MATHEMATICS OF COMPUTATION, VOLUME 29, NUMBER 130

APRIL 1975, PAGES 620-647

New Primality Criteria and Factorizations of 2™d= 1

By John Brillhart, D. H. Lehmer and J. L. Selfridge

Abstract. A collection of theorems is developed for testing a given integer N for

primality. The first type of theorem considered is based on the converse of Fer-
mât 's theorem and uses factors of N — 1. The second type is based on divisibility
properties of Lucas sequences and uses factors of N + 1. The third type uses fac-
tors of both N — 1 and N + 1 and provides a more effective, yet more complica-
ted, primality test. The search bound for factors of N ± 1 and properties of the
1 2
hyperbola N = x —y are utilized in the theory for the first time.
A collection of 133 new complete factorizations of 2m ± 1 and associated
numbers is included, along with two status lists: one for the complete factoriza-
tions of 2m ± 1 ; the other for the original Mersenne numbers.

1. Introduction. The theory of testing a given odd integer TV for primality by

some converse of Fermat's theorem, or by its generalization in Lucas sequences, was
begun in 1876 by Lucas ([9], [10, p. 302]).
Since that time, this theory has gradually been developed by various writers
(Proth [15], Lucas [11], Pocklington [14], Lehmer [6], [7], [8], Robinson [18],
Brillhart and Selfridge [4], Williamsand Zarnke [21], Riesel [17] ) in the direction of
reducing the amount of calculation needed to complete a primality test on N.
In Sections 2 through 7 of the present paper, this purpose is carried considerably
further. The contents of these sections are the following:
Section 2 contains two theorems in which A^ — 1 is completely factored. Theo-
rem 1 was given earlier in [4]. Theorem 2, which is somewhat unfamiliar, is an im-
provement on Theorem 1 (see Kraitchik [5] ). In the latter theorem, the condition
a(N-i )/2 = _ j (moci /v) is used (see [18]) rather than the usual test that TVis a "pseudo-
prime base a."
Section 3 contains five theorems and three corollaries which use only partial factor-
izations of TV- 1. Theorem 3 is a strengthening of a theorem of Proth [15]. Theorem 4
and Corollary 1 are familiar. Theorem 5 is new and is an advance over the old theory in
that the factored portion of TV- 1 need only be about TV1^ before the primality test can
be completed. Corollary 3 brings the direct search bound for factors of TV- 1 into the
theory for the first time. Theorem 7 uses thn bound to construct an improved version of
Theorem 5. Ordinarily, representing N numerically as a difference of squares is used for

Received January 28, 1974.

AMS (MOS) subject classifications (1970). Primary 10A25; Secondary 10A35.
Key words and phrases. Primality testing, factorization of 2m ± 1, Lucas sequences, converse
of Fermat's theorem. „Copyright
., ©„ 1975.
,.,, American
, .. „ .. .. _,,
Mathematical Society

620

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m + 1 621

the purpose of factoring a composite TV. However, this representation is used in

a new way to establish the primality test in Theorem 7. It also appears indirectly
in the proofs of Theorems 5,17, and 19.
Section 4 contains a resume of properties of Lucas sequences that are needed
for the theoretical developments in Sections 5-7.
Sections 5 and 6 exactly parallel Sections 2 and 3 in that they contain com-
parable theorems in which factors of TV+ 1 are used instead of those of TV— 1.
That such a parallel development is possible rests on Theorem 16, which is due to
Michael Morrison [12]. The discovery of this theorem came as a surprise, since,
previously, it had been thought that the theory using the factors of TV+ 1 was
considerably more complicated.
Section 7 contains two theorems and a corollary which utilize factorizations
of both TV— 1 and TV+ 1. A considerable advantage is gained thereby since the
amount of factorization needed to test TV for primality is substantially reduced.
Theorem 21 is unusual in that it does not deal directly with the prime factors of
TV± 1, but rather with the primes dividing algebraic factors of these numbers.
The final section of the paper contains a discussion of numerical results, a
listing of which is given in three tables. In particular, 133 complete factorizations
of 2m ± 1 and associated numbers are given, along with a status table showing
which numbers of these forms have been completely factored. A current status
table for the Mersenne numbers IP — 1, p < 257, is also included.
It should be noted that many of the theorems in this paper are stated in
more detail and generality than may be needed for some applications. In such
applications, some of the variables can be set to their minimum values, and minor
terms can often be dropped. The generality in the theorems may be of use in
certain cases and has been given to delimit more carefully the theoretical results.

2. Theorems Requiring a Complete Factorization of TV— 1. As it sometimes

happens, a complete factorization of TV— 1 can be found without difficulty. For
example, if TVhas a special form such as TV= 3 • 2m + 1, or if by chance TV— 1
possesses only small prime factors which can be discovered almost immediately by
direct search, the complete factorization is at hand. In these cases, because of
the uncomplicated nature of the theorems in this section, as well as Theorem 3
in the next section, a simple program can be written to carry out the primality
testing which does not require much memory space. Such a program, however,
requires more running time than one based on later sections, but may be more
suitable for use in small computers where memory space is limited (see Selfridge
and Guy [20]).
By way of notation, the symbol TVwill denote an odd integer > 1, and p,
q, and n (as well as p¡, q¡, and n¡) will denote primes throughout the rest of this
paper. The expression "TVis a psp base a" will be used for a number TVwhich

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 622 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

satisfies the congruence aN~l = 1 (mod TV), 1 <a < TV- 1, i.e., TVis a "pseudoprime"
base a. (Since a is chosen in advance, it is extremely rare that TVis composite when it
is found to be a psp base a.)
Theorem 1. Let TV— 1 = n p¡ '. If for each p¡ there exists an a¡ such that TVis
a psp base a¡,but a¡ p'^ 1 (mod TV), then TVis prime.
Proof. Let e¡ be the order of a¡ (mod TV). Since e¡ | TV- 1, but e( '('(TV- l)lp¡,
then pi '\e¡. But for each i, e¡ | 4>(N), so that p¡ '\ 0(TV),which implies TV- 1 |0(TV).
Hence, TVis prime. Q.E.D.
Remarks. 1. Theorem 1 indicates that if for any p¡ a base a¡ can be found for
which both hypotheses are satisfied, then that p¡ is settled once and for all. (See [4, p.
89].) This is in contrast to the somewhat less satisfactory situation in earlier theorems
(see Lehmer [6] and Lucas [11]) where a single base a is used for which the hy-
potheses must be satisfied for all p¡.
2. The computations for each p¡ can be done efficiently by calculating

(1) a^-\)lPi _b^ l (mod N^ and then bP, _ j (mod Ny

3. In practice a good strategy for choosing the a¡ is the following:

(i) Find «j by the quadratic reciprocity law so that (a,/TV) = — 1.
(ii) Use a1 for successive p¡ as long as (1) is satisfied. (For each p¡ for which the
base is not changed, it is of course not necessary to compute the second part of (1).)
(in) Whenever (1) is not satisfied, change the base according to (i), returning to
a previous base, if possible, to avoid having to recompute the second part of (1).
The next theorem is an improvement over Theorem 1 in that slightly less calcula-
tion is required to complete the primality test.
Theorem 2. Let TV- 1 = n pt '. If for each p¡ there exists an a¡ such that

(2) «f-')'2H- 1 (mod TV),

but (for p, > 2),

(3) a¡N-l)l2p'*-l (mod TV),

then TVis prime.
Proof. Congruence (2) implies TVis a psp for each base a¡. For each p¡ > 2, if
a(N-l )/2P/ s b¡ (mod Nl then a(N- 1),pt sb2pi (mod N). for> if b2 m !
(mod TV)for some /, then, since pf is odd, —1 = a\N~l^l2 = b¡ ' = b¡ (mod TV),which
contradicts (3). Hence, TVis prime by Theorem 1. Q.E.D.

3. Theorems in Which TV- 1 is Partially Factored. In the special case where a

prime factor of TV— 1 exceeds s/Ñ/2 - 1, the next theorem, which is a strengthening
of a theorem of Proth [15], provides a primality test involving less computing than
Theorem 2.
Theorem 3. Let TV- 1 = mp, where p is an odd prime such that 2p + 1 > \/TV.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 new primality criteria and factorizations of 2m± 1 623

