> Congratulations to Jens K. Andersen who has just broken the existing

Thanks.

> records for generalised bitwins of lengths 3-6 and has found the

> first generalised bitwin of length 7.

> The bitwin tables are maintained by Henri Lifchitz and can be found

> at http://ourworld.compuserve.com/homepages/hlifchitz/

A generalized BiTwin with k links (length k+1) is k+1 twin primes on the form

n*b^i+/-1 for k+1 consecutive positive values of i, and b>2.

My new records follow. Each took less than a GHz week.

Two links:

72731043^2*31#^6*1723#/27527471*(16788518784000*17#*19#*43#*400#/164923)^i+/-1

for i = 1,2,3 (1006,1203,1401 digits)

Three links:

66*630^5*22218733^2*19#*317#/561857*(8084448001600*13#^3*197#/12863477)^i+/-1

for i = 1,2,3,4 (261,360,459,558 digits)

Four links: (1250561*151#+423642234015)*4^i+/-1

for i = 1,2,3,4,5 (67,67,68,68,69 digits)

Five links: (526583*83#+27663009*19#)/4*4^i+/-1

for i = 1,2,3,4,5,6 (39,39,40,40,41,42 digits)

Six links: 1394855870347655081*13#*4^i+/-1

for i = 2,3,4,5,6,7,8 (24,25,26,26,27,27,28 digits)

The GMP library and PrimeForm/GW were used for prp testing.

PrimeForm/GW proved the primes above 100 digits.

Marcel Martin's Primo proved the remaining primes.

My usual tuplet sieve was used for 4, 5 and 6 links.

Here b was fixed at 4 for convenience to my existing code.

2 and 3 links were found with an algorithm designed for generalized BiTwins:

Choose large numbers A and C with lots of divisors.

Compute a pool of twins on the form a*A*C+/-1, where a divides A.

Compute a second pool of twins on the form c*A^2*C+/-1, where c divides C.

Every twin combination a*A*C+/-1 and c*A^2*C+/-1 is now a generalized BiTwin:

n*b^1+/-1 and n*b^2+/-1, where n = a^2*C/c and b = c*A/a.

For each twin combination, check whether it extends.

If n*b^3+/-1 happens to be a twin then there are 2 links.

3 links requires n*b^4+/-1 to be a twin as well.

For 2 links, I chose A = 2^10*3^6*5^3*7^2*13#*19#*43#*400#, C = 31#^6*1723#.

For 3 links: A = 2^6*3^4*5^3*7^2*13#*23#*197#, C = A with 317# instead of 197#.

A and C must be relatively large to have enough divisors a and c for the two

twin pools.

I estimate the algorithm could be used on 4 links with some work, but I'm not

planning to do it.

5 links appears unrealistic at a reasonable size.

Henri Lifchitz' record page was getting long (60 screenfuls on my system) and

still growing. I suggested shortening it. It seems he now only lists BiTwins

with k links when they are above the record for k+1 links. This currently means

only I am listed for generalized BiTwins above 1 link. If anybody is curious

about the old page, it is here for a while:

http://hjem.get2net.dk/jka/math/BiTwinRec.htm

--

Jens Kruse Andersen