Conference PaperPDF Available

# Factorization of a 768-Bit RSA Modulus

Authors:

## Abstract and Figures

This paper reports on the factorization of the 768-bit number RSA-768 by the number field sieve factoring method and discusses some implications for RSA.
Content may be subject to copyright.
Factorization of a 768-bit RSA modulus
version 1.3, January 24, 2010
Thorsten Kleinjung1,
Kazumaro Aoki2, Jens Franke3, Arjen K. Lenstra1, Emmanuel Thomé4,
Joppe W. Bos1, Pierrick Gaudry4, Alexander Kruppa4, Peter L. Montgomery5,6,
Dag Arne Osvik1, Herman te Riele6, Andrey Timofeev6, and Paul Zimmermann4
1EPFL IC LACAL, Station 14, CH-1015 Lausanne, Switzerland
2NTT, 3-9-11 Midori-cho, Musashino-shi, Tokyo, 180-8585 Japan
3University of Bonn, Department of Mathematics, Beringstraße 1, D-53115 Bonn, Germany
4INRIA CNRS LORIA, Équipe CARAMEL - bâtiment A, 615 rue du jardin botanique,
F-54602 Villers-lès-Nancy Cedex, France
5Microsoft Research, One Microsoft Way, Redmond, WA 98052, USA
6CWI, P.O. Box 94079, 1090 GB Amsterdam, The Netherlands
Abstract. This paper reports on the factorization of the 768-bit number RSA-768 by the num-
ber ﬁeld sieve factoring method and discusses some implications for RSA.
Keywords: RSA, number ﬁeld sieve
1 Introduction
On December 12, 2009, we factored the 768-bit, 232-digit number RSA-768 by the number
ﬁeld sieve (NFS, [19]). The number RSA-768 was taken from the now obsolete RSA Challenge
list [37] as a representative 768-bit RSA modulus (cf. [36]). This result is a record for factoring
general integers. Factoring a 1024-bit RSA modulus would be about a thousand times harder,
and a 768-bit RSA modulus is several thousands times harder to factor than a 512-bit one.
Because the ﬁrst factorization of a 512-bit RSA modulus was reported only a decade ago
(cf. [7]) it is not unreasonable to expect that 1024-bit RSA moduli can be factored well within
the next decade by an academic eﬀort such as ours or the one in [7]. Thus, it would be prudent
to phase out usage of 1024-bit RSA within the next three to four years.
The previous record NFS factorization was that of the 663-bit, 200-digit number RSA-200
(cf. [4]), established on May 9, 2005. That 663-bit and the current 768-bit record NFS fac-
torizations should not be confused with record special NFS (SNFS) factorizations, the most
recent of which was the complete factorization of the 1039-bit number 21039 1, obtained in
the spring of 2007 (cf. [2]). Although 21039 1is much bigger than RSA-768, its special form
made it an order of magnitude easier to factor than RSA-768.
The following eﬀort was involved. We spent half a year on 80 processors on polynomial selec-
tion. This was about 3% of the main task, the sieving, which was done on many hundreds of
machines and took almost two years. On a single core 2.2 GHz AMD Opteron processor with
2 GB RAM per core, sieving would have taken about ﬁfteen hundred years. This included a
generous amount of oversieving, to make the most cumbersome step, the matrix step, more
manageable. Preparing the sieving data for the matrix step took a couple of weeks on a few
processors, the ﬁnal step after the matrix step took less than half a day of computing, but
took about four days of intensive labor because a few bugs had to be ﬁxed.
It turned out that we had done about twice the sieving strictly necessary to obtain a usable
matrix, and that the extra data allowed generation of a matrix that was quite a bit easier than
anticipated at the outset of the project. Although we spent more computer time on the sieving
than required, sieving is a rather laid back process that, once running, does not require much
care beyond occasionally restarting a machine. The matrix step, on the other hand, is a more
subtle aﬀair where a slight disturbance easily causes major trouble, in particular if the problem
is by its sheer size stretching the available resources. Thus, our approach to overspend on an
easygoing part of the computation led to a matrix that could be handled relatively smoothly,
thereby saving us considerable headaches. More importantly, and another reason behind the
oversieving, the extra sieving data allow us to conduct various experiments aimed at getting
a better understanding about the relation between sieving and matrix eﬀorts and the eﬀect
on NFS feasibility and overall performance. This is ongoing research, the results of which will
be reported elsewhere. All in all, the extra sieving cycles were well spent.
For 21039 1the matrix step was, for the ﬁrst time, performed on a number of diﬀerent clusters:
as reported in [2], the main part of the calculation was run in parallel at two locations on four
clusters. This is possible due to our usage of the block Wiedemann algorithm for the matrix
step (cf. [9]), where the main calculation consists of two consecutive sequences of matrix times
vector multiplications. As discussed in [2], a greater degree of independent parallelization leads
to a variety of challenges that need to be overcome before larger problems can be tackled
(such as, for instance, 1024-bit moduli). Here we set a ﬁrst step toward a scalable solution by
showing how greater ﬂexibility can be achieved in the number and contributions of the various
independent clusters. As a result a nine times harder (than for 21039 1) matrix step was solved
in less than four months running up to eight independent jobs on clusters located in France,
Japan, and Switzerland. During one of the substeps about a terabyte of memory was needed.
These ﬁgures imply that much larger matrices are already within reach, leaving preciously
little doubt about the feasibility by the year 2020 of a matrix required for a 1024-bit NFS
factorization. As part of the experiments mentioned above we also intend to study if a single
large cluster would be able to handle such matrices using the block Lanczos algorithm (cf. [8]).
Compared to block Wiedemann this has advantages (a shorter, single sequence of iterations
and no tedious and memory-hungry central Berlekamp-Massey step [40]) but disadvantages
as well (it cannot be run on separate clusters and each iteration consists of a multiplication
by both a matrix and its transpose).
The factorization problem and the steps taken to solve it using NFS are described in Section 2.
The factorization is presented in Section 2.5. Implications for moduli larger than RSA-768 are
brieﬂy discussed in the concluding Section 3. Appendix A presents some of the details of the
sieving approach that we used, and Appendix B describes a new twist of the block Wiedemann
algorithm that makes it easier to share large calculations among diﬀerent parties.
There are many aspects to an eﬀort as described in this paper that are not interesting from a
scientiﬁc or research point of view, but that are nevertheless crucial for its success. Although
NFS can be run as a BOINC project (cf. [30]), we chose not to do so. Instead, we gathered
a small but knowledgeable group of contributors who upfront committed themselves to ﬁnish
the project, while dedicating at least a known minimum of computational resources for a
substantial period of time. This allowed us to target a reasonable completion date, a date
2
that we managed to meet easily despite our ample oversieving. Although diﬀerent sieving-
clients do not need to communicate, each client needs to communicate a fair amount of data
to the central storage location. This required a bit more organizational eﬀorts than expected,
occasional recovery from mishaps such as unplugged network cables, switched oﬀ servers, or
faulty raids, and a constantly growing farm of backup drives. We do not further comment
on these more managerial issues in this article, but note that larger eﬀorts of this sort would
beneﬁt from full-time professional supervision.