// there exists an a for which a(N~' )/2 = - 1 (mod TV),but aml2 £ -1 (mod TV),
then TVis prime.
Proof. Let e be the order of a (mod TV). Then e | TV- 1. But, using the same
argument as in the proof of Theorem 2, am £ 1 (mod TV), so e \ (TV- l)/p. Hence,
p | e, and since e | 0(TV), then p | </>(TV).
Also,
0(TV)|/Vn(/z¿
- 1) = (mp + l)n(H(.- 1),
so p | n(«(. - 1), or p | n, - 1 for some i, say /' = 1. Thus, «, = 1 (mod 2p). But
TV= 1 (mod 2p), which implies TV/«X= 1 (mod 2p). On the other hand, since «, >
2p + 1 > \/TV, then 1 < TV/Wj< VtV"< 2p + 1. Therefore, the only possibility for
TV/Hjis 1, so TVis prime. Q.E.D.
Remark. This theorem reduces the amount of testing because the prime factors
of m can be ignored. Also, note that p need not be the largest prime divisor of TV- 1,
as TV= 31 and p = 3 shows.
Throughout the rest of this paper the notation TV— 1 = F1Rl will be used,
where F1 is the even factored portion of TV- 1, R x is > 1, and (Fl, R x) = 1.
Theorem 4 (Pocklington [14]). If for each prime p¡ dividing Ft there exists
an a¡ such that TVis a psp base a¡ and (a¡ ' — 1, TV) = 1, then each prime divisor
of N is = 1 (mod F¡).
Proof. Let « be a prime divisor of TV,and e{ be the order of a¡ (mod n). Then
«,111-1. Also, of'1 = 1 (mod n), so e¡ |TV- 1. On the other hand, (a?~l)IPi -
1, n) = 1, so e¡ -f (TV— \)¡p¡, which implies p¡ ' \e¡, where p¡ ' IlFx. Hence, for each
i, p"''|« - 1, so that F, |n - 1. Q.E.D.
Remark (R. DeVogelaere). In verifying the hypotheses of this theorem, only
one GCD computation is necessary: First find an a¡ such that a¡ ' — 1 = b¡ £ 0
(mod TV) for each /; then calculate the product II b¡ = c (mod TV); and finally, if c ^
0, compute d = (c, TV).
If d i= 1, then TVis composite and a factor has been found. Also, if c = 0, then
some b¡ has a prime factor in common with TV.
For convenience of reference put:
(I) . For each prime p¡ dividing Fl there exists an a¡ such that TVis a psp base a¡
and{af-ï)lp'-\,N)=\.
Corollary 1. Assume (I). If Fx > \/N, then TVis prime.
Remark. Corollary 1 is an improvement over Theorem 2 in that the primality
test can be completed as soon as the factored part of TV- 1 exceeds the unfactored
part. This saving in time is offset only to a slight degree by the amount of computing
needed to calculate the required GCD's. It will be the main goal of the rest of this
paper to continue to reduce the amount of auxiliary factorization, as in this case,
through the introduction of various conditions which require a small amount of com-
puting time as compared to the factoring time eliminated. In this regard, the next
theorem is a considerable improvement on Corollary 1, since TV— 1 need only be fac-
tored to the point where FI > (TV/2)1/3 rather than Fx > s/Ñ. A further reduction is

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 624 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

possible if m is chosen to be > 1. The cost of this reduction is at most the time needed
to calculate (r2 — 8s)Vl and the trial division of XFj + 1 into TVfor m — 1 values of X.
Theorem 5. Assume (I) and let m be > 1. When m > 1, assume further that
\Fl + 1 \ Nfor 1< \<m. If

(4) TV< (mFx + 1) [2F2 + (r - m)F1 + 1],

where r and s are defined by R¡ = (TV- 1)/Fj = 2Fls + r, 1 < r < 2Fv then TVis
prime if and only if s = 0 or r2 - 8s ¥=D. (r^O since R1 is odd.)
Proof. The theorem will be proved in the equivalent form: TVis composite if and
only if s =¿0andr2 - 8s = D.
(i) (=>). From Theorem 4 it follows that all factors of TVare 1 (mod F{).
Thus, since TVis composite,

(5) TV= (cFt + \)(dFl + 1), c,d>m.

Also, Ft is odd and Fl is even, so the equation

(6) Rl={N-\)¡Fl=cdFl+c + d

implies that c + d is odd, so cd is even. Hence, from

(7) cdF1 +c +d = Rl =2FjS+r

it follows that
(8) c + d=r (mod 2Fj),
where c + d - r > 0, since r is the least positive remainder (mod 2F1 ). On the other
hand, (c - m) (d - m) > 0 implies cd > m(c + d) - m2, so that

(mFl + 1) [2F2 + (r - m)F1 + 1] >N = cdF2 + (c + d)Fl + 1

> [m(c + d)- m2] F2 + (c + d)Fl + 1

= (mFi + l){[(c + d)-m]Fl + 1}.

Thus, 2F2 + (r - m)Fl + 1 > [(c + d) - m] F, + 1, or c + d - r < 2FX. Com-
bining this result with (8) gives c + d = r. Thus, from (7) it follows that 2s = cd ¥= 0.
Finally, r2 - 8s = (c + d)2 - 4cd = (c - d)2.
(ii) (<=). With s + 0 and, say, r2 - 8s = t2, then
N = FiR1 + 1 = F!(2F,s + r) + 1 = [(r2 - r2)F2/4] + rFx + 1

where the factors on the right are > 1, since s =£ 0. Q.E.D.

Remarks. 1. In the factorization in (ii), if m > 1, the two factors are prime; for
if TV= (cFj + l)(dFj + l)(eFj + 1), where c, d, e > m > 2, then (4) is contradicted.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 625

To see this, it is sufficient to consider the smallest values of the coefficients, i.e.,
when c = d = e = m. Then
TV= (mFj + l)3 = (wFj + l)[m2F2 + 2/nF, + I] > (mF1 + 1)[4F,2 + 2mFl + 1]

> (mFl + l)[2F2 + (r + 2m)Fl + 1] > (mF1 + 1)[2F2 + (r - m)Fl + 1].

This argument does not hold when m = 1.
2. Note that the right side of (4) is composite, so the inequality is sharp. (Cf.
[18, Theorem 10], where F1 = 2n.)
3. The choice of m in the hypothesis is arbitrary. It would usually be chosen
large enough to ensure that (4) is satisfied. Increasing the size of m for this purpose, of
course, must be weighed against further factoring of TV— 1 to try to increase the size of
Fj. Differentiating the right side, f(m), of (4) with respect to the real variable m (with
Fj and r constant) gives the critical value m = Fl + r/2. Thus, 1 < m < Fl +r/2 and
the largest TVthat can be tested by Theorem 5 is less than the integer f(Fx + r/2) =
(F2 +rFj2+ l)2.
4. The coefficient 2 in (4) arises because 2 divides cd in (6). In general, if it can
be shown that some odd integer g also divides cd, then the coefficient 2 in (4) can be re-
placed by 2g. The 2 in the definition of r and s must also be replaced by 2g. For
example, if TV= - 1 (mod 3), then in (5) one of the factors, say cFx + 1, must be
= 1 (mod 3). Thus, 3 | cFx, and since 3 \FX, 31c, i.e., 31cd.
Also, if TV= — 1 (mod 5), and it is known that 5 is a quadratic residue of TV,
then since 5 -f"FX, 5 \cd. If TV= - 1 (mod 8), and 2 is a quadratic residue of TV,then
81cdFx. But since N - 1 = - 2 (mod 8), 2IIFx, which implies 4 \cd (instead of 2
dividing cd). Similarly, if TV= 3 (mod 8), and - 2 is a quadratic residue of TV,then
8|cdF,,2llF,,and 4|cd.
It should be observed that the above conditions, when they hold, can be com-
bined to give a larger leading coefficient in (4). (These observations are due to Michael
Morrison.)
Theorem 6. Let n be a prime divisor of TV. If N is a psp base a, and

(9) (/»- \,N)=\,

then n = 1 (mod p), where p is some prime divisor of Rx depending on n.
Proof Let e be the order of a (mod n). Then e\n — 1. Also, since TVis a psp
base a, it follows that e|TV- 1 = FjFj. But from (9), aFl f- 1 (mod «), so e \ F,.
Hence, (e, Rx) > 1, i.e., there exists a prime p such that p\e and p|F,. Thus, p\n —
1. Q.E.D.
For convenience of reference put:
(II) For some a, N is a psp base a and (a i — 1, TV) = 1.
Remark. The exponent in (II) has the same form as the exponent in (I), so in a
program, (I) and (II) can be treated as a single test by considering Rx as the final
"prime" factor of TV- 1.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 626 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

Corollary 2. Assume (I) and (II), and let n be a prime divisor of N. Then n =
1 (mod pFx), where p is some prime divisor of Rx depending on n.
Proof. Since (Fx, Rx) = 1, the corollary follows from Theorems 4 and 6. Q.E.D.
Corollary 3. Assume (I) and (II). If all the prime factors of Rx are > Bx
and BXFX> \/W, then TVis prime.
Proof. From Corollary 2, n — 1 > pFx > BXFX> %/TV,which implies TVis prime.
Q.E.D.
Remark. The new feature on Corollary 3 is that Bx appears in the inequality for
TV. The number Bx is quite different from Fx, since Fx contains the "discovered"
factors of TV— 1, while Bx gives the information (not immediately verifiable) that the
prime factors of Rx are greater than or equal to Bx. (This latter assumes that no factor
of TV— 1 has been overlooked, as it might be if the computer were not working
properly.)
The next theorem, which improves on Corollary 3, uses formulas relating to the
hyperbola x2 —y2 = TV,in a way similar to what was done implicitly in the proof of
Theorem 5.
Lemma 1. If either 0<a<b < \/TV or y/Ñ < Z>< a, then b + TV/Z?
<a+
TV/a.
Proof. The conclusion follows from (a-1 - ¿>_1)(TV- ab) > 0. Q.E.D
Theorem 7. Assume (I) and (II), and also that the prime factors of Rx are > Bx.
If

