Today’s topic is an always rich one. It was suggested by aajohannas, who so far as I know has’t got an active blog or other project. If I’m mistaken please let me know. I’m glad to mention the creative works of people hanging around my blog.
An old Sydney Harris cartoon I probably won’t be able to find a copy of before this publishes. A couple people gather around an old fanfold-paper printer. On the printout is the sequence “1 … 2 … 3 … 4 … 5 … ” The caption: ‘Bizarre sequence of computer-generated random numbers’.
Randomness feels familiar. It feels knowable. It means surprise, unpredictability. The upending of patterns. The obliteration of structure. I imagine there are sociologists who’d say it’s what defines Modernity. It’s hard to avoid noticing that the first great scientific theories that embrace unpredictability — evolution and thermodynamics — came to public awareness at the same time impressionism came to arts, and the subconscious mind came to psychology. It’s grown since then. Quantum mechanics is built on unpredictable specifics. Chaos theory tells us even if we could predict statistics it would do us no good. Randomness feels familiar, even necessary. Even desirable. A certain type of nerd thinks eagerly of the Singularity, the point past which no social interactions are predictable anymore. We live in randomness.
And yet … it is hard to find randomness. At least to be sure we have found it. We might choose between options we find ambivalent by tossing a coin. This seems random. But anyone who was six years old and trying to cheat a sibling knows ways around that. Drop the coin without spinning it, from a half-inch above the table, and you know the outcome, all the way through to the sibling’s punching you. When we’re older and can be made to be better sports we’re fairer about it. We toss the coin and give it a spin. There’s no way we could predict the outcome. Unless we knew just how strong a toss we gave it, and how fast it spun, and how the mass of the coin was distributed. … Really, if we knew enough, our tossed coin would be as predictably as the coin we dropped as a six-year-old. At least unless we tossed in some chaotic way, where each throw would be deterministic, but we couldn’t usefully make a prediction.
Our instinctive idea of what randomness must be is flawed. That shouldn’t surprise. Our instinctive idea of anything is flawed. But randomness gives us trouble. It’s obvious, for example, that randomly selected things should have no pattern. But then how is that reasonable? If we draw letters from the alphabet at random, we should expect sometimes to get some cute pattern like ‘aaaaa’ or ‘qwertyuiop’ or the works of Shakespeare. Perhaps we mean we shouldn’t get patterns any more often than we would expect. All right; how often is that?
We can make tests. Some of them are obvious. Take something that generates possibly-random results. Look up how probable each of those outcomes is. Then run off a bunch of outcomes. Do we get about as many of each result as we should expect? Probability tells us we should get as close as we like to the expected frequency if we let the random process run long enough. If this doesn’t happen, great! We can conclude we don’t really have something random.
We can do more tests. Some of them are brilliantly clever. Suppose there’s a way to order the results. Since mathematicians usually want numbers, putting them in order is easy to do. If they’re not, there’s usually a way to match results to numbers. You’ll see me slide here into talking about random numbers as though that were the same as random results. But if I can distinguish different outcomes, then I can label them. If I can label them, I can use numbers as labels. If the order of the numbers doesn’t matter — should “red” be a 1 or a 2? Should “green” be a 3 or an 8? — then, fine; any order is good.
There are 120 ways to order five distinct things. So generate lots of sets of, say, five numbers. What order are they in? There’s 120 possibilities. Do each of the possibilities turn up as often as expected? If they don’t, great! We can conclude we don’t really have something random.
I can go on. There are many tests which will let us say something isn’t a truly random sequence. They’ll allow for something like Sydney Harris’s peculiar sequence of random numbers. Mostly by supposing that if we let it run long enough the sequence would stop. But these all rule out random number generators. Do we have any that rule them in? That say yes, this generates randomness?
I don’t know of any. I suspect there can’t be any, on the grounds that a test of a thousand or a thousand million or a thousand million quadrillion numbers can’t assure us the generator won’t break down next time we use it. If we knew the algorithm by which the random numbers were generated — oh, but there we’re foiled before we can start. An algorithm is the instructions of how to do a thing. How can an instruction tell us how to do a thing that can’t be predicted?