2 Factoring RSA-768
2.1 Factoring using the Morrison-Brillhart approach
A composite integer ncan be factored by ﬁnding integer solutions x, y of the congruence
of squares x2y2mod n, and by hoping that nis factored by writing it as a product
gcd(xy, n)·gcd(x+y, n). For a random such pair the probability is at least 1
2that a
non-trivial factor of nis found in this way. The Morrison-Brillhart approach [25] solves the
equation x2y2mod nby combining congruences of smooth squares. In this section this is
explained and, very rudimentarily, how NFS applies the same basic idea.
A non-zero integer uis b-smooth if the prime factors of |u|are at most b. Each b-smooth
integer ucorresponds to the (π(b) + 1)-dimensional integer vector v(u)of exponents of the
primes bin its factorization, where π(b)is the number of primes band the “+1” accounts
for inclusion of the exponent of 1. The factor base consists of the primes at most equal to
the smoothness bound.
Let nbe a composite integer, ba smoothness bound, and ta positive integer. Suppose that
π(b) + 1 + tdiﬀerent integers vhave been found such that the least absolute remainders
r(v) = v2mod nare b-smooth. Because the corresponding (π(b) + 1)-dimensional vectors
v(r(v)) are linearly dependent, at least tindependent subsets Sof the v’s can be found using
linear algebra such that, for each of these subsets, PvSv(r(v)) is an all-even vector. Each
subset Sthen leads to a solution x=QvSvand y=pQvSr(v)to x2y2mod nand
thus, overall, tchances of at least 1
2to factor n.
If v’s as above are generated by picking random integers and hoping that the least absolute
remainder modulo nof their square is smooth, the search for the proper v’s depends on the
smoothness probability of integers of the same order of magnitude as n. That is Dixon’s
random squares method (cf. [11]), which has the advantage that its expected runtime can
rigorously be proved. It was preceded by Morrison and Brillhart (cf. [25]) who had already
obtained a much higher smoothness probability, and thus a much faster – but heuristic –
factoring method, by using continued fractions to generate quadratic residues modulo nof
roughly half the size of n, i.e., of order n1/2. Richard Schroeppel with his linear sieve was the
ﬁrst, in about 1976, to combine essentially the same high smoothness probability with a fast
sieving-based method to recognize the smooth numbers (cf. [33, Section 6]). He was also the
ﬁrst to analyze the resulting expected runtime (cf. [17, Section 4.2.6]). Later this culminated
in Carl Pomerance’s quadratic sieve (cf. [33,34]) which, like Schroeppel’s method, relies on
fast sieving based recognition of smooth quadratic residues of rough order n1/2.
3
Factoring methods of this sort that rely on smoothness of residues of order nθ(1) can be shown
to be stuck at expected runtimes of the form
e(c+o(1))(ln n)1/2(ln ln n)1/2,
for positive constants cand asymptotically for n→ ∞. The number ﬁeld sieve [19] was the
ﬁrst, and so far the only, practical factoring method to break through the barrier of the
ln n-exponent of 1
2. This is achieved by looking at more contrived congruences that involve
smoothness of numbers of order no(1), for n→ ∞, that can, as usual, be combined into a
congruence of squares x2y2mod n. As a result, the number ﬁeld sieve factors a composite
integer nin heuristic expected time
e((64/9)1/3+o(1))(ln n)1/3(ln ln n)2/3,
asymptotically for n→ ∞. It is currently the algorithm of choice to factor numbers without
special properties, such as those on the RSA Challenge List [37]. Of this list, we concentrate
on the number RSA-768, a 768-bit RSA modulus with 232-digit decimal representation
123018668453011775513049495838496272077285356959533479219732245215172640050726
365751874520219978646938995647494277406384592519255732630345373154826850791702
6122142913461670429214311602221240479274737794080665351419597459856902143413.
Similar to Schroeppel’s linear sieve, the most important steps of NFS are sieving and the matrix
step. In the former relations are collected, congruences involving smooth values similar to the
smooth r(v)-values above. In the latter linear dependencies are found among the exponent
vectors of the smooth values. Unlike Schroeppel’s method, however, NFS requires two non-
trivial additional steps: a pre-processing polynomial selection step before the sieving can start,
and a post-processing square root step to convert the linear dependencies into congruences of
squares. A rough operational description of these steps as applied to RSA-768 is given in the
remainder of this section. For an explanation why these steps work, the reader may consult
one of the many expositions on NFS in the literature (cf. [19,20,35]).
2.2 Polynomial selection
Let nbe a composite integer to be factored. In NFS relations are given by coprime pairs of
integers (a, b)with b > 0such that two integer values v1(a, b)and v2(a, b)that depend on a, b
are simultaneously smooth, v1(a, b)with respect to some bound b1and v2(a, b)with respect to
some bound b2. The values v1(a, b)and v2(a, b)are deﬁned as follows. Let f1(X), f2(X)Z[X]
be two irreducible integer polynomials of degrees d1and d2, respectively, with a common
root mmodulo n, i.e., f1(m)f2(m)0 mod n. For our simpliﬁed exposition we assume
that f1and f2are monic, despite the fact that the actual f1and f2will not be monic. Then
vk(a, b) = bdkfk(a/b)Z. Suﬃciently many more than π(b1) + π(b2)+2 relations lead to
enough chances to factor n(cf. [19]), as sketched in the next two paragraphs.
Let Q(αk) = Q[X]/(fk(X)) for k= 1,2be two algebraic number ﬁelds. The elements ak
Z[αk]have norm vk(a, b)and belong to the ﬁrst degree prime ideals in Q(αk)of (prime) norms
equal to the prime factors of vk(a, b). These prime ideals in Q(αk)correspond bijectively to the
4
pairs (p, r mod p)where pis prime and fk(r)0 mod p: excluding factors of fk’s discriminant,
a ﬁrst degree prime ideal corresponding to such a pair (p, r mod p)has norm pand is generated
by pand rαk.
Because fk(m)0 mod n, there are two natural ring homomorphisms φk:Z[αk]Z/nZfor
k= 1,2that map Pdk1
i=0 aiαi
k(with aiZ) to Pdk1
i=0 aimimod nand for which φ1(a1)
φ2(a2) mod n. Finding a linear dependency modulo 2 among the vectors consisting of
the exponents of the primes in the b1-smooth v1(a, b),b2-smooth v2(a, b)pairs (and after,
at some stage, slightly extending the vectors to make sure that the square norms lead to
squares of algebraic numbers, cf. [1]), subsets Sof the set of relations are constructed such
that Q(a,b)S(ak)is a square σkin Q(αk), both for k= 1 and for k= 2. With φ1(σ1)
φ2(σ2) mod nit then remains to compute square roots τk=σ1/2
kQ(αk)for k= 1,2to ﬁnd
a solution x=φ1(τ1)and y=φ2(τ2)to x2y2mod n.
It is easy to ﬁnd polynomials f1and f2that lead to smoothness of numbers of order no(1), for
n→ ∞. Let d1Nbe of order (3 ln n
ln ln n)1/3, let d2= 1, let mbe an integer slightly smaller
than n1/d1, and let n=Pd1
i=0 nimiwith 0ni< m be the radix mrepresentation of n. Then
f1(X) = Pd1
i=0 niXiand f2(X) = Xmhave coeﬃcients that are no(1) for n→ ∞, they
have mas common root modulo n, and for the particular choice of d1the values a,bthat suﬃce
to generate enough relations are small enough to keep bd1f1(a/b)and bd2f2(a/b)of order no(1)
as well. Finally, if f1is not irreducible, it can be used to directly factor nor, if that fails, one of
its factors can be used instead of f1. If d1>1and d2= 1 we refer to “k= 1 as the algebraic
side and “k= 2 as the rational side. With d2= 1 the algebraic number ﬁeld Q(α2)is simply
Q, the ﬁrst degree prime ideals in Qare the regular primes and, with f2(X) = Xm, the
element a2of Z[α2]is abm =v2(a, b)Zwith φ2(a2) = abm mod n.
