duplicate hunting, pt. 13
Andrew Plewe
aplewe at sbcglobal.net
Thu May 10 00:34:58 CEST 2007
Nick Hobson wrote:
> Suppose p and q are consecutive primes. Let
>
> a = sqrt((p^2 + q^2)/2 - 1). (1)
>
> It is easy to see that if q = p + 2 then a is an integer. Is the converse
> true?
>
> If we consider the smallest prime q > p + 2 which makes a an integer, we
> generate the following sequence, where 0 is used if there is no such q:
>
> 11, 263, 59, 23, 101, 109, 0, 151, 193, 79, 269, 277, 311, 0, 179, 83,
> 83003, 479, 487, 181, 563, 571, 613, 1201, 157, 141509, 739, 773, 479
Don Reble replied:
> I get p=19,q=1278886952463697; p=127,q=6858037981;
> but I find no q for p=47.
There is no q for p=47. For this case, (1) simplifies to the Pell-type
equation
q^2 - 2 a^2 = -2207. (2)
I claim that for every integer solution of this, q is divisible by either
3, 5, 7, or 11, so it can't be prime.
Equation (2) describes a hyperbola. Given one point (q,a) on the a>0
branch, we can construct infinitely many others (q', a') on the same branch
by writing
q' + a' sqrt(2) = (q + a sqrt(2)) (3 + 2 sqrt(2))^n (3)
where n is an integer. Every point on the a>0 branch is equivalent in this
way to exactly one with -q < a <= q. Combining that inequality with
equation (2) implies 33 < sqrt(2207/2) <= a <= sqrt(2207) < 47. Checking
the values of a in that range produces just two integer solutions:
(q,a) = (45,46) and (q,a) = (-45,46). So every integer solution of (2)
with a>0 is given by
(q,a) = (45 + 46 sqrt(2)) (3 + 2 sqrt(2))^n (4)
or
(q,a) = (-45 + 46 sqrt(2)) (3 + 2 sqrt(2))^n (5)
for some integer n. Note that
(3 + 2 sqrt(2))^12 = 768398401 + 543339720 sqrt(2)
= 1 + 1155 (665280 + 470424 sqrt(2))
So if
q' + a' sqrt(2) = (q + a sqrt(2)) (3 + 2 sqrt(2))^12 (6)
then q' == q (mod 1155). Note that 1155 = 3*5*7*11. So to verify that the
q's given by (4) and (5) are always divisible by 3, 5, 7, or 11, it suffices
to check 12 consecutive values in each sequence:
n q from (4) a from (4) q is divisible by
-----------------------------------------------------
0 45 46 3, 5
1 319 228 11
2 1869 1322 3, 7
3 10895 7704 5
4 63501 44902 3
5 370111 261708 7
6 2157165 1525346 3, 5
7 12572879 8890368 11
8 73280109 51816862 3, 7
9 427107775 302010804 5
10 2489366541 1760247962 3
11 14509091471 10259476968 7
n q from (5) a from (5) q is divisible by
-----------------------------------------------------
0 -45 46 3, 5
1 49 48 7
2 339 242 3
3 1985 1404 5
4 11571 8182 3, 7
5 67441 47688 11
6 393075 277946 3, 5
7 2291009 1619988 7
8 13352979 9441982 3
9 77826865 55031904 5
10 453608211 320749442 3, 7
11 2643822401 1869464748 11
So q is never prime.
Dean Hickerson
dean at math.ucdavis.edu
Jon Schoenfield wrote today asking a question that I think Magma
could answer - maybe someone could help?
[Becuase these questions reduce to finding integer solutions
to cubic equations.]
Neil
>From jonscho at hiwaay.net Wed May 9 20:16:01 2007
>From: "Jon Schoenfield" <jonscho at hiwaay.net>
>To: <njas at research.att.com>, <zakseidov at yahoo.com>,
> "Camillia Smith Barnes" <cammie at math.harvard.edu>
>Subject: FYI: Conjecture re A116108, A116145, et al.
>Date: Wed, 9 May 2007 19:15:43 -0500
>
>Regarding A116108 ("Squares that are equal to the sum of 3 consecutive
>cubes"), A116145 ("Squares that are equal to the sum of 5 consecutive
>cubes"), and similar sequences that could be defined as "Squares that are
>equal to the sum of 2k+1 consecutive cubes" (for a given positive integer
>k), I suspect that each of the sequences is finite and very short, although
>I don't know how to prove it....
>
>If we define c as the central value to be cubed, so that, e.g., in A116108,
>each value in the sequence can be written as
>
>(c-1)^3 + c^3 + (c+1)^3 = 3c^3 + 6c = 3c(c^2 + 2)
>
>(so c-1 equals the variable "m" in A116108's Comment field), then the
>sequence's terms, 0, 9, 36, and 41616, correspond to c = 0, 1, 2, and 24.
>After an exhaustive search up through c = 4*10^8, I can say that the next
>term exceeds 1.92*10^26 -- if it exists at all, which I doubt, but can't
>prove ....
>
>Defining c similarly in A116145 (so c-2 = m) gives
>
>(c-2)^3 + ... + (c+2)^3 = 5c^3 + 30c = 5c(c^2 + 6)
>
>and the sequence's terms, 0, 100, 225, 99225, 4708900, 8643600, correspond
>to c = 0, 2, 3, 27, 98, and 120.
