Column space collects feasible outputs $\mathbf{b}$

Fix a matrix $A$. The column space is every output you can reach: all vectors $\mathbf{b}$ such that $A\mathbf{x}=\mathbf{b}$ for some $\mathbf{x}$ . Equivalently, it is the span of the columns of $A$, thought of as arrows in the codomain.

Existence of solutions is a geometric membership question: is $\mathbf{b}$ inside that span? If not, no amount of algebra will manufacture an exact solution. Rank equals the dimension of the column space: how many independent output directions the map actually uses.

Recipe language helps: columns are ingredients, linear combinations are recipes, and the column space is the menu of dishes you can cook. Coordinates in $\mathbf{x}$ are recipe amounts. Picture $\mathbf{b}$ as a target arrow and ask whether it lies in the span of the column arrows .

Rank-nullity for $A:\mathbb{R}^n\to\mathbb{R}^m$ states $\mathrm{rank}(A)+\mathrm{nullity}(A)=n$: domain dimension splits into visible output directions plus invisible inputs that map to zero. If $\mathbf{b}$ lies outside the column space, no $\mathbf{x}$ can reach it; if it lies inside and the null space is nontrivial, infinitely many $\mathbf{x}$ share the same output. Rank is the dimension of that reachable menu, not the number of columns printed on the page .