We know that we can apply [[EROs]] to any augmented matrix into [[REF]]. Suppose the system has $n$ equations and $m$ variables, and let $k$ be the number of non-zero rows in REF. Also suppose the system is consistent: then the REF has no row of the form $[0~0~0~\dots~1]$. {{ ::ge-obs.png?nolink&800 |}} * $k\le n$, because there are only $n$ rows in the whole matrix * $k$ is precisely the number of [[leading variables]]. So $k$ is no bigger $m$, the total number of variables; in symbols, we have $k\le m$. * All the other variables are free variables, so $$ \text{$m-k$ is the number of free variables.} $$ What does this tell us about the set of solutions? For example, how many solutions are there? === Observation 1: free variables and the number of solutions === For consistent systems, this shows that: * //either// $k=m$; * so $m-k=0$ * there are no free variables * the system has one solution and no more * We say it has a **unique solution**. * //or// $k0$ * there is at least one free variable * so the system has **infinitely many solutions** (one for each value of each free variable) * The number of free variables, $m-k$, is called the [[wp>dimension]] of the solution set. === Observation 2: systems with fewer equations than variables === For consistent systems where $n