The ECM method is a probabilistic method, and can be viewed in some sense
as a generalization of the P-1 and P+1 method, where we only require that
P+t+1 is smooth, where t depends on the curve we use and satisfies
|t| <= 2*P^(1/2) (Hasse's theorem). The optimal B1 and B2 bounds have
to be chosen according to the (usually unknown) size of P. The following
table gives a set of nearly optimal B1 and B2 pairs, with the corresponding
expected number of curves to find a factor of given size (column "-power 1"
does not take into account the extra factors found by Brent-Suyama's exten-
sion, whereas column "default poly" takes them into account, with the poly-
nomial used by default: D(n) means Dickson's polynomial of degree n):
digits D optimal B1 default B2 expected curves
N(B1,B2,D)
-power 1 default poly
20 11e3 1.9e6 74 74 [x^1]
25 5e4 1.3e7 221 214 [x^2]
30 25e4 1.3e8 453 430 [D(3)]
35 1e6 1.0e9 984 904 [D(6)]
40 3e6 5.7e9 2541 2350 [D(6)]
45 11e6 3.5e10 4949 4480 [D(12)]
50 43e6 2.4e11 8266 7553 [D(12)]
55 11e7 7.8e11 20158 17769 [D(30)]
60 26e7 3.2e12 47173 42017 [D(30)]
65 85e7 1.6e13 77666 69408 [D(30)]
Table 1: optimal B1 and expected number of curves to find a
factor of D digits with GMP-ECM.
After performing the expected number of curves from Table 1, the
probability that a factor of D digits was missed is exp(-1), i.e.,
about 37%. After twice the expected number of curves, it is exp(-2),
i.e., about 14%, and so on.
Example: after performing 8266 curves with B1=43e6 and B2=2.4e11
(or 7553 curves with -dickson 12), the probability to miss a 50-digit
factor is about 37%.

и
,
он самый быстрый. Для полупростых видимо ECM всегда хуже QS, хотя для чисел
стоит это аккуратно перепроверить.
(используются числа до корня квадратного, а они влезают в регистры и медленная длинная арифметика не нужна).
(первое попавшееся) он 25-digit собрался перебирать минут 12 (за 5м перебрал около половины и переключился на QS).
было достаточно гладким. Может быть есть какие то признаки, и зная
..