Although with the above polynomials NFS achieves the asymptotic runtime given in the
previous section, the amount of freedom in the choice of m,f1, and f2means that from many
diﬀerent choices the best may be selected. Here we keep interpretation of the term “best”
intuitive and make no attempt to deﬁne it precisely. When comparing diﬀerent possibilities,
however, it is usually obvious which one is better. Most often this is strongly correlated with
the smoothness probability over the relevant range of a, b pairs, and thus with the size of
the coeﬃcients and the number of roots modulo small primes, smoothness properties of the
leading coeﬃcients, and the number of real roots. Sieving experiments may be carried out to
break a tie.
Given this somewhat inadequate explanation of what we are looking for, the way one should
be looking for good candidates is the subject of active research. Current best methods involve
extensive searches, are guided by experience, helped by luck, and proﬁt from patience. Only
one method is known that produces two good polynomials of degrees greater than one (namely,
twice degree two, cf. [5]), but its results are not competitive with the current best d1>1,
d2= 1 methods which are all based on reﬁnements described in [15] of the approach from [24]
and [26] as summarized in [7, Section 3.1]. A search of three months on a cluster of 80 Opteron
cores (i.e., 3
12 ·80 = 20 core years), conducted at BSI in 2005 already and thus not including
the idea from [16], produced three pairs of polynomials of comparable quality. We used the
5
pair
f1(X) = 265482057982680X6
+ 1276509360768321888X5
5006815697800138351796828X4
46477854471727854271772677450X3
+ 6525437261935989397109667371894785X2
18185779352088594356726018862434803054X
277565266791543881995216199713801103343120,
f2(X) = 34661003550492501851445829X1291187456580021223163547791574810881
from which the common root mfollows as the negative of the ratio modulo RSA-768 of the
constant and leading coeﬃcients of f2. The leading coeﬃcients factor as 23·32·5·72·11 ·
17 ·23 ·31 ·112 877 and 13 ·37 ·79 ·97 ·103 ·331 ·601 ·619 ·769 ·907 ·1 063, respectively.
The discriminant of f1equals 212 ·32·52·13 ·17 ·17 722 398 737 ·c273, where c273 denotes a
273-digit composite number that probably does not have a factor of 30 digits or less and that
is expected to be square-free. The discriminant of f2equals one. A renewed search at EPFL in
the spring of 2007 (also not using the idea from [16]), after we had decided to factor RSA-768,
produced a couple of candidates of similar quality, again after spending about 20 core years.
Following the algorithm from [15], polynomials with the following properties were considered:
the leading coeﬃcient of f2allowed 11 (ﬁrst search at BSI) or 10 (second search at EPFL)
prime factors equal to 1 mod 6 with at most one other factor <215.5, and the leading coeﬃcient
of f1was a multiple of 258 060 = 22·3·5·11·17·23. Overall, at least 2·1018 pairs of polynomials
were considered.
Given the polynomials, smoothness bounds have to be selected. Also, we need to specify a
sieving region Sof Z×Z>0where the search for relations will be conducted. We conclude this
section with a discussion of this ﬁnal part of the parameter selection process.
During any practical search for smooth numbers, one always encounters near misses: candidates
that are smooth with the exception of a small number of prime factors that are somewhat larger
than the smoothness bound. Not only can these large primes, as these factors are commonly
referred to, be recognized at no or little extra cost, they also turn out to be useful in the
factorization process. As a result the smoothness bounds can be chosen substantially lower
than they would have to be chosen otherwise and relation collection goes faster while requiring
less memory. On the negative side, it is somewhat harder to decide whether enough relations
have been found – as the simple criterion that more than π(b1) + π(b2) + 2 are needed is no
longer adequate – but that is an issue of relatively minor concern. In principle the decision
requires duplicate removal, repeated singleton removal, and counting, all brieﬂy touched upon
at the end of Section 2.3. In practice it is a simple matter of experience.
Let bdenote the upper bound on the large primes that we want to keep, which we assume
to be the same for the algebraic and rational sides. Thus, bmax(b1, b2). We say that a
non-zero integer xis (bk, b)-smooth if with the exception of, say, four prime factors between
bkand b, all remaining prime factors of |x|are at most bk. Consequently, we change the
deﬁnition of a relation into a coprime pair of integers a, b with b > 0such that bd1f1(a/b)is
(b1, b)-smooth and bd2f2(a/b)is (b2, b)-smooth.
6
Selection of the smoothness bounds b1,b2, and b, of the search area Sfor a, b, and how
many primes between b1or b2and bwill be permitted and looked for is also mostly guided
by experience, and supported by sieving experiments to check if the yield is adequate. For
RSA-768 we used smoothness bounds b1= 11 ·108,b2= 2 ·108and b= 240 on cores with
at least 2 GB RAM (which turned out to be the majority). On cores with less memory (but
at least a GB RAM, since otherwise it would not be worth our trouble), b1= 4.5·108and
b2= 108were used instead (with the same b). For either choice it was expected that it would
suﬃce to use as sieving region the bounded subset Sof Z×Z>0containing about 11 ·1018
coprime pairs a, b with |a| ≤ 3·109·κ1/26.3·1011 and 0< b 3·1091/21.4·107.
Here κ= 44 000 approximates the skewness of the polynomial f1, and serves to approximately
minimize the largest norm v1(a, b)that can be encountered in the sieving region. Although
prime ideal norms up to b= 240 were accepted, the parameters were optimized for norms
up to 237. Most jobs attempted to split after the sieving algebraic and rational cofactors up
to 2140 and 2110, respectively, only considering the most promising candidates (cf. [14]). As
far as we know, this was the ﬁrst NFS factorization where more than three large primes were
allowed on the algebraic side.
2.3 Sieving
With the parameters as chosen above, we need to ﬁnd relations: coprime pairs of inte-
gers (a, b)Ssuch that bd1f1(a/b)is (b1, b)-smooth and bd2f2(a/b)is (b2, b)-smooth. In
this section we describe this process.
A prime pdividing fk(r)is equivalent to (rmod p)being a root of fkmodulo p(as already
noted above with the ﬁrst degree prime ideals, and disregarding prime factors of fk’s discrimi-
nant). Because d2= 1, the polynomial f2has one root modulo pfor each prime pnot dividing
the leading coeﬃcient of f2, and each such prime pdivides f2(j)once every pconsecutive val-
ues of j. For f1the number of roots modulo pmay vary from 0(if f1has no roots modulo p)
to d1(if f1has d1linear factors modulo p). Thus, some primes do not divide f1(j)for all j,
whereas other primes pmay divide f1(j)a total of d1times for every pconsecutive j-values.
All (prime,root) pairs with prime at most b1for f1and at most b2for f2can be precomputed.