Algorithms seem, briefly, to offer a way to tell whether we do have a good random sequence, though. We can describe patterns. A strong pattern is easy to describe, the way a familiar story is easy to reference. A weak pattern, a random one, is hard to describe. It’s like a dream, in which you can just list events. So we can call random something which can’t be described any more efficiently than just giving a list of all the results. But how do we know that can’t be done? 7, 7, 2, 4, 5, 3, 8, 5, 0, 9 looks like a pretty good set of digits, whole numbers from 0 through 9. I’ll bet not more than one in ten of you guesses correctly what the next digit in the sequence is. Unless you’ve noticed that these are the digits in the square root of π, so that the next couple digits have to be 0, 5, 5, and 1.
We know, on theoretical grounds, that we have randomness all around us. Quantum mechanics depends on it. If we need truly random numbers we can set a sensor. It will turn the arrival of cosmic rays, or the decay of radioactive atoms, or the sighing of a material flexing in the heat into numbers. We trust we gather these and process them in a way that doesn’t spoil their unpredictability. To what end?
That is, why do we care about randomness? Especially why should mathematicians care? The image of mathematics is that it is a series of logical deductions. That is, things known to be true because they follow from premises known to be true. Where can randomness fit?
One answer, one close to my heart, is called Monte Carlo methods. These are techniques that find approximate answers to questions. They do well when exact answers are too hard for us to find. They use random numbers to approximate answers and, often, to make approximate answers better. This demands computations. The field didn’t really exist before computers, although there are some neat forebears. I mean the Buffon needle problem, which lets you calculate the digits of π about as slowly as you could hope to do.
Another, linked to Monte Carlo methods, is stochastic geometry. “Stochastic” is the word mathematicians attach to things when they feel they’ve said “random” too often, or in an undignified manner. Stochastic geometery is what we can know about shapes when there’s randomness about how the shapes are formed. This sounds like it’d be too weak a subject to study. That it’s built on relatively weak assumptions means it describes things in many fields, though. It can be seen in understanding how forests grow. How to find structures inside images. How to place cell phone towers. Why materials should act like they do instead of some other way. Why galaxies cluster.
There’s also a stochastic calculus, a bit of calculus with randomness added. This is useful for understanding systems where some persistent unpredictable behavior is there. It comes, if I understand the histories of this right, from studying the ways molecules will move around in weird zig-zagging twists. They do this even when there is no overall flow, just a fluid at a fixed temperature. It too has surprising applications. Without the assumption that some prices of things are regularly jostled by arbitrary and unpredictable forces, and the treatment of that by stochastic calculus methods, we wouldn’t have nearly the ability to hedge investments against weird chaotic events. This would be a bad thing, I am told by people with more sophisticated investments than I have. I personally own like ten shares of the Tootsie Roll corporation and am working my way to a $2.00 rebate check from Boyer.
Given that we need randomness, but don’t know how to get it — or at least don’t know how to be sure we have it — what is there to do? We accept our failings and make do with “quasirandom numbers”. We find some process that generates numbers which look about like random numbers should. These have failings. Most important is that if we could predict them. They’re random like “the date Easter will fall on” is random. The date Easter will fall is not at all random; it’s defined by a specific and humanly knowable formula. But if the only information you have is that this year, Easter fell on the 1st of April (Gregorian computus), you don’t have much guidance to whether this coming year it’ll be on the 7th, 14th, or 21st of April the next year. Most notably, quasirandom number generators will tend to repeat after enough numbers are drawn. If we know we won’t need enough numbers to see a repetition, though? Another stereotype of the mathematician is that of a person who demands exactness. It is often more true to say she is looking for an answer good enough. We are usually all right with a merely good enough quasirandomness.
Boyer candies — Mallo Cups, most famously, although I more like the peanut butter Smoothies — come with a cardboard card backing. Each card has two play money “coins”, of values from 5 cents to 50 cents. These can be gathered up for a rebate check or for various prizes. Whether your coin is 5 cents, 10, 25, or 50 cents … well, there’s no way to tell, before you open the package. It’s, so far as you can tell, randomness.
My next A To Z post should be available at this link. It’s coming Tuesday and should be the letter ‘S’.