Fri, 14 Sep 2018
A while back I said I wanted to memorize all the prime numbers under 1,000, because I am tired of getting some number like 851 or 857, or even 307, and then not knowing whether it is prime.
The straightforward way to deal with this is: just memorize the list. There are only 168 of them, and I have the first 25 or so memorized anyway.
But I had a different idea also. Say that a set of numbers from !!10n!! to !!10n+9!! is a “decade”. Each decade contains at most 4 primes, so 4 bits are enough to describe the primes in a single decade. Assign a consonant to each of the 16 possible patterns, say “b” when none of the four numbers is a prime, “d” when only !!10n+1!! is prime, “f” when only !!10n+3!! is, and so on.
Now memorizing the primes in the 90 decades is reduced to memorizing 90 consonants. Inserting vowels wherever convenient, we have now turned the problem into one of memorizing around 45 words. A word like “potato” would encode the constellation of primes in three consecutive decades. 45 words is only a few sentences, so perhaps we could reduce the list of primes to a short and easily-remembered paragraph. If so, memorizing a few sentences should be much easier than memorizing the original list of primes.
The method has several clear drawbacks. We would have to memorize the mapping from consonants to bit patterns, but this is short and maybe not too difficult.
More significant is that if we're trying to decide if, say, 637 is prime, we have to remember which consonant in which word represents the 63rd decade. This can be fixed, maybe, by selecting words and sentences of standard length. Say there are three sentences and each contains 30 consonants. Maybe we can arrange that words always appear in patterns, say four words with 1 or 2 consonants each that together total 7 consonants, followed by a single long word with three consonants. Then each sentence can contain three of these five-word groups and it will be relatively easy to locate the 23rd consonant in a sentence: it is early in the third group.
Katara and I tried this, with not much success. But I'm not ready to give up on the idea quite yet. A problem we encountered early on is that we remember consonants not be how words are spelled but by how they sound. So we don't want a word like “hammer” to represent the consonant pattern h-m-m but rather just h-m.
Another problem is that some constellations of primes are much more common than others. We initially assigned consonants to constellations in order. This assigned letter “b” to the decades that contain no primes. But this is the most common situation, so the letter “b” tended to predominate in the words we needed for our mnemonic. We need to be careful to assign the most common constellations to the best letters.
Some consonants in English like to appear in clusters, and it's not trivial to match these up with the common constellations. The mapping from prime constellations to consonants must be carefully chosen to work with English. We initially assigned “s” to the constellation “☆•☆☆” (where !!10n+1, 10n+7,!! and !!10n+9!! are prime but !!10n+3!! is not) and “t” to the constellation “☆☆••” (where !!10n+1!! and !!10n+3!! are prime but !!10n+7!! and !!10n+9!! are not) but these constellations cannot appear consecutively, since at least one of !!10n+7, 10n+9, 10n+11!! is composite. So any word with “s” and “t” with no intervening consonants was out of play. This eliminated a significant fraction of the entire dictionary!
I still think it could be made to work, maybe. If you're interested
in playing around with this, the programs I wrote are available on
Github. The mapping
from decade constellations to consonant clusters is in