(10) TV< (BXFX+ 1) [2F2 + (r - BX)FX + 1],

where r and s are defined by Rx = 2Fxs + r, 1 < r < 2FX, then TVis prime if and
only ifs = 0orr2-8s^D.
Proof. The theorem will be proved in the equivalent form: TVis composite if
and only if s + 0 and r2 - 8s = D.
(i) (=>). From Theorem 4 all the factors of TVare 1 (mod Fx ). Since TVis
composite, it can be written asTV= nw = x2 -y2 = (x -y)(x +y) = (cFx + l)(dFx + 1),
c, d > 1, where n is the smallest prime factor of TVand w > 1. Then TV= cdF2 + (c + d)Fx
+ 1 and 2x = (c + d)Fx +2. But R, = cdF\ + c + d, and since R x is odd and F, is
even, then c + d is odd, so that cd is even, say cd = 2g. Then TV= 2gF\ + 2x - 1,
so 2x = F1Ä1 + 2 - 2gF\ = Fx(2Fxs + r) + 2 - 2gF\ = (s - g)2F\ + rFx + 2.
Let X = s - g. Then from rFx + 2 < Ft(2F, - 1) + 2 < 2F2 it follows, since x >
0, that 0 < 2x = 2XF2 + rFx + 2 < 2F2(X + 1), so that X > 0. On the other hand,
2x=n+w=n+ N/n, and from Corollary 2, n = 1 (mod pFj), son > pFj + I >
BXFX + 1. Hence, using Lemma 1 and (10), 2XF2 + rF, + 2 = 2x = n + N/n <
(T^F, + 1) + TV/(FjFj + 1) < (£,F, + 1) + 2F2 + (r - Ax)Fj + 1 = 2F,2 4- rFx
+ 2. Consequently, X < 1. Thus, X = 0 and rFx + 2 = 2x = (c + d)Fx + 2, which
implies r = c 4- d. Then 2Fts + r = Rx —cdFx + c + d gives 2s = cd ^ 0.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 627

Finally, r2 - 8s = (c + d)2 - 4cd = (c - d)2.

(ii) (•*=). The proof is the same as Theorem 5(ii). Q.E.D.
Remark. If it happens that Rx is a pseudoprime but Bx is not large enough for
(10) to be satisfied, then a primality investigation can be carried out on Rx itself (see
Brillhart [3, p. 448]). If it can be shown that Rx is prime, then the theorems of Sec-
tion 2 can be used to show TVis prime. If, however, it is difficult to show that Rx is
prime, Theorem 4 can at least be used (with the factors of Rx - 1) to establish a lower
bound for the prime factors of Rx, which, if it exceeds Bx, can replace Bx in Theorem 7.

4. Lucas Sequences. The primality theory which was established in the preceding
sections was based on factoring TV- 1. In this section and the two that follow, a
primality theory is developed which depends on factoring TV+ 1.
Central to the TV+ 1 theory are the divisibility properties of certain second order
recurring sequences known as Lucas sequences. These properties, which contain Fer-
mat's theorem as a special case, will be reviewed here along with several other results that
apply to the later development. Some of the more familiar results will be given with-
out proof (see Lucas [10] ).
The Lucas sequences {Uk} and {Vk} are defined recursively by the formulas:

Uk+2=PUk+l-QUk, k>0, u0 = o, ux = \,

Vk+2=PVk+l-QVk, k>0, V0 = 2, VX=P,

where P and Q are integers such that D = P2 — 4ß =/=0. (In case several sequences,
defined by P¡ and Q¡, are used, the notation {U^} and {V^} will be employed.)
If a and ß are the (unequal) roots of x2 - Px + Q = 0, then the members of
these sequences can be expressed in terms of a and ß by the equations:

Uk = (ak - ßk)/(a - ß) and Vk = ak + ßk, k > 0.

From these formulas four useful identities can be derived:

(ii) u2k = ukvk,

(12) DU2 = V2k - 2Qk,

(13) V2 -DU2k = 4Qk,

(14) 2Vr+s=VrVs+DUrUs.
In what follows the notation et will be used for the value of the Jacobi symbol
(D/t).
The main divisibility properties of these sequences are contained in the theorems
and corollaries which follow.
Theorem 8. (a) Ifp^2Q,thenUD_e y p =0 (mod p).
(b) Ifp\2QD,then V =2Q^-^p)n (modp).

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 628 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

Remark. Theorem 8(a) is the generalization of Fermat's theorem mentioned earlier.

As such, it could also be used as a test for compositeness: IfTV'f'g andTV'f'iJ^ , then
N
TVis composite. (Fermat's theorem can be obtained from Theorem 8(a) in the following
way: Let p be an odd prime such that p \a(a — 1). Consider the Lucas sequence with
OL= a and0 = 1, soD = (a - l)2. Then ep = 1 andap_1 - 1 = (a - \)U x =0
(mod p).)
Theorem 9. Ifp\2QD, then P\U(p_e )/2 if and only if (Q/p) = 1.
Proof. Identity (12), Theorem 8(b), and Euler's criterion give

DU2p_£p)l2 = Vp_€p - 2Q(p-eP)l2 =2Q(1~eP)l2 - 2(Q/p)Qil-eP)l2

= 2Q(l'ePy2{l-(Q/p)} (modp),
from which the theorem immediately follows. Q.E.D.
Corollary 4. Ifp\2QD, then p\ V^p_e )/2 if and only if (Q/p) = - 1.
Proof. This follows from Theorem 8, (11), Theorem 9, and (13). Q.E.D.
From Corollary 4 a test for compositeness can also be obtained.
Corollary 5. Suppose N^QDand that (Q/N) - - I. IfN^V{N_e )/2,
then TVis composite.
Remark. The residues of U and Vm (mod TV), which must be computed in
these theorems, can be computed with about triple the work of computing a power
(mod TV). An efficient method for calculating Vm (mod TV) is discussed in detail in
Lehmer [8, p. 129]. To compute Um (mod TV) one can use the formulas: U2k =
UkVk and V2k = Vk - 2Qk for doubling the subscript, and U2k+X =
(PU2k + V2k)/2 and V2k+X = (DU2k + PV2k)/2 for a "side-step" of 1. The sequence of
doublings and side-steps to be followed is easily obtained from the binary expansion of m.
Theorem 8 shows that an odd prime p, not dividing Q, will divide at least one
term of {Uk}, namely Ue . The least positive k such that p\Uk is called the "rank
of apparition" of p (or just "rank") and is denoted here by p(p). (If several Lucas
sequences {U^} are being employed, then p¡(p)will denote rank in {Uj¡.')}.) This
notation will also designate the rank of a composite number.
Theorem 10. Suppose p\2Q and that pa IIUp(p), a > 1. Then Pa+ßWmp(p)
if and only if p^Wm.
Remark. If a prime p divides Q but does not divide P, then p-\Uk, k> \.
When (TV,Q) = 1, the following formula for p(TV) can be obtained from Theorems
8(a) and 10:
P(N)= LCM [pKKmaX(7ra,',0)l

where TV= ITf=1rij* and n^W^.y

Theorem 11. Suppose (TV,Q) = 1. Then
(a) p(N) exists.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 629

(b) N\Ukifandonlyifp(N)\k.
It will be convenient to introduce a function, similar to the Euler 0 function,
which will be of use in deriving the primality theorems.
Definition. If (TV,D) = 1 and TV= W¡=, n¡*, let

^(N,D)=2y-°f=i
n («/-o»?'-1-
'
(This function is not a generalization of the Euler function, because of the power of 2
in front of the product.)
Theorem 12. // (TV,D) = 1, then t//(TV,
D) = TV- ejy */aw? on/y //TVz'sprime.
Proof. (•*=). Clear from the definition of \¡i.
(=>). The statement will be proved in the equivalent form:
If TVis composite, then <//(TV,
D) =£TV- eN.
Case 1. s = 1, i.e., TV= n7, ?> 2. 77ie«
i//(TV,
D) = (n - e„)n7" ' = TV- TVe„/wJ=N- eN.

Case 2. s > 2. In this case

W D) = 2U fc(/i,- e„>J<"-1< 2 n W*i+ O"?''"1

/=i ' i=i

= 2TVn^(l+i-)<2Tv(|)(|)...<f <TV_1. Q.E.D.

