Projective tensor product

From HandWiki

In functional analysis, an area of mathematics, the projective tensor product of two locally convex topological vector spaces is a natural topological vector space structure on their tensor product. Namely, given locally convex topological vector spaces X and Y, the projective topology, or π-topology, on XY is the strongest topology which makes XY a locally convex topological vector space such that the canonical map (x,y)xy (from X×Y to XY) is continuous. When equipped with this topology, XY is denoted XπY and called the projective tensor product of X and Y.

Definitions

Let X and Y be locally convex topological vector spaces. Their projective tensor product XπY is the unique locally convex topological vector space with underlying vector space XY having the following universal property:[1]

For any locally convex topological vector space Z, if ΦZ is the canonical map from the vector space of bilinear maps X×YZ to the vector space of linear maps XYZ; then the image of the restriction of ΦZ to the continuous bilinear maps is the space of continuous linear maps XπYZ.

When the topologies of X and Y are induced by seminorms, the topology of XπY is induced by seminorms constructed from those on X and Y as follows. If p is a seminorm on X, and q is a seminorm on Y, define their tensor product pq to be the seminorm on XY given by (pq)(b)=infr>0,brWr for all b in XY, where W is the balanced convex hull of the set {xy:p(x)1,q(y)1}. The projective topology on XY is generated by the collection of such tensor products of the seminorms on X and Y.[2][1] When X and Y are normed spaces, this definition applied to the norms on X and Y gives a norm, called the projective norm, on XY which generates the projective topology.[3]

Properties

Throughout, all spaces are assumed to be locally convex. The symbol X^πY denotes the completion of the projective tensor product of X and Y.

  • If X and Y are both Hausdorff then so is XπY;[3] if X and Y are Fréchet spaces then XπY is barelled.[4]
  • For any two continuous linear operators u1:X1Y1 and u2:X2Y2, their tensor product (as linear maps) u1u2:X1πX2Y1πY2 is continuous.[5]
  • In general, the projective tensor product does not respect subspaces (e.g. if Z is a vector subspace of X then the TVS ZπY has in general a coarser topology than the subspace topology inherited from XπY).[6]
  • If E and F are complemented subspaces of X and Y, respectively, then EF is a complemented vector subspace of XπY and the projective norm on EπF is equivalent to the projective norm on XπY restricted to the subspace EF. Furthermore, if X and F are complemented by projections of norm 1, then EF is complemented by a projection of norm 1.[6]
  • Let E and F be vector subspaces of the Banach spaces X and Y, respectively. Then E^F is a TVS-subspace of X^πY if and only if every bounded bilinear form on E×F extends to a continuous bilinear form on X×Y with the same norm.[7]

Completion

In general, the space XπY is not complete, even if both X and Y are complete (in fact, if X and Y are both infinite-dimensional Banach spaces then XπY is necessarily not complete[8]). However, XπY can always be linearly embedded as a dense vector subspace of some complete locally convex TVS, which is generally denoted by X^πY.

The continuous dual space of X^πY is the same as that of XπY, namely, the space of continuous bilinear forms B(X,Y).[9]

Grothendieck's representation of elements in the completion

In a Hausdorff locally convex space X, a sequence (xi)i=1 in X is absolutely convergent if i=1p(xi)< for every continuous seminorm p on X.[10] We write x=i=1xi if the sequence of partial sums (i=1nxi)n=1 converges to x in X.[10]

The following fundamental result in the theory of topological tensor products is due to Alexander Grothendieck.[11]

Theorem — Let X and Y be metrizable locally convex TVSs and let zX^πY. Then z is the sum of an absolutely convergent series z=i=1λixiyi where i=1|λi|<, and (xi)i=1 and (yi)i=1 are null sequences in X and Y, respectively.

The next theorem shows that it is possible to make the representation of z independent of the sequences (xi)i=1 and (yi)i=1.

Theorem[12] — Let X and Y be Fréchet spaces and let U (resp. V) be a balanced open neighborhood of the origin in X (resp. in Y). Let K0 be a compact subset of the convex balanced hull of UV:={uv:uU,vV}. There exists a compact subset K1 of the unit ball in 1 and sequences (xi)i=1 and (yi)i=1 contained in U and V, respectively, converging to the origin such that for every zK0 there exists some (λi)i=1K1 such that z=i=1λixiyi.

