Anyone hoping for an answer besides the 29th of February either suspects I’m doing some clickbait thing, maybe talking about that time Sweden didn’t quite make the transition from the Julian to the Gregorian calendar, or realizes I’m talking about days of the week. Are 29ths of February more likely to be a Sunday, a Monday, what?

The reason this is a question at all is that the Gregorian calendar has this very slight, but real, bias. Some days of the week are more likely for a year to start on than other days are. This gives us the phenomenon where the 13th of months are slightly more likely to be Fridays than any other day of the week. Here “likely” reflects that, if we do not know a specific month and year, then we can’t say which of the seven days the calendar’s rules give us for the date of the 13th.

Or, for that matter, if we don’t know which leap year we’re thinking of. There are 97 of them every 400 years. Since 97 things can’t be uniformly spread across the seven days of the week, how are they spread?

This is what computers are for. You’ve seen me do this for the date of Easter and for the date of (US) Thanksgiving. Using the ‘weekday’ function in Octave (a Matlab clone) I checked. In any 400-year span of the Gregorian calendar — and the calendar recycles every 400 years, so that’s as much as we need — we will see this distribution:

Leap Day will be a

this many times

Sunday

13

Monday

15

Tuesday

13

Wednesday

15

Thursday

13

Friday

14

Saturday

14

in 400 years

Through to 2100, though, the calendar is going to follow a 28-year period. So this will be the last Saturday leap day until 2048. The next several ones will be Thursday, Tuesday, Sunday, Friday, Wednesday, and Monday.

Last week was another light week of work from Comic Strip Master Command. One could fairly argue that nothing is worth my attention. Except … one comic strip got onto the calendar. And that, my friends, is demanding I pay attention. Because the comic strip got multiple things wrong. And then the comments on GoComics got it more wrong. Got things wrong to the point that I could not be sure people weren’t trolling each other. I know how nerds work. They do this. It’s not pretty. So since I have the responsibility to correct strangers online I’ll focus a bit on that.

Robb Armstrong’s JumpStart for the 13th starts off all right. The early Roman calendar had ten months, December the tenth of them. This was a calendar that didn’t try to cover the whole year. It just started in spring and ran into early winter and that was it. This may seem baffling to us moderns, but it is, I promise you, the least confusing aspect of the Roman calendar. This may seem less strange if you think of the Roman calendar as like a sports team’s calendar, or a playhouse’s schedule of shows, or a timeline for a particular complicated event. There are just some fallow months that don’t need mention.

Things go wrong with Rob’s claim that December will have five Saturdays, five Sundays, and five Mondays. December 2019 will have no such thing. It has four Saturdays. There are five Sundays, Mondays, and Tuesdays. From Crunchy’s response it sounds like Joe’s run across some Internet Dubious Science Folklore. You know, where you see a claim that (like) Saturn will be larger in the sky than anytime since the glaciers receded or something. And as you’d expect, it’s gotten a bit out of date. December 2018 had five Saturdays, Sundays, and Mondays. So did December 2012. And December 2007.

And as this shows, that’s not a rare thing. Any month with 31 days will have five of some three days in the week. August 2019, for example, has five Thursdays, Fridays, and Saturdays. October 2019 will have five Tuesdays, Wednesdays, and Thursdays. This we can show by the pigeonhole principle. And there are seven months each with 31 days in every year.

It’s not every year that has some month with five Saturdays, Sundays, and Mondays in it. 2024 will not, for example. But a lot of years do. I’m not sure why December gets singled out for attention here. From the setup about December having long ago been the tenth month, I guess it’s some attempt to link the fives of the weekend days to the ten of the month number. But we get this kind of December about every five or six years.

This 823 years stuff, now that’s just gibberish. The Gregorian calendar has its wonders and mysteries yes. None of them have anything to do with 823 years. Here, people in the comments got really bad at explaining what was going on.

So. There are fourteen different … let me call them year plans, available to the Gregorian calendar. January can start on a Sunday when it is a leap year. Or January can start on a Sunday when it is not a leap year. January can start on a Monday when it is a leap year. January can start on a Monday when it is not a leap year. And so on. So there are fourteen possible arrangements of the twelve months of the year, what days of the week the twentieth of January and the thirtieth of December can occur on. The incautious might think this means there’s a period of fourteen years in the calendar. This comes from misapplying the pigeonhole principle.

