Cosets: parallel copies of a subspace
Fix a subspace U of V. For each vector v, the [[coset|coset]] v + U = { v + u : u in U } is the parallel translate of U passing through v. These cosets are the affine subspaces parallel to U: U itself is the coset 0 + U (passes through the origin), and any other coset is a shifted, off-origin copy.
V = R^2, U = the x-axis = { (t, 0) }.
coset (0,0) + U = the x-axis itself (y = 0)
coset (0,3) + U = horizontal line y = 3
coset (5,3) + U = ALSO the line y = 3 (since (5,3)-(0,3) in U)
Key fact: v + U = w + U <=> v - w in U.
Each coset is labeled by its height y; the x-coordinate is forgotten.The quotient space: cosets become the new vectors
Now the bold move: declare each coset to be a single new vector. The set of all cosets is the [[quotient-space|quotient space]] V/U. Add and scale them by reaching down to representatives: (v + U) + (w + U) = (v + w) + U and a(v + U) = (av) + U. The zero vector of V/U is the coset U itself — so *inside the quotient, everything in U has been collapsed to zero*.
Dimension and what the quotient 'measures'
dim(V/U) = dim(V) - dim(U) (the codimension of U) V = R^2, U = x-axis (dim 1): dim(V/U) = 2 - 1 = 1. V/U is a line: its 'vectors' are the heights y, exactly R. Compare a complement W with V = U (+) W: the quotient map W -> V/U, w |-> w + U is an isomorphism. So V/U behaves like ANY complement of U -- but without you having to CHOOSE one.
This is the real virtue of quotients over complements. A complement W is a *choice* (recall: never unique); the quotient V/U is *canonical* — built once and for all from U, no arbitrary pick required. That is why quotients power the isomorphism theorems: the kernel of a linear map is collapsed, and what survives, V/ker, is forced to match the image. Next guide makes that 'forced to match' precise.