The Equation F(x) + M( x)G(l/x)
= 0
and Homogeneous Biadditive Forms
C. T.zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
Ng* zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
Department
of Pure Mathematics
University
of W aterloo
Waterloo,
Ontario,
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
Canada N2L 3Gl
Submitted by Robert Guralnick
ABSTRACT
We determine all additive F, G and multiplicative M satisfying the functional
equation F(r) + M(x)G(l/ r)
= 0 on a commutative field k of characteristic + 2.
Let U, V be k-vector spaces. A functional (form) 7’: U + k is functionally homogeneous if, for some scalar function M: zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONM
k + k, T(h) = M(A)T(u) for all X E k, u E U.
A simple application of the equation in the title leads to a complete description of all
biadditive forms T: V X V+ k which are functionally homogeneous. Some results of
Baker, Gleason, and Kurepa concerning the Halperin problem on quadratic forms are
generalized and unified.
1.
INTRODUCTION
Let k be a field. We assume, once and for all, that k is commutative
characteristic
all x, y E k, and is multiplicative
purpose
with
f 2. A map zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
f:k + k is additive if f(x + y) = f(x)+ f(y) for
if f(xy)
= f(x)f(y)
for all x, y E k. The
of this paper is to determine all additive F, G and multiplicative
on the field k satisfying the functional
F(r)+
for all x E k * = k \ (0).
M
equation
M (x)G(l/x)
=0
(FE)
Since the value of M at x = 0 plays no role in (FE),
*Supported by NSERC of Canada Grant A 8212.
LINEAR
ALGEBRA
AND ITS APPLICATIONS
93:2.X%279 (1987)
255
0 Elsevier Science PublishingCo., Inc., 1987
52 V anderbilt Ave., New York, NY 10017
00243795/87/$3.50
256 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
C. zyxwvutsrqponm
T. zyxwvutsrqpon
NG
and since every map M: zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
k * + k with the multiplicativity M( xy) = M(x)M(y)
for all X, y E k * can be extended to a multiplicative map on k by defining
M(0) = 0, there is no loss of generality in assuming
M(0) = 0.
(1.0.1)
In his work on quadratic functionals, Kurepa [14] came across the
functional equation (FE) with M(x) = x2 on the reals [w
. Through this
equation he obtained the general form of functionals Q on R-vector spaces
satisfying the parallelogram law and the homogeneity Q(Ax) = X20(x), thus
answering a question raised by Israel Halperin in 1963 in Paris. With this
result over R he further solved [15] the problem on vector spaces over the
complex C or the field of quatemions under the homogeneity Q(Ax) =
1X]2Q(~). VrbovH [ 171 managed to solve (FE) with F = G and M(X) = ]x12 on
C and gave a new proof for the result of Kurepa on complex spaces. Baker [6]
enriched Kurepa’s result on complex spaces and solved the problem under
the homogeneity Q( Ax) = x” Q(x), or Q( hx) = x20(x). Gleason [9] determined all functionals Q on k-vector spaces satisfying the parallelogram law
and the homogeneity Q(Xx) = x” Q(x). D avison [8] extended the result of
Gleason to modules over a ring under appropriate homogeneity on Q, which
incidently reduces to Q(Ax) = X2Q(x) when the ring happens to be k. Here
we generalize the result of Baker, Gleason, and Kurepa not by broadening the
scope of k, but by allowing a very general notion of homogeneity on Q. To be
precise, we lay down some definitions and terminology.
Let U, V be vector spaces over the field k (commutative and is of
characteristic
# 2). A biadditive zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPON
form T on V is a map of V X V into k
which is additive in each of its two variables. There is a one to one
correspondence between sy mmetric biadditive forms S on V and functionals
Q : V + k satisfying the parallelogram law Q(x + y ) + Q(x - y ) = 2Q(x) +
2Q(y), which is provided by Q(x) = S(x,x) and 4S(x,y) = Q(x+ y) - Q(x - y)
[5, 111. We refer to Q as the diagonal of S, and S as the sy mmetric
polarization of Q. The diagonal of a biadditive form T always satisfies the
parallelogram law, and T is referred to as a polarization of its diagonal. A
functional f: U + k is (functionally ) homogeneous if, for some scalar function M: k + k, f(Au) = M(X)f(u) for all A E k, u E U, and we shall say that
f is M-homogeneous [lo]. A functional Q: V + k satisfying the parallelogram
law is M-homogeneous if, and only if, its symmetric polarization S on
U = V zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
x V is M-homogeneous. Thus the study of (functionally) homogeneous
Q satisfying the parallelogram law is reduced to the study of (functionahy)
homogeneous symmetric biadditive forms. A functional f: V X V + k is
(functionully )
bihomogeneous if there exist a pair of functions M ,, M ,: k + k
HOMOGENEOUS
257
BIADDITIVE FORMS
such that f(Xx, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFE
py ) = M ,(h)M ,(p)f(x,y )
for all X, p E k, x,y E V. If so, we
say that f is (M,, Ma>bihomogeneous.
Obviously, zyxwvutsrqponmlkjihgfedcbaZY
(M,, Ma)-bihomogeneity
implies M,M,-homogeneity.
A functional
f: V X V + zyxwvutsrqponmlkjihgfedcbaZYX
k is homogeneous
if,
and only if, both its symmetric part and its skew-symmetric
part are. Hence
the symmetric part and skew-symmetric
part of a bihomogeneous
f: V X V
+ k are both homogeneous. Generalizing the Halperin problem, we may ask
the following question:
For given M: k + k, and k-vector space V, is every
symmetric
M-homogeneous
biadditive T: V X V + k
the symmetric
part of some bihomogeneous
biadditive form?
(1.0.2)
We shall prove that M of interest must be one of the following three
exclusive types: M = C#B~
where $J is a morphism on k, M = $xj where C#B
and
# are distinct morphisms on k, or M = $6 where + is a nontrivial embedding
of k into a quadratic field extension k(G). The structure of M-homogeneous
biadditive forms will be made clear so that many questions concerning them
can be readily answered.
Aczel observed the presence of real derivations
in his work on the
fundamental
equation of information
and its generalization
[2,3]. It is not
surprising
that (FE) also finds its application
in information
theory [16].
There is much interesting reading material in [l].
2.
ON THE EQUATION
(FE) WITH
F = G
LEMMA 2.1. Let additive A z 0 and multiplicative M on k satisfy the
equation
A(x)+
M(x)A(l/x)
= 0
(2.1.1)
for all x # 0 in k. Then the following hold for all x, y E k:
M(x)A(x-‘y)+
M(1) = 1,
A(xy)
M(y)A(y-lx)
M(-r)=M(r),
=B(x)A(y)+A(x)B(y),
=O,
and
x, y # 0,
A(l)=O,
(2.1.1.E)
(2.1.2)
(2.1.3)
258 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
C. zyxwvutsrqponml
T. NG zyxwvutsr
where
B(x)
= 2-‘[1+
M (x)
- M (l-
X)])
(2.1.4)
B is additive,
B(xY)
=
B(+(Y)+
B(x)
cAb)A(y)
forsome
constant
- M (x)B(l/x)
= 0,
x # 0,
M(x) =
Proof.