Here’s the trouble. January 2019 started on a Tuesday. This implies that January 2020 starts on a Wednesday. January 2025 also starts on a Wednesday. But January 2024 starts on a Monday. You start to see the pattern. If this is not a leap year, the next year starts one day of the week later than this one. If this is a leap year, the next year starts two days of the week later. This is all a slightly annoying pattern, but it means that, typically, it takes 28 years to get back where you started. January 2019 started on Tuesday; January 2020 on Wednesday, and January 2021 on Friday. the same will hold for January 2047 and 2048 and 2049. There are other successive years that will start on Tuesday and Wednesday and Friday before that.

Except.

The important difference between the Julian and the Gregorian calendars is century years. 1900. 2000. 2100. These are all leap years by the Julian calendar reckoning. Most of them are not, by the Gregorian. Only century years divisible by 400 are. 2000 was a leap year; 2400 will be. 1900 was not; 2100 will not be, by the Gregorian scheme.

These exceptions to the leap-year-every-four-years pattern mess things up. The 28-year-period does not work if it stretches across a non-leap-year century year. By the way, if you have a friend who’s a programmer who has to deal with calendars? That friend hates being a programmer who has to deal with calendars.

There is still a period. It’s just a longer period. Happily the Gregorian calendar has a period of 400 years. The whole sequence of year patterns from 2000 through 2019 will reappear, 2400 through 2419. 2800 through 2819. 3200 through 3219.

(Whether they were also the year patterns for 1600 through 1619 depends on where you are. Countries which adopted the Gregorian calendar promptly? Yes. Countries which held out against it, such as Turkey or the United Kingdom? No. Other places? Other, possibly quite complicated, stories. If you ask your computer for the 1619 calendar it may well look nothing like 2019’s, and that’s because it is showing the Julian rather than Gregorian calendar.)

Except.

This is all in reference to the days of the week. The date of Easter, and all of the movable holidays tied to Easter, is on a completely different cycle. Easter is set by … oh, dear. Well, it’s supposed to be a simple enough idea: the Sunday after the first spring full moon. It uses a notional moon that’s less difficult to predict than the real one. It’s still a bit of a mess. The date of Easter is periodic again, yes. But the period is crazy long. It would take 5,700,000 years to complete its cycle on the Gregorian calendar. It never will. Never try to predict Easter. It won’t go well. Don’t believe anything amazing you read about Easter online.

Michael Jantze’s The Norm (Classics) for the 15th is much less trouble. It uses some mathematics to represent things being easy and things being hard. Easy’s represented with arithmetic. Hard is represented with the calculations of quantum mechanics. Which, oddly, look very much like arithmetic. even has fewer symbols than has. But the symbols mean different abstract things. In a quantum mechanics context, ‘A’ and ‘B’ represent — well, possibly matrices. More likely operators. Operators work a lot like functions and I’m going to skip discussing the ways they don’t. Multiplying operators together — B times A, here — works by using the range of one function as the domain of the other. Like, imagine ‘B’ means ‘take the square of’ and ‘A’ means ‘take the sine of’. Then ‘BA’ would mean ‘take the square of the sine of’ (something). The fun part is the ‘AB’ would mean ‘take the sine of the square of’ (something). Which is fun because most of the time, those won’t have the same value. We accept that, mathematically. It turns out to work well for some quantum mechanics properties, even though it doesn’t work like regular arithmetic. So holds complexity, or at least strangeness, in its few symbols.

There were some more comic strips which just mentioned mathematics in passing.

Brian Boychuk and Ron Boychuk’s The Chuckle Brothers rerun for the 11th has a blackboard of mathematics used to represent deep thinking. Also, it I think, the colorist didn’t realize that they were standing in front of a blackboard. You can see mathematicians doing work in several colors, either to convey information in shorthand or because they had several colors of chalk. Not this way, though.

Mark Leiknes’s Cow and Boy rerun for the 16th mentions “being good at math” as something to respect cows for. The comic’s just this past week started over from its beginning. If you’re interested in deeply weird and long-since cancelled comics this is as good a chance to jump on as you can get.

