It’s been another strong week for mathematics in the comic strips. The 15th particularly was a busy enough day I’m going to move its strips off to the next Reading the Comics group. What we have already lets me talk about shapes, and statistics, and what randomness can do for you.
Carol Lay’s Lay Lines for the 11th of October turns the infinite-monkeys thought-experiment into a contest. It’s an intriguing idea. To have the monkey save correct pages foils the pure randomness that makes the experiment so mind-boggling. However, saving partial successes like correct pages is, essentially, how randomness can be harnessed to do work for us. This is normally in fields known, generally, as Monte Carlo methods, named in honor of the famed casinos.
Suppose you have a problem in which it’s hard to find the best answer, but it’s easy to compare whether one answer is better than another. For example, suppose you’re trying to find the shortest path through a very complicated web of interactions. It’s easy to say how long a path is, and easy to say which of two paths is shorter. It’s hard to say you’ve found the shortest. So what you can do is pick a path at random, and take its length. Then make an arbitrary, random change in it. The changed path is either shorter or longer. If the random change makes the path shorter, great! If the random change makes the path longer, then (usually) forget it. Repeat this process and you’ll get, by hoarding incremental improvements and throwing away garbage, your shortest possible path. Or at least close to it.
Properly, you have to sometimes go along with changes that lengthen the path. It might turn out there’s a really short path you can get to if you start out in an unpromising direction. For a monkey-typing problem such as in the comic, there’s no need for that. You can save correct pages and discard the junk.
Geoff Grogan’s Jetpack Junior for the 12th of October, and after, continues the explorations of a tesseract. The strip uses the familiar idea that a tesseract opens up to a vast, nearly infinite space. I’m torn about whether that’s a fair representation. A four-dimensional hypercube is still a finite (hyper)volume, after all. A four-dimensional cube ten feet on each side contains 10,000 hypercubic feet, not infinitely great a (hyper)volume. On the other hand … well, think of how many two-dimensional squares you could fit in a three-dimensional box. A two-dimensional object has no volume — zero measure, in three-dimensional space — so you could fit anything into it. This may be reasonable but it still runs against my intuition, and my sense of what makes for a fair story premise.
Ernie Bushmiller’s Nancy for the 13th of October, originally printed in 1955, describes a couple geometric objects. I have to give Nancy credit for a description of a sphere that’s convincing, even if it isn’t exactly correct. Even if the bubble-gum bubble Nancy were blowing didn’t have a distortion to her mouth, it still sags under gravity. But it’s easy, at least if you already have an intuitive understanding of spheres, to go from the bubble-gum bubble to the ideal sphere. (Homework question: why does Sluggo’s description of an octagon need to specify “a figure with eight sides and eight angles”? Why isn’t specifying a figure with eight sides, or eight angles, be enough?)
Jon Rosenberg’s Scenes From A Multiverse for the 13th of October depicts a playground with kids who’re well-versed in the problems of statistical inference. A “statistically significant sample size” nearly explains itself. It is difficult to draw reliable conclusions from a small sample, because a small sample can be weird. Generally, the difference between the statistics of a sample and the statistics of the broader population it’s drawn from will be smaller the larger the sample is. There are several courses hidden in that “generally” there.
“Selection bias” is one of the courses hidden in that “generally”. A good sample should represent the population fairly. Whatever’s being measured should appear in the sample about as often as it appears in the population. It’s hard to say that’s so, though, before you know what the population is like. A biased selection over-represents some part of the population, or under-represents it, in some way.
“Confirmation bias” is another of the courses. That amounts to putting more trust in evidence that supports what we want to believe, and in discounting evidence against it. People tend to do this, without meaning to fool themselves or anyone else. It’s easiest to do with ambiguous evidence: is the car really running smoother after putting in more expensive spark plugs? Is the dog actually walking more steadily after taking this new arthritis medicine? Has the TV program gotten better since the old show-runner was kicked out? If these can be quantified in some way, and a complete record made, it’s typically easier to resist confirmation bias. But not everything can be quantified, and even so, differences can be subtle, and demand more research than we can afford.
On the 15th, Scenes From A Multiverse did another strip with some mathematical content. It’s about the question of whether it’s possible to determine whether the universe is a computer simulation. But the same ideas apply to questions like whether there could be a multiverse, some other universe than ours. The questions seem superficially to be unanswerable. There are some enthusiastic attempts, based on what things we might conclude. I suspect that the universe is just too small a sample size to draw any good conclusions from, though.
Dan Thompson’s Brevity for the 14th of October is another anthropomorphized-numerals joke.