Bernoulli numbers B_k with denominators of the form 2^n - 2

54 views
Skip to first unread message

Tomasz Ordowski

unread,
Jul 20, 2026, 4:03:00 AMJul 20
to seq...@googlegroups.com
Hello! 

Conjecture: 
Denominator(B_k) = 2^n - 2 if and only if 
n = 2^m + 1 for 0 <= m <= 5 and k is a power of 2. 
Cf. https://oeis.org/A219196  > 1.  

Best, 

Tom Ordo 

Tomasz Ordowski

unread,
Jul 20, 2026, 2:15:30 PMJul 20
to seq...@googlegroups.com
PS. Let us define (using nice symmetry): 
Primes p such that 2^p - 2 is squarefree
and for every prime factor q of 2^p - 2
we have q - 1 | p - 1 or p - 1 | q - 1.  
Cf. A099596 - OEIS (here p odd). 

Geoffrey Caveney

unread,
Jul 20, 2026, 3:26:17 PMJul 20
to seq...@googlegroups.com
The "only if...k is a power of 2" condition is false: counterexamples include the denominators of B_14, B_26, B_34, B_38, B_62, etc. = 6, the denominators of B_68, B_76, B_124, etc. = 30, and the denominators of B_304, B_496, B_608, etc. = 510. One might rather conjecture that, for each given denominator value 2^n - 2, the values of k that are powers of 2 are all smaller than the values of k that are not powers of 2. 

Based on the comments in A219196 and A074777, it seems that your conjecture in general, with the above condition removed, may be equivalent to the conjecture that there do not exist any Fermat primes > 65537. The fact that the values 2, 6, 30, 510, 131070, and 8589934590 are each the product of 2 and the first m Fermat primes would also explain the squarefree and q - 1 | p - 1 or p - 1 | q - 1 conditions that you mention.


--
You received this message because you are subscribed to the Google Groups "SeqFan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to seqfan+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/seqfan/CAF0qcNMUJz4nv_-Q1nZp2WP1Gfo2t_ftiicQjg-zfr76hhkVog%40mail.gmail.com.

Tomasz Ordowski

unread,
Jul 20, 2026, 4:03:30 PMJul 20
to seq...@googlegroups.com
Yes, Geoffrey, thank you for clarifying my conjecture, 
but the PS issue needs better justification as to whether 
such symmetric odd primes p are exactly A099596. 
Primes p such that 2^p - 2 is squarefree
and for every prime factor q of 2^p - 2
we have q - 1 | p - 1 or p - 1 | q - 1. 

Tomasz Ordowski

unread,
Jul 23, 2026, 3:48:27 AM (14 days ago) Jul 23
to seq...@googlegroups.com
PS. I found an exception; 2^137-2 is not square-free. 
However, my definition is interesting in itself and 
could be the name of a new OEIS sequence.
Primes p such that 2^p - 2 is squarefree
and for every prime factor q of 2^p - 2
we have q - 1 | p - 1 or p - 1 | q - 1. 

Tomasz Ordowski

unread,
Jul 24, 2026, 5:50:42 AM (13 days ago) Jul 24
to seq...@googlegroups.com

Hello Everyone! 

It is known that the denominator of the Bernoulli number B_{2n} is of the form 2^m - 2 if and only if m = 2^k + 1 for k = 1, 2, 3, 4, 5. 

Indeed, by the von Staudt-Clausen theorem, 
Den(B_{2n}) = Product_{p-1 | 2n} p. 
Hence Den(B_{2n}) = 2 * Product_{F_i -1 | 2n} F_i, where the product runs over the Fermat primes F_i. Since F_5 = 2^(2^5) + 1 is  composite, the only possible Fermat-prime factors are 3, 5, 17, 257, and 65537, whose product is 
2 * 3 * 5 * 17 * 257 * 65537 = 2^33 - 2 = 2^(2^5+1) - 2. 

Now define a(n), for n > 0, as the largest integer k such that 
2^(2^k+1) - 2 divides Den(B_{2n}). 

It follows immediately that a(n) <= 5 for every n. 

Moreover, the condition 
2^(2^k+1) - 2 divides Den(B_{2n}) 
is equivalent to 
2^(k+1) divides 2n, 
that is, 
2^k divides n. 
Therefore, 
a(n) = min(v_2(n), 5), 
where v_2(n) denotes the exponent of 2 in n. 