>
>Having run similar tests for squares that are equal to the sum of 2k+1
>consecutive cubes (for positive integer values of k), and noting that
>
>(c-k)^3 + ... + (c+k)^3 = ... = (2k+1)c(c^2 + 2k(2k+1)(2k+2))
>
>it looks to me as though, for _most_ values of k, there will exist only four
>values in the sequence (as I believe is the case for A116108): the ones
>corresponding to c = 0, k, k+1, and 2k(2k+1)(2k+2). The only exceptions
>I've found (through k = 50) are as follows (and if there's any kind of
>pattern here, I haven't yet figured it out):
>
> k = 2: c = 27, 98 (as noted above)
> k = 6: c = 150
> k = 7: c = 32, 57967
> k = 8: c = 17, 128
> k = 10: c = 24, 154
> k = 16: c = 49
> k = 17: c = 242
> k = 19: c = 130
> k = 22: c = 198
> k = 28: c = 1653
> k = 31: c = 4, 248, 868, 1152
> k = 34: c = 115
> k = 36: c = 16900, 26973
> k = 38: c = 182
> k = 43: c = 275
> k = 45: c = 4830
> k = 48: c = 23716, 33708
>
(end)
At 4:43 PM -0700 5/9/07, Dean Hickerson wrote:
>Nick Hobson wrote:
>
>> Suppose p and q are consecutive primes. Let
>>
>> a = sqrt((p^2 + q^2)/2 - 1). (1)
>>
>> It is easy to see that if q = p + 2 then a is an integer. Is the converse
>> true?
>>
>> If we consider the smallest prime q > p + 2 which makes a an integer, we
>> generate the following sequence, where 0 is used if there is no such q:
>>
>> 11, 263, 59, 23, 101, 109, 0, 151, 193, 79, 269, 277, 311, 0, 179, 83,
>> 83003, 479, 487, 181, 563, 571, 613, 1201, 157, 141509, 739, 773, 479
>
>Don Reble replied:
>
>> I get p=19,q=1278886952463697; p=127,q=6858037981;
>> but I find no q for p=47.
>
>There is no q for p=47. For this case, (1) simplifies to the Pell-type
>equation
>
> q^2 - 2 a^2 = -2207. (2)
>
>I claim that for every integer solution of this, q is divisible by either
>3, 5, 7, or 11, so it can't be prime.
>
>Equation (2) describes a hyperbola. Given one point (q,a) on the a>0
>branch, we can construct infinitely many others (q', a') on the same branch
>by writing
>
> q' + a' sqrt(2) = (q + a sqrt(2)) (3 + 2 sqrt(2))^n (3)
>
>where n is an integer. Every point on the a>0 branch is equivalent in this
>way to exactly one with -q < a <= q. Combining that inequality with
>equation (2) implies 33 < sqrt(2207/2) <= a <= sqrt(2207) < 47. Checking
>the values of a in that range produces just two integer solutions:
>(q,a) = (45,46) and (q,a) = (-45,46). So every integer solution of (2)
>with a>0 is given by
>
> (q,a) = (45 + 46 sqrt(2)) (3 + 2 sqrt(2))^n (4)
>or
> (q,a) = (-45 + 46 sqrt(2)) (3 + 2 sqrt(2))^n (5)
>
>for some integer n. Note that
>
> (3 + 2 sqrt(2))^12 = 768398401 + 543339720 sqrt(2)
>
> = 1 + 1155 (665280 + 470424 sqrt(2))
>
>So if
>
> q' + a' sqrt(2) = (q + a sqrt(2)) (3 + 2 sqrt(2))^12 (6)
>
>then q' == q (mod 1155). Note that 1155 = 3*5*7*11. So to verify that the
>q's given by (4) and (5) are always divisible by 3, 5, 7, or 11, it suffices
>to check 12 consecutive values in each sequence:
>
>n q from (4) a from (4) q is divisible by
>-----------------------------------------------------
>0 45 46 3, 5
>1 319 228 11
>2 1869 1322 3, 7
>3 10895 7704 5
>4 63501 44902 3
>5 370111 261708 7
>6 2157165 1525346 3, 5
>7 12572879 8890368 11
>8 73280109 51816862 3, 7
>9 427107775 302010804 5
>10 2489366541 1760247962 3
>11 14509091471 10259476968 7
>
>n q from (5) a from (5) q is divisible by
>-----------------------------------------------------
>0 -45 46 3, 5
>1 49 48 7
>2 339 242 3
>3 1985 1404 5
>4 11571 8182 3, 7
>5 67441 47688 11
>6 393075 277946 3, 5
>7 2291009 1619988 7
>8 13352979 9441982 3
>9 77826865 55031904 5
>10 453608211 320749442 3, 7
>11 2643822401 1869464748 11
>
Dean, I agree with your analysis. Don Reble and I have been working on
this problem off-list. I have computed a 1000 terms and Don has computed
even more. I have done all my work with recursive equations instead of the
radicals you use. For instance, after the q(0) and q(1) terms are known,
the second column of your tables can be computed using the recursion
To prove that there is no possible solution for a prime such as 47, only
the first 6 terms need to be computed. Using this method, it can be shown
that there is no possible solution for the primes 47, 443, 1867, 6883,
13417, and 13907.
Tony
More information about the SeqFan
mailing list