(2.1.1.E)
W ith
(2.1.5)
a multiplicative
B’(X)
c E k,
zyxwvutsrq
(2.1.6)
(2.1.7)
-CA’(X).
(2.1.8)
M, the equivalence
between
(2.1.1)
and
is evident.
Since A # 0 in (2.1.1),
the multiplicative
map
M cannot
be identically
zero on k *. This implies M(1) = 1. Replacing
x by - x in (2.1.1) and use the
fact that A is odd, we obtain M( - x)A(l/ x)
= M(x)A(l/ x).
By fixing
x = r0 # 0 with A(l/ x,) # 0, we get M( -x0) = M(x,).
Since M is multiplicative, M( - x) = M( - x,)M(x/ x,)
= M(r,)M(x/ x,)
= M(x)
for all r
E k. Putting in (2.1.1) x = 1, we get 2A(l) = 0; and as char(k) f 2, A(1) = 0
follows. This proves (2.1.2).
Consider the simple algebraic identity
(l-n:)-l(l-y)~‘-(l-x)~‘=(l-x)-‘xy(l-y)~’+Y(l-Y)
l>
(2.1.9)
which holds for all x f 1, y f 1 in k, and apply to it the additive map A term
by term to get
A[(l-r)-‘(l-y)-‘]
-A[(l-x)-r]
=A[(l-x)-‘xy(l-y)-‘]+A[&-Y)-r]
(2.1.10)
for all x # 1, y # 1. We multiply (2.1.10) by M ((lx)(1 - y)); use the
additivity of A, the multiplicativity of M, and Equations (2.1.1.E) and
HOMOGENEOUS
(2.1.2);
259 zyxwvutsrqp
BIADDITIVE FORMS
and carry out the following sequence of computations:
- A[(1 - x)(1 - y)] + M(l=-
y)A(l
-x)
M(xy)A[(l-x)x-‘y-‘(l-y)]
-M((l-~)Y)A[Y-‘(l-y)],
A(x)+A(y)-A-M(l-y)A(x)
= -&+y)A(x-‘y-‘-x-‘-y-‘+l)-M((l-x)y)A(yP’-1)
= - M(xy)A(r-‘yP’)+M(y)i+)A(x-‘)
+ zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCB
M (x)M (y )A(y - ‘)
- M (l+W Y)A(Y- ‘),
A(x)+
A(Y) - A
= A(ry )
- M (l-
- M (y )A(r)
y )A(x)
- M (r)A(y )+
M (l-
~)A(Y).
Solving for the A(xy) term from the last equation, and using the assumption
char(k) # 2, we get (2.1.3) for all x, y # O,l, where B is defined by (2.1.4).
Equation (2.1.3) also holds when x = 0,l or y = 0,l because of (l.O.l), which
implies B(0) = 0 and B(1) = 1.
The additivity (2.1.5) of B follows from that of A f 0 in (2.1.3). We
compute A(xyz) first as A((xy)z) and then as A(x(yz)) using (2.1.3), and
compare the results. This comparison and A # 0 lead to (2.1.6) [12].
We verify (2.1.7) with the following computations:
M(x)B(x-‘)=2-‘M(x)[l+M(x&-M(l-x-l)]
=2P[M(x)+M(xx-‘)44(x(1-r-‘))]
=2-r[M(x)+l-M(x-l)]
=2-94(x)+144(1-x)]
=B(x).
The following computations use (2.1.1), (2.1.3), (2.1.6), and (2.1.7):
M(x)A(y)
= M(x)A(r+(xy))
= - A(x)[B(x)B(y )+
= M(x)A(x-l)B(xy)+
cA(x)A(y )l
+ B(~)[A(~)B(Y)+B(~)A(Y)~
= [B’(X)- cA2(x)]A(y ).
M (r)B(x- ‘)A(xy )
C. zyxwvutsrqponml
T. zyxwvutsrqpon
NG
260
Then the final elimination of the factor A(y) yields (2.1.8)
Equation (2.1.8) holds when 1c= 0 because of (1.0.1).
for x # 0.
n
THEOREM 2.2. Let additive A # 0 and multiplicative
M on k satisfy the
functional equution (2.1.1) A(r)+ M(x)A(l/ x)
= 0 for all x # 0. Let c E k
be the unique constant in Lemma 2.1 such that (2.1.6) and (2.1.8) hold.
Then, based on the nature of c, we can describe A and M as follows:
Case (i).
Suppose
c = 0. Then for some (field)
morphism
+: k + k,
+ + 0,
A is a nontrivial
Here, a @derivation
+derivation
is an additive
(2.2.1.a)
and M = $I~.
map A satisfying
A(xy) = +(x)A( y) +
for all x, Y E k.
Case (ii).
Suppose c # 0 and c = a2 for some a E k. Then for some
morphisms +,$:k-,k,
$,I+Q#O, and +ZI/J, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSR
+(y)A(x)
A= (2a)-‘(+ -rC/)
Case (iii).
Suppose c # 0 and a = 6
ding +: k -+ k(a), (Pz &
A=(2a)-‘(c)--s)
Here the conjugate
x+ya=r-yaforallx,yEk.
operation
and
(2.2.1.b)
M=+$.
4 k. Then for some (field)
and
on the field
(2.2.l.c)
M=+s.
extension
embed-
k(a)
is defined
by
Conversely, if A and M are given by (2.2.l.a), (2.2.l.b), or (2.2.1.~) for
some constant c E k, then A # 0 is additive, M is multiplicative, and (2.1.1) is
satisfied.
Proof.
Let additive A # 0 and multiplicative M satisfy (2.1.1), and let
c E k be as given in Lemma 2.1. There are three cases to consider (cf. [7]).
Case (i): Suppose c = 0. Then (2.1.6) reduces to the multiplicativity of B.
Thus $I = I3 is a field morphism on k. From (2.1.3) where A f 0, we get
+ # 0 and that A is a @derivation. Equation (2.1.8) reduces to M = G2 as
claimed.
Case (ii): Suppose c # 0, and c = a2 with a E k. Then a f 0. Equations
(2.1.3) and (2.1.6) correspond to the multiplicativity of $ = B + uA and
$ = B - aA. Since B and A are additive, so are + and I/ J.Hence 9, and #
HOMOGENEOUS
281 zyxwvutsrq
BIADDITIVE FORMS
are field morphisms on k. Since a, A # 0 and char(k) # 2, we have $J# 4.
