2) d, and similarly [using assumption (A3)]
large random problems we find that d' > d,
1 ) (~ l 2)~+I (d k 1 )2,2,
1) (d +
1)2~2.
(69)
For polynomials in Zp[x], assumption (A4) permits us to restrict our attention to
singleword primes; hence all coefficient operations can be performed in some fixed
amount of time. For Algorithm U, let dl =
O(Fi)
and d2 = 0(F2), and suppose
dl >_ d2 > 0. Then, by essentially the same argument that led to (63),
T~ (U) ~ d, (< 
d3),
(70)
where da is the degree of the GCD. For the Chinese remainder algorithm, we again
use the notation of Section 4.8. Let d~ and d2 be the degrees of m~ and m2, respec
tively. By (70), the time to compute c in Step (1) is codominant with
did2.
Since
Steps (2) and (3) can also be performed within this bound, we have
Tp (CRA) ~
&d2.
(71)
5.5
ALGORITmr C.
In analyzing the computing time for Algorithm C, we shall
use the notation of Section 2.4, and we shall assume that Step (4) is performed by
means of the subresultant PRS algorithm (Section 3.6). Let F~' and F2' be the given
nonzero polynomials in v variables, and let
(1, d)
bound their dimension vectors.
Let C(v, l, d) denote the maximum computing time for Algorithm C, and let
C~(v, l, d)
denote the time for the ith step. We shall omit the analyses of several
steps which obviously make no contribution to the final bound.
Step (1).
Since this step involves at most 2d +
1 GCD's of polynomial co
efficients, we have
C~(v, l, d) ~ d.C(v 
1, l, d).
(72)
!
Step
(2).
By (67), the time required to divide Cl or c2 into a coefficient of F~
or F2' '
"
~
2
1)2~2.
is dominated bv 1 (4 +
Since there are at most, 2 (d +
1 ) such divisions,
we have
C~ ,~, 12 (d +
1 )2~1.
(73)
Step (4).
Clearly most of the work in this step is in the pseudodivisions which
yield prem(Fi_2, Fi_~) for i = 3, • • • , k. As in Section 3.1, let dl denote the degree
of F~ in the main variable. Then by assumption (A1), we have d~ = d2 t 2 
i,
for i = 3, • • • , k, and therefore k < d2 4 2. Furthermore, by (20) the degree of F~
in any auxiliary variable cammt exceed
d(d~ 4
d2) _<
2d 2.
Similarly, by (22),
ignoring the logarithm terms in accordance with assumption (A3), the integer
length of Fi cannot exceed
l(d~ 4 d2) < 2ld.
Hence we can bound the time for a
single pseudodivision by replacing l by
2hl, d~
by d, 5 by 1, and d by 2 d 2 in (69).
Multiplying the result by d, which bounds the number of pseudodivisions, and
applying some obvious simplifying inequalities, we obtain
C4 ,~, l 2 (d +
1 )4~22~3~.
(74)
Step
(6).
To bound the time for computing
fk/O,
we can replace v by v 
1,
12 by l, d2 by d, 13 by
21d,
and da by 2d 2 in
(67); the result is dominated
Journal of the Association for Computing Machinery, Vol. 18, No. 4, October 1971
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
500
W.S.
BROWN
by 12 (d +
1 )3v22v. To bound the time for dividing this quantity into a coefficient
of Fk, we can replace v by v 
1, 12 by
21d, d2
by 2 d ~, 13 by l, and d3 by d in (67),
with the same result. Multiplying by d +
2, which bounds the number of such
divisions, we find
C6 £
t ~ (d +
1)3v12v.
(75)
Step (7).
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '13
 MRR
 Math, Coefficient, F~

Click to edit the document details