Corollary 6. //(TV, D) = 1, then N - eN\i//(TV,Í)) implies that TVz'spriwe.
Proof. UN is composite, then i//(TV,D) < TV- 1 in Case 2 of the above proof.
In Case 1, TV- e^ |TV- TVen/« implies en = — 1. However, in that case n1 ± 1 \ny +
n7_1, which is impossible when 7 > 2. Q.E.D.
Corollary 7. //(TV, Q/J) = 1, then p(N)\\p(N, D).
Proof. The condition (TV,QD) = 1 implies TVhas a rank. Thus
p(TV)= LCM [p(H>max(^-^0>]

which divides
LCM[(«,.- e„>7/_1] = 2 LCM[H(n,- e_ )«7'_1],
which divides

2 fl Hn¡ - e„.)«/'_1= WTV,

D). Q.E.D.
i=l '

5. Theorems Requiring a Complete Factorization of TV+ 1. With the preparation

in the last section it is now possible to prove a collection of theorems based on the
factorization of TV4- 1. These theorems, which are proved in this and the next section,
exactly parallel Theorems 1—7.
Lemma 2. Let {Uk} be a Lucas sequence for which (D/N) = - 1 and N\UN+ x.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 630 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

Then (TV,QD) = 1, i//(TV,

D) is defined, and TVhas a rank which dividesTV+ 1.
Proof. Since the Jacobi symbol (D/N) =£0, it follows that (TV,D) ~ 1. If there
were a prime n dividing both TVand Q, it would follow from D = P2 —4Q that n \ P,
since n -\D. But then the remark following Theorem 10 would imply n, and therefore
TV,has no rank, contrary to the fact that TV|UN+ x. Therefore, (TV,Q) = 1. The
remainder of the conclusion follows from the definition of i//(TV,D) and Theorem 11.
Q.E.D.
Theorem 13. Let TV4- 1 = IT(jr.', and consider the set U of Lucas sequences
{Uk} with the given discriminant D for which the Jacobi symbol (D/N) = — 1. If for
each q¡ there exists a Lucas sequence in U such that N\ uff, but N-^Uy^+ .. , then
TVis prime.
Proof. It is clear from Lemma 2 that p¡(N)\N 4- 1. But p,.(TV)>(TV4- l)lqt, so
¿?f''lp,(TV).
ByCorollary7, pi(N)\\p(N,D)for all/. Thisimplies<jf''| \J/(N,D). Thus,
TV4- 11t//(TV,
D), so TVis prime by Corollary6. Q.E.D.
Remarks. 1. This theorem corresponds to Theorem 1 in that it allows for a
change to another sequence with the same discriminant if TV|UrA, ... for some q-.
As such, it constitutes an improvement over the earlier theorem in which a single se-
quence with P = 1 was employed (see [8, p. 128]).
2. From one Lucas sequence with Px, Qx, and D, another with the same D can
be obtained by setting P2= Px 4-2 and Q2 = Px 4- Qx 4- 1. (It is necessary to check
that (TV,ß,) = 1.)
The next theorem improves on Theorem 13 in that only V's (with smaller sub-
scripts) are calculated in the primality test (see the remark following Corollary 5),
(also see Theorem 3, p. 128 in [8]).
Theorem 14. Let TV4- 1 = ilqt ' and consider the set \J of Lucas sequences
{Vk^} with the given discriminant D for which the Jacobi symbol (D/N) = — 1. If for
each q¡ there exists a sequence in \J such that

O5) TVI7(,'>
JV| K(AT+l)/2'
but (for qt > 2)

1 K(AT+1)/2V
then TVis prime.
Proof. From (11) and (15) it follows for each i that TV|UJp+x, so p,(TV)exists
and p,(TV)|TV
4- 1 by Theorem 11(b). Also, for each qf > 2, TVt ^$4.,),,.; for, if
Wl U(N+Xyq. for some i, then from Theorem 11(b),

(17) NIU0)

where s = (q¡ - l)/2. But then, using (15), (14), and (17), it follows that

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 631

n = 9 I/O') = ? pO')
U ~ ZK(AT+l)/2 ZK[i(iV+l)/<îI.+ (iV+l)/2rî(.]

= i/(0 T/(i) + n/y(0 r/(0

Vs(N+l)/qiy{N+l)l2q¡ T UUs{N+ 1 )/q, (JV+ 1 )/2qf

= FSv+i ),</$+, )/2«,. (modiV).

Now, (TV,^i]^+ j w ) = 1; for, if a prime divided both numbers, it would divide

uHN+i)/q- by (17)>and so by (13)would divide ö- But byljdmm& 2>(N>Q) = *•

Hence, TV|V^+X)l2 , which contradicts (16). Thus,TVis prime by Theorem 13. Q.E.D.

6. Theorems in WhichTV4- 1 is Partially Factored.

Theorem 15. Let TV4- 1 = mq, where q is an odd prime such that 2q - 1 >
\/TV. // there exists a Lucas sequence {Vk} of discriminant D with (D/N) = - 1 for
which TV| V(N+ j y2, but N \ Vm/2,then TVis prime.
Proof. From (11) it follows that TV|UN+ x, so p(TV)exists and p(TV)| TV4- 1 by
Theorem 11(b). Also, using the same argument as in the proof of Theorem 14, TV'f'
U(N+i)/q >so P(TV)-f(TV
4- l)/q. Hence, q \p(N), and since p(TV)|i//(TV,
D) by Corollary
7,q\\p(N,D). But

V/(TV,
D)\Nn (»,■
- e ) = (mi?- 1) ft (»/- e .),
/=i ' /=i
so (71FI¿_j («¿ - en ), or q | «¿ - en for some /', say i = 1. Thus « j = en (mod 2<7).
Also,TV= - 1 (mod 2q), so N/nx = - en (mod 2q). But «j > 2q - 1 > \/TV,which
implies 1 <TV/«j <\/TV <2q — 1. Thus, the only possibility in the interval
[1, 2(7- 1) is that TV/«j= 1, i.e., TVis prime. Q.E.D.
Throughout this section the notation TV4- 1 = F2R2 will be used, where F2 is
the even factored portion of TV4- \,R2 is > 1, and (F2, R2) = 1.
Theorem 16 (Morrison [12]). Consider the set U of Lucus sequences {U^}
with the given discriminant D for which (D/N) = — 1. If for each prime q¡ dividing F2
there exists a Lucas sequence in U such that N\U^+ and (U\ll, , TV) = 1, then
each prime divisor n of N is = en (mod F2).
Proof. It is clear from Lemma 2 that p¡(N)\N 4- 1, which implies p¡(n)\N + 1.
Since n \ t/g+, , Theorem 11(b) implies p,.(«) \(N 4- \)lq{. Thus, q\l \ Pi(n),
where q. ' IIF2. Also, p¡(n)\n - en, so q¡ ' |n - e„ for all i, that is, F2 \n - en. Q.E.D.
For convenience of reference put:
(III) For each prime q¡ dividing F2 there exists a Lucas sequence {iftO} with
discriminantD for which (D/N) = - 1, TV|UJp+i, and (U$+ x)/q , N) = 1.
Corollary 8. Assume (III). If F2 > V?V 4- 1, then TVis prime.
Proof. n + l>n-en>F2> \¿N + 1, which implies TVis prime. Q.E.D.
In what follows the notation Fx = F,/2 andF2 = F2\2 will be used.
Theorem 17. Assume (III) and let m be > \. When m > I, then assume further

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 632 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

that\F2 ± If TV,1 <\<m. If

TV< (mF2 - 1) [2F2 + (m - \r\)F2 4- 1],
where r and s are defined by R2 = 2F2s + r, \r\ < F2, then TVis prime if and only if
s = 0 or r2 4- 8s ^ D.
Proof. The theorem will be proved in the equivalent form: TVis composite if and
only if s =¿0andr2 4- 8s = D.
(i) (=>). Since TV= - 1 (mod F2), it follows from Theorem 16 that TV=
(cF2 — \)(dF2 4- 1), c, d > m. Also, R2 is odd and F2 is even, so the equation F2 =
(TV4- 1)/F2 = cdF2 + c — d implies that c - d is odd, so cd is even. Hence, from

(18) cdF2+c - d = R2=2F2s+r

it follows that

(19) c-d =r (mod 2F2).

On the other hand, (c - m)(d + m)> 0 implies that cd> (d - c)m + m2, so that

(mF2 - l)[2Ff + (m - r)F2 + 1] > (mF2 - 1)[2F2 4- (m - \r\)F2 4- 1]

> TV= cdF] + (c - d)F2- 1> [(d - c)m 4- m2]F2

+ (c - d)F2 - 1 = (mF2 - 1) [(d - c 4- m)F2 4- 1].

Thus, 2F2 +(m - r)F2 4- 1 > (d - c + m)F2 4- 1, or

(20) -2F2+r<c -d.

Also, (c + m)(d —m)> 0 implies cd > (c - d)m 4- m2, so that

(mF2 + 1)[2F22 + (m+ r)F2 -\\> (mF2 - 1)[2F2 + (m - \r\)F2 + 1]

>TV = cdF2 + (c - d)F2 - 1 > [(c - d)m + m2]F2 + (c - d)F2 - 1

= (mF2 + 1) [(c - d + m)F2 - 1].

Thus, 2F| +(m + r)F2 - 1 > (c - d + m)F2 - l,oxc -d<r + 2F2.

Combining this result with (19) and (20) gives c - d = r.
Thus, from (18) it follows that 2s = cd J= 0. Finally, r2 + 8s = (c - d)2 +
4cd = (c + d)2.
(ii) (<=). With s ^ 0 and, say, r2 + 8s = t2, then
TV= F2R2 - 1 = F2(2F2s 4- r) - 1

= [(í-r)F24-l][(í4-r)F2-l],
where the factors on the right are > 1, since s =£ 0. Q.E.D.
Remark. The value of r in Theorem 17 is chosen to be the absolutely least re-
mainder because c - d may well be negative.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2 ± 1 633

Theorem 18. Let n be a prime divisor of N. If for some Lucas sequence {Uk}
for which (D/N) = - 1, TV|UN+, and

(21) (UF2,N)=l,

then n = en (mod q), where q is some prime divisor of R2 depending on n.