From the definition of + and I/J,we get immediately A = (2a)-l(+
- JI) and
B = 2-‘($I + JI). Knowing both B and A in terms of + and J/, Equation
(2.1.8) then delivers M = +#.
Case (iii): This case is very much like (ii). W e also define + = B + UA and
4 = B - aA, and observe that 4 = 6, the conjugate of +.
The converse is easy to verify.
COROLLARY 2.3.
x z 0 with additive
A = D,
of
The general solution
A(r)+
zyxwvutsrqponmlkjihgfedcbaZYXW
M (x)A(l/x)
= 0 for all
A # 0 and multiplicative M on the reaZs R is given by
a nontrivial real derivation,
and M(x)
= x2,
(2.3.1.a)
or
A=a-‘Im$
and
(2.3.1.b)
M(x)=~+(x)~~,
where u # 0 is a real constant, + : R -+ C is a nontrivial embedding of the
reals into the complex numbers, Im $I is the imaginary part of $J, and 1.1
denotes the usual norm on Q=.
Proof.
Take Theorem 2.2 with k = R. The only (nonzero) morphism
(p: R + R is given by G(X) = x, the identity map. Thus (2.2.1.a) reduces to
(2.3.1.a). Case (ii) and (2.2.1.b) will not materialize, as there are no such
distinct morphisms (P and 4. In case (iii), c < 0, say c = - u2 with (I E R,
u # 0. Thus A’ = UA and M form a solution of (2.1.1) with c’ = - 1, and we
can take a’ = J-1
= i. The field extension R(i) is C, and (2.2.1.~) reduces
to A’ = (Zi)-l(+
- 4) = Im+ and M = j+12. Thus A = a-‘A’ = u-l Im$, and
n
(2.3.1.b) is established.
3.
ON THE FUNCTIONAL
LEMMA 3.1.
EQUATION (FE) W ITH F = - G
Let additive B # 0 and multiplicative
M a
k satisfy
the
equution
B(x) - M(x)B(l/x)
= 0
(3.14
262 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
C. zyxwvutsrqponm
T. zyxwvutsrqpon
NG
for all x z 0 in k. Then the following
hold for all x, y E k:
and
M (l)=1
B(x)
M(-r)=M(x),
= B(1)2-‘[1+
lf we further assume,
M (x)
X)].
(3.1.3)
that
= 1,
(3.1.4)
also hold fm all x, y E k:
M (x)
[2B2(x)
(3.12)
- M (l-
just for convenience,
B(1)
then the following
=o, zyxwvutsrqponmlkjihgfedcbaZY
X, y # 0,
(3.1.1E)
-M(y)B(y-‘x)
M(x)B(x-‘y)
- B(x2)]
[2B2(y)
[2B2(x)
= 2P(x)
- B(y2)]
- B(X2),
= 2B2(xy)
- B(X2)] B(y)
(3.1.5)
(3.1.6)
- B(x2y2),
= 2B(x)B(xy)
(3.1.7) zyxwvutsrqponm
- B(r2y),
b(X2)- B2(41
- zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJI
wB(Y)12=
DbY)
(3.1.8)
x [B(Y2)
zyxwvutsrqponmlkjihgfedcbaZYXWVUTS
- B2(Y)l)
[BbY)
-
=
Proof.
(3.1.1.E)
e4B(Y)l
Mu4 -
B(Y2)
M is multiplicative,
Since
- JwB(Y)l
%4Wl[
is obvious. The property
same way (2.1.2)
[WY)
(3.1.9)
B2(Y)].
the equivalence
(3.1.2)
follows from (2.1.1),
-
between
follows from (3.1.1)
with the exception
(3.1.1)
in exactly
and
the
that we cannot say
much about B(1).
If we take the identity
similar calculations
obtain
with
(3.1.3).
y=2-‘,
(2.1.9),
apply B to it term by term,
to those done in Lemma
In fact, for the current purpose,
namely
2(1-
r)-‘-
we
it is sufficient to take (2.1.9)
2 = 2x(1 - x)-l.
can drop the factor 2 prior to applying
and follow
2.1, which lead to (2.1.3),
Since char(k)
# 2, we
B to it term by term. The following
HOMOGENEOUS
BIADDITIVE
263 zyxwvutsrq
FORMS
sequence of computations shows all the details:
B[(l-x)-l]
M(l-r)B[(l-Lx-‘]
-B(l)=B[x(l-x)-l],
-M(l-r)B(l)=M(l-x)R[(l-r)-lr],
B(l-
x) - M(l-
x)B(l)
= M(x)B[x_‘(1
B(1) - B(x)
- M(l-
+3(l)
= M(x)B(x_1)
B(1) - B(x)
- M(l-
x)B(l)
= B(x)
-x)],
- M(x)B(l),
- M(r)B(l).
Solving for the B(x) term in the last equation gives (3.1.3) for x Z 0,l; it also
holds for x = 0,l by (1.0.1) and (3.1.2).
Since I3 # 0, from (3.1.3) we get B( 1) + 0. As the equation (3.1.1) is linear
in B, we may assume the normalization (3.1.4) for convenience.
We verify (3.1.5) with the following calculations using (3.1.3) and (3.1.4):
2B2(x)-B(x2)=2-‘[1+M(r)-M(1-x)]2
-2-‘[l+M(Xs)-M(l-x2)]
= 2-‘[1+
P(z)+
- 2M(l-
M2(1-
x)+2M(x)
x) - 2M(r)M(l-
x)]
-2-‘[l+Ms(X)-M(l-x)M(l+x)]
= M(x)+2-‘M(l-
x)
x[ -2-2M(x)+M(l-x)+M(l+x)]
= M(x)+2-‘M(l-
Lx){ [l-r M(l-
x) - M(x)]
- [1+ M( -x)
= M(x)+2-‘M(l-
x)[2B(l
= M(x)+2-‘M(l-
x)[2B(l)
-x)
- M(lS
- 2B( -x)
x)] - 2)
- 21
- 21
= M(x).
Equation (3.1.6) is but a reminder that M in (3.1.5) is multiplicative. We
replace y in (3.1.6) by 1 - y and subtract the resulting equation from (3.1.6)
to obtain (3.1.7).
C.T.zyxwvutsrqpo
NG zyxwvutsrq
264
Replace y in (3.1.7) by y2 to get [2B2(x) - B(x2)]B(y2)
- B(x2y2). We subtract it from (3.1.6) to get
[2P(x)-B(x2)][B2(y)-
= 2B(x)B(xy2)
B(y2)]=B2(xy)-B(x)B(ry2).
