calculation we have

(1 ak = -Ck Cia + - Cj + CL CS .
a
(23)
This relation is obtained by first substituting the expressions for Pim
and Plj in terms of Pjk and P1I, respectively, by means of (16) and then
by employing the results in (22) and (21). Finally, by (20) and the form
of the curvature tensor Q1kl we obtain
(QJk0) = Ckj Cla + Clk C1a + Cjl Cka
and hence without difficulty
QIkl = O.

Theorem 3. All three covariant derivatives of the torsion tensor QJk based
on the affine connections P6a, Q,cp r6c, respectively, are zero.
On differentiating the equation (16) with respect to a' in a system of
canonical coordinates and evaluating the resultant expression at the origin
of this coordinate system we obtain the relation

117o 4Cjk Cal +-Cak

4 Cjl +
4
Cja Ckal-

Taking cognizance of the fact that (rP,)o = 0, it is easy to arrive at the

result
Q2k;l = 0,
where d4k;j denotes the covariant derivative of &23k based on the sym-
metric connection rc. On the basis of these results and the fact that
(P'p)o= 1/2C!.p, one can show that the covariant derivatives of Q2k
based on the asymmetric connections Pt, and Q4 , also vanish.
4. The Curvature Tensor of the Symmetric Connection.-Let us now
prove the following theorem.
Theorem 4. The covariant derivative Bkklm of the curvature tensor B.k1
based on the symmetric affine connection 1',, vanishes.
To prove this theorem we first observe that

(Bjkl)0 =I=1rilk br;1'

With the aid of (22), (23) and the identities (9) satisfied by the structural
constants Cap, we are led to the formula
(Bjkl)0 = 1/4Caj Cl ((24)
VOL. 14, 1928 MA THEMA TICS: A.D. MICHAL 751

From the transformation (18) we see that

ak= (k) when a' = q

ba-i ~ ~ **(25)
ak= Ak) when a = q.

The tensor law of transformation of BI,, coupled with the results (24),
(25) and the fact that the origin of canonical coordinates is arbitrary
leads us to the following expression for the curvature tensor in an arbi-
trary coordinate system
BJa= 1/4CPa C'g7 (r 4a)4jca 4$ ) (26)
On differentiating equations (26) with respect to am when the equations
are referred to a system of canonical coordinates a' and evaluating the
resultant expression at the origin of this system, we obtain by virtue of
(21) and the identities (9) the fact that
(8~~?BikI -o 0
bam
The truth of theorem 4 is now obvious.
5. Replacement Theorem.-The purpose of this paragraph is to prove
the theorem.
Theorem 5. Any tensor differential invariant (relative or absolute)

( )ai bajbaa*
Y

of the group vectors a can by a mere replaceiment of the arguments be put

into the form
where
T,': r, W(.() a e(a)j a e,a3jk)
and
,a)j =
I/2C"a e(,) A') (27)
(a)jk - 1/12(C aX C ry + C Io A, Ak(yi) (28)
To prove this theorem we first need to show that the first tensor ex-
tension ra)j (covariant derivative based on rsa) of a is given by formula
(27). In a canonical coordinate system

(f ) =2
Since (a)j is a tensor such that
(W(a)j)o =(
752 MA THEMA TICS: A.D. MICHAL PROC. N. A. S.

we can arrive at formula (27) by making use of an argument that was

employed in deriving expression (26) for the curvature tensor Bikz.
We next need to show that the second tensor extension ea)jk (not the
second covariant derivative) of a is given by (28). By Veblen's normal
co6rdinatel2 methods it is not difficult to show that
_____l 1 br
~2k\++
((a)jk = 3+ Ca) + a' + * (29)

where the symbol * stands for terms that vanish with ri Now

j= -2 (A a + bA
and hence by calculation we find
(W a)V ( 1 62(2j)
akJ = 2 YJ)ad/o -
2 Va?ako)
+ -C.k
8 C'oj + -CAk
8 C.ya* (30)
Consequently,
F
(?r~++
LC(a) k aj+ +
PA
baaC)J
~~)~~.
- (,a ak + bakOaa
+ ~2
ataJ
(3)1)
o

Now recalling that

(Wa=jk)o -4 C^a Cjk,
i.e.,
/r.j brak 1
(ak - aC /0 = 4 CaaCjIk
we obtain the relation
.-1
(2aa§))
_

~a~baj
_)=
(ba2&a
0
)O
ac?bak +-2
_

2
C'a
_j
Cjfk + 4 Caj Cak+-C4
o
1
" cc
-
ak C's j. (32)
On expressing (b2.j) 1bak6aa)o and (82Xk) /bacbai)o in terms of
(620it)/baiJak)o means of (32), relations (29) and (31) yield without
by
difficulty the following result

(ea)jk)o = -121 (Caj Ci, k

+ CQk C;7). (33)
Formula (28) is an immediate consequence of this result on the basis of
the method used in establishing formula (26).
VOL. 14, 1928 MATHEMATICS: A. D. MICHAL 753
The final step in the proof of theorem 5 consists in showing the validity
of the replacement process. Since the procedure is identical with that
used in the various replacement theorems of the geometry of paths,'3
we spare the reader the calculations.
6. Integral Invariants.14-The following theorem is of importance in
the calculation of invariants of continuous groups by integrations in the
group manifold.
Theorem 6. The functionals

IA' dalda2... da
Af (34)
and
Jb| i) |dal da2...dar (35)

are integral invariants of the parameter-groups R2 and R1, respectively.

A necessary and sufficient condition that (34) be an integral invariant
of the parameter-group R2 is that the determinant
system of partial differential equations
IA")
satisfy the

7(a) 0 (36)
Clearly

e(b)

= - |&I(A)) a(b) (b)

This result coupled with the relation (12) enables one to verify that the
I I
determinant A() is a solution of (36). Similar methods lead to the
I
result that the determinant ! s) is a solution of the system
f +f
a a
0.

1 E. Cartan, J. Math. p. Appi., 6 (1927), pp. 1-119.

2 E. Cartan, Ann. Ecole Normal, 40 (1923), pp. 325-412.
' The repetition of an index in a term will be used to denote summation with respect
to this index over the integral values 1, 2, ..., r. We shall also understand that free
indices can take on any integral value 1, 2, .. ., r.
4 L. Bianchi, Lezioni Sulla Teoria Dei Gruppi Continui Finiti Di Trasformazioni,
1918.
' L. Bianchi, loc. cit., pp. 297-304.
6 L. P. Eisenhart, these PROca6DINGS, pp. 24-250.
7 L. Bianchi, loc. cit., pp. 270-272, p. 300.
754 MATHEMATICS: A. D. MICHAL
8 L. P. Eisenhart, "Non-Riemannian Geometry," Amer. Math. Soc. Colloq. Publ., 8
(1927), Chapter I, Paragraph 2.
0. Veblen and T. Y. Thomas, Trans. Amer. Math. Soc., 25 (1923), pp. 551-580.
10 P. Appell, Traite de Mecanique Rationnelle, Vol. V (1926), p. 169.
11 Such a system is clearly a special kind of geodesic coordinate system. We refrain
from using this terminology, however, on the ground that the coefficients of the linear
terms in (18) are not the Kronecker deltas.
12 Q. Veblen and T. Y. Thomas, loc. cit., pp. 569-573. For a delightful introduction
to the general subject of tensor analysis the reader is referred to 0. Veblen, Invariants
of Quadratic Differential Forms, 1927.
la T. Y. Thomas and A. D. Michal, Annals Math., 28 (1927), p. 199.
14 A. D. Michal, Trans. Amer. Math. Soc., 29 (1927), pp. 612-646.