Proof. By Lemma 2 and Theorem 8(a), p(n)\n - en and p(n)|TV 4- 1 = F2R2.
But (21) implies p(n)\F2, so (p(n), R2) > 1, i.e., there exists a prime q such that
q\p(n) and q\R2. Hence, q\n - en. Q.E.D.
As a further abbreviation put:
(IV) For some Lucas sequence {Uk} for which (D/N) = - 1, N\UN+X and
(U(N+i)/R2,N)=l.
Remark. As in (II), the subscript of U is written to suggest (III) and (IV) can be
computed together, R2 being treated as the final "prime" factor of TV4- 1.
Corollary 9. Assume (III) and (IV), and let n be a prime divisor of N. Then
n = en (mod qF2), where q is some prime divisor of R2 depending on n.
Proof. Since (F2, F2) = 1, the corollary follows from Theorems 16 and 18.
Q.E.D.
Corollary 10. Assume (III) and (IV). If all the prime factors of R2 are > B2
and B2F2 > -y/TV4- 1, then TVis prime.
Proof, n + \>n - en >qF2 >B2F2 >\/N+ 1, which implies TVis prime. Q.E.D.
Theorem 19. Assume (III) and (IV), and also that the prime factors of R2 are
>B2. If
(22) N<(B2F2 - 1)[2F22 4- (B2 - \r\)F2 4- 1],

where r and s are defined by R2 = 2F2s + r, \r\ < F2, then TVis prime if and only if
s = Oorr2 4- 8s ¥=D .
Proof. The theorem will be proved in the equivalent form: TVis composite if and
only ifs^Oandr2 + 8s = D.
(i) (=*). Let n be a prime factor of TV,and write TV= nw, w > 1. Then from
Corollary 9, n = en (mod qF2), and since TV= - 1 (mod qF2), w = - en (mod qF2).
Then TV= (cF2 + e„)(dF2 - en), where c, d>B2. Also, F2 is odd and F2 is even, so

F2 = (TV4- 1)/F2 = cdF2 4- en(d - c),

implies d — c is odd, so cd is even. Hence, from

(23) cdF2 + en(d - c) = R2 = 2F2s + r

it follows that
en(d-c) = r (mod 2F2).

On the other hand,

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 634 JOHN brillhart, d. h. lehmer and j. l. selfridge

(c - B2)(d 4- B2) > 0 implies cd > (d - c)B2 + B\

and
(c 4- B2)(d - B2) > 0 implies cd > (c - d)B2 4- B2.

These together imply cd > ± en(d - c)B2 + B\. Now using (22),

(F2F2-l)[2F224-(52-r)F24-l]

> (B2F2 - 1)[2F2 4- (B2 - \r\)F2 4- 1]

> TV= cdF2 + en(d - c)F2 - 1

> [- en(d - c)B2 + B\ ] Fl 4- e„(d - c)F2 - 1

= (B2F2 - 1) {[- en(d - c) + B2] F2 4- 1}.

Therefore,
2F2+B2-r>-en(d - c)+B2, or - 2F2 + r < e„(d - c)
Also,
(F2F2 4- 1) [2F2 + (B2 + r)F2 - 1]

> (B2F2 - 1)[2F22 + (B2 - \r\)F2 4- 1]

> TV= cdF\ + en(d - c)F2 - 1

> [en(d - c)B2 4- B\ ] F\ 4- en(d - c)F2 - 1

= (B2F2 + l){[e„(d-c) + B2]F2-l}.

Thus,
2F2+B2+r>en(d - c)+B2, or 2F2 + r > en(d - c).

Hence, r = en(d - c) and from (23), 2s = cd ^ 0. Also,

r2 4- 8s = (d - c)2 4- 4cd = (c + d)2.

(ii) (<=). Same as Theorem 17(ii). Q.E.D.

7. Combined Theorems. As was mentioned in the introduction, a considerable

advantage is gained by combining the information obtained from factoring both N — 1
and TV4- 1. This advantage lies as usual in reducing the total amount of factoring time
by a trade-off with less time-consuming, nontentative tests (such as a GCD) (see [8]).
Of the two theorems given here, Theorem 20 and its corollary have proven to be
quite useful when other primality tests could not be applied. Theorem 21 treats the
case in which TV± 1 can be factored algebraically into possibly rather large pieces, each
of which has been factored to a certain extent (see [6, p. 329]).
Theorem 20. Assume (l)-(W),and suppose the prime factors of Rx and R2
are respectively > Bx and B2. Define r and s by Rx = F2s 4- r, 0 < r < F2, and let

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 635

G = max(ßj Fx + \,B2F2~\, mFxF2 + rFx + I), m>\.

Further, in the case that G = wFjF2 4- rFx 4- 1, assume (XFXF2 + rFx + \)^TV,
5q < X < m, where dr0 is the Kronecker delta. (Note: When r = 0 and m = 1, the X
interval is empty.)
//TV < G(BXB2FXF24- 1), then TVis prime.
Proof (by contradiction). Assume TVis composite, say TV= nw, n prime and
w > 1. Then Corollary 2 gives
(24) "-1 (mod pFj),

where p|Fj, and w =«w =TV = FjF, 4- 1 = 1 (mod pFj). Thus,

(25) w>pFx 4-1 >BXFX + 1.

Similarly, Corollary 9 gives
(26) « = e„ (mod (7F2),

where q\R2, and w = wne„ = TVe„= (F2R2 - l)en = - en (mod qF2). Also,

(27) nw = N = FXRX + I = Fx(sF2 + r) + I = rFx + 1 (mod FXF2),

where rFx 4- 1 < FjF2 4- 1, or more sharply, /Fj 4- 1 < FjF2 - 1, i.e., rFx 4- 1 is the
least positive remainder (mod FXF2).
Case 1.- en = 1. Combining (24) and (26) gives

(28) n s 1 (mod pqFxF2),

since (Fj, F2) = 2. Hence,

n>pqFxF2 + 1 >BXB2FXF2 + 1.
Also, « s l (mod FjF2) from (28). Combining this with (27) gives w = nw = rFx +
1 (mod FjF2), which implies w > mFxF2 4- rFt 4- 1. On the other hand, w = - \
(mod qF2) implies

w>qF2 - 1 >£2F2 - 1.

These results with (25) give w > G. Thus finally, TV= wn > G(BXB2FXF24- 1), which
is a contradiction. Hence, TVis prime.
Case 2. en = — 1. This case is the same as Case 1 with the roles of n and w
reversed and (25) changed to read: n >BXFX 4- 1. Q.E.D.
Remarks. 1. In practice TV- 1 and TV4- 1 can be factored simultaneously; for
if a trial divisor d for TV4- 1 leaves a remainder t + 0, then d will divide TV- 1 if and
only if t = 2.
2. Usually Bx = B2 when the factoring of TV- 1 and TV4- 1 is done by the
method of Remark 1. These factoring bounds may be different, however, if the form
of TVpermits algebraic factorization, and the algebraic factors are investigated separately.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 636 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

3. If the main inequality of the hypothesis is not satisfied at some point in the
factorization of TV± 1, there are three ways to increase the size of the product on the
right of the inequality: increase Bx and B2; find more factors of TV± 1 (thereby in-
creasing Fj or F2); increase the size of m. What strategy is adopted will, of course, de-
pend on the amount of increase needed to satisfy the inequality. An excellent example
of the use of this theorem will be found in the next section where the factorizations of
three Mersenne numbersMx61, Mxgi, and Af241 are shown to be complete. From
these examples, it becomes clear that none of the other hypotheses of Theorem 20 need
to be verified until the inequality on TVhas been satisfied, i.e., the auxiliary testing,
which is needed to complete the primality test, is done only after enough factoring data
have been obtained. (This, of course, is true for the other theorems in this paper.)
Thus, conditions (I)-(IV) are usually referred to as "final tests."
4. The special case when r = 0 occurs when F2 \RX, which implies F2 is odd.
Also, F21TV- 1, and since F2 |TV4- 1, then F2 12. Thus, F2 = 1. This case will occur
if and only if TV= 4k + 1 and TV4- 1 has no "small" odd prime factors.
Corollary 11. Assume (I)-(IV) and that the prime factors of both Rx and R2
are>B =BX = B2.
(a) // B > (N/F2F2)1 /3, then TVis prime.
(b) IfB> (N/FxF2y/3, then TVis prime.
Proof, (a) N<B3F2F2 <BFX(B2FXF2 4- 1) < G(B2FXF2 4- 1). (Note here
that only the first argument in the definition of G is used. Since the third argument in
this definition is not used at all in this theorem, no divisibility testing is needed in the
hypothesis of the corollary.)
(b) First observe in the proof of Theorem 20 that p and q are both > B, and
since p + q, pq> B(B 4- 2). Thus, the inequality following (28) can be written n >
B(B 4- 2)FjF2 4- 1. Consequently, when B = Bx = B2, the inequality in the theorem
can be strengthened to read TV< G[B(B + 2)FXF2 4- 1]. Then
TV< B3FXF¡ < (BF2 - 1) [B(B + 2)F,F2 4- 1]

< G[B(B + 2)FjF2 4- 1]. Q.E.D.