Given a (prime,root) pair (p, r)for fk, the prime pdivides bdkfk(a/b)if a/b rmod p, i.e.,
for a ﬁxed bwhenever a=rb mod p. This leads to the line sieving approach where per b-
value for each (p, r)pair for f1(or f2) one marks with “ p all relevant a-values (of the form
rb +ip for iZ). After this sieving step, the good locations are those that have been marked
for many diﬀerent p’s. These locations are remembered, after which the process is repeated
for f2(or f1), and the intersection of the new good locations with the old ones is further
studied, since those are the locations where relations may be found. With 1.4·107b-values
to be processed (the lines) and about 2·κtimes that many a-values per b, the lines can be
partitioned among almost any number of diﬀerent processors a factoring eﬀort of this limited
scale can get access to. This is how the earliest implementations of distributed NFS sieving
worked.
A more eﬃcient but also more complicated approach has gained popularity since the mid 1990s:
the lattice sieve as described in [32]. For a (prime,root) pair q= (q, s)deﬁne Lqas the lattice
of integer linear combinations of the 2-dimensional integer (row-)vectors (q, 0),(s, 1) Z2,
and let Sqbe the intersection of Sand Lq. Fix a (prime,root) pair q= (q, s)for, say, f1. The
7
special prime q(as this lattice deﬁning large prime was referred to in [32]) is typically chosen
of the same order of magnitude as b1, possibly bigger – but smaller than b. It follows that
qdivides bd1f1(a/b)for each (a, b)Sq. Lattice sieving consists of marking, for each other
(prime,root) pair pfor f1for which the prime is bounded by b1, the points in the intersection
LpSq. The good locations are remembered, after which the process is repeated for all the
(prime,root) pairs for f2with prime bounded by b2, and the intersection of the new good
locations with the old ones is further studied. In this way relations in Sqare found. Thus, for
each of these relations it is the case that qdivides v1(a, b), and the lattice sieve is repeated
for other (prime,root) pairs quntil enough relations have been found.
A relation thus found and involving some prime q06=qdividing v1(a, b)may also be found
when lattice sieving with q0= (q0, s0). Thus, unlike line sieving, when lattice sieving duplicates
will be found: these have to be removed (while, obviously, keeping one copy).
In practice one ﬁxes bounds Iand Jindependent of qand deﬁnes Sq={iu +jv :i, j
Z,I/2i < I/2,0< j < J }, where u, v form a basis for Lqthat minimizes the norms
v1(a, b)for (a, b)Sq. Such a basis is found by partially reducing the initial basis (q, 0),(s, 1)
for Lqsuch that, roughly speaking, the ﬁrst coordinate is about κtimes bigger than the second,
thereby minimizing the norms v1(a, b)for (a, b)Sq, cf. skewness of the sieving region S.
Actual sieving is carried out over the set {(i, j)Z×Z>0:I/2i < I/2,0< j < J},
interpreted as Sqin the above manner.
For RSA-768 we used I= 216 and J= 215 which implies that we did lattice sieving over
an area of size roughly 231 2·109(we did not use any line sieving). With b1= 11 ·108
and b2= 2 ·108, the majority of the primes that are being sieved with can be expected to
hit Sqonly a few times. Thus, for any pthat one will be sieving with, only a few of the
j-values (the lines) will be hit, unlike the situation in plain line sieving where the lines are
so long that each line will be hit several times by each prime. It implies that, when lattice
sieving, one cannot aﬀord to look at all lines 0< j < J per p, and that a sieving method
more sophisticated than line sieving must be used. This sieving by vectors, as it was referred
to in [32], was ﬁrst implemented in [13] and used for many factorizations in the 1990s, such
as those reported in [10] and [7]. We used a particularly eﬃcient implementation of sieving by
vectors, the details of which have never appeared in print (cf. [12]). They are summarized in
Appendix A.
Lattice sieving with sieving by vectors as above was done for most of the about 480 million
(prime,root) pairs (q, s)for special qs between 450 million and 11.1 billion (and some spe-
cial qs below 450 million, with a smaller b1-value) by eight contributing parties during the
period August 2007 until April 2009. Table 1 lists the percentages contributed by the eight
participants. Scaled to a 2.2 GHz Opteron core with 2 GB RAM, a single (prime,root) pair
(q, s)was processed in less than a hundred seconds on average and produced about 134 rela-
tions, for an average of about four relations every three seconds. This average rate varies by
a factor of about two between both ends of the special qrange that we considered.
In total 64 334 489 730 relations were collected, each requiring about 150 bytes. Compressed
they occupied about 5 terabytes of disk space, backed up at various locations. Of these rela-
tions, 27.4% were duplicates. Duplicate removal can simply be done using bucket sorting (on
b-values, for instance) followed by application of UNIX commands sort and uniq, but may
be complicated by odd data formats, and may be done in any way one deems convenient. We
8
name of contributor relations matrix
contribution stage 1 stage 2 stage 3 total
Bonn (University and BSI) 8.14%
CWI 3.44%
EPFL 29.78% 34.3% 100% 78.2% 51.9%
INRIA LORIA (ALADDIN-G5K) 37.97% 46.8% 17.3% 35.0%
NTT 15.01% 18.9% 4.5% 13.1%
Scott Contini (Australia) 0.43%
Paul Leyland (UK) 0.69%
Heinz Stockinger (Enabling Grids for E-sciencE) 4.54%
Table 1: Percentages contributed (with matrix stages 1, 2, and 3 contributing 3
5th, 0, and 2
5th to total).
used hashing. Most uniqueing was done during the sieving. Overall it took less than 10 days
on a 2.66 GHz Core2 processor with ten 1TB hard disks. After inclusion of 57 223 462 free
relations (cf. [19]) the uniqueing resulted in 47 762 243 404 relations involving 35 288 334 017
prime ideals.
Given the set of unique relations, those that involve a prime ideal that does not occur in
any other relation, the singletons, cannot be part of a dependency and can thus be removed.
Recognizing singletons can be done using a simple hash table with a count (counting “0”,
“1”, “many”). Doing this once reduced the set of relations to 28 984 986 047 elements, with
14 498 007 183 prime ideals. However, removal of singletons usually leads to new singletons,
and singleton removal must be repeated until no more relations are removed. Note that the ﬁrst
(usually high volume) runs can be made to accept diﬀerent prime ideals with the same hash to
lower memory requirements for the hash table. Once singletons have been removed, we have
to make sure that the number of remaining relations is larger than the total number of prime
ideals occurring in them, so that there will indeed be dependencies among the exponent vectors.
After a few more singleton removals 24 615 168 385 relations involving at most 9 976 671 468
prime ideals were left.
At this point further singleton removal iterations were combined with clique removal, i.e.,
search of combinations with matching ﬁrst degree prime ideals of norms larger than bk. Re-
moval of almost all singletons together with clique removal (cf. [6]) led to 2 458 248 361 relations
involving a total of 1 697 618 199 prime ideals, and still containing an almost insigniﬁcant num-
ber 604 423 of free relations. This not only clearly indicated that we had done enough sieving,
but also gave us lots of ﬂexibility while creating the matrix. In the next section it is described
how dependencies were found. Overall singleton removal and clique ﬁnding took less than 10
days on the same platform as used for the uniqueing.