Consequently, the sequence is periodic with the least period 
P = 2^15 = 32768, 
since min(v_2(n + 2^15), 5) = min(v_2(n), 5) for every n, while no smaller power of two has this property. 

Has this sequence, or the simple description a(n) = min(v_2(n), 5), appeared before in the literature or in OEIS? 

Are there any  interesting generalizations that come to mind? 

Best, 
Tom Ordo 


M F Hasler

unread,
Jul 24, 2026, 8:37:26 AM (12 days ago) Jul 24
to seq...@googlegroups.com


On Fri, Jul 24, 2026, 05:50 Tomasz Ordowski <tomaszo...@gmail.com> wrote:

Hello Everyone! 

It is known that the denominator of the Bernoulli number B_{2n} is of the form 2^m - 2 if and only if m = 2^k + 1 for k = 1, 2, 3, 4, 5. 


As Geoffrey wrote, the "if" part is not true.  "A if B" means B => A.
But you want A => B.

You must say 
" if the denominator of B(2n) is of the form 2^m – 2, then m = ... ".
Or, "the denominator... can be ... ONLY IF ..."

Because, consider for example 
m = 3 = 2¹ + 1.
Now tell me, is the denominator of B(2n) of the form 2^m – 2 or not ?

You can't tell !

Because your "necessary and sufficient condition" actually doesn't tell anything about *the* denominator of B(2n).
It tells something about the number 2^m–2, 
if that number is a denominator of (some) B(2n).


Maybe it's easier to see better why the first is absurd, by considering a simpler example of the same form:
" the least prime divisor of n is 2m+1 if and only if 2m+1 is prime."

That's exactly the form of your statement.
But you mean "only if m = ..., then it can be ..."

- Maximilian

Tomasz Ordowski

unread,
Jul 24, 2026, 9:38:05 AM (12 days ago) Jul 24
to seq...@googlegroups.com

Yes, Maximilian, formally you are right, but this inaccuracy does not affect the rest of my post, which I recommend reading.

If the denominator of the Bernoulli number B_{2n} is of the form 2^m - 2, then m = 2^k + 1 for k = 1, 2, 3, 4, 5. 

Indeed, by the von Staudt-Clausen theorem, 
Den(B_{2n}) = Product_{p-1 | 2n} p. 
Hence Den(B_{2n}) = 2 * Product_{F_i -1 | 2n} F_i, where the product runs over the Fermat primes F_i. Since F_5 = 2^(2^5) + 1 is  composite, the only possible Fermat-prime factors are 3, 5, 17, 257, and 65537, whose product is 
2 * 3 * 5 * 17 * 257 * 65537 = 2^33 - 2 = 2^(2^5+1) - 2. 

Now define a(n), for n > 0, as the largest integer k such that 
2^(2^k+1) - 2 divides Den(B_{2n}). 

It follows immediately that a(n) <= 5 for every n. 

Moreover, the condition 
2^(2^k+1) - 2 divides Den(B_{2n}) 
is equivalent to 
2^(k+1) divides 2n, 
that is, 
2^k divides n. 
Therefore, 
a(n) = min(v_2(n), 5), 
where v_2(n) denotes the exponent of 2 in n. 

Consequently, the sequence is periodic with the least period 
P = 2^15 = 32768, 
since min(v_2(n + 2^15), 5) = min(v_2(n), 5) for every n, while no smaller power of two has this property. 

Has this sequence, or the simple description a(n) = min(v_2(n), 5), appeared before in the literature or in OEIS? 

Are there any interesting generalizations that come to mind? 

Best, 
Tom Ordo 

--
You received this message because you are subscribed to the Google Groups "SeqFan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to seqfan+un...@googlegroups.com.

Tomasz Ordowski

unread,
Jul 25, 2026, 5:25:11 AM (12 days ago) Jul 25
to seq...@googlegroups.com
PS. Primes p such that 2^p - 2 is squarefree and 
if prime q | 2^p - 2, then q - 1 | p - 1 or p - 1 | q - 1.
Cf. A334797 - OEIS (add square-free requirement). 
Reply all
Reply to author
Forward
0 new messages