Theorem 21. Let TV- 1 = Wi=xF*'' and TV4- 1 = Usi=xsf', where R¡ and S,
are not necessarily prime, and (R¡, F) = (S¡, S.) = 1, / ¥=j. Suppose the prime fac-
tors of R¡ and S¡ are respectively greater than B¡ and C¡. Let B = n¿_ j B. ' and C =
nj_ x C, '. Assume (II) and (IV) are satisfied respectively for each R¡ and S¡ (where
not necessarily the same base or Lucas sequence is used). Let G = max(F 4- 1, C - 1).
//TV < G(BC/2 4- 1), then TVis prime.
Proof. If TVis not prime, then TV= nw, where n is prime and w > 1. Let a(. be
the base used for R¡ in (II) and suppose the order of a,- (mod n) is e¡. Then e¡ |TV- 1,
but e¡ \(N — l)/Rj• Hence, there is a prime divisor p(. of F(. which divides e¡ to R¡'s
full power in TV- 1; i.e., p¡ '\e¡. But e'An - 1. Thus, since (R¡, R¡) =1, i =£j,
ïlri=iPi'\n - 1. Also, w = nw = TV= 1 (mod ITJLj p. '). On the other hand, if

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 637

{£/£')} is the sequence used for S¡ in (IV) and p¡(n) is the rank of« in {U^}, then by
Lemma 2, p¿(«)|TV 4- 1, but pi(n)-\'(N 4- 1)/S¡. Thus there is a prime divisor q. of S¡
which divides p((«) to S¡% full power in TV4- 1; i.e., q¡ '\p¡(n). But p.(n)\n — en , so
since(S,, S;) = 1, ITJ=1(¡rf'|n - e„. Also,

w = ennw = enN e„ (modft (zf'j.

Case 1. en = 1. In this case

« = 1 (modft (/f'j,
so since (TV- 1,TV4- 1) = 2,

2n = 2 (modft pV ft «?') and w - l (modft ?"'

(Note: p¡ and <jrfmay be odd for all i.) Hence,

«>^(ftrf')fi rf'+1>^(n ^') n ^+1 =f +1

and
w> n pf' +1 > n a*' +1 = 5 +1.
i=i /=i
Also, w = - 1 (mod n^=, q¡ '), so

w>ft iff-l>n cf'-l=C-l.

/=i (=i
Thus, TV= nw> (BC/2 4- l)max(F 4- 1, C - 1) = G(FC/2 4- 1), a contradiction.
Case 2. e_ = — 1. This case is the same as Case 1 with the roles of n and w
reversed. Q.E.D.
Remark. An example for which Theorem 21 might be of use is:
Let TVbe a pseudoprime of the form (a '28 4- l)/257. Then

TV- 1 =(a128 -256)/257

= (a16 - 2)(a16 4- 2)(a16 - 2a8 4- 2)(a16 4- 2a8 4- 2)(a64 4- 16)/257;

8. Numerical Results. The 131 complete factorizations given in Table 1 are the
results obtained by the authors over the last seven years on numbers of the form 2m
± 1, 22r ± 2r + 1, and 22'-1 ± 2r 4- 1 (see [4, p. 87]). (Note that factorizations of
both the primitive and algebraic parts of 2447 - 1 and 2471 — 1 appear in Table 1
and Section 9.)
In Table 1, all factors listed are prime. Those preceding a colon are algebraic;
those following a colon are primitive. An asterisk indicates the factor was first dis-
covered by R. M. Merson.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

TABLE 1. Complete Factorizations

1. 294 + 247 + 1 7 : 4375578271-646675035253258729

2. 2101 _ 251 + , 5 : 9491060093'53425037363873248657
3. 2101 + 251 + 1 : 809-5218735279937-600503817460697
4. ,102 251 + 1 3-19 : 123931-26159806891-27439122228481
5. 2103 + 1 3 : 4l5l4l630193-8l4276708l771726l71
6. 2104 _ 252 + 1 241 : 8Ü159375948762099254554456081
7. ,106 253 + i 3 ; 6043-4475130366518102084427698737
8. ,109 255 + 1 5 : 74323515777853-1746518852140345553
9. ,112 256 + 1 97-673 : 2017-25629623713-1538595959564161
10. 2"" -2-"257 + 1 = 3-19"2 : I9177458387940268II63497666122II
,111

11. - 259 + 1 = 3 : 13099-4453762543897-1898685496465999273

12. 2118 + 259 + 1 = 7 : l8408l-2798994l729-9213624084535989031
13- 2119 + , = 3-43-43691 : 823679683-143162553165560959297
14. 2119 + 260 + 1 = 5-29-26317 : 9521-l8292898984l569l6l56396l01
15. 2120 _ 260 + 1 = 1433.38737 : i68692292721'469775495C6243496l
2121 _ 26l + , = 2113 : 3389-9l96l>40369625840108070l4809213
16.
17. 2121 + 1 = 3-683 : 117371-11054184582797800455736061107
2121 + 26l + 1 = 5.3g7 . 1339272539833668386958920468400193
18.
19. „122 261 + 1 = 3 : 1772303994379887829769795077302561451

20. 2122 + 261 + x = 7 : 367-55633-37201708625305146303973352041

21. 2124 + 1 = 17 : 290657-377O20264l-ll4l629l80401976895873
22. ol25 _ , = 31-601-1801 : 2690898o60Ol-47io883l68879506O01
23- 2125 + 1 = 3-11-251-4051 : 229668251-5519485418336288303251
24. ,126 ,63 + 1 = 3-87211 379-119827-127391413339
•56202143607667

25. 2127 + 1 = 3 : 5671372782OI5641O577229IOI238628O35243

26. 2127 +2+1=5: 18797-72118729-2792688414613
.898835788O5OI
27- ,128 2M + 1 - 769.442499826945303593556473164314770689
28. ,129 265 + 1 13-173-101653-500177 :
• 5951631966296685834686149
29. ,131 266 + 1 5 : 6428ll237-2745098l89'3085446954097694273O9
30. ,131 + 3 : 1049-4744297*-l8233ll2868l20778l78439i8l36ll

*Merson factor

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1
639
TABLE 1 (Continued)

31. ,131 + 2 66 + 1 : 269665073-810791440841-12450751815271172041

32. ,133 - 267 + 1 5-29-229-457 : 1597
•449329386292232535250647435097

33. ,133 + 1 = 3-43-174763 : 4523-10678829O443848295284382097O33

34. ,133 + 267 + 1 113-525313 : 2129-126848469231149

•679253585011429

35. ,136 - 2 + 1 = 241 : 8l6l-40932193*-l467129352609

•737539985835313
36. ,136 + 1 = 257-383521 : 2368179743873-373200722470799764577

37. ,137 - 269 + 1 = : I89O6I.9215257079II8405873906173308863627OI

38. ,137 + 1 =3 1097-15619-32127963626435681

•105498212027592977
,l3t ,69 + 1 = 3-19 : 6113142872404227834840443898241613032969
39.
40. ,138 + 2 69^ + 1 = 73 79903-634569679-2232578641663
•42166482463639
41. ,139 270 + 1 1408349-1573677 4913-492717674609
-1276366OO54721

42. ,139 - 1 : 5625767248687-1238761322O52O8335762278423601

43- ,139 + 1 = 3 : 450b937*-5154263952466l795300074l74250365699
44. ,139 + 270 + 1 : 557-1251163891299967635860272509229764287909
45. ,140 + 1 = 17-61681-15790321 : 84179842077657862OII86788968I

46. ,141 + 271 + ] 13-140737471578113 : 5641

•270097268484167653999069

47. ,142 271 + 1 3 : 5113-17467-102241

•203525545766301306933226271929

48. ,143 272 + 1 = 53-157-2113 : 958673-661521349351105339668937661297

49. ,143 1 23-89-8191 : 724153-158822951431
.5782172113400990737

50. ,143 + 1 3-683-2731 : 2003-6156182033-10425285443

-15500487753323

51. ,145 273 + 1 = 41-536903681 168781

•12004541501954811085302214141

52. ,145 = 31.233.1103-2089 :

2679895157783862814690027494144991
53- ,145 + 1 = 3-H-59-3033169 : 7553921*
•999802854724715300883845411

54. ,145 + 273 + 1 = 52-107367629 : 17401-244716883381

•3902095192430070721

*Merson factor

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 640 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

TABLE 1 (Continued)
55. 2li<7 + 274 + 1 = 13-113-1429-4981857697937 :
17059410504738323992180849
56. 2149 +1 = 3 : 1193-650833-38369587*
•7984559573504259856359124657

57- 2150 - 275 + 1 = 3-19-18837001 : 4714696801

•281941472953710177758647201

58. 2153 + 277 + 1 = 5-109-409-3061-13669-26317 : 613-318194713

•238495197879143209

59- 215 - 277 + 1 = 3-67-5419-20857 : 14323

•70180796165277040349245703851057
60. 215 + 277 + 1 =.72-337-599479 : 463
•4982397651178256151338302204762057
61. 2155 - 278 + 1 = 52,868l-49477 : 37201-87421-52597081*
•24865899693834809641
62. 2155 + 1 = 3-H-715827883 : 11161-5947603221397891
•29126056043168521