That’s the mathematically-themed comic strips for last week. All my Reading the Comics essays should be at this link. I’ve traditionally run at least one essay a week on Sunday. But recently that’s moved to Tuesday for no truly compelling reason. That seems like it’s working for me, though. I may stick with it. If you do have an opinion about Sunday versus Tuesday please let me know.

Don’t let me know on Twitter. I continue to have this problem where Twitter won’t load on Safari. I don’t know why. I’m this close to trying it out on a different web browser.

And, again, I’m planning a fresh A To Z sequence. It’s never to early to think of mathematics topics that I might explain. I should probably have already started writing some. But you’ll know the official announcement when it comes. It’ll have art and everything.

If I’d started pondering the question a week earlier I’d have a nice timely post. Too bad. Shouldn’t wait nearly a year to use this one, though.

My love and I got talking about early and late Easters. We know that we’re all but certainly not going to be alive to see the earliest possible Easter, at least not unless the rule for setting the date of Easter changes. Easter can be as early as the 22nd of March or as late as the 25th of April. Nobody presently alive has seen a 22nd of March Easter; the last one was in 1818. Nobody presently alive will; the next will be 2285. The last time Easter was its latest date was 1943; the next time will be 2038. I know people who’ve seen the one in 1943 and hope to make it at least through 2038.

But that invites the question: what dates are most likely to be Easter? What ones are least? In a sense the question is nonsense. The rules establishing Easter and the Gregorian calendar are known. To speak of the “chance” of a particular day being Easter is like asking the probability that Grover Cleveland was president of the United States in 1894. Technically there’s a probability distribution there. But it’s different in some way from asking the chance of rolling at least a nine on a pair of dice.

But as with the question about what day is most likely to be Thanksgiving we can make the question sensible. We have to take the question to mean “given a month and day, and no information about what year it is, what is the chance that this as Easter?” (I’m still not quite happy with that formulation. I’d be open to a more careful phrasing, if someone’s got one.)

When we’ve got that, though, we can tackle the problem. We could do as I did for working out what days are most likely to be Thanksgiving. Run through all the possible configurations of the calendar, tally how often each of the days in the range is Easter, and see what comes up most often. There’s a hassle here. Working out the date of Easter follows a rule, yes. The rule is that it’s the first Sunday after the first full moon after the spring equinox. There are wrinkles, mostly because the Moon is complicated. A notional Moon that’s a little more predictable gets used instead. There are algorithms you can use to work out when Easter is. They all look like some kind of trick being used to put something over on you. No matter. They seem to work, as far as we know. I found some Matlab code that uses the Easter-computing routine that Karl Friedrich Gauss developed and that’ll do.

Problem. The Moon and the Earth follow cycles around the sun, yes. Wait long enough and the positions of the Earth and Moon and Sun. This takes 532 years and is known as the Paschal Cycle. In the Julian calendar Easter this year is the same date it was in the year 1485, and the same it will be in 2549. It’s no particular problem to set a computer program to run a calculation, even a tedious one, 532 times. But it’s not meaningful like that either.

The problem is the Julian calendar repeats itself every 28 years, which fits nicely with the Paschal Cycle. The Gregorian calendar, with different rules about how to handle century years like 1900 and 2100, repeats itself only every 400 years. So it takes much longer to complete the cycle and get Earth, Moon, and calendar date back to the same position. To fully account for all the related cycles would take 5,700,000 years, estimates Duncan Steel in Marking Time: The Epic Quest To Invent The Perfect Calendar.

Write code to calculate Easter on a range of years and you can do that, of course. It’s no harder to calculate the dates of Easter for six million years than it is for six hundred years. It just takes longer to finish. The problem is that it is meaningless to do so. Over the course of a mere(!) 26,000 years the precession of the Earth’s axes will change the times of the seasons completely. If we still use the Gregorian calendar there will be a time that late September is the start of the Northern Hemisphere’s spring, and another time that early February is the heart of the Canadian summer. Within five thousand years we will have to change the calendar, change the rule for computing Easter, or change the idea of it as happening in Europe’s early spring. To calculate a date for Easter of the year 5,002,017 is to waste energy.