In Iigbt of (3.1.7), we can replace the factor B(xy2) in the last term to get
bB2(x)- B(X2)1
[B2(Y) = B2bY)
B(Y2)1
- W{2B(YMY4
- [2B2(Y)
- ~(Y2)]W}.
After simplication we obtain (3.1.8).
Equation (3.1.9) is the symmetric polarization of (3.1.8) in the variable r.
n
THEOREM 3.2. Zf
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
additive B # 0 and multiplicatiue M satisfy the equution (3.1.1) B(x) - M (x)B(l/x)
= 0 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQP
forall x # 0 in k, then they zyxwvutsrqponmlkjihg
have one
of the following three exclusive forms:
Case (i). Forsome
stunt b#O ink.
(field)
morphism
B=b+
Case (ii).
Forsomemorphisms
sameconstantb~Oink,
and
Case (iii).
For some constant
b # 0 in k,
B=b2-‘(++G)
Here 4: k 4 k(a) is an embedding
(the conjugate of +).
and
cek
+ZO,
andsomecon-
(3.2.1.a)
M=+2.
+,J/:k+k,
B=b2-‘(++#)
constant
+:k+k,
$,$#O
and
(3.2.1.b)
M=+#.
with a=&Gk,
and
and @#II/,
and for some
M=+$.
(3.2.1.~)
of k into its field extensiun k(a),
Conuersely, if B and M are of the forms (3.2.la), (3.2.l.b)
then B # 0 is additive, M is multiplicative,
and (3.1.1) holds.
up# 6
or (3.2.l.c),
Proof.
Let B # 0, and M be a solution of (3.1.1). Then by Lemma 3.1,
B(1) f 0. We can normalize B and assume B(1) = 1. We need to show that
HOMOGENEOUS BIADDITIVE FORMS
265
b = B(1) = 1. Consider the
(3.2.l.a), (3.2.l.b), or (3.2.1.~) must hold with zyxwvutsrqponmlkjihgfedcbaZYXW
equations (3.1.2) to (3.1.9) and begin with the last. There are two exclusive
possibilities:
Case 1. Suppose B(y’) - B2(y) = 0 for all y E zyxwvutsrqponmlkjihgfedcbaZYXW
k. This is equivalent to
the symmetric biadditive map B(xy) - B(x)B(y) = 0, i.e., B is multiplicative. Thus I#I= B is a (field) morphism on zyxwvutsrqponmlkjihgfedcbaZYXWVUT
k and + # 0. Equation (3.1.5) then
yields M(x) = +“ (x). This confirms (3.2.1.a) with b = 1.
Case 2. Suppose for some ya, I?(yt) - B2(y0) # 0. Then by fixing such
a y in (3.1.9), we get
B(uv) - B(u)B(v) =cA(u)A(v),
all
u,vek,
(3.2.2)
is a nonzero constant in k, and A(u) =
where c = [B(yt) - B2(y0)]-’
B( uy,) - B( u)B( y,,). Since B(W) - B( u)B( v) does not vanish identically,
we obtain from (3.2.2) that
A is additive
and
A # 0.
(3.2.3)
Putting (3.2.2) back in (3.1.7), we have
- zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONM
[B(x2)B(y)+ cA(x2)A(y)l.
It can be simplified to 0 = 2cA(x)B(r)A(y)
factor cA( y), we obtain A(r2) = 2A(x)B(x).
- cA(x2)A(y).
Canceling the
Polarizing it symmetrically, we
get
Aby)
From (3.1.5) and (3.2.2)
= A(+(Y)+
BAG.
(3.2.4)
cA2(r).
(3.2.5)
we get
M(x) = B2(x) At this point, we have constructed
equations (2.1.2) to (2.1.8). Referring
Theorem 2.2 which correspond to the
and obtain either (3.2.1.b) or (3.2.l.c),
The converse is easy to verify.
an A and reinstated the system of
to cases (ii) and (iii) in the proof of
case c + 0, we observe the form of B
with b = 1.
W
C. zyxwvutsrqponmlk
T. NC
266
The deliberations that led to Corollary 2.3 from Theorem
repeated for Theorem 3.2 with k = W, and give the following:
2.2 can be
COROLMY 3.3.
The general solution of B(x) - M(r)B(l/x)
= 0 for all
x # 0 with additive B # 0 and multiplicative
M on the reals Iw is given by
B=bx
B=bRe$
and
and
M(;r)=r2
(3.3.1.a)
(3.3.1.b)
M(x)=l$(x)12,
where b + 0 is an a&tray
real constant, + : IR + C is a nontrivial embedding, Re $J is the real part of +, and 1.1 denotes the usual norm on Q=.
4.
ON THE EQUATION
(FE) AND ITS GENERAL
SOLUTION
We consider the functional equation
F(x)
+ M(x)G(l/ x)
= 0
forall
x#Oin
k,
(FE)
where F, G : k + k are additive and M: k + k is multiplicative. As pointed
out in the introduction, there is no loss of generality in assuming (1.0.1)
M(0) = 0, and to avoid trivial cases, we suppose that
(4.0.1)
F,G,M#O.
In particular, M(r) # 0 and M(x-‘)
x by x-l in (FE) and arrive at
G(x)
+ M(x)F(l/x)
= M(x)-’
= 0
for all x # 0 in k. We replace
forall
x+Oin
k.
(4.0.2)
Forming the sum and difference of (FE) and (4.0.2), we transform them to
the system of equations
M(x)A(l/ x)
= 0,
(4.0.3)
B(x) - M(x)B(l/ x)
= 0,
(4.0.4)
A(x)+
HOMOGENEOUS
BIADDITIVE
267 zyxwvutsrqp
FORMS
where A = 22’(F + G) and B = 2-‘(F - G) are again additive. Because of
(4.0.1), either A # 0 or B # 0. We use Theorem 2.2 and Theorem 3.2 to
obtain the solution of the system (4.0.3) and (4.0.4), which in turn gives the
general solution of (FE) where zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPON
F = A + B and G = A - B. Their combined
use leads to the following result. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONM
THEOREM 4.1.
Let additive F, G : k + k and multiplicative M : k + k be
rwnzero maps satisfy ing (FE).
Then they have one of the following three
exclzrsive representations:
(i) For some (field)
morphism $J: k - + k, cp# 0, some $derivation
which is an additive map satisfy ing D(xy ) = +(x)D(y )+
+(y )D(r)
D,
fm all
x, y in k, and a constant b E k with (D, b) f (O,O), we have
F=D+b$,
(ii)
M =G~.
