Let d be a C*-algebra, and ~r the real Banach space of all hermitian elements in ~4 which is partially ordered by the cone dh + = {a'a; aest}. If d contains an identity e, then e also serves as an order-unit for sCh; indeed the order interval [ -e, e] is precisely the unit ball in sr h. If d does not contain an identity then, by a classical theorem of Segal, it has an approximate identity: a monotone increasing net (ez) of elements in dh + of norm less than 1 such that limeaa=a=limaea for each a in ~r It is natural to ask if such (e~) has to be an approximate order-unit, that is, by definition, if the gauge of the union 0 [-ea, e~] of order intervals is the norm of sr h. Unfortunately, this is not the case; for example, in the C*-algebra of all complex null sequences under the pointwise algebraic operations, and conjugation (as involution), if let e,=(1 .... ,1,0,0 .... ), then (e.;n=l,2 .... ) is an approximate identity but not approximate order-unit. However, conversely, we do have the following result (John Pickford has also obtained this result).
Theorem 1. If (ea) is an approximate order-unit for ~h, then it is also an approximate identity for d.Proof Let x~d h of norm 1, and let 5>0. By Segal's theorem (cf. [2, 1.7.2]), there exists uedh + with I[u[[ <1 such that J[x-xu[[ <5. Take 2 0 such that eao>=u and consider all 2 => 2 o. Then e~ > e > u 0_ e-ez __< e-u and so 9
< x(e-e~)x < x(e -u)x,where e is an identity adjoint to d. Hence O< flx(e-ea)xlf < Nx(e-u)xl