We probably don’t need it anyway, though. The differences between any blocks of 532 years are, I’m going to guess, minor things. I would be surprised if the frequency of any date’s appearance changed more than a quarter of a percent. That might scramble the rankings of dates if we have several nearly-as-common dates, but it won’t be much.

So let me do that. Here’s a table of how often each particular calendar date appears as Easter from the years 2000 to 5000, inclusive. And I don’t believe that by the year we would call 5000 we’ll still have the same calendar and Easter and expectations of Easter all together, so I’m comfortable overlooking that. Indeed, I expect we’ll have some different calendar or Easter or expectation of Easter by the year 4985 at the latest.

For this enormous date range, though, here’s the frequency of Easters on each possible date:

Date

Number Of Occurrences, 2000 – 5000

Probability Of Occurence

22 March

12

0.400%

23 March

17

0.566%

24 March

41

1.366%

25 March

74

2.466%

26 March

75

2.499%

27 March

68

2.266%

28 March

90

2.999%

29 March

110

3.665%

30 March

114

3.799%

31 March

99

3.299%

1 April

87

2.899%

2 April

83

2.766%

3 April

106

3.532%

4 April

112

3.732%

5 April

110

3.665%

6 April

92

3.066%

7 April

86

2.866%

8 April

98

3.266%

9 April

112

3.732%

10 April

114

3.799%

11 April

96

3.199%

12 April

88

2.932%

13 April

90

2.999%

14 April

108

3.599%

15 April

117

3.899%

16 April

104

3.466%

17 April

90

2.999%

18 April

93

3.099%

19 April

114

3.799%

20 April

116

3.865%

21 April

93

3.099%

22 April

60

1.999%

23 April

46

1.533%

24 April

57

1.899%

25 April

29

0.966%

If I haven’t missed anything, this indicates that the 15th of April is the most likely date for Easter, with the 20th close behind and the 10th and 14th hardly rare. The least probable date is the 22nd of March, with the 23rd of March and the 25th of April almost as unlikely.

And since the date range does affect the results, here’s a smaller sampling, one closer fit to the dates of anyone alive to read this as I publish. For the years 1925 through 2100 the appearance of each Easter date are:

Date

Number Of Occurrences, 1925 – 2100

Probability Of Occurence

22 March

0

0.000%

23 March

1

0.568%

24 March

1

0.568%

25 March

3

1.705%

26 March

6

3.409%

27 March

3

1.705%

28 March

5

2.841%

29 March

6

3.409%

30 March

7

3.977%

31 March

7

3.977%

1 April

6

3.409%

2 April

4

2.273%

3 April

6

3.409%

4 April

6

3.409%

5 April

7

3.977%

6 April

7

3.977%

7 April

4

2.273%

8 April

4

2.273%

9 April

6

3.409%

10 April

7

3.977%

11 April

7

3.977%

12 April

7

3.977%

13 April

4

2.273%

14 April

6

3.409%

15 April

7

3.977%

16 April

6

3.409%

17 April

7

3.977%

18 April

6

3.409%

19 April

6

3.409%

20 April

6

3.409%

21 April

7

3.977%

22 April

5

2.841%

23 April

2

1.136%

24 April

2

1.136%

25 April

2

1.136%

If we take this as the “working lifespan” of our common experience then the 22nd of March is the least likely Easter we’ll see, as we never do. The 23rd and 24th are the next least likely Easter. There’s a ten-way tie for the most common date of Easter, if I haven’t missed one or more. But the 30th and 31st of March, and the 5th, 6th, 10th, 11th, 12th, 15th, 17th, and 21st of April each turn up seven times in this range.

The Julian calendar Easter dates are different and perhaps I’ll look at that sometime.

September 1999 was a heck of a month you maybe remember. There that all that excitement of the Moon being blasted out of orbit thanks to the nuclear waste pile up there getting tipped over or something. And that was just as we were getting over the final new episode of Mystery Science Theater 3000‘s first airing. That episode was number 1003, Merlin’s Shop of Mystical Wonders, which aired a month after the season finale because of one of those broadcast rights tangles that the show always suffered through.

