complementary subspace

A complement of a subspace U in V is another subspace W that pairs with U to fill the whole space directly: V = U ⊕ W. Concretely, W must satisfy U + W = V and U ∩ W = {0}. You can picture U as a chosen plane through the origin and W as any line through the origin that is not in that plane — together they reconstruct all of space, with no redundancy.

The reassuring theorem is that complements always exist (in finite dimensions, and with Zorn's lemma in general): take any basis of U, extend it to a basis of V, and the new basis vectors span a complement. So no subspace is ever 'stuck' — there is always a partner that completes it to a direct sum. This is one of the quiet workhorses that makes abstract linear algebra so frictionless compared to, say, the theory of modules, where complements can fail.

The crucial subtlety is non-uniqueness. A subspace has MANY complements, not one. The x-axis in R^2 is complemented by the y-axis, but equally by the line y = x, or y = 2x, or any line through the origin other than the x-axis itself. There is no canonical complement; choosing one is a genuine choice. The only thing forced is the DIMENSION of the complement: dim W = dim V - dim U always, even though W itself is not determined.

Complements are the same data as projections. Each decomposition V = U ⊕ W gives a projection P onto U along W (send u + w to u), and every projection arises this way from its image and kernel. So 'pick a complement' and 'pick a projection onto U' are two phrasings of one act. This dictionary is why complements quietly underlie least squares, conditional expectation, and every place you split a vector into a 'wanted' and 'discarded' part.