For a Hilbert space , the linear span of the matrix-coefficient functionals is a predual of . The unit ball is compact in the resulting weak operator topology by the Tychonoff theorem, and the compact norming dual-pair criterion identifies with the dual of that span.
For each place of a number field , let be the corresponding completion, and for finite let be its valuation ring. The adele ring is the restricted product
Its restricted product topology has basic open sets , where every is open and at all but finitely many finite places.
First take . The neighborhood
of zero meets the diagonal copy of only in zero: a rational number lying in every is an integer, and the only integer in the indicated real interval is zero. Thus is discrete in .
Every rational adele is congruent modulo to an element of
Indeed, the finitely many negative -adic principal parts can be removed simultaneously by subtracting a rational number, using the Chinese remainder theorem; subtracting an integer then moves the real component into . This set is compact by the compactness of , the compactness of every , and the Tychonoff theorem. Its image covers the quotient, so is compact.
Now choose a -basis of the number field . The given topological isomorphism
identifies the additive pair with . A finite product of discrete subgroups is discrete, and
is compact.
The idele group is
where the distinguished subgroup at a finite place is . It carries the corresponding restricted product topology on the idele group. The inclusion is continuous: the inverse image of a basic adelic open set is locally a product of open subsets of , and outside finitely many places every idele component already belongs to .
It is not a homeomorphism onto its image. Let be the th rational prime and define the idele to equal at the place over and everywhere else. In the adele topology, : the difference is zero at every fixed place once is large, while at the single moving place. In the idele topology the sequence does not converge to , because the open neighborhood
contains no : its -component has positive valuation and is not a unit. Hence the inverse of on its image is not continuous.
Give each nonempty finite set the discrete topology. The product space is compact by the Tychonoff theorem. For each , the compatibility condition defines a closed subset .
These sets have the finite intersection property. Indeed, for finitely many conditions choose an index above every index occurring in them, choose any , and use the transition maps from to define all required coordinates; choose the remaining coordinates arbitrarily. Compactness therefore gives
This is the nonemptiness theorem for inverse limits of finite sets.
Goldstine theorem states that the canonical image of the closed unit ball of a normed vector space is weak-star dense in .
The Banach-Alaoglu theorem states that is compact in the weak-star topology. To prove it, map each to its values in
Every factor is compact, so Tychonoff theorem makes compact. The image of is cut out by the closed linearity conditions
and is therefore closed in . The product topology restricted to this image is exactly pointwise convergence on , namely the weak-star topology. Hence the ball is compact.
The estimate
shows . For nonzero , the rank-one operator
has norm one and attains equality. The zero cases are immediate.
Embed the operator unit ball into
by . The product is compact by Tychonoff theorem. A pointwise limit of these coordinates is a bilinear form satisfying
By the stated representation theorem for bounded bilinear forms, for a unique operator with . The image is therefore closed and compact. Its product topology is precisely the weak operator topology .
The linear span separates operators, and the preceding compactness lets part (a) identify isometrically with . Thus is a dual Banach space. This is the operator predual from matrix coefficients.