Dear Uncle Colin,

I’ve got a matrix, and I’m not afraid to use it. It’s (3542)

Apparently, it has invariant lines. Those, I’m afraid of.

How do I find them?

-- Terrors About Rank, Safely Knowing Inverses

Hi, TARSKI!

An invariant line of a transformation is one where every point on the line is mapped to a point on the line – possibly the same point.

We can write that algebraically as Mx=X, where x=(xmx+c) and X=(XmX+c). Our job is to find the possible values of m and c.

So, for this example, we have:

(3542)(xmx+c)=(XmX+c)

And now it gets messy. We have two equations which hold for any value of x:

3x5(mx+c)=X

4x+2(mx+c)=mX+c

Substituting for X in the second equation, we have:

4x+2(mx+c)=m(3x5(mx+c))+c

… which tidies up to …

(2m4)x+2c=(5m2+3m)x+(5m+1)c

… or …

(5m2m4)x+(5m+1)c=0, for all x (*).

Aside, on the difference between variables and constants

There are three letters in that equation, m, c and x. For a long while, I thought “letters are letters, right? bits of algebraic furniture you can move around.” This isn’t true. In fact, there are two different flavours of letter here.

The m and the c are constants: numbers with specific values that don’t change. The x, on the other hand, is a variable, a letter that can mean anything we happen to find convenient.

Back to work

Considering x=0, this can only be true if either 5m+1=0 or c=0, so let’s treat those two cases separately.

If m=15, then equation (*) becomes 185x=0, which is not true for all x; m=15 is therefore not a solution.

Instead, if c=0, the equation becomes (5m2m4)x=0, which is true if x=0 (which it doesn’t, generally), or if (5m2m4)=0, which it can; it factorises as (5m+4)(m1)=0, so m=45 and m=1 are both possible answers when c=0.

So the two equations of invariant lines are y=45x and y=x.

Just to check: if we multiply M by (5,4), we get (35,28), which is also on the line y=45x. Similarly, if we apply the matrix to (1,1), we get (2,2) – again, it lies on the given line.

(It turns out that these invariant lines are related in this case to the eigenvectors of the matrix, but sh. Let’s not scare anyone off.)

-- Uncle Colin

* Edited 2019-06-08 to fix an arithmetic error. Thanks to Tom for finding it!