G=D- b+,
For some morphisms +,$:k+k,
(p,$# O
and @ # I/J,
(4.1.1.a)
and some
constants cl, cz E k with (c,, c2) # (O,O),
F = cl+ - czJl>
(iii)
k(a)
G=c,+-c,$,
Forsomeconstantc~kwitha=6Ek,
M =+# .
someembedding
(4.1.1.b)
+:k+
with $I f 5, and some nonzero constant X E k(a),
F=h++
Here, conjugacy
A+,
G=
- % # A&
M =&j.
(4.1.l.c)
in k(a) is defined by x + y a= x - y a for all x, y E k.
The converse is also true.
Proof.
Either A or B in (4.0.3) and (4.0.4) must be nonzero. If A = 0, or
if B = 0, then (4.1.1.a) to (4.1.1.~) hold because of Theorem 3.2, or Theorem
2.2. Suppose A # 0 and B # 0 in (4.0.3) and (4.0.4). Then Theorem 2.2 and
Theorem 2.3 are both applicable, along with their supporting lemmas.
Compare (2.1.4) with (3.1.3), and observe that the B in the system (2.1.2) to
(2.1.8) corresponding to Equation (4.0.3) coincides with the B in (4.0.4)
when the latter is normalized. Hence (4.0.4) after normalization coincides
with (2.1.7), and so (4.0.4) can be regarded as part of (4.0.3). In this case,
there is no need to develop (3.1.5) to (3.1.9), and we can obtain the form of
A and M from Theorem 2.2, and at the same time obtain B from its proof.
Namely, B = (p with (2.2.l.a),
I?= 2-‘(+ + 4) with (2.2.l.b), and B =
C. zyxwvutsrqponml
T. NG
268
2-‘($I + 6) with (2.2.1.~). We can put in a constant factor b # 0 next to the
above forms of B to compensate the normalization. The final result for
F = A + B and G = A - B is precisely (4.1.1.a) to (4.1.1.~) with D # 0,
b#O,
c,#fc,,
A=2-‘(a-‘+b).
Th us in ah situations, F, G, and M
must be of the forms (4.1.1.a) to (4.1.1.~). No solutions wilI be lost if we
restrict X to those of the form X = b or A = 2-‘(a-’
+ b) where b E k (cf.
Remark 4.4). The converse is easy to justify.
n
We combine Corollary 2.3 and Corollary 3.3 to give the general solution
of (FE) over the real field: zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJI
COROLLARY 4.2.
The general solution of (FE) with
F, G and multiplicative
M on the reals 08 is given by
F(x)
= D(r)
+ bx,
G(x)
F=b,Im++b,Re+,
= D(x)
- br
and
G=b,Im$-b,Re$
nonzero additive
M(x) =x2
and
(4.2.1.a)
M=l+l’.
(4.2.1.b)
Here D is a derivation on W, b E R is a constant, and (D, b) # (0,O);
+: R -+ C is a nontrivial embedding, b,, b, E R are constants, and (b,, b,)
+ (0,O).
uniqueness
result for
(i) For given M, the morphism
(4.1.1.a) is unique.
(ii) For given M, the distinct
M = C#B#in (4.1.1.b) are unique up
{ C#B,
I/J} is unique.
(iii) Let us define the following
kl&Zk}
c k*:
c-c’
ToeuchcEX,
of Theorem 4.1, we have the
representations
(4.1.l.a)
(4.1.l.b),
and
Under the hypothesis
THEOREM 4.3.
following
(4.1.l.c):
iff
C#J
( # 0) on k generating M via M = +2 in
morphisms 9, +!J( # 0) generating M via
to a permutation, i.e., the unordered set
equivalence
c=d2c’
letS(c)={(F,G,M)jF=A++&G=
relation on the set X = {c E
forsome
(4.3.1)
d Ek.
-&#B-&
M=&
269
HOM OGENEOUS BIADDITIVE FORM S
us zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
in (4.1.1.~)) be the associated family of solutions of (FE) over all X and
+. It turns out that S(c) and S(c’) are either identical or disjoint, according
to whether c or c’ are equivakmt or not. Furthermore, zyxwvutsrqponmlkjihgfedcbaZ
forgiven M and c E X,
the embedding
+: k - - , k(G),
generating M via M = &
in (4.1.l.c),
is
unique up to conjugation.
Proof.
(i): Suppose 9, +’ are morphisms such that M = # = (+‘)‘. Then
the symmetric biadditive maps +(x)+(y)
and +‘(x)+‘(y)
have equal diago-
nal, and thus #(x)$(y)
=+‘(x)+‘(y)
for all x, y E k. Since +,#
# 0, we
have ~(1) = cp’(1) = 1, and so by setting y = 1 we get 4(r) p+‘(r)
for all
r E k, i.e., C$= $‘. This proves the uniqueness of $I in the representation
M = (p”.
(ii): Suppose M = +J, = +‘\I/‘.This implies the symmetric biadditive maps
+(X)+(Y)+
J/(~)+(Y)
and +‘(X)+‘(Y)+
#‘(x)+‘(~)
are equal. Specifying
y = 1, we get C#I
+ \c,= $I’ + JI’. This linear dependence between the four
nonzero multiplicative functions (p, \I/,+‘, #’ implies they cannot be pairwke
distinct; some two of them must be equal. Since + # 4, 9’ # #‘, and
$I + # = +’ + 1+5’,we must have (c#J’,I/‘) = (+, #) or ( I+!J
#I),
, proving that the
distinct morphisms (p, # generating M via M = +Ic, are unique up to a
permutation.
(iii): W ith reasonings analogous to (ii) above, we conclude that if nonzero
+, +‘: k + k(A)
+’ #
J,
are morphisms such that M = +? = +‘&‘, where + # 4 and
then I#J
= +’
or Cp= 3.
This proves the uniqueness of (p up to
conjugation in the representation M = ~$6 when c is fixed. W e now proceed
to examine the families S(c) over all c E X.
Suppose S(c) and S(c’) meet. Then there exist embeddings +: k + k(h)
and +‘:k+k(@ )
such that M =&=+‘$.
W e write #=B+Afi
and
9’ = B’ + A’@ ,
where (B, A) and (B’, A’) : k + k are additive pairs satisfy-
ing the relations (2.1.3) to (2.1.8). In view of (2.1.4) and (2.1.8), we have
B = B’ and B2 - cA2 = BT2 - c’At2; i.e, B = B’ and cA2 - c’Ar2. Since A, A’
# 0, the relation cA2 = c’A’~ implies c - c’, where -
is defined by (4.3.1).
This proves that S(c) and S(c’) meet only if c - c’.
Suppose c - c’; we proceed to show that S(c) = S( c’). By symmetry it is
sufficient to show S(c) c S(c’). Let (F, G, M ) E S(c) be arbitrarily given.
