Today’s term ended up being a free choice. Nobody found anything appealing in the D’s to ask about. That’s all right.
I’m still looking for topics for the letters G through M, excluding L, if you’d like in on those letters.
And for my own sake, please check out the Playful Mathematics Education Blog Carnival, #121, if you haven’t already.
I have to specify. There’s a bunch of mathematics concepts called `distribution’. Some of them are linked. Some of them are just called that because we don’t have a better word. Like, what else would you call multiplying the sum of something? I want to describe a distribution that comes to us in probability and in statistics. Through these it runs through modern physics, as well as truly difficult sciences like sociology and economics.
We get to distributions through random variables. These are variables that might be any one of multiple possible values. There might be as few as two options. There might be a finite number of possibilities. There might be infinitely many. They might be numbers. At the risk of sounding unimaginative, they often are. We’re always interested in measuring things. And we’re used to measuring them in numbers.
What makes random variables hard to deal with is that, if we’re playing by the rules, we never know what it is. Once we get through (high school) algebra we’re comfortable working with an ‘x’ whose value we don’t know. But that’s because we trust that, if we really cared, we would find out what it is. Or we would know that it’s a ‘dummy variable’, whose value is unimportant but gets us to something that is. A random variable is different. Its value matters, but we can’t know what it is.
Instead we get a distribution. This is a function which gives us information about what the outcomes are, and how likely they are. There are different ways to organize this data. If whoever’s talking about it doesn’t say just what they’re doing, bet on it being a “probability distribution function”. This follows slightly different rules based on whether the range of values is discrete or continuous, but the idea is roughly the same. Every possible outcome has a probability at least zero but not more than one. The total probability over every possible outcome is exactly one. There’s rules about the probability of two distinct outcomes happening. Stuff like that.
Distributions are interesting enough when they’re about fixed things. In learning probability this is stuff like hands of cards or totals of die rolls or numbers of snowstorms in the season. Fun enough. These get to be more personal when we take a census, or otherwise sample things that people do. There’s something wondrous in knowing that while, say, you might not know how long a commute your neighbor has, you know there’s an 80 percent change it’s between 15 and 25 minutes (or whatever). It’s also good for urban planners to know.
It gets exciting when we look at how distributions can change. It’s hard not to think of that as “changing over time”. (You could make a fair argument that “change” is “time”.) But it doesn’t have to. We can take a function with a domain that contains all the possible values in the distribution, and a range that’s something else. The image of the distribution is some new distribution. (Trusting that the function doesn’t do something naughty.) These functions — these mappings — might reflect nothing more than relabelling, going from (say) a distribution of “false and true” values to one of “-5 and 5” values instead. They might reflect regathering data; say, going from the distribution of a die’s outcomes of “1, 2, 3, 4, 5, or 6” to something simpler, like, “less than two, exactly two, or more than two”. Or they might reflect how something does change in time. They’re all mappings; they’re all ways to change what a distribution represents.
These mappings turn up in statistical mechanics. Processes will change the distribution of positions and momentums and electric charges and whatever else the things moving around do. It’s hard to learn. At least my first instinct was to try to warm up to it by doing a couple test cases. Pick specific values for the random variables and see how they change. This can help build confidence that one’s calculating correctly. Maybe give some idea of what sorts of behaviors to expect.
But it’s calculating the wrong thing. You need to look at the distribution as a specific thing, and how that changes. It’s a change of view. It’s like the change in view from thinking of a position as an x- and y- and maybe z-coordinate to thinking of position as a vector. (Which, I realize now, gave me slightly similar difficulties in thinking of what to do for any particular calculation.)
Distributions can change in time, just the way that — in simpler physics — positions might change. Distributions might stabilize, forming an equilibrium. This can mean that everything’s found a place to stop and rest. That will never happen for any interesting problem. What you might get is an equilibrium like the rings of Saturn. Everything’s moving, everything’s changing, but the overall shape stays the same. (Roughly.)
There are many specifically named distributions. They represent patterns that turn up all the time. The binomial distribution, for example, which represents what to expect if you have a lot of examples of something that can be one of two values each. The Poisson distribution, for representing how likely something that could happen any time (or any place) will happen in a particular span of time (or space). The normal distribution, also called the Gaussian distribution, which describes everything that isn’t trying to be difficult. There are like 400 billion dozen more named ones, each really good at describing particular kinds of problems. But they’re all distributions.