Today’s glossary term is one that turns up in many areas of mathematics. But these all share some connotations. So I mean to start with the easiest one to understand.
Suppose you are somewhere. Most of us are. Where is something else?
That isn’t hard to answer if conditions are right. If we’re allowed to point and the something else is in sight, we’re done. It’s when pointing and following the line of sight breaks down that we’re in trouble. We’re also in trouble if we want to say how to get from that something to yet another spot. How can we guide someone from one point to another?
We have a good answer from everyday life. We can impose some order, some direction, on space. We’re familiar with this from the cardinal directions. We say where things on the surface of the Earth are by how far they are north or south, east or west, from something else. The scheme breaks down a bit if we’re at the North or the South pole exactly, but there we can fall back on pointing.
When we start using north and south and east and west as directions we are choosing basis vectors. Vectors are directions in how far to move and in what direction. Suppose we have two vectors that aren’t pointing in the same direction. Then we can describe any two-dimensional movement using them. We can say “go this far in the direction of the first vector and also that far in the direction of the second vector”. With the cardinal directions, we consider north and east, or east and south, or south and west, or west and north to be a pair of vectors going in different directions.
(North and south, in this context, are the same thing. “Go twenty paces north” says the same thing as “go negative twenty paces south”. Most mathematicians don’t pull this sort of stunt when telling you how to get somewhere unless they’re trying to be funny without succeeding.)
A basis vector is just a direction, and distance in that direction, that we’ve decided to be a reference for telling different points in space apart. A basis set, or basis, is the collection of all the basis vectors we need. What do we need? We need enough basis vectors to get to all the points in whatever space we’re working with.
(If you are going to ask about doesn’t “east” point in different directions as we go around the surface of the Earth, you’re doing very well. Please pretend we never move so far from where we start that anyone could notice the difference. If you can’t do that, please pretend the Earth has been smooshed into a huge flat square with north at one end and we’re only just now noticing.)
We are free to choose whatever basis vectors we like. The worst that can happen if we choose a lousy basis is that we have to write out more things than we otherwise would. Our work won’t be less true, it’ll just be more tedious. But there are some properties that often make for a good basis.
One is that the basis should relate to the problem you’re doing. Suppose you were in one of mathematicians’ favorite places, midtown Manhattan. There is a compelling grid here of streets running north-south and avenues running east-west. (Broadway we ignore as an implementation error retained for reasons of backwards compatibility.) Well, we pretend they run north-south and east-west. They’re actually a good bit clockwise of north-south and east-west. They do that to better match the geography of the island. A “north” street runs about parallel to the way Manhattan’s long dimension runs. In the circumstance, it would be daft to describe directions by true north or true east. We would say to go so many streets “north” and so many avenues “east”.
Purely mathematical problems aren’t concerned with streets and avenues. But there will often be preferred directions. Mathematicians often look at the way a process alters shapes or redirects forces. There’ll be some directions where the alterations are biggest. There’ll be some where the alterations are shortest. Those directions are probably good choices for a basis. They stand out as important.
We also tend to like basis vectors that are a unit length. That is, their size is 1 in some convenient unit. That’s for the same reason it’s easier to say how expensive something is if it costs 45 dollars instead of nine five-dollar bills. Or if you’re told it was 180 quarter-dollars. The length of your basis vector is just a scaling factor. But the more factors you have to work with the more likely you are to misunderstand something.
And we tend to like basis vectors that are perpendicular to one another. They don’t have to be. But if they are then it’s easier to divide up our work. We can study each direction separately. Mathematicians tend to like techniques that let us divide problems up into smaller ones that we can study separately.
I’ve described basis sets using vectors. They have intuitive appeal. It’s easy to understand directions of things in space. But the idea carries across into other things. For example, we can build functions out of other functions. So we can choose a set of basis functions. We can multiply them by real numbers (scalars) and add them together. This makes whatever function we’re interested in into a kind of weighted average of basis functions.
Why do that? Well, again, we often study processes that change shapes and directions. If we choose a basis well, though, the process changes the basis vectors in easy to describe ways. And many interesting processes let us describe the changing of an arbitrary function as the weighted sum of the changes in the basis vectors. By solving a couple of simple problems we get the ability to solve every interesting problem.
We can even define something that works like the angle between functions. And something that works a lot like perpendicularity for functions.
And this carries on to other mathematical constructs. We look for ways to impose some order, some direction, on whatever structure we’re looking at. We’re often successful, and can work with unreal things using tools like those that let us find our place in a city.