Can. J. Math., Vol. XXVI, No. 4, 1974, pp.

794-799

ON RINGS WITH INVOLUTION

I. N . H E R S T E I N

This paper is dedicated in memory of Adrian Albert.

In this note we prove some results which assert that under certain conditions
the involution on a prime ring must satisfy a form of positive definiteness. As
a consequence of the first of our theorems we obtain a fairly short and simple
proof of a recent theorem of Lanski [3]. In fact, in doing so we actually gener-
alize his result in that we need not avoid the presence of 2-torsion. One can
easily adapt Lanski's original proof, also, to cover the case in which 2-torsion
is present. This result of Lanski has been greatly generalized in a joint work
by Susan Montgomery and ourselves [2].
Let R be a ring, and let * be an involution on R. We let
5 = jx Ç R\x* = x} and K = {x Ç R\x* = — x},
and refer to the elements of S and K as the symmetric and skew elements,
respectively, of R. We shall also use the notation Z, or sometimes Z(R), to
denote the center of R. Finally, we call a ring R a domain, even when it is not
commutative, if ab = 0 in R forces a = 0 or b = 0.
LEMMA 1. Let R be a ring with involution *, and let U 9e 0 be an ideal of R
such that U* = U. If U P\ S = 0, that is, U has no non-zero symmetric elements,
then U3 = 0.
Proof. If 0 7± x e U then x* <E U* = U, hence x + x* £ U C\ S = 0. Thus
x* = —x for every x £ U. In particular, since U C\ S = 0, we have that
2x = 0, for x in U, forces x = 0. Now if x G U, x* = — x whence (x 2 )* = x2.
By our hypothesis U P\ S = 0 we are led to x2 = 0 for every x G U. Since U
is without 2-torsion, we get that U3 = 0.
We now prove
THEOREM 1. Let R be a prime ring with involution * such that no non-zero
element of S is nilpotent. Then either
(1) xx* = 0 in R implies x = 0, or
(2) S C Z and R is an order in F2, the ring of all 2 X 2 matrices over afield F.
Proof. Suppose that xx* = 0 for some x ^ 0 in R; we want to show that
S C Z and R is an order in F2.
Now x*5x C S, and from xx* = 0 every element in x*5x has square 0.
Consequently, from the hypothesis on R, we have x*5x = 0. If r Ç R then

Received October 12, 1972 and in revised form, November 3, 1973. This research was sup-
ported by NSF Grant GP-29269.
794