Time moves on, and strange things happen, and show co-creator and first host Joel Hodgson got together a Kickstarter and a Netflix deal. The show’s Season Eleven is supposed to air starting the 14th of April, this year. The natural question: how long will we go, then, between new episodes of Mystery Science Theater 3000? Or more generally, how do you work out how long it is between two dates?

The answer is dear Lord under no circumstances try to work this out yourself. I’m sorry to be so firm. But the Gregorian calendar grew out of a bunch of different weird influences. It’s just hard to keep track of all the different 31- and 30-day months between two events. And then February is all sorts of extra complications. It’s especially tricky if the interval spans a century year, like 2000, since the majority of those are not leap years, except that the one century year I’m likely to experience was. And then if your interval happens to cross the time the local region switched from the Julian to the Gregorian calendar —

So my answer is don’t ever try to work this out yourself. Never. Just refuse the problem if you’re given it. If you’re a consultant charge an extra hundred dollars for even hearing the problem.

All right, but what if you really absolutely must know for some reason? I only know one good answer. Convert the start and the end dates of your interval into Julian Dates and subtract one from the other. I mean subtract the smaller number from the larger. Julian Dates are one of those extremely minor points of calendar use. They track the number of days elapsed since noon, Universal Time, on the Julian-calendar date we call the 1st of January, 4713 BC. The scheme, for years, was set up in 1583 by Joseph Justus Scalinger, calendar reformer, who wanted for convenience an index year so far back that every historically known event would have a positive number. In the 19th century the astronomer John Herschel expanded it to date-counting.

Scalinger picked the year from the convergence of a couple of convenient calendar cycles about how the sun and moon move as well as the 15-year indiction cycle that the Roman Empire used for tax matters (and that left an impression on European nations). His reasons don’t much matter to us. The specific choice means we’re not quite three-fifths of the way through the days in the 2,400,000’s. So it’s not rare to modify the Julian Date by subtracting 2,400,000 from it. The date starts from noon because astronomers used to start their new day at noon, which was more convenient for logging a whole night’s observations. Since astronomers started taking pictures of stuff and looking at them later they’ve switched to the new day starting at midnight like everybody else, but you know what it’s like changing an old system.

This summons the problem: so how do I know many dates passed between whatever day I’m interested in and the Julian Calendar 1st of January, 4713 BC? Yes, there’s a formula. No, don’t try to use it. Let the fine people at the United States Naval Observatory do the work for you. They know what they’re doing and they’ve had this calculator up for a very long time without any appreciable scandal accruing to it. The system asks you for a time of day, because the Julian Date increases as the day goes on. You can just make something up if the time doesn’t matter. I normally leave it on midnight myself.

So. The last episode of Mystery Science Theater 3000 to debut, on the 12th of September, 1999, did so on Julian Date 2,451,433. (Well, at 9 am Eastern that day, but nobody cares about that fine grain a detail.) The new season’s to debut on Netflix the 14th of April, 2017, which will be Julian Date 2,457,857. (I have no idea if there’s a set hour or if it’ll just become available at 12:01 am in whatever time zone Netflix Master Command’s servers are in.) That’s a difference of 6,424 days. You’re on your own in arguing about whether that means it was 6,424 or 6,423 days between new episodes.

If you do take anything away from this, though, please let it be the warning: never try to work out the interval between dates yourself.

So my question from last Thursday nagged at my mind. And I learned that Octave (a Matlab clone that’s rather cheaper) has a function that calculates the day of the week for any given day. And I spent longer than I would have expected fiddling with the formatting to get what I wanted to know.

It turns out there are some days in November more likely to be the fourth Thursday than others are. (This is the current standard for Thanksgiving Day in the United States.) And as I’d suspected without being able to prove, this doesn’t quite match the breakdown of which months are more likely to have Friday the 13ths. That is, it’s more likely that an arbitrarily selected month will start on Sunday than any other day of the week. It’s least likely that an arbitrarily selected month will start on a Saturday or Monday. The difference is extremely tiny; there are only four more Sunday-starting months than there are Monday-starting months over the course of 400 years.

