2.16.5 Tensor Infinite Algebraic Limitation

Tensor Infinite Algebraic Limitation is the collection of specific ways in which identities and constructions that hold automatically for tensors over finite-dimensional vector spaces fail, or require additional hypotheses, once the underlying vector space is infinite-dimensional, covering the breakdown of reflexivity, the mismatch between algebraic and continuous duals, and the gap between the algebraic tensor product and the completed tensor products used in analysis. These are not mere technical inconveniences; each limitation marks a point where a genuinely different mathematical object or extra structure becomes necessary to recover something resembling the finite-dimensional statement.

Failure of Reflexivity

The Finite-Dimensional Fact

In finite dimension, every vector space V is naturally isomorphic to its double dual V**, the dual of its dual space, via the evaluation map that sends v to the functional "evaluate at v" on V*. This isomorphism is what allows type (1, 1) tensors to be identified with linear operators, and more generally allows contravariant and covariant tensor slots to be treated symmetrically.

The Infinite-Dimensional Breakdown

For an infinite-dimensional vector space, the evaluation map V → V** is still injective but generally fails to be surjective; V embeds into V** as a proper subspace, and elements of V** that are not evaluation functionals genuinely exist. A space for which the evaluation map is surjective is called reflexive, and reflexivity is a special, non-automatic property that must be separately verified (Hilbert spaces are reflexive; many Banach spaces, such as the space of bounded sequences, are not).

Algebraic Dual Versus Continuous Dual

Two Different Objects Sharing One Name

Once V carries a topology, typically from a norm, two different notions of "dual space" arise: the algebraic dual, consisting of all linear functionals, and the continuous (or topological) dual, consisting only of those linear functionals that are continuous with respect to the topology. In finite dimension every linear functional is automatically continuous, so the two duals coincide; in infinite dimension the algebraic dual is strictly larger than the continuous dual.

Consequences for Tensor Constructions

Tensor constructions in analysis, such as those used to represent bilinear forms or operators, are almost always built from the continuous dual, not the algebraic dual, precisely because the continuous dual is the one compatible with the topology and with limiting operations. A tensor theory that uses the algebraic dual by default will produce objects too large and too poorly behaved for the analytic applications that motivated the construction in the first place.

The Algebraic Tensor Product Is Usually Too Small

What the Algebraic Construction Misses

The algebraic tensor product V ⊗ W of two infinite-dimensional normed spaces consists only of finite sums of simple tensors, but many natural bilinear objects, such as the kernel of an integral operator or a general bounded operator between Hilbert spaces, correspond to elements that can only be approximated by, but are not equal to, any finite sum of simple tensors.

Completions Recover the Missing Objects

Equipping V ⊗ W with a suitable norm, such as the projective norm, the injective norm, or (for Hilbert spaces) the norm induced by extending the inner product, and then completing with respect to that norm produces a strictly larger space in which such limits genuinely exist. Different choices of norm generally produce different, inequivalent completions, so unlike the finite-dimensional case, "the" tensor product of two infinite-dimensional normed spaces is not unique without specifying which norm is used.

where the left side is the algebraic tensor product and the right side denotes its completion with respect to a chosen cross norm α, generally a proper containment.

Distinguishing Operator Classes That Coincide in Finite Dimension

Trace-Class, Hilbert-Schmidt, and Bounded Operators

In finite dimension, every linear operator has a well-defined trace, is automatically compact, and belongs to every reasonable operator class simultaneously. In infinite dimension, operators split into strictly nested classes, trace-class operators, Hilbert-Schmidt operators, compact operators, and bounded operators, each corresponding to a different completion of the tensor product V ⊗ V* under a different norm, and an operator lying in one class need not lie in a smaller one.

This strict nesting, with no reverse inclusions, has no counterpart in finite dimension, where all four classes are identical.

Diagram of the Limitation Hierarchy

Summary of the Underlying Pattern

A Single Recurring Cause

Each of these limitations traces back to the same underlying cause: finite-dimensional linear algebra allows dual and double-dual identifications, uniqueness of the tensor product, and equality of operator classes because sums, limits, and dualities are all automatically well-behaved when only finitely many dimensions are involved. Once infinitely many dimensions are present, every one of these automatic identifications must instead be established, or explicitly replaced by a topologically refined substitute, on a case-by-case basis.