I get to start this week with another request. Today’s Leap Day Mathematics A To Z term is a famous one, and one that I remember terrifying me in the earliest days of high school. The request comes from Gaurish, chief author of the Gaurish4Math blog.
Lewis Carroll didn’t like the matrix. Well, Charles Dodgson, anyway. And it isn’t that he disliked matrices particularly. He believed it was a bad use of a word. “Surely,” he wrote, “[ matrix ] means rather the mould, or form, into which algebraical quantities may be introduced, than an actual assemblage of such quantities”. He might have had etymology on his side. The word meant the place where something was developed, the source of something else. History has outvoted him, and his preferred “block”. The first mathematicians to use the word “matrix” were interested in things derived from the matrix. So for them, the matrix was the source of something else.
What we mean by a matrix is a collection of some number of rows and columns. Inside each individual row and column is some mathematical entity. We call this an element. Elements are almost always real numbers. When they’re not real numbers they’re complex-valued numbers. (I’m sure somebody, somewhere has created matrices with something else as elements. You’ll never see these freaks.)
Matrices work a lot like vectors do. We can add them together. We can multiply them by real- or complex-valued numbers, called scalars. But we can do other things with them. We can define multiplication, at least sometimes. The definition looks like a lot of work, but it represents something useful that way. And for square matrices, ones with equal numbers of rows and columns, we can find other useful stuff. We give that stuff wonderful names like traces and determinants and eigenvalues and eigenvectors and such.
One of the big uses of matrices is to represent a mapping. A matrix can describe how points in a domain map to points in a range. Properly, a matrix made up of real numbers can only describe what are called linear mappings. These are ones that turn the domain into the range by stretching or squeezing down or rotating the whole domain the same amount. A mapping might follow different rules in different regions, but that’s all right. We can write a matrix that approximates the original mapping, at least in some areas. We do this in the same way, and for pretty much the same reason, we can approximate a real and complicated curve with a bunch of straight lines. Or the way we can approximate a complicated surface with a bunch of triangular plates.
We can compound mappings. That is, we can start with a domain and a mapping, and find the image of that domain. We can then use a mapping again and find the image of the image of that domain. The matrix that describes this mapping-of-a-mapping is the one you get by multiplying the matrix of the first mapping and the matrix of the second mapping together. This is why we define matrix multiplication the odd way we do. Mapping are that useful, and matrices are that tied to them.
I wrote about some of the uses of matrices in a Set Tour essay. That was based on a use of matrices in physics. We can describe the changing of a physical system with a mapping. And we can understand equilibriums, states where a system doesn’t change, by looking at the matrix that approximates what the mapping does near but not exactly on the equilibrium.
But there are other uses of matrices. Many of them have nothing to do with mappings or physical systems or anything. For example, we have graph theory. A graph, here, means a bunch of points, “vertices”, connected by curves, “edges”. Many interesting properties of graphs depend on how many other vertices each vertex is connected to. And this is well-represented by a matrix. Index your vertices. Then create a matrix. If vertex number 1 connects to vertex number 2, put a ‘1’ in the first row, second column. If vertex number 1 connects to vertex number 3, put a ‘1’ in the first row, third column. If vertex number 2 isn’t connected to vertex number 3, put a ‘0’ in the second row, third column. And so on.
We don’t have to use ones and zeroes. A “network” is a kind of graph where there’s some cost associated with each edge. We can put that cost, that number, into the matrix. Studying the matrix of a graph or network can tell us things that aren’t obvious from looking at the drawing.