2.4 The matrix step
Finding dependencies among the rows of a sparse matrix using block Wiedemann or block
Lanzcos, currently the two favorite methods, takes time proportional to the product of the
dimension and the weight (i.e., number of non-zero entries) of the matrix. Merging is a generic
term for the set of strategies developed to build a matrix for which close to optimal performance
can be expected for the dependency search of one’s choice. It is described in [6]. We ran about
10 merging jobs, aiming for various optimizations (low dimension, low weight, best-of-both,
etc.), which each took a couple of days on a single core per node of a 37 node 2.66 GHz
9
Core2 cluster with 16 GB RAM per node, and a not particularly fast interconnection network.
The matrix that we ended up using was generated by a 5-day process running on two to
three cores on the same 37-node cluster, where the long duration was probably due to the
larger communication overhead. It produced a 192 796 550 ×192 795 550-matrix of total weight
27 797 115 920 (thus, on average 144 non-zeros per row). Storage of the matrix required about
105 gigabytes. When we started the project, we expected dimension about a quarter billion
4times harder than what we managed
to ﬁnd (thanks to our large amount of oversieving).
We obtained various other matrices that would all lead to worse performance for the next step,
determining dependencies. For instance, for the full collection of relations we found a matrix
with about 181 million rows and average weight about 171 per row, which is only slightly
worse. Restriction to the set of 1 296 488 663 relations (involving fewer than 1 321 104 619
prime ideals) satisfying the much smaller b-value of 234 still led, after singleton removal, to
more relations (905 325 141) than prime ideals (894 248 046), as required, and ultimately to a
252 735 215×252 734 215-matrix of weight 37 268 770 998. For b= 235 there were 3 169 194 001
relations on at most 2 620 011 949 prime ideals, of which 2 487 635 021 relations on 1 892 034 766
prime ideals remained after singleton removal, and which led to a 208 065 007 ×208 064 007-
matrix of weight 31 440 035 830. Neither matrix was competitive.
We used the block Wiedemann algorithm [9] with block width 8·64 to ﬁnd dependencies
modulo 2 among the approximately 193 million rows resulting from the ﬁltering. The details
of this algorithm can be found in [9,40] and [2, Section 5.1]. Below we give a higher level
explanation of the three basic steps of the Wiedemann algorithm (cf. [18, Section 2.19]).
Given a non-singular d×dmatrix Mover the ﬁnite ﬁeld F2of two elements and bFd
2,
we wish to solve the system Mx =b. The minimal polynomial Fof Mon the vector space
spanned by b,Mb,M2b,. . . has degree at most d, so that
F(M)b=
d
X
i=0
FiMib= 0.
From F0= 1 it follows that x=Pd
i=1 FiMi1b, so it suﬃces to ﬁnd the Fi’s.
Denoting by mi,j the jth coordinate of the vector Mib, it follows from the above equation
that for each jwith 1jdthe sequence (mi,j)
i=0 satisﬁes a linear recurrence relation of
order at most ddeﬁned by the coeﬃcients Fi: for any t0and 1jdwe have that
d
X
i=0
Fimi+t,j = 0.
It is well known that 2d+ 1 consecutive terms of an order dlinear recurrence suﬃce to
compute the coeﬃcients generating the recurrence, and that this can be done eﬃciently using
the Berlekamp-Massey method (cf. [22,40]). Each particular jmay lead to a polynomial of
smaller degree than F, but taking, if necessary, the least common multiple of the polynomials
found for a small number of diﬀerent indices j, the correct minimal polynomial will be found.
Summarizing the above, there are three major stages: a ﬁrst iteration consisting of 2dmatrix
times vector multiplication steps to generate 2d+ 1 consecutive terms of the linear recurrence,
the Berlekamp-Massey stage to calculate the Fi’s, and ﬁnally the second iteration consisting
10
of dmatrix times vector multiplication steps to calculate the solution using the Fi’s. For large
matrices the two iterations, i.e., the ﬁrst and the ﬁnal stage, are the most time consuming.
In practice it is common to use blocking, taking advantage of the fact that on 64-bit machines
64 diﬀerent vectors bover F2can be processed simultaneously, at little or no extra cost
compared to a single vector (cf. [9]), while using the same three main stages. If the vector ¯
b
is 64 bits wide and in the ﬁrst stage, the ﬁrst 64 coordinates of each of the generated 64 bits
wide vectors Mi¯
bare kept, the number of matrix (M) times vector (¯
b) multiplication steps
in both the ﬁrst and the ﬁnal stage is reduced by a factor of 64 compared to the number of
Mtimes bmultiplication steps, while making the central Berlekamp-Massey stage a bit more
cumbersome. It is less common to take the blocking a step further and run both iteration
stages spread over a small number n0of diﬀerent sequences, possibly run on diﬀerent clusters
at diﬀerent locations; in [2] this was done with n0= 4 sequences run on three clusters at two
locations. If for each sequence one keeps the ﬁrst 64 ·n0coordinates of each of the 64 bits
wide vectors they generate during the ﬁrst stage, the number of steps to be carried out (per
sequence) is further reduced by a factor of n0, while allowing independent and simultaneous
execution on possibly n0diﬀerent clusters. After the ﬁrst stage the data generated for the n0
sequences have to be gathered at a central location where the Berlekamp-Massey stage will
be carried out.
While keeping the ﬁrst 64 ·n0coordinates per step for each sequence results in a reduction
of the number of steps per sequence by a factor of 64 ·n0, keeping a diﬀerent number of
coordinates while using n0sequences results in another reduction in the number of steps for
the ﬁrst stage. Following [2, Section 5.1], if the ﬁrst 64 ·m0coordinates are kept of the 64
bits wide vectors for n0sequences, the numbers of steps become d
64·m0+d
64·n0= ( n0
m0+ 1) d
64·n0
and d
64·n0for the ﬁrst and third stage, respectively and for each of the n0sequences. The
choices of m0and n0should be weighed oﬀ against the cost of the Berlekamp-Massey step
with time and space complexities proportional to (m0+n0)3
n0d1+o(1) and (m0+n0)2
n0d, respectively
and for d→ ∞, and where the exponent “3” may be replaced by the matrix multiplication
exponent (our implementation uses “3”).
When actually running the ﬁrst stage of block Wiedemann in this way using n0diﬀerent
sequences, the eﬀect of non-identical resources used for the diﬀerent sequences quickly becomes
apparent: some locations ﬁnish their allotted work faster than others (illustrated in Fig. 1). To
keep the fast contributors busy and to reduce the work of the slower ones (thereby reducing
the wall-clock time), it is possible to let a quickly processed ﬁrst stage sequence continue
for ssteps beyond (n0
m0+ 1) d
64·n0while reducing the number of steps in another ﬁrst stage
sequence by the same s. As described in Appendix B, this can be done in a very ﬂexible
way, as long as the overall number of steps over all ﬁrst stage sequences adds up to n0·
(n0
m0+ 1) d
64·n0. The termination points of the sequences in the third stage need to be adapted
accordingly. This is easily arranged for, since the third stage allows much easier and much
wider parallelization anyhow (assuming checkpoints from the ﬁrst stage are kept). Another
way to keep all parties busy is swapping jobs, thus requiring data exchanges, synchronization,
and more human interaction, making it a less attractive option altogether.
For our matrix with d193·106we used, as in [2], m0= 2n0. But where n0= 4 was used in [2],
we used n0= 8, thereby quadrupling the Berlekamp-Massey runtime and doubling its memory
compared to the matrix from [2], on top of the increased runtime and memory demands caused
by the larger dimension of the matrix. On the other hand, the compute intensive ﬁrst and third
11
(a) First stage contributions. (b) Final shot of third stage bookkeeping tool.
