Reference Request on the existence of $k$ satisfying $P(\Phi_d(2))^k \gt \Phi_d(2)$ for all $d$ I am working my way through the literature regarding the following conjecture:  There is a positive integer $k$ such that for all positive integers $d$, 
$$P(\Phi_d(2))^k \gt \Phi_d(2).$$
I am looking at factors of Mersenne numbers $2^d-1$, so using factors from cyclotomic polynomials $\Phi_d(x)$ is a first step.  $P(n)$ comes from the literature and represents the largest prime factor of integer $n$ or $1$ if $n$ has no prime factors.  Examining the tables at the Cunningham project suggest that $k\gt 3$, but I have not found any theory to resolve the conjecture.
What references address this conjecture?  Note that much of the literature focuses on $P(2^d -1)$ and not on the quantity above.  Also I am interested in what is known about the number of factors of $\Phi_d(2)$, or the rough size of $P(\Phi_d(p))$ for $p$ prime, so references that come close to prime factors of cyclotomic values are welcome.
Gerhard "Primarily Interested In The References" Paseman, 2016.04.09.
 A: Given a positive integer $k$, the natural density of the set of positive
integers $n$ whose largest prime factor is smaller than the $k$-th root of $n$
is estimated by the value of the Dickman function $\rho$ at $k$,
cf. also Page 106 in
David M. Bressoud, Factorization and Primality Testing.
  Springer-Verlag, 1989.
For all $k$ the value of this function is strictly positive.
Therefore if instead of the numbers $\Phi_d(2)$ you would consider all 
integers, regardless of which $k$ you choose, the set of counterexamples
to your conjecture would have positive natural density.
Now, your numbers $\Phi_d(2)$ are not arbitrary integers. In particular, 
their prime divisors satisfy certain congruence conditions, which prevents
the numbers from having very many very small prime factors. The effect of 
these congruence conditions is quite big as long as the numbers are
relatively small, but it is likely asymptotically negligible when $d$ goes
to infinity. -- However, as Lucia already wrote, what is presently really
known here does not suffice to disprove your conjecture.
Finally, obtaining evidence for or against your conjecture by means of
computation is likely difficult -- already for $k = 6$ one would statistically
expect one counterexample every $50000$ values of $d$ or so, hence finding one
is likely to require factoring integers with several thousands of decimal digits.