But an arbitrary month is different from an arbitrary November. It turns out Novembers are most likely to start on a Sunday, Tuesday, or Thursday. And that makes the 26th, 24th, and 22nd the most likely days to be Thanksgiving. The 23rd and 25th are the least likely days to be Thanksgiving. Here’s the full roster, if I haven’t made any serious mistakes with it:

November

Will Be Thanksgiving

22

58

23

56

24

58

25

56

26

58

27

57

28

57

times in 400 years

I don’t pretend there’s any significance to this. But it is another of those interesting quirks of probability. What you would say the probability is of a month starting on the 1st — equivalently, of having a Friday the 13th, or a Fourth Thursday of the Month that’s the 26th — depends on how much you know about the month. If you know only that it’s a month on the Gregorian calendar it’s one thing (specifically, it’s 688/4800, or about 0.14333). If you know only that it’s a November than it’s another (58/400, or 0.145). If you know only that it’s a month in 2016 then it’s another yet (1/12, or about 0.08333). If you know that it’s November 2016 then the probability is 0. Information does strange things to probability questions.

It’s well-known, at least in calendar-appreciation circles, that the 13th of a month is more likely to be Friday than any other day of the week. That’s on the Gregorian calendar, which has some funny rules about whether a century year — 1900, 2000, 2100 — will be a leap year. Three of them aren’t in every four centuries. The result is the pattern of dates on the calendar is locked into this 400-year cycle, instead of the 28-year cycle you might imagine. And this makes some days of the week more likely for some dates than they otherwise might be.

This got me wondering. Does the 13th being slightly more likely imply that the United States Thanksgiving is more likely to be on the 26th of the month? The current rule is that Thanksgiving is the fourth Thursday of November. We’ll pretend that’s an unalterable fact of nature for the sake of having a problem we can solve. So if the 13th is more likely to be a Friday than any other day of the week, isn’t the 26th more likely to be a Thursday than any other day of the week?

And that’s so, but I’m not quite certain yet. What’s got me pondering this in the shower is that the 13th is more likely a Friday for an arbitrary month. That is, if I think of a month and don’t tell you anything about what it is, all we can say is it chance of the 13th being a Friday is such-and-such. But if I pick a particular month — say, November 2017 — things are different. The chance the 13th of November, 2017 is a Friday is zero. So the chance the 26th of December, 2017 is a Thursday is zero. Our calendar system sets rules. We’ll pretend that’s an unalterable fact of nature for the sake of having a problem we can solve, too.

So: does knowing that I am thinking of November, rather than a completely unknown month, change the probabilities? And I don’t know. My gut says “it’s plausible the dates of Novembers are different from the dates of arbitrary months”. I don’t know a way to argue this purely logically, though. It might have to be tested by going through 400 years of calendars and counting when the fourth Thursdays are. (The problem isn’t so tedious as that. There’s formulas computers are good at which can do this pretty well.)

But I would like to know if it can be argued there’s a difference, or that there isn’t.

I’ve got enough mathematics comics for another roundup, and this time, the subjects give me reason to dip into ancient days: one to the most famous, among mathematicians and astronomers anyway, of Greek shipwrecks, and another to some point in the midst of winter nearly seven thousand years ago.

Eric the Circle (November 15) returns “Griffinetsabine” to the writer’s role and gives another “Shape Single’s Bar” scene. I’m amused by Eric appearing with his ex: x is practically the icon denoting “this is an algebraic expression”, while geometry … well, circles are good for denoting that, although I suspect that triangles or maybe parallelograms are the ways to denote “this is a geometric expression”. Maybe it’s the little symbol for a right angle.

Jim Meddick’s Monty (November 17) presents Monty trying to work out just how many days there are to Christmas. This is a problem fraught with difficulties, starting with the obvious: does “today” count as a shopping day until Christmas? That is, if it were the 24th, would you say there are zero or one shopping days left? Also, is there even a difference between a “shopping day” and a “day” anymore now that nobody shops downtown so it’s only the stores nobody cares about that close on Sundays? Sort all that out and there’s the perpetual problem in working out intervals between dates on the Gregorian calendar, which is that you have to be daft to try working out intervals between dates on the Gregorian calendar. The only worse thing is trying to work out the intervals between Easters on it. My own habit for this kind of problem is to use the United States Navy’s Julian Date conversion page. The Julian date is a straight serial number, counting the number of days that have elapsed since noon Universal Time at what’s called the 1st of January, 4713 BCE, on the proleptic Julian calendar (“proleptic” because nobody around at the time was using, or even imagined, the calendar, but we can project back to what date that would have been), a year picked because it’s the start of several astronomical cycles, and it’s way before any specific recordable dates in human history, so any day you might have to particularly deal with has a positive number. Of course, to do this, we’re transforming the problem of “counting the number of days between two dates” to “counting the number of days between a date and January 1, 4713 BCE, twice”, but the advantage of that is, the United States Navy (and other people) have worked out how to do that and we can use their work.

