Running it backwards
Every system of linear equations is secretly the same question. Take
and write it as a single transformation applied to an unknown vector:
Now read it aloud in the language you have built: which starting vector, after this motion, lands on ?
When you can just play it in reverse
If the motion did not destroy anything, there is another linear transformation that undoes it, called the inverse and written . Run then and every vector returns home, so together they do nothing at all:
The identity matrix is the do-nothing motion: its columns say lands on , and on .
With an inverse in hand, the system is solved by playing the motion backwards from the target: . Our example has , so it works and gives . Check it: and .
When it is impossible
Watch the visual collapse the plane onto a line. Many different starting vectors now land on the same point.
Ask the reverse question at one of those points and it has no single answer — several inputs are equally valid. A function cannot return several answers, so no inverse transformation exists.
This is exactly the case , exactly the case where rank dropped. Three descriptions, one phenomenon.
And the collapse leaves a specific fingerprint: a whole set of vectors got crushed onto the origin. Naming that set is the last piece of Act III.