pacific journal of mathematics

Vol. 177, No. 1, 1997

TWO GENERALIZATIONS OF THE GLEASON-KAHANE-ZELAZKO THEOREM Erik Christensen

In this article we obtain 2 generalizations of the well known Gleason-Kahane-Zelazko Theorem. We consider a unital Banach algebra A, and a continuous unital linear mapping ' of A into Mn(C ) { the n n matrices over C . The rst generalization states that if ' sends invertible elements to invertible elements, then the kernel of ' is contained in a proper two sided closed ideal of nite codimension. The second result characterizes this property for ' in saying that '(A ) is contained in GLn(C ) if and only if for each a in A and each natural number k: k k
inv

trace( (

'a

)) = trace( ( ) )

'a

1. Introduction. The results are based on Aupetits work 1], where he proves that if ' is

surjective and '(Ainv ) is contained in GLn (C ) then ' is a Jordan homomorphism. Some of the key ingredients in Aupetits proof are some relations which involve ' and the trace. We quote 1, Rel (3), p. 15] trace('(xy)) = trace('(x)'(y)): The results in this paper are based on some elaboration of the identity above. Aupetits result is proved via a Liouville Theorem for harmonic functions. Since we could not get the reference quoted in 1] for this result, we have included a slightly di erent argument, which on the other hand is an application of the Hadamard Factorization Theorem, very much in the same way as used in 3, 4].

2. Notation and basic results.

We will consider a unital Banach algebra A over the complex numbers, and let Ainv denote the set of invertible elements in A. For a natural number n we let Mn (C ) denote the n n matrices over C and we will let tr( ) denote the usual trace on Mn (C ), which satis es tr(I ) = n. We remind the reader that a functional f on an algebra B is called a trace if f (ab) = f (ba) for all
27

28

ERIK CHRISTENSEN

a; b from B. Moreover { up to scalar multiples { there exists only one trace on Mn (C ).

The following lemma is well known, but we do not have an exact reference.

Lemma 2.1. There exists a positive real r(0 < r < 1) such that for all x in Mn (C ) with kx ? I k < r we have j det(x) ? 1j < 1 and Log(det(x)) = tr(Log(x)):
Proof. The existence of r(0 < r < 1) such that kx ? I k < r ) j det(x) ? 1j < 1 follows from the continuity of the determinant. Suppose now x in Mn (C ) is chosen such that kx ? I k < r < 1 then the power series for Log(1 + u) in the circle fu 2 C j juj < 1g converges for both (det(x) ? 1) and (x ? I ) so the expressions make sense. The equality is easily obtained when x is represented in a Jordan normal form.

We start by recapturing the basic results from 1] in Theorem 3.1, and then we present our extensions.
linear mapping of A into Mn (C ). If '(Ainv ) GLn (C ) then: (i) 8 a; b 2 A : det('(ea eb )) = det(e'(a) e'(b) ); (ii) 8 a; b 2 A : tr('(ab)) = tr('(a)'(b)) = tr('(ba)); (iii) 8 a; b 2 A : det('(ab)) = det('(a)'(b)):
f (z ) = det('(eza )e?z'(a) ):

3. Main results.

Theorem 3.1. Let A be a unital Banach algebra ' a continuous unital

Proof. For an a in A we de ne an entire function f (z ) by

As usual f (z ) 6= 0 for all z and the order p of f satis es p 1 since

jf (z)j k'kn exp(jzjn(kak + k'(a)k)): By Hadamards Factorization Theorem 2, p. 291; 5 p. 250] we have f (z ) =
e
+

but f (0) = 1 so f (z ) = e z . Following Lemma 2.1 we get that for some positive real r we have

z,

8 z ; jzj < r : z = tr(Log(I + z'(a) + O(z ))) ? z tr('(a))

2

= O(z 2 ):