Then there exist constant A and an embedding @ : k + k(6),
I# #
I 4, such
that (4.1.1.~) holds. W e write $I = B + Afi, where B, A are additive maps
on k. Then the relations (2.1.3) to (2.1.8) hold. Since c - c’, we can write
c = d2c’ for some d E k. W e define B’ = B, A’ = ~54. Then (2.1.3) to (2.1.8)
hold for B’, A’, c’, and M . Thus 9’ = B’ + A’@
defines an embedding of k
into k(c),
and M =+‘i.
Let h=X,+h,fi
with X,,X,Ek;
then FA+ +q
2(A,B
is equivalent to F=
+ c’. dh,.
dA) = 2(A,B’
2(h,B+
&,A).
+ c’. dX,.A’),
But c =d2c’,
and so F =
which is equivalent to
C. zyxwvutsrqponml
T. NC
270
F = 2~’ + A’+‘, where A’ is defined by x’ = X 1 + dh,@ .
Similarly, G =
- A+ - h6 translates into G = -E+’ - 22. Obviously +’ ~2 and x’ # 0.
Hence (F, G, M ) E S( c’). This proves S(c) c S( c’) as claimed.
n
REMARK4.4. We pointed out in the proof of Theorem 4.1 that there is
no loss of solutions to (FE) if we restrict A E k(a) to be of the form X = 2-‘b
or X = 22’(a-’
+ b) where b E k. By allowing X E k(u) to be free of this
restriction, we have created some redundancy in the representation (4.1.1.~).
The reason for creating the redundancy in the first place is so that we can
now remove it by a suitable choice of c. To be precise, in light of Theorem
4.3, we can restrict the choice of c E k with fi P k to those that are not
mutually equivalent under the definition (4.3.1).
For example, Corollary 4.2 is formulated with this idea in mind. When
k=[W,wehaveX={cEkl~4IW}=]-00,0[andthereisonlyoneequivalence class on X, as c - - 1 for alI c E X. It is sufficient to consider just one
field extension W( - 1) = 6.
In general, the set Y = { d2 1d E k * } is a subgroup of k * under multiplication, and it is the image of k * under the multiplicative map d + d ‘. The
set k * is the disjoint union of X and Y. The equivalence classes on X under
the equivalence relation (4.3.1) are members of the quotient k */Y with Y
removed. A choice of representatives from each of the equivalence classes on
X corresponds to a lifting I (restricted to (k */Y )\{ Y }) from the quotient
group k */Y back to k *. When k is a finite field, the Lagrange theorem
implies ]Y 1= 2- ‘1k * 1and k */Y is cyclic of order 2, and so there is only one
equivalence class on X. Hence for finite fields, it is sufficient to use one field
extension in generating solutions in (4.4.l.c). In general, we can restrict the
choice of c in (4.1.1.~) to the image of an arbitrarily fixed lifting 1.
5.
ON THE EQUATIONS
THEIR INVERSIONS
(2.1.1) AND (3.1.1)
WITH
GIVEN
M AND
In retrospect, we solved (FE) by a reduction to the equations (2.1.1) and
(3.1.1), which are tied by a natural duality. We have shown in Corollary 2.3
the presence of nontrivial additive A and multiplicative M other than
derivations and the square. In the course of solving (2.1.1) and (3.1.1), we
have established some important correlations between the maps A, B, and
M , which are contained in Lemma 2.1 and Lemma 3.1. Some of these served
only a passing role; e.g. (2.1.2), (3.1.2), (3.1.4), (3.1.6) to (3.1.9). The rest
deserve more attention, as they offer a basis for the construction of the
functions, and provide transformation rules between them.
271
HOMOGENEOUS BIADDITIVE FORMS
In many applications the map M is given and we want to solve for A and
B in (2.1.1) and (3.1.1). Conversely, given A # 0 or B # 0, we can reconstruct M. The functional relations reported in Lemma 2.1 and Lemma 3.1
can be used effectively in such computations, as summarized in Figures 1
and 2. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
EXAMPLE 5.1.
To find the general solution of F(z) + l.z12G(Z) = 0 on C
with additive F, G, we simply follow the flow chart in Figure 1 beginning
with M(z) = 1~1~.The logical steps in the reasoning are listed below:
M = 1~1~; B,(z) = 2-‘(2 + 5); B, is additive;
the general solution of B is B = C(Z + Z), with arbitrary c E C;
B,,(zi) - @(~a) # 0 is true, say with ~a_= i;
A,(z) = B,(zi) - B,(z)&,(i)
= 22’(zi +zi) = 2-‘i(z - 5);
the general solution of A is A = c( z - Z), with arbitrary c E C.
Therefore the general solution of (FE) with M(z) = lz12 on C is given by
F = A + B = az - bZ, G(z) = A - B = bz - aZ, where a, b E C are arbitrary
constants.
6.
REPRESENTATION
OF FUNCTIONALLY
BIADDITIVE
FORMS
HOMOGENEOUS
THEOREM 6.1. Let V be a vector space over k. Let T:V XV+
nonzero M-homogeneous biadditive form for some M: k + k, i.e.
k be a
forall
XEk,
x,y~V.
(6.1.1)
T(h zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
Xy) = M(A)T(x,y)
Then M is multiplicative,
and is of one of the following
three exclusive types:
M = c$~
where
C$: k + k is a nonzero ( field)
M=NJ
where
C#B,
II/: k + k are distinct rumzero morphisms,
M=&
where
+: k ---*k(G) is a nontrivial embedding of k
into an extension of k obtained
by adjoining a square root which is not in k.
morphism,
(6.1.2.1)
(6.1 .Z.II)
(6.1.2.111)
Furthermore,
4.3).
such representations
of M are essentially
unique (cf. Theorem
C. T. NG
w
A=0 and B=O
are the general
solutions of (A) and (B)
is the general
solution of (B)
w
BO is a field
morphism and
A=a B,-derivation
is the general solution
of (A)
Yes
FIG. 1. Flow chart for finding the general solutions of (A) A(x)+ M(x)A(l/x)
= 0 or (B) B(r) - M(r)B(l/r)
= 0 with additive A, B when multiplicative
M is
given.
HOMOGENEOUS
273
BIADDITIVE FORMS
Given E!
Given A
There exists
no solution M
M,(x)
= 2B’o (x) - B,lx2)
No
)
There exists
no solution M
1Yes
M = M, is the unique
solution
for
= 0
FIG. 2. Flow chart for finding the general solution of A(x) + M( r)A(l/r)
or B(x) - M( r)B(l/r)
= 0 with multiplicative M when additive A + 0 or B f 0 is
given.
C. zyxwvutsrqponm
T. NG
274
W e fix arbitrarily x0, y,, E V such that T(x,, yO)# 0. Since T( Xp.
