Nuclear Spaces and Tensor Products — Mathematics Knowledge System

Overview

A nuclear space is a locally convex topological vector space with exceptional approximation properties: every continuous linear map from it to any Banach space is nuclear (i.e., can be written as a sum of rank-one operators with summable coefficients). Nuclear spaces arise naturally in analysis -- the space of smooth functions C^inf(M), the Schwartz space S(R), and spaces of analytic functions are nuclear. They are exactly the spaces for which the injective and projective tensor products coincide, making them fundamental in the theory of distributions and quantum field theory.

Intuition

Nuclearity is a strong compactness condition: nuclear spaces behave 'like finite-dimensional spaces' in that any continuous linear map factors through a Hilbert space in a very efficient way (the factoring map is Hilbert-Schmidt). Practically: if you can represent sections of a space by rapidly converging series of smooth functions, the space is likely nuclear. The key consequence is that tensor products are well-behaved: for nuclear spaces, there is a canonical (unique) complete tensor product, enabling distribution theory to work cleanly.

Formal Definition

Definition

A locally convex space E is nuclear if for every continuous seminorm p on E there exists a stronger continuous seminorm q such that the natural map E_q -> E_p (between the completion of E/ker(q) and E/ker(p)) is a nuclear operator. Equivalently, E is nuclear if every continuous linear map from E to any Banach space is nuclear. A nuclear operator T: E -> F between Banach spaces is one expressible as T(x) = sum_n lambda_n * f_n(x) * y_n with (lambda_n) in l^1, (f_n) equibounded in E*, and (y_n) bounded in F.

T: E \to F \text{ nuclear} \iff T(x) = \sum_n \lambda_n f_n(x) y_n,\quad \sum_n |\lambda_n| < \infty

Nuclear operator

E \text{ nuclear} \iff E \hat{\otimes}_\pi F \cong E \hat{\otimes}_\varepsilon F \text{ for all locally convex } F

Nuclearity via tensor norms

\|T\|_1 = \inf \left\{ \sum_n |\lambda_n| : T = \sum_n \lambda_n f_n \otimes y_n \right\}

Nuclear norm

Notation

Notation	Meaning
$E \hat{\otimes}_\pi F$	Projective tensor product completion
$E \hat{\otimes}_\varepsilon F$	Injective tensor product completion
$\\|T\\|_1$	Nuclear norm of operator T

Theorems

Theorem 1: Grothendieck's Nuclear Tensor Product Theorem

A locally convex space E is nuclear if and only if for every locally convex space F, the natural map E tensor_pi F -> E tensor_eps F (from the projective to the injective tensor product) is an isomorphism. For nuclear E, there is a unique reasonable tensor product completion, denoted E otimes F.

Theorem 2: Nuclearity of Function Spaces

The following are nuclear spaces: the Schwartz space S(R^n), the space of smooth functions C^inf(M) on a compact manifold M, the space of analytic functions on an open subset of C^n, and the space of entire functions of exponential type. The spaces L^p (for 1 <= p <= infinity) and C(K) for infinite compact K are NOT nuclear.

Theorem 3: Kernel Theorem

For nuclear spaces E, F, every continuous bilinear form B: E x F -> C corresponds to a unique element K in (E otimes F)* = E* otimes F*. Equivalently, every continuous linear map T: E -> F* is an integral operator: T corresponds to a distribution kernel K in (E otimes F)* such that (T*phi)(psi) = K(phi otimes psi).

Worked Examples

1
S(R) is topologised by the family of seminorms ||f||_{m,n} = sup_x |x^m f^{(n)}(x)| for m, n >= 0.
2
For any (m,n), the identity map S(R) -> S_{m,n} (Hilbert space completion under ||·||_{m,n}) factors through S(R) -> S_{m+2,n+2} -> S_{m,n}, where the second map is Hilbert-Schmidt (hence nuclear).
$\mathrm{id}: \mathcal{S}(\mathbb{R}) \to \mathcal{S}_{m,n} \text{ factors through a Hilbert-Schmidt map}$
3
The Hilbert-Schmidt property follows because the eigenvalues of the harmonic oscillator decay polynomially: lambda_k ~ k^{-r} for any r, which is square-summable.

✓ Answer

S(R) is nuclear: the connecting maps between its Hilbert space seminorm completions are Hilbert-Schmidt (in particular nuclear).

Practice Problems

Hardfree response

Explain the significance of the Schwartz kernel theorem and how nuclearity of S(R^n) is essential for it.

Common Mistakes

Common Mistake

Thinking nuclear spaces must be Banach spaces.

Nuclear BANACH spaces must be finite-dimensional. Nuclear spaces are locally convex but generally NOT Banach spaces -- they are typically Frechet or LF spaces like S(R) or C^inf(M).

Quiz

Which of the following is a nuclear space?

L^2(R)The Schwartz space S(R)l^inf (bounded sequences)C([0,1]) with the sup norm

Historical Background

Nuclear spaces were introduced by Alexandre Grothendieck in his 1955 doctoral thesis 'Produits tensoriels topologiques et espaces nucleaires', submitted in Nancy. The thesis, written while Grothendieck was working in algebraic topology and before his move to algebraic geometry, revolutionised functional analysis by providing a categorical framework for tensor products of locally convex spaces and characterising nuclear spaces as those where the tensor norms coincide.

1955
Grothendieck introduces nuclear spaces in his thesis on tensor products
Alexander Grothendieck
1960s
Gelfand-Shilov develop the theory of nuclear spaces in the context of generalised functions
Israel Gelfand, Georgiy Shilov

Summary

A nuclear space is a locally convex space where all continuous linear maps to Banach spaces are nuclear operators.
Key examples: S(R^n), C^inf(M), spaces of analytic functions. Key non-examples: all infinite-dimensional Banach spaces.
Nuclearity is equivalent to the projective and injective tensor products coinciding.
The Schwartz kernel theorem relies on nuclearity of S(R^n) to identify continuous operators with distribution kernels.

References

BookGrothendieck, A. Produits Tensoriels Topologiques et Espaces Nucleaires. AMS Memoirs, 1955.
BookTreves, F. Topological Vector Spaces, Distributions and Kernels. Academic Press, 1967.
WebsiteWikipedia -- Nuclear space