Fig. 1: Contributions to sequences 0-7 with blue indicating INRIA, orange EPFL, and pink NTT.
stages could be split up into twice as many independent jobs as before. For the ﬁrst stage on
average (8
16 + 1)193·106
64·8565 000 steps needed to be taken per sequence (for 8 sequences), for
the second stage the average was about 193·106
64·8380 000 steps. The actual numbers of steps
varied, approximately, between 490 000 and 650 000 for the ﬁrst stage and between 300 000
and 430 000 for the third stage. The calculation of these stages was carried out on a wide
variety of clusters accessed from three locations: a 56-node cluster of 2.2GHz dual hex-core
AMD processors with Inﬁniband at EPFL (installed while the ﬁrst stage was in progress),
a variety of ALADDIN-G5K clusters in France accessed from INRIA LORIA, and a cluster
of 110 3.0GHz Pentium-D processors on a Gb Ethernet at NTT. A comprehensive overview
of clusters and timings is given in Appendix D. It follows from those timings that doing the
entire ﬁrst and third stage would have taken 98 days on 48 nodes (576 cores) of the 56-node
EPFL cluster.
The ﬁrst stage was split up into eight independent jobs run in parallel on those clusters, with
each of the eight sequences check-pointing once every 214 steps. Running a ﬁrst (or third)
stage sequence required 180 GB RAM, a single 64 bits wide ¯
btook 1.5 gigabytes, and a single
mimatrix 8 kilobytes, of which 565 000 were kept, on average, per ﬁrst stage sequence. Each
partial sum during the third stage evaluation required 12 gigabytes.
The central Berlekamp-Massey stage was done in 17 hours and 20 minutes on the 56-node
EPFL cluster (with 16 GB RAM per node) mentioned above, while using just 4 of the 12
available cores per node. Most of the time the available 896 GB RAM suﬃced, but during a
central part of the calculation more memory was needed (up to about 1 TB) and some swapping
occurred. The third stage started right after completion of the second stage, running as many
jobs in parallel as possible. The actual bookkeeping sheet used is pictured in Fig. 1b. Fig. 1a
pictures the ﬁrst stage contributions apocryphally but accurately. Calendar time for the entire
block Wiedemann step was 119 days, ﬁnishing on December 8, 2009. The percentages of the
contributions to the three stages are given more precisely in Table 1.
2.5 That’s a bingo7
There is nothing new to be reported for the square root step, except for the resulting factor-
ization of RSA-768. Nevertheless, and for the record, we present some of the details.
7“Is that the way you say it? “That’s a bingo?”
“You just say “bingo”.” (cf. [39])
12
As expected the matrix step resulted in 512 = 64 ·8linear dependencies modulo 2 among the
exponent vectors. This was more than enough to include the quadratic characters, which were
not included in the matrix, at this stage (cf. [1]). As as result, the solution space was reduced
from 512 to 460 elements, giving us 460 independent chances of about 1
2to factor RSA-768.
In the 52 = 512 460 diﬀerence, a dimension of 46 can be attributed to prime ideals dividing
the leading coeﬃcients or the discriminant that were not included in the matrix.
The square roots of the algebraic numbers were calculated by means of the method from [23]
(see also [31]), which uses the known factorization of the algebraic numbers into small prime
ideals of known norms. The implementation based on [3] turned out to have a bug when
computing the valuations for the free relations of the prime ideals lying above the divisor
17 722 398 737 >232 of the discriminant of f1. Along with a bug in the quadratic character
calculation, this delayed completion of the square root step by a few (harrowing) days. Another
approach would be to write down the algebraic number as a product over the a1in a
dependency and to compute the square root using a mathematics software package. Although
the latter method is attractively simple to program and has been used for small scale NFS
factorizations, it was not considered for the present calculation due to the less appealing size
of the numbers involved. We estimate that writing down the algebraic number would have
required about 36 gigabytes, and that the calculation would ﬁt in 50 to 100 GB RAM. We
did the calculation on the rational side: on a single core of a 2.83GHz Xeon E5440 computing
the six gigabyte rational number took two hours, extracting the squareroot half an hour.
Once the bugs were located and ﬁxed, it took two hours using the hard disk and one core on
each of twelve dual hex-core 2.2GHz AMD processors to compute the exponents of all prime
ideals for eight solutions simultaneously. Computing a square root using the implementation
from [3] took one hour and forty minutes on such a dual hex-core processor. The ﬁrst one (and
four of the other seven) led to the factorization, found at 20:16 GMT on December 12, 2009:
RSA-768 = 3347807169895689878604416984821269081770479498371376856891
2431388982883793878002287614711652531743087737814467999489 ·
3674604366679959042824463379962795263227915816434308764267
6032283815739666511279233373417143396810270092798736308917.
Both factors have 384 bits and 116 decimal digits. The factorizations of the factors ±1can be
found in Appendix C.
3 Concluding remarks
It is customary to conclude a paper reporting a new factoring record with a preview of coming
attractions. For the present case it suﬃces to remark that our main conclusion has already
been reported in [2, Section 7], and was summarized in the ﬁrst paragraph of the introduction
to this paper: at this point factoring a 1024-bit RSA modulus looks more than ﬁve times
easier than a 768-bit RSA modulus looked back in 1999, when we achieved the ﬁrst public
factorization of a 512-bit RSA modulus. Nevertheless, a 1024-bit RSA modulus is still about
one thousand times harder to factor than a 768-bit one. If we are optimistic, it may be possible
to factor a 1024-bit RSA modulus within the next decade by means of an academic eﬀort on
13
the same limited scale as the eﬀort presented here. From a practical security point of view
this is not a big deal, given that standards recommend phasing out such moduli by the end
Another conclusion from our work is that we can quite conﬁdently say that if we restrict our-
selves to an open community, academic eﬀort as ours and unless something dramatic happens
in factoring, we will not be able to factor a 1024-bit RSA modulus within the next ﬁve years
(cf. [29]). After that, all bets are oﬀ.
The ratio between sieving and matrix time was almost 10. This is probably not optimal if
one wants to minimize the overall runtime. But, as already alluded to in the introduction,
minimization of runtime may not be the most important criterion. Sieving is easy, and doing
more sieving may be a good investment if it leads to a less painful matrix step. We expect
that the relations collected for RSA-768 will enable us to get a better insight in the trade-oﬀ
between sieving and matrix eﬀorts, where also the choice of the bound bmay play a role. This
is a subject for further study that may be expected to lead, ultimately, to a recommendation
for close to optimal parameter choices – depending on what one wants to optimize – for NFS
factorizations of numbers of 700 to 800 bits.
Our computation required more than 1020 operations. With the equivalent of almost 2000
years of computing on a single core 2.2GHz AMD Opteron, on the order of 267 instructions
were carried out. The overall eﬀort is suﬃciently low that even for short-term protection of
data of little value, 768-bit RSA moduli can no longer be recommended. This conclusion is
the opposite of the one arrived at on [38], which is based on a hypothetical factoring eﬀort of
six months on 100 000 workstations, i.e., about two orders of magnitude more than we spent.