Topology of bi-bounded convergence

Let 𝔅X and 𝔅Y denote the families of all bounded subsets of X and Y, respectively. Since the continuous dual space of X^πY is the space of continuous bilinear forms B(X,Y), we can place on B(X,Y) the topology of uniform convergence on sets in 𝔅X×𝔅Y, which is also called the topology of bi-bounded convergence. This topology is coarser than the strong topology on B(X,Y), and in (Grothendieck 1955), Alexander Grothendieck was interested in when these two topologies were identical. This is equivalent to the problem: Given a bounded subset BX^Y, do there exist bounded subsets B1X and B2Y such that B is a subset of the closed convex hull of B1B2:={b1b2:b1B1,b2B2}?

Grothendieck proved that these topologies are equal when X and Y are both Banach spaces or both are DF-spaces (a class of spaces introduced by Grothendieck[13]). They are also equal when both spaces are Fréchet with one of them being nuclear.[9]

Strong dual and bidual

Let X be a locally convex topological vector space and let X be its continuous dual space. Alexander Grothendieck characterized the strong dual and bidual for certain situations:

Theorem[14] (Grothendieck) — Let N and Y be locally convex topological vector spaces with N nuclear. Assume that both N and Y are Fréchet spaces, or else that they are both DF-spaces. Then, denoting strong dual spaces with a subscripted b:

  1. The strong dual of N^πY can be identified with Nb^πYb;
  2. The bidual of N^πY can be identified with N^πY;
  3. If Y is reflexive then N^πY (and hence Nb^πYb) is a reflexive space;
  4. Every separately continuous bilinear form on Nb×Yb is continuous;
  5. Let L(Xb,Y) be the space of bounded linear maps from Xb to Y. Then, its strong dual can be identified with Nb^πYb, so in particular if Y is reflexive then so is Lb(Xb,Y).

Examples

  • For (X,𝒜,μ) a measure space, let L1 be the real Lebesgue space L1(μ); let E be a real Banach space. Let LE1 be the completion of the space of simple functions XE, modulo the subspace of functions XE whose pointwise norms, considered as functions X, have integral 0 with respect to μ. Then LE1 is isometrically isomorphic to L1^πE.[15]

See also

Citations

  1. 1.0 1.1 Trèves 2006, p. 438.
  2. Trèves 2006, p. 435.
  3. 3.0 3.1 Trèves 2006, p. 437.
  4. Trèves 2006, p. 445.
  5. Trèves 2006, p. 439.
  6. 6.0 6.1 Ryan 2002, p. 18.
  7. Ryan 2002, p. 24.
  8. Ryan 2002, p. 43.
  9. 9.0 9.1 Schaefer & Wolff 1999, p. 173.
  10. 10.0 10.1 Schaefer & Wolff 1999, p. 120.
  11. Schaefer & Wolff 1999, p. 94.
  12. Trèves 2006, pp. 459-460.
  13. Schaefer & Wolff 1999, p. 154.
  14. Schaefer & Wolff 1999, pp. 175-176.
  15. Schaefer & Wolff 1999, p. 95.

References

Further reading

  • Diestel, Joe (2008). The metric theory of tensor products : Grothendieck's résumé revisited. Providence, R.I: American Mathematical Society. ISBN 978-0-8218-4440-3. OCLC 185095773. 
  • Grothendieck, Grothendieck (1966) (in fr). Produits tensoriels topologiques et espaces nucléaires. Providence: American Mathematical Society. ISBN 0-8218-1216-5. OCLC 1315788. 
  • Pietsch, Albrecht (1972). Nuclear locally convex spaces. Berlin, New York: Springer-Verlag. ISBN 0-387-05644-0. OCLC 539541. 
  • Wong (1979). Schwartz spaces, nuclear spaces, and tensor products. Berlin New York: Springer-Verlag. ISBN 3-540-09513-6. OCLC 5126158.