63. 2157 - 1 = : 85213320i«60726444l67'l654058017289

■2134387368610417

64. 2158 - 279 + 1 = 3 : 647011-13664473*

•13775694692898492184744709216599873
65. 2159 - 28° + 1 = 13-15358129-586477649 : 207973
•30007459254393181618012897
66. 2159 + 280 + 1 = 5-1801439824104653 : 10177
•7971862004867103303293462593
67. 2l6° +1 = 641-6700417 : 3602561*
•9445568495348456305599183855808I
68. 2161 - 281 + 1 = 113-277-30269 : 3221-169373-209160253
•27037028118448801270021

69. 2161 - 1 = 47'127-17848l : 1289'3l88767-450760^4553

•14808607715315782481

70. 2161 +1 = 3-43-2796203 :

81034674927597923271498OO361564410265219
71. 2161 + 281 + 1 = 5-29-1013-1657 : 1933-298817-115927640417
•179351574736387915177

72. 2 5 + 2 3 + 1 = 13-41-61-2113-312709-415878438361 :
391249826881-13379250952 981
73. 2167 - 1 : 2349023-prime
74. 2 7 +1 = 3 : prime
75. 2168 - 2 + 1 = 433-38737 : l009-2ll69-2627857*-269389009
•1475204679190128571777

76. 2171 - 286 + 1 = 5-109-229-457-275415303169 : 4598533*

•41435606371227835355919073

*Merson factor

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 641

TABLE 1 (Continued)
2174 + 2 "^ + 1 = 73 : prime

2175 _ 288 + x _ m.101.113.8101-7416361 : 701

•243OO659246935171985503227519631OI

2175 +1 = 3-11-43-251-281-4051-86171 : 1051-110251

•347833278451-34010032331525251

2175 + 288 + 1 = 53-29-268501-4739238l :

IO38213793447841940908293355871461401
2177 _ 289 + 1 = 13.5521693.104399276341 : 709-12037
•2995240087117909078735942093

2177 + 289 + 1 = 5-1181-3541-157649-174877 : 31153-5397793*

94789873-20847858316750657
2183 - 292 + 1 = 1 3456749-667055378149 : 508c
4209508589941 -19125556519918081
2185 - 293 + 1 = 41-593-231769777
593-231769777 : 1392776941
4964166554103541-1258710725115650761
2185 +1 = 3-11-1777-25781083 : 1481-28136651*
•7784293653978876O854C61833087328I
2189 - 295 + 1 = 5-29-109-14449-246241-40388473189 : 757
•456376431053626339473533320957
29 + 295 + 1 = 13-37-H3-1429-279073-11875O098349 :
304832756195865229284807891468769
2190 _ 295 + 1 = 3 • 331-571-160465489 : IIOI81I
•1565399070589631354726923722004116936
lQl
2 y +1 = 3 : prime
2195 _ 298 + 1 = 5*521-1321-l6l3-3121-2l84l-5l48l'34110701 :
2341-723447661-892527899379324l
2195 + 298 + 1 = i32-4l-53-6l-l57-313-l249-i08l4098955868l :
468781-7 20453772 427518446437641
2196 +1 = 17-15790321 : 7057-273617*1007441
•375327457-1405628248417-364565561997841
2197 - 1 : 7487-prime
1QQ
2 " +1 = 3 : prime

2200 _ 2100 +1= 241-4562284561 : prime

201 _ 2101 + 1 = I3.i5i52453'9739278030221 : 3217-192961
-214473433-71848008781-175132692529
201 + 2101 + 1 = 5.269-42875177-2559066073 : 10453
-132661-15704900959651293774270521395753

22°3 _ 1 = 127-233-1103-2089 : 136417-121793911

•II348O55580883272OIIO9O856O5317536III3
2205 +i = 3.II.83.8831418697 : prime

*Merson factor

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 642 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

TABLE 1 (Continued)

100. 2205 + 2103 + ! „ 4i2.181549-12112549 821-269896441

82777720757144341.758399801407611361
101. ,207 ,104 + 1 = 13-37-277'30269*54l5624023749 : 829
•853669-26785337l49-4968l708ll09150685921
102. 2207 + 2104 + 1 = 5-109-1013-1657-70334392823809 : 3313
•18217-318781-6542857-25395382141805460457

103. ,213 ,107 + 1 = 5-569-148587949-5585522857 : 266677

•1396429*'18369973*-40524027877
•20111008087273

104. ,213 + 2 107 + 1 = 13-4999465853-472287102421 : 853-189997

•2646185328486854129693169911139349

105. ,215 + 2 + 1 = 52-175921776558l : 370661-1952201*

•4538991421-260125854015641
•1401345270171101

106. ,217 2109 + 1 113-5581-384773 : prime

107- 2220 + 1 17-353-6l68l-29315424l7 : 109121-148721
•34O46760O1-110354657O8081
•2546717317681681
222 111
108. 2 + 2 + 1 73 : 1999-10657'l6983l-123876l*-36085879*
•199381087-698962539799*4096460559560875111
109- 2225 + 2113 + 1 = 53.i09'i8l-1321-54O01'639Ol«268501-l3334701
6957 01-30711639849O301-62 92198401
110. ,231 ,116 + 1 = 13-H3-1429-2113-8317-312709-76O96559910757
3931002956111648245378728475226109181
111. 2231 + 2116 + 1 = 5-29-397-14449-4327489-869467061-3019242689
365212445341097287826412838353955921
112. ,233 : 1399-135607-622577-prime
113. ,237 119
+ 1 = 5-317-381364611866507317969 : 151681-prime
114. ,239 479-1913-5737•176383«134000609-prime
115. ,241 22000409-prime
116. ,255 ,128 + 1 = 13 41-6l-137-953-1326700741
7226904352843746841 : 51001-2949879781
611787251461-15455023589221
117- ,255 + 1 = 3' 11-307-331-2857-6529-43691
268314230360653526II : 12241
4l856298635756l-51366l49455494753931
118. ,255 + 2 12E + 1 = 5' 409-1021-1321-3061-4421-13669-26317
5508OI.2365OO6I : 15571321
42515530888344717190444817256OI
119- ,272 ,136 + 1 = 97-673 prime

*Merson factor

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2 ± 1 643

TABLE 1 (Continued)

120. 2273 + 2137 + 1 5-29-1093 -1613-3121-14449.21841

•8861085190774909 : 1948129
•3194753987813988499397428643895659569

121. 2283 - 2142 + 1 5 : prime

122 . 2285 - 2143 + 1 52-229-457-1321-54721-275415303169

•276696631250953741 : 185821-247381
•3996146881 • 23480412082098913326841
123. 2298 + 2lit9 + 1
prime

124. 2313 + 1 prime

^14 157
125. 2J + 2 J' + 1 prime

126. 2315 + 2158 + 1 13-37-41-61-113-1429-7416361-29247661

•Il8750098349-104l8l5865690l8l :
1711081-430839361
•1736945952990905777323344246I

127. 2318 - 2159 + 1 3-19 : prime

128. ,356 + x 17 : prime

129. ,563 2282 + 1 prime

130. 2613 _ 2307 + 1 prime

131. ,691 ,346 + prime

TABLE 2. Completed Factorizations

2m - 1, m odd: m = I-I67,171,175-183,189,195,197,201,203 ,207 ,

225,231,233,239,241,255,261,315,333,447,471,521,607,1279,2203,
2281,3217,4253,4423,9689,9941,11213,19937.
2m + 1: m = 0-150,153-156,158-162,165-168,170,171,174,
175,177,178,l80,l82,l83,l85,l86,l89-192,194-196,198,199,201,202,
204-207,210,213,214,218,220,222,226,230,231,234,237,238,242,246,
250,252,254,255,258,262,266,270,278,282,285,286,290,294,300,306,
313,318,322,330,342,350,354,356,378,390,402,408,414,426,462,477,
510,566.
2 - 2r + 1, m = 2r - 1: m = 1-147 ,151-155,159,l6l,165,167,171,
17 5,177,183,185,189,195,201,207,213,217,231,237-241,255,283,285,
353,367,457,563,613,691.
2m + 2r + 1, m = 2r-l: m = 1-135,139-147,153,157-165,171,175,
177,189 ,195,201,207,213,215,225,231,255,273,283,315,379.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 644 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

