In his book “The Man Who Mistook His Wife For A Hat” (1985) Oliver Sacks describes an intriguing case of savant syndrome. He tells the story about his encounter with the twins John and Michael, who had been in institutions since childhood, variously diagnosed as autistic, psychotic or severely retarded. Others before Sacks had already investigated these boys and found out that they were very good in calendar calculating. For a random given date they could quickly tell which day of the week it is. Sacks however discovered a far more unusual ability of the twins when he observed them in 1966. On a certain day he notices that they are exchanging six-figure numbers and seem to be very pleased with the numbers they receive from each other. Sacks notes down the numbers and when he comes home, takes out his ‘tables of powers, factors, logarithms and primes’ and finds out that all the numbers, mentioned by the twins, are in fact prime! Truly remarkable, because the twins were not able to perform simple calculations and did not seem to understand multiplication and division.
The next day Sacks returns with a book full of prime numbers and starts to play the game along. But he starts out with a prime number consisting of eight figures. After half a minute thinking (or longer) the twins start to smile. This is to Sacks a clear sign that they are both pleased with this new ‘toy’ (a bigger prime) and happy to find out that Sacks had understood their game. In return John mentions, after five minutes thinking, a nine-figure number and his brother follows suit. Sacks now picks a ten-figure prime from his booklet, which in a while gets responded by a twelve-figure number. As Sacks’s precious book only contains primes up to ten-figure numbers, he retracts from the game. One hour later the brothers are exchanging twenty-figure numbers!
I had already encounterd this story years ago, but when I came across it again recently (while reading the book “We are our brains” by Dick Swaab), I noticed that there were some sceptical questions raised considering the credibility of Sacks’s story. More than enough reason for me to dig a bit deeper into this!
In 2006 Makato Yamaguchi wrote an article in which he questions the report of Sacks. The most important question he asks is which book Sacks used. If the book would indeed contain all the primes consisting of ten figures or less, it would have to list more than 455 million numbers, which is of course infeasible. Maybe his booklet contained only a few ten-figure primes? Yamaguchi couldn’t find any book available in 1966 which contains any list of this sort. Also Sacks couldn’t tell him which book it had been. Nor was he able to recollect which numbers were involved. All his original annotations and the book are lost. Sacks would admit however that his book may only have contained primes up to eight figures.
If you read the description of events by Sacks carefully (read it yourself), you’ll notice he actually only mentions the primeness of the six-figure numbers which he checked at home and the numbers he chose himself from the book. He further states that he assumes that the twenty-figure number is a prime. But he didn’t check whether the numbers called out by the twins larger than six figures were actually prime. Of course this wouldn’t have been an easy task in 1966 without the access to ready available computer programs we have nowadays. But it seems that Sacks didn’t even try to get certainty; he just states it would have been difficult.
In a postscript to the chapter about the twins (at least in the Dutch version I have), Sacks mentions another algorithm to test numbers for primeness. But his description only shows that he has no real mathematical knowledge that matters in this case. Sacks sticks to a romantic idea that the twins had a special sensibility to numbers, that they could somehow pick out prime numbers in vast ocean of ordinary numbers. The idea that someone could ‘see’ the primeness of a big number without doing some calculations, is something I don’t buy, however. Perhaps some idiot savants might be faster in recognizing prime numbers, but that doesn’t imply that they are using other methods. Let alone methods, which cannot be described as an algorithm.
Can we rely upon the first part of the story: were the twins really exchanging six-figure primes? Or is there another more reasonable explanation for this at first sight astonishing performance? Let’s start by pointing out that the calculation skills of the twins were not as bad as Sacks writes: the researchers (Horwitz a.o.), who established (in 1965) that they could hardly perform calculations, in 1969 wrote they had at least the capability of adding three-figure numbers. As far as I know Horwitz only describes the calendar calculation. Unfortunately I didn’t have access to their articles to see if they mention anything about their skills with prime numbers.
How hard is it anyway to give six-figure primes? Between 100,000 and 999,999 there are 68,906 primes. What is the chance of choosing a prime while keeping away from those numbers which are obviously not prime? It would be silly to try an even number or one which ends in ‘5’; those are of course divisible by 2 and 5 respectively. Many will recall the trick to find out if a number is divisible by 3: you just add the figures of the number, repeat that over en over again with the result, until you can easily see if the remaining number is divisible by 3. If so, the original number is also divisible by three. An example: 561,251 gives as sum of its figures 20, in the next step this gives 2, which is not divisible by three. And therefore we can conclude that 561,251 is not divisible by 3.