[Link] Published online by Cambridge University Press

 RINGS WITH INVOLUTION 795

x*(r* + r)x = 0, hence x*r*x = — x*rx, which is to say, x*rx is skew for every
r € R. Because R is a prime ring, for some r G R, k = x*rx ^ 0. The element
& is skew, and k2 = 0. Thus k cannot be symmetric, in consequence of which
the characteristic of R is not 2.
Since 0 = k2 = — kk*, as above, we have that kSk = 0. If s Ç 5, then
ks — sk £ S. Moreover, since &2 = 0, and since 5, s2 are both in 5, and since
kSk = 0, we see that (ks — sk)2 = 0. By our hypothesis on S we must have
that ks — sk = 0 for all s £ S. Therefore k commutes with every element in 5,
the su bring of R generated by 5.
By its definition, S is a subring of R. Moreover, iï s £ S and y Ç R then
sy — y s = sy + y* s — (y + y*) s is in 5. From this it follows that 5 is a Lie
ideal of R. Since R is 2-torsion free and has no nilpotent ideals, by [1, Lemma
1.3], either S C Z or S contains a non-zero ideal U or R. This latter possibility
implies that k centralizes the non-zero ideal U; in a prime ring this forces k
to be in Z. However, since k 5* 0 and k2 = 0, this is in contradiction to the
fact that k is an element in the center of a prime ring. Hence we are left with
the possibility S C Z.
We claim that in this situation R must be an order in F2. Since S C. Z and
the elements of Z are not zero divisors in R, we can localize R at the non-zero
elements of S to obtain a ring
T = {r/s\r e R, s ^ 0 6 5}.
J" is clearly a prime ring with involution; moreover, its non-zero symmetric
elements are invertible in T. We assert that T is simple. If V 9e 0 is an ideal
of T, then U = VV* ^ 0, and U* = U. By Lemma 1, U has a non-zero
symmetric element; this element being invertible in T, we obtain that U = T.
Since V D U,we have that F = 7\ In short, JH is simple. Since the symmetric
elements of T lie in the center of T, by [1, Theorem 1.6] we have that T is
4-dimensional over its center. Since T has zero divisors—recall that xx* = 0—
it is not a division ring. Thus T = F2, the 2 X 2 matrices over a field F. By
construction, R is an order in T. This completes the proof of Theorem 1.
Before proceeding to a skew analogue of Theorem 1, we use Theorem 1 to
derive a result of Lanski [3]. First we dispose of the prime case for his theorem.
LEMMA 2. Let Rbe a prime ring with involution in which ab ^ 0 if a ^ 0 and
b 9e 0 are in S. Then either
(1) R is a domain, or
(2) S C Z and R is an order in F2, the 2 X 2 matrices over a field F.
Proof. Suppose that R is not an order in F2. By Theorem 1, if xx* = 0
then x — 0 in R. Suppose that uv = 0, u, v G R. Then (u*u)(vv*) = 0; since
both u*u and vv* are in 5, either u*u = 0 or w* = 0. Hence either u = 0 or
v = 0. Thus i? must be a domain.
We wish to push Lemma 2 further, to characterize semi-prime rings with

[Link] Published online by Cambridge University Press

 796 I. N. HERSTEIN

involution in which ab 9^ 0 if a 9^ 0 and b 9^ 0 are in S. We already know the

story if R is prime—this is Lemma 2. So we may assume that there are ideals
A ^ 0 and B 9* 0 in R such that AB = 0.
We claim that A*A = 0. We certainly do have A*AB = 0, hence
0 = (A*AB)* = B*A*A. Since R is semi-prime, this relation implies that
A*AB* = 0. Thus (A* A ) (B + B*) = 0. If A*A 9* 0, then, since R is semi-prime,
by Lemma 1 there is an element u 5e 0 £ A*A such that u* = u. Likewise, in
B + B* there is an elements 9* 0 such thatv* = v. From (A*A) (B + B*) = 0
we have uv = 0, contradicting our hypothesis on the elements of S. In short,
whenever AB = 0 with A, B non-zero ideals of R, then A*A = 0. Note that
since R is semi-prime, from A*A = 0 we get i H i * = 0.
Let i f = {x G R\xA = 0} ; M is an ideal of R and M D A*. Since MA = 0,
by the argument above, if* i f = 0. Because R is semi-prime, this yields that
MM* = 0. If Mx = 0 then ^4*x = 0 since M D A*, hence x*A = 0, and so
x* £ M. In other words, M* is precisely the annihilator in R of M.
We claim that R/M* is a domain. First note that M has no non-zero nilpotent
elements. For if u2 = 0, where u G M, then since zm* Ç MM* = 0, and w*w G
APitf = 0, we get (u + u*)2 = 0. Since 5 has no nilpotent elements, u + w* = 0,
and s o n = — u* is in ikf P\ ikf* = 0. We assert further that M is a domain.
For if uv = 0 with u, v Ç jkf, then since (W) 2 = 0, zw = 0. Now (u + u*)
(v + v*) = 0 results because uv = 0, W*A* = {vu)* = 0 and wfl* = 0 and
u*v = 0 since they lie in MM* = 0 and M*M = 0 respectively. Our hypo-
thesis on 5 forces u + u* = 0 or v + v* = 0. Because M P\ M* = 0 these
latter relations yield u = 0 or v = 0. In short, ikf is a domain.
Suppose that xy G if*. If mi, m2 are in Af, then (mix) (ym2) £ M C\ M* = 0.
However, i f is a domain and m^x, ym2 are in M\ therefore m\X = 0 or ym* = 0.
This immediately leads to Mx = 0 or y M = 0; but if y M = 0 then, by the
semi-primeness of R, My = 0. In other words, xy G M* implies Mx = 0 or
My = 0; since M* is the annihilator of M these translate into: xy G M*
implies x G M* or y G if*, which is to say, R/M* is a domain. Similarly,
i ^ / i f is a domain.
Since i f P\ if* = 0, R is a subdirect sum of i ^ / i f and R/M*; moreover,
the involution * on R interchanges the components of this subdirect sum. We
have proved all the pieces of
THEOREM 2 (Lanski). If R is a semi-prime ring with involution such that
ab 9e 0 for a ^ 0, b 9e 0 in Sy then
(1) R is a domain, or
(2) S C Z and R is an order in F2, the 2 X 2 matrices over a field F} or
(3) R is a subdirect sum of a domain and its opposite, with the involution
being the exchange involution.
Note that if R is any ring with involution satisfying the hypotheses of
Theorem 2 on 5, then the nil radical N of R must satisfy Ns = 0 (Lemma 1).
So we could describe R via N3 = 0 and the structure of R/N given in Theorem