TABLE 3. Mersenne Status List

M = 2b 1, p prime, p < 257

P

Character of

2,3,5,7,13,17,19,31,61,89,107,127 Prime

(All other p under 172), Composite and completely

179,181,197,233,239,241 factored

173,191,193,211,223,229,251 Cofactor is composite

199,227,257 Composite but no factor

known

Table 2 shows which numbers of the above forms have been completely factored.
(Also from Table 2 it is not difficult to discover that 2 500 1,2 600 1,2700-1,
,816 1020 1 have been completely factored.) Table 3 gives the present
1, and 2
status of the "original" Mersenne numbers M = 2P — 1, p a prime < 257.
(The eight new factorizationsofMp are for p = 137, 139, 149, 157, 167, 197, 239,
and 241.)
Several different methods were used to complete the factorization of those num-
bers in Table 1 whose cofactors were composite. Notable examples are:
(i) The cofactors of 2139 - 1, 2205 4- 2103 4- 1, and 2255 4- 1 were factored
by a continued fraction method on the IBM 360/91 at the Campus Computing Net-
work at UCLA (see Morrison and Brillhart [13]). The times required for these factor-
izations were 80, 15, and 12 minutes respectively.
(Ü) 2101 4- 2S1 4- 1,2109 - 255 4- 1,2136 4- l,and 2137 4 1 were factored
by representing their composite cofactors as a difference of squares, using the delay-
line sieve DLS 127 at UC, Berkeley. (2136 4- 1 is particularly notable, having run on
DLS 127 for 2600 hours (!) before it factored.)
(hi) 2102 — 251 4- 1 was factored by expressing its cofactor as a sum of two
squares in two different ways on DLS 127.
(iv) 2131 4-2664- 1,2157- l,and2185 2 4-1 were completed on DLS
127 as in (ii) only after a new prime factor was found using idle time on the CDC 6400
at UC, Berkeley. Most surprising among these is the Mersenne number 2157 — 1,
which split unexpectedly into four factors.
Those numbers having a pseudoprime cofactor for some base a > 2 (see [4, p. 91])
were proved to be prime by some primality test (see Sections 2, 3, or 5). Of special
interest are the Mersenne numbers Mxbn, Mxgi, M239, and M24x, which were tested
using Corollary 11.
To illustrate the use of this corollary, the details for MX61 and M24X are given
here.
(a) Let

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 645

TV= M167/2349023 = 79638304766856507377778616296087448490695649,

a number of 44 digits. TVis a pseudoprime base 13. Also,

TV- 1 =25 • 11 • 37 • 167 • Rx,

where Rx is composite with no factor < 2 • 106. Further,TV 4- 1 = 2 • 33 • 52 ■
1381 • 3167 • R2, where R2 is composite with no factor < 2 • 106. Thus,
Fx = 25 • 11 • 37 • 167 = 2175008 > 2 • 106,
sof\ > 106,and
F2 = 2 • 33 • 52 • 1381 • 3167 = 5904396450 > 5 • 109.
Hence, with B = 2 • 106, the inequality in Corollary 11 (b) is satisfied, since B3FXF2
> (2 • 106)3106(5 • 109)2 > 1044 >TV.
The final tests (I)-(IV) required only a few seconds to show TVwas prime. The
single Lucas sequence P = 1,0= 13 was used in (III) and (IV).
(b) Let
TV=M241/22000409
= 160619474372352289412737508720216839225805656328990879953332340439,
a number of 66 digits. TVis a pseudoprime base 13. Also, TV-1=2* 241 • 21221 •
Rx and TV4- 1 = 23 • 32 • 5 • 23 • 643 • 96763 • 4975177 • 17944799 • R2.
ThenFj = 10228522 and F2 = 45993638617007146424985960. Hence, with B =
21221, TVis prime by Corollary 11(b). One Lucas sequence with P = 1,0 = 5 was
used in the final tests in (III) and (IV).
It is worth mentioning that the factorization of 21S7 - 1, along with the factor-
izations of 2109 ± 1 in [4], finish the 3 factorizations that were left incomplete in
Robinson [19] ; in fact, all numbers attempted there (except F8,Fg, . . .) have now
been completely factored.
Several final comments are in order. The cofactors of F9 and F10, the ninth and
tenth Fermât numbers, have been tested for pseudoprimality, and are both composite.
The tests were run twice with complete agreement in the remainders.
In [4, p. 87], it was stated that "in general nothing but frustration can be
expected to come from an attack on a number of 25 or more digits, even with the
speeds available in modern computers." In view of the increase in speed of computers
and the developments in factorization methodology (see [13]), a number of 40 digits
can now be factored in about 50 minutes on, say, the IBM 360/91. Thus, the above
quote should now be changed to read "50 or more digits."

9. Two Other Factorizations. The following "most wanted" Mersenne factoriza-

tions are due to R. Schroeppel at MIT (see [1]), who found them using essentially the
continued fraction method discussed in [13].
2137 - 1 = 32032215596496435569 • 5439042183600204290159,
2149 - 1 =86656268566282183151 • 8235109336690846723986161.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 646 JOHN BRILLHART, D. H. LEHMER AND J. L. SELFRIDGE

10. Acknowledgements. The authors would like to express their appreciation to

Emma Lehmer, who has greatly assisted in obtaining the numerical results of this paper,
and also to Michael Morrison, whose contributions to this paper have materially influenced
the direction of its development and the character of its results. They would also like
to thank Daniel Shanks for his valuable suggestions, Peter Weinberger and Alex Hurwitz
for their programming and computing assistance, and Patrick Morton for his assistance
in completing the manuscript. Finally, they would like to express their gratitude to the
directors of the following computer centers, who made available the time to obtain the
numerical results of this paper: Campus Computing Facility, UCLA (IBM 7094, IBM
360/91), Bell Telephone Laboratories, Holmdel, N. J. (IBM 7094), The Computer Cen-
ter, UC, Berkeley(IBM7094, CDC6400).

Department of Mathematics
University of Arizona
Tucson, Arizona 8S721

Department of Mathematics
University of California
Berkeley, California 94720

Department of Mathematics
Northern Illinois University
DeKalb, Illinois 60115

1. M. BEELER, R. W. GOSPER & R. SCHROEPPEL, Artificial Intelligence Memo 239,

MIT, Artificial Intelligence Laboratory, 1972, p. 13.
2. J. BRILLHART, "Concerning the numbers 22p + 1, p prime," Math. Comp., v. 16, 1962,
pp. 424-430. MR 26 #6100.
3. J. BRILLHART, "Some miscellaneous factorizations," Math. Comp., v. 17, 1963, pp.
447-450.
4. J. BRILLHART & J. L. SELFRIDGE, "Some factorizations of 2n ± 1 and related results,"
Math Comp., v. 21, 1967, pp. 87-96; Corrigendum, ibid., p. 751. MR 37 #131.
5. M. KRAITCHIK, Theorie des Nombres, vol. 2, Gauthier-Villars, Paris, 1926, p. 135.
6. D. H. LEHMER, "Tests for primality by the converse of Fermat's theorem," Bull. Amer.
Math. Soc, v. 33, 1927, pp. 327-340.
7. D. H. LEHMER, "An extended theory of Lucas functions," Ann. of Math., v. 31, 1930,
pp. 419-448.
8. D. H. LEHMER, "Computer technology applied to the theory of numbers," Studies in
Number Theory, MAA Studies in Math., vol. 6, Prentice-Hall, Englewood Cliffs, N. J., 1969, pp.
117-151. MR 40 #84.
9. E. LUCAS, "Considerations nouvelles sur la théorie des nombres premiers et sur la division
géome'trique de la circonférence en parties égales," Assoc. Franc. Avancement Sei. C. R., v. 6,
1877, p. 162.
10. E. LUCAS, "Theorie des fonctions numériques simplement périodiques," Amer. J. Math.,
v. 1, 1878, pp. 184-240, 289-321.
11. E. LUCAS, "Théorie des nombres, Tome I: Le Calcul des Nombres Entiers, le Calcul
des Nombres Rationnels, la Divisibilité Arithmétique, Librairie Scientifique et Technique, Albert
Blanchard, Paris, 1961, pp. 423, 441. MR 23 #A828.
12. M. A. MORRISON, "A note on primality testing using Lucas sequences," Moth. Comp.,
v. 29, 1975, pp. 181-182.
13. M. A. MORRISON & J. BRILLHART, "A method of factoring and the factorization of
F7," Math. Comp., v. 29, 1975, pp. 183-205.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use

 NEW PRIMALITY CRITERIA AND FACTORIZATIONS OF 2m ± 1 647

14. H. C. POCKLINGTON, "The determination of the prime or composite nature of large

numbers by Fermat's theorem, " Proc. Cambridge Philos. Soc, v. 18, 1914-16, pp. 29—30.
15. E. PROTH, "Théorèmes sur les nombres premiers," C.R. Acad. Sei. Paris, v. 87, 1878,
p. 926.
16. H. RIESEL,£nBo/comPri'mra/,Studentlitteratur, Lund, Sweden, 1968, pp. 44-65.
MR 42 #4507.
17. H. RIESEL, "Lucasian criteria for the primality of N = h • 2n - 1," Math. Comp.,\. 23,
1969, pp. 869-875. MR 41 #6773.
18. R. M. ROBINSON, "The converse of Fermat's theorem," Amer. Math. Monthly, v. 64,
1957, pp. 703-710. MR 20 #4520.
19. R. M. ROBINSON, "Some factorizations of numbers of the form 2" ± l,"MTAC,v. 11,
1957, pp. 265-268. MR 20 #832.
20. J. L. SELFRIDGE & R. K. GUY, Primality Testing with Applications to Small Machines,
Research Paper #121, Dept. of Math., Univ. of Calgary, Canada, 1971.
21. H. C. WILLIAMS & C. R. ZARNKE, "A report on prime numbers of the forms M =
(6a + l)22m_1 - 1 and Ai' = (6a - l)22m - I," Math. Comp., v. 22, 1968, pp. 420-422.
MR 37 #2680.

License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use