nuclear space



A nuclear vector space is a locally convex topological vector space that is as far from being a normed vector space as possible. Any map from a nuclear space into a normed vector space is compact, whence the only normed nuclear spaces are finite dimensional.

Nuclear spaces have very good properties with regard to topological tensor product?s and duality.


To define a nuclear space we need to start with the concept of a nuclear map, first between Banach spaces.

Let EE and FF be Banach spaces. Let (E,F)\mathcal{L}(E,F) be the Banach space of continuous linear maps EFE \to F. Let E *E^* denote the dual Banach space of EE. Let E *˜FE^* \widetilde{\otimes} F denote the completion of the projective tensor product of E *E^* and FF. The bilinear map E *×F(E,F)E^* \times F \to \mathcal{L}(E,F) extends to a continuous linear map E *˜F(E,F)E^* \widetilde{\otimes} F \to \mathcal{L}(E,F) (which might not be injective).


Let EE and FF be Banach spaces. A linear map f:EFf \colon E \to F is nuclear if it lies in the image in (E,F)\mathcal{L}(E,F) of the completion of the projective tensor product E *˜FE^* \widetilde{\otimes} F.

From the notion of nuclear maps between Banach spaces we can define nuclear maps between arbitrary LCTVS. In essence, a linear map between arbitrary LCTVS is nuclear if it factors through a nuclear map of Banach spaces.

To make this precise, we need to recall how to associate Banach spaces to certain subsets of an LCTVS. Let EE be an LCTVS and UEU \subseteq E a convex circled 00-neighbourhood. Then we can define a Banach space E˜ U\widetilde{E}_U as follows: as UU is convex and circled, its Minkowski functional is a semi-norm on EE. The quotient E UE/kerUE_U \coloneqq E/\ker U is therefore a normed vector space. As UU is a 00-neighbourhood, the quotient mapping defines a continuous linear function EE UE \to E_U. We define E˜ U\tilde{E}_U to be the Banach completion of E UE_U.

There is a dual notion. Let FF be an LCTVS and BFB \subseteq F a convex, circled, and bounded subset of FF. Let F BF_B be the span of BB in FF. Then BB is absorbing in F BF_B and so its Minkowski functional is defined. If FF is Hausdorff then BB cannot contain a linear subspace and thus F BF_B is a normed vector space. We cannot complete F BF_B to a Banach space but it might so happen that it is one. As BB is bounded, the inclusion F BFF_B \to F is continuous.

(There is no danger of confusing the two notations since if EE admits a bounded 00-neighbourhood then it is a normed vector space.)

Now we say that a continuous linear map f:EFf \colon E \to F is bounded if for some 00-neighbourhood UU of EE (which we may take to be circled and convex), f(U)f(U) is bounded in FF. In which case, ff factors through a continuous map f U,B:E UF Bf_{U,B} \colon E_U \to F_B where BFB \subseteq F is bounded and contains f(U)f(U).


Let EE and FF be LCTVS. A linear map f:EFf \colon E \to F is nuclear if there exists a convex circled 00-neighbourhood, say UU, in EE and a convex circled bounded, say BB, in FF with F BF_B complete such that f(U)Bf(U) \subseteq B and the associated map f U,B:E˜ UF Bf_{U,B} \colon \tilde{E}_U \to F_B is nuclear.

The following characterisation of nuclear maps is often helpful.


A linear map f:EFf \colon E \to F is nuclear if and only if it is of the form:

u(x)= n=1 λ nf n(x)y n u(x) = \sum_{n=1}^\infty \lambda_n f_n(x) y_n

where n=1 |λ n|<\sum_{n=1}^\infty {|\lambda_n|} \lt \infty, {f n}\{f_n\} is an equicontinuous sequence in E *E^*, and {y n}\{y_n\} is a sequence in FF contained in a convex, circled, bounded subset BB such that F BF_B is complete.

Now that we have the notion of a nuclear map, we can define a nuclear space.


A LCTVS EE is nuclear if it has a base \mathcal{B} of convex circled 00-neighbourhoods such that for VV \in \mathcal{B} the canonical mapping EE˜ VE \to \tilde{E}_V is nuclear.


  1. The following are equivalent:
    1. EE is nuclear,
    2. Every continuous linear map of EE into any Banach space is nuclear,
    3. Every convex, circled 00-neighbourhood UU contains another, say VV, such that the canonical map E˜ VE˜ U\tilde{E}_V \to \tilde{E}_U is nuclear.
  2. Every bounded subset of a nuclear space is precompact.
  3. The completion of a nuclear space is a nuclear space.
  4. A nuclear space is a projective limit of p\ell^p spaces (in particular, of Hilbert spaces).
  5. Nuclearity is inherited by the following constructions: subspaces, separated quotients, arbitrary products, locally convex direct sum of a countable family, projective limits, countable inductive limits.
  6. The projective tensor product (and its completion) of two nuclear spaces is nuclear.


  1. Any finite dimensional vector space is nuclear,
  2. For a finite dimensional smooth manifold MM, C (M)C^\infty(M) is nuclear,
  3. The space of rapidly decreasing functions on n\mathbb{R}^n is nuclear,
  4. The product an arbitrary number of copies of \mathbb{R} is nuclear,
  5. The direct sum of a countable number of copies of \mathbb{R} is nuclear,
  6. The direct sum of \mathbb{R}-copies of \mathbb{R} is not nuclear.


  • A. Grothendieck, Produits tensoriels topologiques et espaces nucléaires, Memoirs of the Amer. Math. Soc. 16, 190 pp. and 140 pp. (1955).
  • Alexander Grothendieck, Résumé des résultats essentiels dans la théorie des produits tensoriels topologiques et des espaces nucléaires, Annales de l’institut Fourier 4 (1952), p. 73-112, numdam

Last revised on June 9, 2013 at 12:31:35. See the history of this page for a list of all contributions to it.