[Link] Published online by Cambridge University Press

 RINGS WITH INVOLUTION 797

2. However, a little more can be said. In order to do this, however, we must

strengthen the hypothesis a little. Instead of insisting that certain elements
are not zero divisors in appropriate subsets we need that they are not zero
divisor on R.
THEOREM 3. Let R be a ring with involution, and suppose N, the maximal nil
ideal of R, is not 0. Then, if either
(1) x — x* is not a zero divisor in R for any x such that x — x* 9* 0 or
(2) x + x* is not a zero divisor in R for any x such that x + x* 9* 0, then
R/N is commutative.
Proof. Since N* = N, if x G N then x ± x* G N, so is nilpotent. So if
condition (1) is satisfied, x = x* for all x £ N; if condition (2) is satisfied,
then x = — x* for all x G N. In both cases we get that ay = y*a for all a G N,
y £R.
Thus if y, z G R, then ayz = (yz)*a = z*y*a = z*ay = azy, which is to say,
a(zy — yz) = 0. Hence N annihilates C, the commutator ideal of R. If C = 0,
R must be commutative, hence so is R/N. If C 9* 0, then since NC = 0,
every element of C is a zero divisor in R. Then, as we did for N, we easily
derive that c(yz — zy) = 0 for all c G C, y, z G R. This leads to C2 = 0.
Hence C C N and so R/N is commutative.
We now prove the skew analogue of Theorem 1—actually the hypothesis
we use is a little less restrictive than that of Theorem 1.
THEOREM 4. Let R be a prime ring with involution * such that no non-zero
element of the form x — x* is nilpotent. Then either
(1) R is an order in F2, the 2 X 2 matrices over a field F, or
(2) xx* = 0 in R implies that x = 0.
Proof. Let
K0 = {x — x*\x G R} ;
if Ko = 0 then every element of R must be symmetric; hence R must be
commutative. As a commutative prime ring, R would be an integral domain,
hence conclusion 2 of the theorem would hold. Thus we may assume that
Ko 9*0.
Suppose that xx* = 0 for some x ^ 0 in R. Then every element in x*K0x
has square 0; however, since x*K0x C K0, by our hypothesis on K0 we have
x*K0x = 0. Thus if r G R, then x*rx = x*r*x, that is, x*Rx C S. Since R is
prime, there is an r G R such that 5 = x*rx 9* 0.
Because s £ S and s2 = 0, as above, we arrive at si£0s = 0. Let k G K0;
then ks + sk G i^o- However,
(&s + s&)2 = fofes + sksk + 5&2s + &s2& = sk2s
since si£0s = 0 and s2 = 0. Thus (ks + s&)4 = (s£2s)2 = 0. Since ks + sk G i£o,
we conclude that ks + sk = 0 for all k G i£o-