Proof.
x0, XP.Y,)
= M ($)T(x,,yJ
and T(Ae~lx~, X.pyO) = M (X)T(P~~, PY~) =
th e multiplicativity M ( AW L)= M ( A)M (p) follows. Since zyxwvutsrqpon
M (~)M (P)T(x,,YlA
VAx,,y,)
T(Xx,,y,)
A-‘y,),
= WX)?‘(x,,
and
the maps
G(X) = - T(x,,
Xy,,)
F, G: k + k induced
satisfy
(FE).
by F(X)
=
Clearly F and G are
additive and nonzero, and so by Theorem 3.2, M must be of one of the cited
w
forms.
The next three theorems disclose the structure of all M -homogeneous
biadditive forms based on the three M types.
THEOREM 6.2.
Let nonzero 9: k + k be a morphism. The following
statements concerning biadditive functions T on a vector space V over k are
equivalent:
(1)
T is G2-homogeneous.
(2) T satisfies
the following
T(hy)
laws:
= D(Lx,y)+
d+W(x>y),
T(x, hy) = - D(b,y)+
where D: k x V x V + k satisfies the following
D(.,x,y)
isa @derivationon
(6.2.1)
$@)T(x,y),
description:
kforeachfixedr,yEV,
D( A,. , . ) is a (+, +)-bihomogeneous
biadditive
form on V
for each fixed X E k.
(3) T can be constructed
through the following
(6.2.2)
(6.2.3)
steps:
(i) Choose an arbitrary basis Q = (e,), E I for V over k.
(ii) Znitiate T on Q X Q (into k) arbitrarily.
(iii) Initiate for each (e,e’) E G x Q a @derivation on k, labeled
by
D( . , e, e’), arbitrarily.
(iv) Extend T and D( A,. , .) above to V X V by the definitions
T(x,y)
= CG(Pj)D(Xi,ei,ej) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJI
ij
-
D(Xtx,y)
=
+(Xi)D(pj,ei,ej)+
G(Ai)G(Pj)T(ei,ej),
(6.2.5)
C~(hi)~(l-lj)D(h,ei,ej),
ij
where x = Xihiei
and y = Cjpjej,
(6-2.4)
Xi, pj E k.
HOMOGENEOUS
BIADDITIVE
275
FORMS
zyxwvutsrqp
Proof. To show that (1) implies (Z), let zyxwvutsrqponmlkjihgfedcbaZYXWVU
T be +2-homogeneous. For each
fixed x, y E V, the functions
(6.2.6)
G (X) = - T(x, Xy )
F(h) = T(hy),
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
satisfy
(FE)
with M = $I~ because F(h) = T(Ax,y ) = ~~(h)T(x, A- ‘y ) =
Thus by Theorem 4.1 and Theorem 4.3, there exist a
@derivation
D and a constant b such that F(h) = D(h) + b+(h)
and
G(X)=D(X)- b+(A).
Since F and G depend on x and y, we explicitly
write F(X) = D(A,x,y )+
b(x,y )+(X)
and G(X) = D(X,x,y ) - b(x,y )+(A).
Hence, from (6.2.6) we get the following laws governing T:
- c$~(A)G(X- ‘).
Thy)
= D(b>y>+
b(x,y)$(X),
T(x,Ay) = - D(X,x,y)+
b(x,y )+(X).
Since D( 1, x, y) = 0 and (p(1) = 1, we get b(x, y ) = T(x, y ), and so
T(hy)
= D(Lx,y)+
Tb, xy> = - D(Lx,y)
+(X)T(x,y)>
+ cp(h)T(x,y),
(6.2.7)
(6.2.8)
proving (6.2.1) and (6.2.2). The additivity of D in its second and third
variables follow from (6.2.7) and the additivity of T in both x and y. What is
left in (6.2.3) is the +-homogeneity of D in x and y. To obtain this we
perform the following calculations using (6.2.7):
Th-y)
= D(kx, Y)+ ~h$‘(x,y)
= @)D(PAY)+
T(hx,y)
G(dD(Lx>y)
+ +(%(~)T(x,y),
(6.2.9)
= D(k /-my) + +(A)T(px,y)
=D(X,~x,y )+~(X)[D(~,x,y )+~(~1)T(x,y )].
(6.2.10)
Comparing the right sides of the above, we get
D(L my)
= @(p)D(b,y),
which is the +-homogeneity of D in its second variable.
+-homogeneity of D in its third variable follows from (6.2.8).
(6.2.11)
Similarly, the
276
C. T. NG
TO show zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
that (2) implies (l), we display the following calculation:
T(Xx, hy)=D(A 9x9hy)++(A)W,
Xy)=~(h)D(h,x,y)+~(X)[-D(X,x,y)
+
+(~)W,y)l = &vw,y).
The equivalence between
(2) and (3) is obvious.
n
THEOREM 6.3. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHG
Let nonmo
$, 4: k + k be distinct mmphisms.
The
following two statements concerning biadditive fm
T on a vector space V
over k are equivalent:
(1) T is q+homogeneous.
(2) T has the decomposition zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
Thy)
=
c,(x,y) - C,(x,y)
(6.3.1)
for some unique biadditive C,, C, : V X V + k, with (+, #>bihomogeneous
C, and (4, +>bihomogeneous
C,.
Proof.
Let T be &!+homogeneous. Then for each fixed x,Y E V, the
functions F(h) =T(Xx,y),
G(X) = - T(x, Xy) satisfy (FE) with M = ++
because F(X) = T(Xx,y) = +(A)$(h)T(x,
A-‘y) = - +(h)#(X)G(h-‘).
By
Theorem 4.1 and Theorem 4.3, there exist constants C,(x,
y) and
zyxwvutsrqponmlkjihgf
C,(x,
y) such
that
WXPY)
=c,(wM~)
T(x, AY) = - Cs(xJ)+O)+
-
C,(x,y)J/(X),
C,(x,y>$(h).
(6.3.2)
(6.3.3)
Since + and 4 are linearly independent, the above equations and the
biadditivity of T imply that C, and Cs are biadditive. We compute T(Apx, y)
under (6.3.2)
first as T(Xp . x, y) = C,(x, y)$(Xp) - Cz(x, y)J/(Ap) =
C,(x, Y)+(~)+(P)
- &(x9 y)$(h)$(p),
and then as T(h. CLX,y) =
C,(/wY)9(A)
- CZ(cLX>Y)W). c om Parison leads to Cr(x, y)+( X)+(p) C&,y M AM pL)
= C1(px,y )$(h)
- C&x,y )# (~).
The independence
of
G(X) and +(A) yields W x,y )
= C1(x,y M (~) ad C&x,y ) = C,(x,y )W ).