Acknowledgements
This work was supported by the Swiss National Science Foundation under grant numbers
200021-119776 and 206021-128727 and by the Netherlands Organization for Scientiﬁc Research
(NWO) as project 617.023.613. Experiments presented in this paper were carried out using the
Grid’5000 experimental testbed, being developed under the INRIA ALADDIN development
action with support from CNRS, RENATER and several universities as well as other funding
bodies (see https://www.grid5000.fr). Condor middleware was used on EPFL’s Greedy net-
work. We gratefully acknowledge sieving contributions by BSI, Scott Contini (using resources
provided by AC3, the Australian Centre for Advanced Computing and Communications), Paul
Leyland (using teaching lab machines at the Genetics Department of Cambridge University),
and Heinz Stockinger (using EGEE, Enabling Grids for E-sciencE). Part of this paper was
inspired by Col. Hans Landa.
References
1. L.M. Adleman, Factoring numbers using singular integers, Proceedings 23rd Annual ACM Symposium on
Theory of Computing (STOC) (1991) 64–71.
2. K. Aoki, J. Franke, T. Kleinjung, A.K. Lenstra, D.A. Osvik, A kilobit special number ﬁeld sieve factor-
ization, Proceedings Asiacrypt 2007, Springer-Verlag, LNCS 4833 (2007) 1–12.
3. F. Bahr, Liniensieben und Quadratwurzelberechnung für das Zahlkörpersieb, Diplomarbeit, University of
Bonn, 2005.
14
4. F. Bahr, M. Böhm, J. Franke, T. Kleinjung, Factorization of RSA-200, May 2005, http://www.loria.fr/
~zimmerma/records/rsa200.
5. J. Buhler, P.L. Montgomery, R. Robson, R. Ruby, Technical report implementing the number ﬁeld sieve,
Oregon State University, Corvallis, OR, 1994.
6. S. Cavallar, Strategies for ﬁltering in the number ﬁeld sieve, Proceedings ANTS IV, Springer-Verlag, LNCS
1838 (2000) 209–231.
7. S. Cavallar, B. Dodson, A. K. Lenstra, P. Leyland, P. L. Montgomery, B. Murphy, H. te Riele, P. Zim-
mermann, et al., Factoring a 512-bit RSA modulus, Proceedings Eurocrypt 2000, Springer-Verlag, LNCS
1807 (2000) 1–18.
8. D. Coppersmith, Solving linear equations over GF(2): block Lanczos algorithm, Linear algebra and its
applications 192 (1993) 33–60.
9. D. Coppersmith, Solving homogeneous linear equations over GF(2) via block Wiedemann algorithm, Math.
of Comp. 62 (1994) 333–350.
10. J. Cowie, B. Dodson. R.M. Elkenbracht-Huizing, A.K. Lenstra, P.L. Montgomery, J. Zayer, A world wide
number ﬁeld sieve factoring record: on to 512 bits, Proceedings Asiacrypt’96, Springer-Verlag, LNCS 1163
(1996), 382–394.
11. J.D. Dixon, Asymptotically fast factorization of integers, Math. Comp. 36 (1981) 255–260.
12. J. Franke, T. Kleinjung, Continued fractions and lattice sieving; Proceedings SHARCS 2005; http://www.
ruhr-uni-bochum.de/itsc/tanja/SHARCS/talks/FrankeKleinjung.pdf.
13. R. Golliver, A.K. Lenstra, K. McCurley, Lattice sieving and trial division, Proceedings ANTS’94, Springer-
Verlag, LNCS 877 (1994) 18–27.
14. T. Kleinjung, Cofactorisation strategies for the number ﬁeld sieve and an estimate for the sieving step
for factoring 1024 bit integers, http://www.hyperelliptic.org/tanja/SHARCS/talks06/thorsten.pdf,
2005.
15. T. Kleinjung, On polynomial selection for the general number ﬁeld sieve. Math. Comp. 75 (2006) 2037–
2047.
16. T. Kleinjung, Polynomial selection, talk presented at the CADO workshop on integer factorization,
17. A.K. Lenstra, Computational methods in public key cryptology, pp. 175–238 in: H. Niederreiter, Coding
theory and cryptology, Singapore University Press, 2002.
18. A.K. Lenstra, H.W. Lenstra, Jr., Algorithms in number theory, chapter 12 in Handbook of theoretical
computer science, Volume A, algorithms and complexity (J. van Leeuwen, ed.), Elsevier, Amsterdam (1990).
19. A.K. Lenstra, H.W. Lenstra, Jr. (editors), The development of the number ﬁeld sieve, Springer-Verlag,
LNM 1554, August 1993.
20. A.K. Lenstra, H.W. Lenstra, Jr., M.S. Manasse, J. Pollard, The factorization of the ninth Fermat number,
Math. Comp. 61 (1994) 319–349.
21. A.K. Lenstra, E. Tromer, A. Shamir, W. Kortsmit, B. Dodson, J. Hughes, P. Leyland, Factoring estimates
for a 1024-bit RSA modulus, Proceedings Asiacrypt 2003, Springer-Verlag, LNCS 2894 (2003) 55–74.
22. J. Massey, Shift-register synthesis and BCH decoding, IEEE Trans. Inf. Th., 15 (1969) 122–127.
23. P.L. Montgomery, Square roots of products of algebraic numbers, http://ftp.cwi.nl/pub/pmontgom/
sqrt.ps.gz.
24. P.L. Montgomery, B. Murphy, Improved polynomial selection for the number ﬁeld sieve, extended abstract
for the conference on the mathematics of public-key cryptography, June 13-17, 1999, the Fields institute,
25. M.A. Morrison, J. Brillhart, A method of factorization and the factorization of F7, Math. Comp. 29 (1975)
183–205.
26. B.A. Murphy, Modelling the Yield of Number Field Sieve Polynomials, Proceedings ANTS III, Springer-
Verlag, LNCS 1423 (1998) 137–147.
27. National Institute of Standards and Technology, Suite B Cryptography: http://csrc.nist.gov/groups/
SMA/ispab/documents/minutes/2006-03/E_Barker-March2006-ISPAB.pdf.
28. National Institute of Standards and Technology, Special Publication 800-57: Recommendation for Key
Management Part 1: General (Revised),
http://csrc.nist.gov/publications/nistpubs/800-57/sp800-57-Part1-revised2_Mar08-2007.pdf.
29. National Institute of Standards and Technology, Discussion paper: the transitioning of cryptographic
algorithms and key sizes,
http://csrc.nist.gov/groups/ST/key_mgmt/documents/Transitioning_CryptoAlgos_070209.pdf.
30. NFS@home, http://escatter11.fullerton.edu/nfs.
15
31. P. Nguyen, A Montgomery-like square root for the number ﬁeld sieve, Proceedings ANTS III, Springer-
Verlag, LNCS 1423 (1998) 151–168.
32. J.M. Pollard, The lattice sieve, 43–49 in [19].
33. C. Pomerance, Analysis and comparison of some integer factoring algorithms, in Computational methods
in number theory (H.W. Lenstra, Jr., R. Tijdeman, eds.), Math. Centre Tracts 154,155, Mathematisch
Centrum, Amsterdam (1983) 89-139.
34. C. Pomerance, The quadratic sieve factoring algorithm, Proceedings Eurocrypt’84, Springer-Verlag, LNCS
209 (1985) 169-182.
