Schauder bases and unconditionality

A sequence of nonzero vectors is a Schauder basis of a Banach space X when every x has a unique scalar expansion whose partial sums converge to x in norm. The associated coefficient functionals are continuous, and the partial-sum operators P_N are bounded projections onto the first N basis vectors. Uniform boundedness gives a common bound on their operator norms, because P_Nx converges for every fixed x. These projections need not be orthogonal, and their norms need not equal one. The supremum of their norms is the basis constant.

Having a Schauder basis forces separability, since finite rational combinations of basis vectors are dense, using rational real and imaginary parts in the complex case. The converse fails: separable Banach spaces without the approximation property cannot have a Schauder basis. Indeed, the uniformly bounded partial-sum projections converge strongly to the identity, and a finite-net argument upgrades this to uniform convergence on each compact set. Thus a basis gives the bounded approximation property, connecting coordinate expansions to the previous lesson without claiming operator-norm convergence on the entire unit ball.

A Schauder basis is unconditional when every permutation of each vector’s expansion converges in norm to the same vector. An equivalent criterion is a uniform bound on the operators selecting arbitrary finite subsets of basis coordinates. Such bounds prevent a collection of far-out terms from becoming large merely because other terms that cancel them were omitted. Sign-change formulations give another equivalent test. Unconditional convergence of vector series does not generally imply summability of the term norms; the finite-dimensional scalar absolute-convergence intuition does not transfer unchanged.

The canonical coordinate bases of ℓᵖ for 1≤p<∞ and of c₀ are unconditional with finite-selection projection norm at most one. The same coordinate vectors do not form a Schauder basis of ℓ∞, since their partial sums cannot uniformly approximate the constant-one sequence. A basis can also be Schauder but conditional, even on c₀, because a different ordered coordinate system can build cancellation into its representation. Thus unconditionality is a property of the chosen basis, not merely of the Banach space’s underlying set or its completeness.

In c₀, let s_n be one on the first n coordinates and zero afterward. For x tending to zero, define coefficients a_n=x_n−x_(n+1). The first N terms of the resulting expansion equal x_j−x_(N+1) in coordinates j≤N and zero afterward. The error is exactly the tail supremum beyond N, so it tends to zero. Comparing successive coordinates also proves uniqueness of the coefficients. This establishes a Schauder basis directly, without appealing to an abstract existence theorem.

Each initial-segment projection has norm at most two, and suitable vectors attain two. Arbitrary coordinate selections behave differently. For a finite alternating vector with entries one, minus one, one, minus one, the odd summing coefficients are all two; selecting just those coefficients accumulates a large first coordinate. Their omitted even companions supplied the cancellation that kept the original vector bounded. The resulting unbounded selection norms prove conditionality. This mechanism explains why stable ordered truncation does not authorize arbitrary pruning or rearrangement of a Banach-space expansion.