[Link] Published online by Cambridge University Press

 798 I. N. HERSTEIN

At this point we divide the argument according as the characteristic of R

is or is not 2.
If the characteristic of R is 2, the above discussion tells us that the element
5 commutes with all elements x + x*, x arbitrary in R. From [4, Theorem 21],
it follows easily that either s £ Z or R is an order in a simple algebra Q which
is 4-dimensional over its center F. Since 5 is nilpotent, it cannot be in the
center of a prime ring. Thus R is an order in Q which is simple and 4-dimen-
sional over its center F. Since Q has zero divisors, we have that Q is isomorphic
to F2y the 2 X 2 matrices over the field F. Hence the characteristic 2 case is
settled.
Suppose, then, that the characteristic of R is not 2. In this case, since
sk + ks = 0 for all k Ç K0, we immediately also have that sk + ks = 0 for
all k Ç K. From this we have that 5 commutes with all k\k2 where fei, k2 G K,
that is, s centralizes K2.
Now K2 is a Lie ideal of R (in fact, so is K02) for if ki, k2 G K and y £ R
then (kxk2)y — y(k1k2) = ki(k2y + y*k2) — (kxy* + yki)k2 and so is in K2.
Thus K2, the subring of R generated by K2, is both a subring and a Lie ideal
of R. By, [1, Lemma 1.3], either K2 C Z or K2 contains a non-zero ideal of
R. In this latter possibility, we would have 5 centralizing a non-zero ideal of
R; in a prime ring this forces 5 to be in Z. Since s2 = 0, this is not possible.
Therefore, K2 C Z.
Since K2 C. Z we immediately have that if a, b £ K then \a = nb 5e 0 for
some X, M G 5 P\ Z.
Localize i£ at 5 P\ Z; if <2 is this localization, then Q is prime, has an involu-
tion and the non-zero skew elements of Q are invertible in Q. From this and
the paragraph above, the skew elements of Q are central multiples of a 9e 0
in K. We claim that Q is simple. For, if U ^ 0 is an ideal of Q, then V = UU*
9* 0 satisfies V* = V. If V contains a skew element b 9^ 0, since & is invertible
in Q we get V = Q, and so Î7 = Q. Thus F contains no non-zero skew elements;
thus V is commutative. However, a commutative ideal in a prime ring must
be in the center. Hence again we end up with an invertible element in V,
whence V = Q and so U — Q. Thus Q is simple. Since its skew elements are
1-dimensional over Z(Q), they cannot generate Q. By [1, Theorem 2.2], Q
must be 4-dimensional over a field. Because R is an order in Q, we have
shown the theorem to be correct.
We close the paper with an immediate corollary to Theorem 3. The result
is a sharpening of Theorem 1.
COROLLARY. Let R be a prime ring with involution * such that no non-zero
element of the form x + x* is nilpotent. Then either
(1) xx* = 0 in R implies x = 0, or
(2) S C Z and R is an order in F2l the ring of all 2 X 2 matrices over afield F.
Proof. If the characteristic of R is not 2, the hypothesis of the corollary

[Link] Published online by Cambridge University Press

 RINGS WITH INVOLUTION 799

implies that of Theorem 1, and so the corollary is true merely by applying

Theorem 1.
On the other hand, if the chacteristic of R is 2, the hypothesis of the corollary
reduces to that of Theorem 4, and the result follows as a consequence of
Theorem 4.
REFERENCES

1. I . N . Herstein, Topics in ring theory (Univ. of Chicago Press, Chicago, 1969).

2. I. N. Herstein and Susan Montgomery, Invertible and regular elements in rings with involution,
J. Algebra 25 (1973), 390-400.
3. Charles Lanski, Rings with involution whose symmetric elements are regular, Proc. Amer. Math.
Soc. 83 (1972), 264-270.
4. Charles Lanski and S. Montgomery, Lie structure of prime rings of characteristic 2 (to appear).

University of Chicago,
Chicago, Illinois;
Weizmann Institute of Science,
Rehovot, Israel

[Link] Published online by Cambridge University Press