## 29 February 2008

### Leap days

Mark Dominus has written an interesting post about leap days, of which today is one. I welcome this turn of events, because it means I don't have to! But I have a few things to add.

The reason that today is February 29 -- and not March 1 like it would be in a normal year -- is because the year is almost a quarter of a day longer than 365 days. So we add an extra day every four years. But not quite -- it's actually more like 365.24 days -- so we leave out one of these extra days every century. But no, it's really more like 365.2425 days -- so we add back in one of the days we got rid of every four centuries. (We did that in 2000, as you may remember. 2000 was a leap year.)

The real number of days per year is something like 365.24219; thus what one wants to do is to find a rational approximation p/q to .24219 and add p leap years every q years. The approximation we use, 97/400, actually isn't that good for a number with such a large denominator; 8/33, a convergent of the continued fraction of .24219, is better. Dominus proposes having leap years in those years congruent to 4, 8, 16, ... , 32 mod 33, a scheme that was also suggested in 1996 by Simon Cassidy on Usenet, complete with the same choice 4, ..., 32 of leap years in each 33-year cycle. The reason for this coincidence is that 1984, 1988, ..., 2012 are leap years in both systems.

And although dividing by 33 is hard, as Dominus points out, it's not that hard to reduce a number modulo 33 in your head. The year abcd (where a, b, c, and d are digits) reduces modulo 99 to ab + cd; for example, 2008 reduces to 28 modulo 99. Then just subtract 33 or 66 as necessary. I'd rather have to do lots of reduction modulo 33 in my head than, say, modulo 37. (And with 33 there's probably some slick Chinese-remainder-theorem method that allows you to reduce a number mod 33 by reducing it mod 3 and 11, both of which are easy.) In any case, there's no reason we couldn't have a calendar that requires more complicated arithmetic than the current one; most people are not in the business of calculating calendars. (And calendars are calculated by computers now; Dominus points out his new rule actually requires less computation than the old one. The 97/400 rule is an artifact of the fact that we work in decimal.)

As is pointed out both by Cassidy and by this page, Jesus is traditionally said to have lived for 33 years; for some people that might lend additional appeal to a 33-year cycle.

Edit (8:23 am): There's also something to be said for a system that includes 31 leap years every 128 years -- have leap years in years divisible by 4, except those divisible by 128 -- this was apparently suggested by Johann Heinrich von Mädler, and would only be off by something like a quarter-second a year. But striving for such accuracy is silly, because the length of the year isn't actually constant.

#### 1 comment:

Anonymous said...

From the position of a computer programmer I'd strongly support the last idea of 31 leap years out of 128. First, it could be easily tested comparing one bitwise "AND" and secondly, I can recall powers of two up to at least 65536 in a split-second, so given the number of years I'll probably live through, it would be pretty easy deal (ya' know, I'd encounter 2048 only). The better-than-nature precision of this scheme is just an added benefit...

I think that in an ideal world the calendars should be designed by programmers... :)