These kind of tricks can also be found for the other small (prime) divisors 7, 11, 13, 17, etc. Because the two brothers were good calendar calculators, they must almost certainly have known how to divide by 7. Or at least have known how to determine the remainder of a division by 7. One of the ways of doing this with 7 goes as follows: take the last two figures of a number and add the double of anything in front of these two figures. The new number is divisible by 7 only if the starting number was divisible by 7 as well. You can repeat this procedure until the point you can easily make out if the remaining number is divisible by 7. Let’s consider our example number 561,251 again: first you get 2 x 5,612 + 51 = 11,275, then 2 x 112 + 75 = 299, then 2 x 2 + 99 = 103 and at last 2 x 1 + 3 = 5, which is not divisible by 7. And therefore 561,251 is also not divisible by 7.
If you exclude the numbers, which are divisible by 2, 3, 5 and 7 using these tricks, you’ll keep 205,714 six-figure numbers. Picking a number at random from this set, gives you a chance of about 33 percent that the chosen number is prime. That’s not a really small chance! If you also master the trick for divisor 11 (which is in fact easier than the trick for 7) this chance rises to 37 percent. If the twins used this approach and Sacks only took a few numbers home to check, the chance that those numbers were all prime by coincidence is not really that small. And we should also think about the way Sacks checked whether these numbers were prime if this unique book of his doesn’t exist.
It would be interesting to know in what way the numbers were called. Probably not as ‘five hundred sixty one thousand two hundred and fifty one’. At least this is not imaginable with the twenty-figure numbers! More likely is that they called out the numbers figure by figure like a telephone number ‘five-six-one-two-five-one’. That could also mean that the twins saw the huge numbers more as a row of figures than as a single number. For applying the tricks for division by small factors it doesn’t really matter and it could be an explanation for the ‘fact’ that they stepped without much trouble from six-figure numbers to eight-figure numbers and even further.
A curious aspect of the story is that John after getting the ten-figure prime from Sacks, answered with a twelve-figure number, and in doing so skipped one with eleven figures. Possibly he thought of an eleven-figure candidate, but found out that it failed one of the division tricks. In that case you can try to add an odd figure at the end and see if this new number works, without too much recalculation. Again an example: does 13,725,097,771 work? No, it’s a pity, it can be divided by 7. Let’s try to add a ‘1’ at the end, which gives us 137,250,977,711. And that number is not divisible by 2,3,5 &7. OK, fine! But note that the number is not prime, because 137,250,977,711 equals 19 x 41,893 x 172,433.
A study by Hermelin and O’Connor (1990, Factors and primes: a specific numerical ability. Psychological Medicine, Vol. 20) shows another ‘idiot savant‘ using a strategy like the one I describe here. He used it to figure out whether four- and five-figure numbers might be primes. He excluded numbers divisible by 3 and 11 and while doing so still made quite some judgment errors.
What could Sacks have done to give his story more credibility? At least he could have tried to call six-figure numbers which are not that simple to unmask as ‘fake’ primes. If he tried for example 254,539 which equals 331 x 769, it would be interesting to see the twins reactions (for clarity 331 and 769 are prime). Maybe they would have been as happy with this composite number as with real primes. That would have been an indication that they were in fact only ruling out small divisors.
We’ll probably never know what the twins were actually capable of, if Sacks cannot explain in more detail what really happened. It’s hard not to get the impression that the story was a bit ‘spiced up’. The calendar calculation skills of the twins were quite impressive, but not unique. Maybe Sacks was tempted to present an even more impressive feat than what already had been written down in existing literature. Why did he actually wait to publish about it until 1985, almost twenty years later? In any case Sacks was clearly not that interested to get better confirmation of his hypothesis. Or maybe he didn’t understand enough about the math involved to test it properly.
Just when I finished writing the original Dutch version of this article, I stumbled upon a comment on a blog concerning the same story. There is in fact a likely candidate for the book Sacks claims he used: D. N. Lehmer, List of Prime Numbers from 1 to 10,006,721. Washington, D. C., Carnegie Institution of Washington, 1914. xvi+133 pp. The book by Lehmer would suffice to do the checks on the six-figure numbers of the twins and there are some 8 figure primes in the book Sacks could have used. From the title of the book I cannot make up if there are any ten-figure primes mentioned, but it might have some appendices for known primes with more figures than the list for which it claims to be complete. I’ve send Sacks an e-mail with this suggestion. Until now I didn’t hear back from him, except for a kind reply by his assistant that my suggestion was passed on to him (he doesn’t use a computer himself it seems).