Similarly, using (6.3.3) we get C,(x, py) = C,(x,Y)$+),
and Ca(x, py) =
n
o
er
words,
C,
is
(I$,
#)-bihomogeneous,
and
C, is (4, +)C,(%Y)@(EL). 1 th
bihomogeneous. The representation (6.3.1) comes from (6.3.2) with X = 1.
This proves that (1) implies (2). The converse is obvious. The uniqueness of
n
C, and C, is due to (6.3.2) and (6.3.3) where + and J, are independent.
THEOREM6.4. Let c E k be such that 6 P k, and let +: k + k(6)
be a
nontrivial embedding,
The following two statements concerning biadditive
HOMOGENEOUS
277 zyxwvutsrqp
BIADDITIVE FORMS
forms T zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
on a vector space V over k are equivalent:
(1) T is &&homogeneous.
(2) T hu.s the unique decomposition
Thy)
where A: V x V + k(h)
(6.4.1)
= zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFE
Nx,y) + &,y)
is ($, &)&homogeneous biudditive.
Proof. By similar arguments to those employed for the previous two
theorems, F(X) = T( Ax,y)
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
and G(A) = - T(x, Xy)zyxwvutsrqponmlkjihgfedcbaZYXWVU
satisfy (FE) with M = $16,
and so by Theorem 4.1 and Theorem 4.3, there exist constants A (x, y)
zyxwvutsrqponmlkjih
E k (6)
such that
Thy)
=
Nd#O)+
~(x,y&i(A),
(6.4.2)
(6.4.3)
T(x, AY) = x(x,y)@(A)+ zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
Aby)?(
The discussion on the uniqueness of (4.1.1.~) in Theorem 4.3 and Remark 4.4
is used here to support the fact that c and the embedding $I: k + k(G) can
be chosen independent of x and y. The rest is a rerun of the proof of
n
Theorem 6.3. The meaning of bihomogeneity is naturally broadened.
REMARK 6.5. A functional T: V X V --) k is M-homogeneous and biadditive if, and only if, its symmetric and skew-symmetric parts are. It is easy to
derive from the above three theorems their symmetric and skew-symmetric
counterparts.
In Theorem 6.2, T is symmetric (skew-symmetric) if, and only if, D( X, . , - )
is skew-symmetric (symmetric). Hence, to construct symmetric (skew-symmetric) T, it is necessary and sufficient that the initializations in (3)(ii), (iii) be
symmetric (skew-symmetric) and skew-symmetric (symmetric) respectively.
In Theorem 6.3, T is symmetric [skew-symmetric] if, and only if, C,(x,y)
+ C,(y,x) = 0 [CXx, y) - C2(y,x) = 01. In Theorem 6.4, T is symmetric
(skew-symmetric) if, and only if, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
A(x, y) = h(y,x) [ A(x, y) = - @y,x)]. Such
decompositions have the same spirit as those results reported in [lo].
One can easily answer questions such as (1.0.2), using the above results.
REMARK6.6. The above result can be routinely applied to determine all
biadditive M-homogeneous transforms T: V X V + W where W is a k-vector
space. All it requires is a coordinatization of W with respect to some basis,
and we pass from the properties of T to those of its coordinate functions.
278
C. T. NG
T: zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
V X V --, k zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHG
is cohomogeneous if there exists a function
N such
T(x, Ay) = N(X)T(x, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPON
y). We can immediately
deduce
from
Theorems 6.2-6.4 that homogeneous biadditive T are all cohomogeneous.
On
A map
that
T(Xz, y) +
the other hand, it is easy to show that cohomogeneity implies homogeneity,
because T(Xx, Ay )+ T(x, A.hy ) = N(A)T(x, Ay ), T(Xx, Ay )+ T(h.Xx,y )=
N(X)T(Xx, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
y)> Vh)[T(h
y) + T(x, Ay)] = N2(A)T(x, y), T(x, A2y ) +
T( A2x, y ) = N( A2)T(x, y ). Subtracting the last from the sum of the first three,
we get the M-homogeneity of T with M (X) = 2- ‘[ N 2(A) - N( X2)]. Thus
cohomogeneity
and homogeneity are equivalent for biadditive forms.
REFERENCES
1
5
10
11
12
13
14
J. Aczkl and J. Dhombres, Characterization
of inner product spaces, in Functional Equations Containing Several Variables, Addison-Wesley, Reading, Mass.,
1985, Chapter 11.
J. Aczel, Derivation and Information Functions (A Tale of Two Surprises and a
Half), Contributions
to Probability, Academic, 1981.
J. Aczirl, Derivations
and functional equations basic to characterizations
of
information measures, C. R. Math. Rep. Acad. Sci. Canada 1:165-168 (1979).
J. Aczel, Functional equations for the trigonometric functions and related equations, in Lectures on Functional Equations and Their Applications,
Academic,
New York, 1966, $3.2.
J. Aczel, The general solution of two functional equations by reduction to
functions additive in two variables and with the aid of Hamel bases, Glas.
Nat.-Fiz. Ashon. 20:65-72 (1965).
J. A. Baker, ,On quadratic functionah
continuous
along rays, Glas. Mat.
3(23):215-229
(1968).
N. Bourbaki, Derivatives, in Elements of Mathematics,
Algebra 1, AddisonWesley, Reading, Mass., 1974, Chapter 3, $10; ibid., Exercise Problem 2 p. 645.
T. M. K. Davison, Jordan derivations and quasi-bilinear forms, Comm. Algebra
12:23-32 (1984).
A. M. Gleason, The definition of a quadratic form, Amer. Math. Monthly
73:1049- 1056 (1966).
K. J. Heuvers, A family of symmetric biadditive nonbilinear functions, Aequationes Math. 29 (1985).
P. Jordan and J. von Neumann, On inner products in linear metric space, Ann.
Math. 36:719-723 (1935).
Pl. Kannappan
and C. T. Ng, On a generalized
fundamental
equation of
information,
Canad. J. Math. 35:862-872 (1983).
H. Kestelman, Automorphisms
of the field of complex numbers, zyxwvutsrqponmlkjihgf
ho c .
London
Math. Sot. (2) 53:1-12 (1951).
S. Kurepa, The Cauchy functional equation and scalar product in vector spaces,
Glas. Mat.-Fiz. Astron. 192336
(1964).
HOMOGENEOUS
15
16
17
BIADDITIVE
FORMS
279
S. Kurepa, Quadratic and sesquilinear functionals, C&s. Mat.-Fiz. Astrm.
20:79-92 (1965).
Gy. Maksa, Solution on the open triangle of the generalized fundamental
equation of information with four unknown functions, Utilitm Math. 21:267-282
(1982).
P. VrbovH, Quadratic functionals and bilinear forms, &.sopis Pest. Mat.
98: 159- 161 (1973).
Receioed 30 April1966;revked 26 August 1986