GLEASON-KAHANE-ZELAZKO THEOREM

29

Hence = 0 and det('(ea )) = det(e'(a) ). In the general case de ne an entire function f (w; z ) in 2 variables by f (w; z ) = det('(ewa ezb )e?w'(a) e?z'(b) ). Let w be xed then the function g(z ) = f (w; z ) is entire, never vanishing, of order 1 and by the previous result g(0) = 1, hence there exists a complex function (w) such that f (w; z ) = g(z ) = e (w)z . By analogy we nd a complex function (z ) such that f (w; z ) = ew (z) . Hence there exists a constant such that f (w; z ) = e wz . On the other hand the function k(z ) = f (z; z ) is easily seen to be of order less than or equal to 1 so = 0 and (i) follows. Applying Lemma 2.1 to both sides in the following identity det ew'(a) det ez'(b) = det ' ewa ezb shows that there exists a positive real r such that for all z; w in

? ?

jzj < r ; jwj < r

with

w tr (' (a)) + z tr (' (b)) = tr Log ' ewa ezb ? ? = tr Log I + w' (a) + z' (b) + wz' (ab) + w2 p1 (w; z ) + z 2 p2 (w; z ) = w tr (' (a)) + z tr (' (b)) + wz tr (' (ab))

? ?

1 ? 2 wz tr (' (a) ' (b) + ' (b) ' (a)) + w p (w; z) + z p (w; z) ;
2 3 2 4

where pi (w; z ) are power series. Hence (ii) follows from the properties of the trace. The relation (iii) is a consequence of (i) since for jz j > kak + kbk we ? have (z ? a) = z (1 ? a ) = z exp Log(I ? a ) and a similar expression for b z z and hence for jz j > kak + kbk we have det('((a ? z )(b ? z ))) = det('(a ? z )) det('(b ? z )): Since the functions involved are entire, we get (iii) for z = 0. The relation (ii) is the basis for the following result.

Theorem 3.2. Let A be a unital Banach algebra and ' a continuous, unital linear mapping of A into Mn (C ). If '(Ainv ) GLn (C ) then A has a proper closed two sided ideal J { of nite codimension { which contains the kernel of '.
Proof. De ne J = fa 2 A j 8 b 2 A : tr('(ab)) = 0g, then J is obviously a closed right ideal, but by the trace property { tr('(ab)) = tr('(ba)) { from (ii) in Theorem 3.1 we see that J is a left ideal as well. The property tr('(ab)) = tr('(a)'(b)) from (ii) above shows that ker ' J . Hence J is

30

ERIK CHRISTENSEN

of nite codimension and since '(I ) = I we get I 2 J and J is a proper = ideal.

Corollary 3.3. Let k; n be natural numbers if ' is a unital linear mapping of Mk (C ) into Mn (C ) which satisfy '(GLk (C )) j GLn (C ), then ' is injective
and k divides n. Proof. Since Mk has no nontrivial ideals, we get ker ' = f0g and ' is injective. By (ii) in Theorem 3.1 we get that trn ' is a trace on Mk (C ) which satis es trn ('(I )) = trn (I ) = n. By the uniqueness (up to scalar multiples) of the trace on Mk (C ) we have for all a in Mk (C ): trn ('(a)) = (n=k) trk (a). Let e be a rank 1 projection in Mk (C ) then ('(e)) (e) = f0; 1g so trn ('(e)) 2 N 0 . On the other hand trn ('(e)) = (n=k) trk (e) = n=k so n=k 2 N .

Corollary 3.4. If '(A ) GLn (C ) then there exists a unital nitedimensional algebra B and a linear mapping is a unital homomorphism.
inv

of Mn (C ) into B such that

'

Proof. Just another formulation of the result ker ' J 6= A, combined with elementary algebra.

The property (ii) from Theorem 3.1 and the theorem above are generalizations of the original Gleason-Kahane-Zelazko Theorem. The following result yields another generalisation as well as a characterization of the mappings ' which satisfy '(Ainv ) GLn (C ).