Finally, Makato Yamaguchi pointed me to a YouTube video of the twins showing their calendar calculation skills. You can check the dates and days of the week yourself, I found these to be correct in all cases.
original version in Dutch: De priemgetal tweeling van Oliver Sacks published December 22nd, 2011
Published on May 9th 2012
17 thoughts to “Oliver Sacks’s Twins and Prime Numbers”
I am reading Marcus du-Sautoy’ Music of the Prime Numbers where I have found the crossrefence on the “savant” twins. I borrowed Sacks’s book from the library to read the relavant chapter and while reading it I got the suspicion that Sacks was wrong. I do not intend to say that he deliberately deceived his readers but his low math knowledge on primes (admitted by himself) let him in the status of meeting magic minds when working with the twins.
I am happy to find your blog on this issue. It convinced me that I am not alone with my suspicion.
Thanks for your article. I had another question about this chapter of the late Sacks’ book. I don’t have any advanced math background and had not learned about some of the math concepts discussed in this chapter. One key sentence was unclear to me: “And if one is thinking of modular arithmetic in general, modular division will produce
neat cyclic patterns only if one uses prime numbers.” Can you explain what he meant or point me to some easy math article online that directly addresses it? After investing so much effort into understanding the rest of the discussion and given how influential this book seems to be it’d be great to understand this too.
That particular sentence is from the following piece of text and to it shows that Sacks just didn’t quite understand it :
For modular arithmetic, you could check out Wikipedia, which is usually quite good on mathematical subjects. Sacks talks about modular division, let’s say a/b modulo m. This is only defined when b and m are relative prime. If m is prime this holds for any b that is not a multiple of m. What this would have to do with calendar arithmetic is unclear to me, because then you’re only interested in the remainder of the division. You could perform calendar arithmetic with a week consisting of 8 days as well. For some week lengths, it would be easier. There is nothing special to our week because 7 is a prime number, as far as I can see.
I came across this today (25 September 2017) and wanted to thank you for your insights and follow-up commentary! I, too, felt uneasy with this particular case study, and was happy to see your thoughts on the matter here.
Oh and another comment since your not dutch, and i can read dutch
in his latest comment the author of the original dutch article did found a book that he might have used to find his numbers. But since he wasn’t into math himself (reading from the dutch comments), he probably didn’t gave the more hard to find prime numbers to these twins.
Anyway there is a reason that calculations are parts of IQ tests, and we do call idiot savants someone with an extreme IQ in a very narrow area
I mention the book in this article in the last but one paragraph 😉
Well i happen to have had once a strange insight into prime numbers like a memory flash.
I never heard of the story you wrote here before.
But i think then that is completely possible.
( i was just drawing some art, while in a sudden i came to new optimal way of displaying prime numbers, while in itself not saying what prime is, but the graph did show some features in an excellent way. )
A few people later got excited by my idea, ..but my wondering have been mainly that it was strange to have a visual inside into it as i wasn’t doing any math science about primes, and had not that much interest in it.
My best solution so far is that the mind has more complex calculation options, ea the ear works logarythmic.
The eyes correct light on frequencies and group colors together; there is knowledge there in grouping, and perspective view from two images requires other complex math, complex if our brains where normal computers, we would need to know about sin and cosin, But our brains are not computers, they are neural networks a principle even more complex then today’s computers, they are analog computers.
So i think at least the tools are there… but we normally have no control on how to use it, for advanced math and advanced math insight.
On the other hand all great people in math, used their head, instead of a computer, so the head is cape-able to do amazing complex math fast. Even NASA relied upon idiot savants in times when some people outperformed computers in calculations. And those people still exist who happen to have extreme capabilities in calculating. I myself once met an astronomer who dared me to test him on calculations with long numbers, he gave immediately an answer.
I think you should find more people like in your article, they are extreme rare to find.
I think those people see numbers and math as landscapes, much alike how the famous chess twins, the polgar sisters, where found to store memories of specific chess positions in brain scans, at locations where normal people would store people’s face. And by doing so their brains could extremely fast judge a chess possition.. an even harder problem then prime numbers where the rules are more simple
I think you really underestimate how fast the number of needed calculations rises with the increase in length of the (potential) prime numbers. The calculations are not in themselves difficult, but you run into time trouble performing them.
since it happened to me, i know about math and primes.