Bill Hind’s kids-sports comic Cleats (November 19, rerun) presents Michael offering basketball advice that verges into logic and set theory problems: making the ball not go to a place outside the net is equivalent to making the ball go inside the net (if we decide that the edge of the net counts as either inside or outside the net, at least), and depending on the problem we want to solve, it might be more convenient to think about putting the ball into the net, or not putting the ball outside the net. We see this, in logic, in a set of relations called De Morgan’s Laws (named for Augustus De Morgan, who put these ideas in modern mathematical form), which describe what kinds of descriptions — “something is outside both sets A and B at one” or “something is not inside set A or set B”, or so on — represent the same relationship between the thing and the sets.

Tom Thaves’s Frank and Ernest (November 19) is set in the classic caveman era, with prehistoric Frank and Ernest and someone else discovering mathematics and working out whether a negative number times a negative number might be positive. It’s not obvious right away that they should, as you realize when you try teaching someone the multiplication rules including negative numbers, and it’s worth pointing out, a negative times a negative equals a positive because that’s the way we, the users of mathematics, have chosen to define negative numbers and multiplication. We could, in principle, have decided that a negative times a negative should give us a negative number. This would be a different “multiplication” (or a different “negative”) than we use, but as long as we had logically self-consistent rules we could do that. We don’t, because it turns out negative-times-negative-is-positive is convenient for problems we like to do. Mathematics may be universal — something following the same rules we do has to get the same results we do — but it’s also something of a construct, and the multiplication of negative numbers is a signal of that.

Mickey Mouse (November 20, rerun) — I don’t know who wrote or draw this, but Walt Disney’s name was plastered onto it — sees messages appearing in alphabet soup. In one sense, such messages are inevitable: jumble and swirl letters around and eventually, surely, any message there are enough letters for will appear. This is very similar to the problem of infinite monkeys at typewriters, although with the special constraint that if, say, the bowl has only two letters “L”, it’s impossible to get the word “parallel”, unless one of the I’s is doing an impersonation. Here, Goofy has the message “buried treasure in back yard” appear in his soup; assuming those are all the letters in his soup then there’s something like 44,881,973,505,008,615,424 different arrangements of letters that could come up. There are several legitimate messages you could make out of that (“treasure buried in back yard”, “in back yard buried treasure”), not to mention shorter messages that don’t use all those letters (“run back”), but I think it’s safe to say the number of possible sentences that make sense are pretty few and it’s remarkable to get something like that. Maybe the cook was trying to tell Goofy something after all.

Gary Delainey and Gerry Rasmussen’s Betty (November 20) mentions the Antikythera Mechanism, one of the most famous analog computers out there, and that’s close enough to pure mathematics for me to feel comfortable including it here. The machine was found in April 1900, in ancient shipwreck, and at first seemed to be just a strange lump of bronze and wood. By 1902 the archeologist Valerios Stais noticed a gear in the mechanism, but since it was believed the wreck far, far predated any gear mechanisms, the machine languished in that strange obscurity that a thing which can’t be explained sometimes suffers. The mechanism appears to be designed to be an astronomical computer, tracking the positions of the Sun and the Moon — tracking the actual moon rather than an approximate mean lunar motion — the rising and etting of some constellations, solar eclipses, several astronomical cycles, and even the Olympic Games. It’s an astounding mechanism, it’s mysterious: who made it? How? Are there others? What happened to them? How was the mechanical engineering needed for this developed, and what other projects did the people who created this also do? Any answers to these questions, if we ever know them, seem sure to be at least as amazing as the questions are.