Theorem 3.5. Let A be a unital Banach algebra and ' a unital continuous
linear mapping into Mn (C ). Then '(Ainv ) GLn (C ) if and only if

8 k 2 N 8 a 2 A : tr ?' ?ak = tr ?'(a)k :

Proof. Suppose rst that '(Ainv ) GLn (C ). Let r be the positive real comming from Lemma 2.1 and let a be in A and z in C , then there exists a positive real r1 ; 0 < r1 < kak?1 such that for jz j < r1

k'(I + za) ? I k < r and k exp('(Log(I + za))) ? I k < r:

By Theorem 3.1 (i) we get for jz j < r1 det('(exp(Log(I + za)))) = det(exp('(Log(I + za))))

GLEASON-KAHANE-ZELAZKO THEOREM

31

so by Lemma 2.1

8 z; jzj < r : tr(Log(I + z'(a))) = tr('(Log(1 + za))):

1

By expanding in power series and comparing terms we get 8 k 2 N : (?1)k?1 k?1 tr ?'(a)k = (?1)k?1 k?1 tr ?'(ak ) ; and the rst part of the proof is complete. Let us now suppose, that for each k in N and any a in A tr('(ak )) = tr('(a)k ). Let b be an element in A and let C denote the abelian unital Banach algebra generated by b and all its resolvents in A. Since C is abelian we get for c; d in C cd = 1 ((c+d)2 ?(c?d)2 ) 4 so 1 ( ) tr('(cd)) = 4 tr ('(c) + '(d))2 ? ('(c) ? '(d))2 = tr('(c)'(d)): Having this identity we may as in the proof of Theorem 3.2 de ne a two sided ideal JC in C by JC = fc 2 C j 8 d 2 C : tr('(cd)) = 0g: Again JC 6= C and (ker ' \ C) JC : Since JC is a proper two sided ideal in C and b is invertible in C, b 2 JC , = and '(b) 6= 0. Let p(x) be a monic polynomial which satis es p('(b)) = 0. The roots for p are divided into two groups L = f 1 ; : : : ; k g and M = f 1; : : : ; l g corresponding to the criteria: for each i we have (b ? i ) is not invertible in C and for each j ; (b ? j ) is invertible in C. Finally, there exist exponents r1 ; : : : ; rk and s1 ; : : : ; sl such that
p(x) =
k Y i=1

(x ?

)ri

1 ! 0Y l s A: @ (x ? j )
j

j =1

In order to link properties of p(b) to properties of p('(b)) = 0 we state and prove that for any polynomial q and any c; d in C we have ( ) tr('(q(c)d)) = tr('(q(c))'(d)) = tr(q('(c))'(d)): The proof of ( ) follows from the proof of the special case where q(x) = xs , s 2 N . Let z 2 C then by assumption

8 z 2 C : tr ?' ?(c + zd)s

+1

= tr ('(c) + z'(d))s+1

32

ERIK CHRISTENSEN

and then by comparing terms tr('(cs d)) = tr ('(c)s '(d)) : Having ( ) we get p(b) 2 JC . Since the elements (b ? j ) are invertible in C Q we get for the polynomial q(x) = k=1 (x ? i )r that q(b) 2 JC . By de nition i of JC and by ( ) we then have
i

8 s 2 N : tr ((q('(b)))s ) = tr (' (q(b)s )) = 0:

The matrix q('(b)) is then nilpotent and then for each in ('(b)) there exists an i 2 f1; : : : kg such that = i 2 (b), and we have proved that ('(b)) (b), so '(Ainv ) GLn (C ).

References
1] B. Aupetit, Une generalisation du theoreme de Gleason-Kahane-Zelazko pour les algebres de Banach, Pac. J. Math., 85 (1979), 11-17. 2] J.B. Conway, Functions of one complex variable, Grad. Texts in Math., Springer, 11 1975. 3] A.M. Gleason, A characterization of maximal ideals, J. Analyse Math., 19 (1967), 171-172. 4] J.P. Kahane and W. Zelazko, A characterization of maximal ideals in commutative Banach algebras, Studia Math., 29 (1968), 339-343. 5] E.C. Titchmarsh, The theory of functions, The Clarendon Press, 1932. Received April 19, 1995 and revised September 18, 1995.
Kobenhavns Universitet / Universitetsparken 5 Dk-2100 Copenhagen O, Denmark /