Prime do follow a pattern, we cant tell what the patern is
ea we cannt say the number x is a patern by linear math.
But the old romans used for example a sieve of 6n+-1
Umlams prime spiral while not a linear math sieve does produce a ‘landscape view’ of prime numbers in which it becomes easier to find for primes. There is more math relating to this with a strong indication (but no proof yet) that there must be some linear relation math related to primes.
But it probably is some kind of not yet known to use, its like someone showing you a fractal but you where not told about the math behind fractals, and yet i would need to believe this guy who claims fractals are part of the shapes in nature… you would say its impossible to calculate it no Math could describe it… until we understood there is new math in fractals..
You mean Ulam’s Spiral, see http://mathworld.wolfram.com/PrimeSpiral.html
It’s important to understand that “it appears to exhibit a strongly nonrandom appearance”, but if you would try to expand the pattern without actual calculations you would identify many numbers as primes which are in fact composite. There are many people who invented patterns which they thought would solve the prime identification problem, but they only work upto a certain number. Most of times there is a fairly small counterexample.
also now that the humain brain is far complexer then todays super computer, idiot savant dont use their brains like you and i use them. Time for 20 number digits… is for the most dense neural network that exists …
shouldnt be that hard. For linear copmuters it will ofcourse takes lots of time, but neural networks are not like that. And the best simulated neiral networks dont even come near our brains.. perhaps a rat brain but not at the speed a real rat thinks.
That our brains work akin a neural network does not help for identifying primality. For pattern recognition it works quite well, but I explained in a previous comment that there is no general pattern for primaility (never found at least by all those mathematicians who are much smarter than myself), so relying on that type of pattern recognition for primality is bound to fail.
If our brains would work like a quantum computer, that could be something different. But as far as I know, there are no indications for that, so I’ll stick to my judgement that it is extremely unlikely that the twins were actually able to see whether a 20 digit number is prime.
I enjoyed your article until reading the very end. I find it appalling that you did not put the new information up higher and then just write your interesting questions as a discussion about whether the twins really had a magical ability. Given that probably 80% of the people who come to your site will not read to the end you have essentially propagated someone else’s slander of an author many of us think highly of. I’m so glad i read to the end otherwise I would take away a deep skepticism of other things Sacks has written. Please consider elevating your endnote or taking out the mean commentary which you yourself have debunked.
I’m a bit baffled by your reaction, I don’t see how anyone can be ‘appalled’ by the way I built up this article. Let me start by stating that were I write: “We’ll probably never know what the twins were actually capable of” I mean their calculation powers by using normal mathematical methods, not ‘magically seeing primes’, which I consider impossible as mentioned in the article.
And how about ‘someone else’s slander’? I think the questions raised by Yamaguchi are legit and Sacks could have reacted in a far more sensible way. This is not slanderous in any way and I don’t get what you mean by “the mean commentary”.
In my opinion the story at best originates from Sacks ‘fooling’ himself by not trying to falsify his far fetched hypothesis, but he might have made it up completely (although I think that not so likely). The fact that a book exists which could explain some of the details in Sacks’s story (if Sacks owned it) makes the first option a bit more likely. But still Sacks has some explaining to do in my opinion.
My skepticism of other writings by Sacks would be raised more by his inadequate reaction to the questions raised about this story than by the likelihood that he has indeed fooled himself in this particular case.
I read the article for the very reason of a previous high impression of Sacks’ book. Re-reading the chapter in question, I found a certain puzzlement about what further commentary it had gained in the following years.
I find nothing mean or slanderous in the article. I would have been tempted to go a bit further and infer that whom I’d previously regarded as a masterful, supple, and richly talented weaver of clinical and popular language into prose was someone also as perhaps a bit of a fabulist; and I do mean ‘bit’. It’s just that I’m now in the same sort of place as I entered on finding that neither the linguistics of Chomsky, nor his incisive deconstruction of the American power establishment, are quite the islands of accuracy that I had earlier charted.
But my impression from the main body of the article was the growing apprehension that Sacks never had any such book of primes in the first place, and I had the opposite reaction to yours—and one of some relief—to the ending: something rather strongly resembling his source in fact was uncovered eventually.
The expression “sadder but wiser” does have specific instances for application. We are compelled by age to become more collators of facts than of heroes.
(that all said, it’s been my experience that such long-winded comments often never survive the WordPress posting-algorithm. Eventually I will learn not to bother.)