35. C. Pomerance, A tale of two sieves, http://www.ams.org/notices/199612/pomerance.pdf.
36. R. Rivest, A. Shamir, L. Adleman, A method for obtaining digital signatures and public key cryptosystems,
Commun. of the ACM, 21 (1978) 120–126.
37. The RSA challenge numbers, formerly on http://www.rsa.com/rsalabs/node.asp?id=2093, now on for
instance http://en.wikipedia.org/wiki/RSA_numbers.
38. The RSA factoring challenge FAQ, http://www.rsa.com/rsalabs/node.asp?id=2094.
http://www.imdb.com/title/tt0361748/quotes.
40. E. Thomé, Subquadratic computation of vector generating polynomials and improvement of the block
Wiedemann algorithm, Journal of symbolic computation 33 (2002) 757–775.
A Sieving by vectors
The aim of this section is to give a short description of the implementation of lattice sieving
described in [12] and used for most of the number ﬁeld sieve factorization records of the
Let vk(a, b) = bdkfk(a/b). Recall that the idea of the method of lattice sieving, which was
introduced by Pollard [32], is to increase the probability of smoothness of v1(a, b)by only
looking at (a, b)-pairs for which v1(a, b)is divisible by some large prime q, called the special q.
Let smod qbe a residue class such that this is the case for asb mod q. One constructs a
reduced base (u, v)of the lattice of all (a, b)Z2with asb mod q. A scalar product adapted
to the skewness of the polynomial pair is used for this reduction. The problem is then to ﬁnd
all pairs (i, j),I /2i < I /2,0< j < J , without common divisor such that v1(a, b)/q and
v2(a, b)are both smooth, with (a, b) = iu +jv. For the sake of simplicity we assume Ito be
even. As mentioned in Section 2.3, for practical values of the parameters, Itends to be much
smaller than the smoothness bounds b1and b2, and it is non-trivial to eﬃciently sieve such
regions.
Pollard proposed to do this by using, for each (prime,root) pair pwith prime pbounded by
the relevant smoothness bound bk, a reduced base of the lattice Γpof pairs (i, j)for which
vk(a, b)for the corresponding (a, b)-pair is divisible by p. In [13] that approach was used for p
larger than a small multiple of I, while avoiding storage of “even, even” sieve locations (and
using line sieving for the other primes). Our approach is similar, but instead uses a truncated
continued fraction expansion to determine a basis B=(α, β),(γ, δ)of Γpwith the following
properties:
aThe numbers βand δare positive.
bWe have I < α 0γ < I and γαI.
Let us assume that Γpconsists of all (i, j)for which iρj mod p, where 0< ρ < p. The case
ρ= 0 and the case where Γpconsists of all (i, j)for which pdivides jare not treated, because
16
they produce just (0,1) and (1,0), respectively, as only coprime pairs. We also assume pI, as
smaller primes are better treated by line sieving. To construct a basis with the above properties,
one takes (i0, j0)=(p, 0),(i1, j1) = (ρ, 1) and puts (i+1, j+1 )=(i1, j1) + r(i, j)with
r=i1
i. Note that (1)+1i0, that ris positive and that the jthus form an increasing
sequence of non-negative numbers. The process is stopped at the ﬁrst with |i|< I . If this
number is odd, we put (α, β)=(i1, j1) + r(i, j), where ris the smallest integer for
which α > I. If is even, we put (γ , δ) = (i1, j1) + r(i, j), where ris the smallest
integer such that γ < I. In both cases, the element of B=(α, β),(γ , δ)not yet described is
given by (i, j).
To explain how to eﬃciently sieve using a basis with these properties, let (i, j)Γpsuch
that I/2i < I/2. We want to ﬁnd the (uniquely determined) (i0, j0)Γpsuch that
I/2i0< I/2,j0> j, and j0is as small as possible. As Bis a basis of Γp, there are integers
dand ewith
(i0, j0)(i, j ) = d(α, β) + e(γ, δ ).
If the numbers dand ewere both diﬀerent from zero, with opposite signs, then condition bon
Bwould force the ﬁrst component of the right hand side to have absolute value I, whereas
our constraints on iand i0force it to have absolute value < I. Since j0j,β, and δare all
positive, we have d0and e0. It is now easy to see that the solution to our problem is:
(d, e) =
(0,1) if i < I/2γ
(1,1) if I/2γi < I/2α
(1,0) if i≥ −I/2α.
To convince oneself of the minimality of j0, one notes that d= 0 leads to a violation of i0< I/2
unless i < I/2γ(i.e., save for the ﬁrst of the above cases) and that e= 0 leads to i0<I/2
unless i≥ −I/2α(i.e., save for the third of the above cases).
To implement this process on a modern CPU, it seems best to take I= 2ιfor some natural
number ι. It is possible to identify pairs (i, j)of integers with I/2i < I/2with integers
xby putting x=j·I+i+I/2. If x0=j0·I+i0+I /2with (i0, j 0)as above, then x0=x+C,
x0=x+A+Cand x0=x+Ain the three cases above, with A=α+I·βand C=γ+I·δ.
The ﬁrst component of a pair (i, j),(α, β )or (γ, δ)is extracted from these numbers by using
a bitwise logical operation, and the selection of the appropriate one of the above three cases
is best done using conditional move instructions.
For cache eﬃciency, the sieving region Sqwas split into areas At,0t<T, of size equal to
the L1-cache size. For primes plarger than that size (or a small multiple thereof), sieving is
not done directly. Instead, the numbers xcorresponding to elements of SqΓpwere calculated
ahead of the sieving process, and their oﬀsets into the appropriate region Atstored in the cor-
responding element of an array Sof Tstacks. To implement the trial division sieve eﬃciently,
the corresponding factor base index was also stored. Of course, this approach may also be
used for line sieving, and in fact was used in [3]. An approach which seems to be somewhat
similar has been described by T. Oliveira e Silva in connection with his implementation of the
Odlyzko-Lagarias-Lehmer-Meissel method.
17
Parallelization is possible in several diﬀerent ways. A topology for splitting the sieving region
among several nodes connected by a network is described in [12]. If one wants to split the task
among several cores sharing their main memory, it seems best to distribute the regions Atand
also the large factor base primes among them. Each core ﬁrst calculates its part of S, for its
assigned part of the large factor base elements, and then uses the information generated by
all cores to treat its share of regions At. A lattice siever parallelized that way was used for
a small part of the RSA-576 sieving tasks, but the code fell out of use and was not used for
the current project. The approach may be more useful today, with many cores per processor
being a standard.
B Unbalanced sequences in block Wiedemann
Before describing the modiﬁcation for unbalanced sequence lengths we give an overview of
Coppersmith’s block version of the Berlekamp-Massey algorithm. To avoid a too technical
description we simplify the presentation of Coppersmith’s algorithm and refer to [9] for details.
The modiﬁcation can also be applied to Thomé’s subquadratic algorithm (cf. [40]) which is
what we did and used. In the following the variables mand ndo not denote the common root
and a number to be factored, but have the same meaning as in Coppersmith’s article. As in
that article, the terms +O(1) are constants depending on mand n. We assume that mand n,
which play the role of 64 ·m0and 64 ·n0in Section 2.4, are much smaller than d.
Let Mbe a d×dmatrix over F2,mn,xkFd
2,1kmand yjFd
2,1jn
satisfying certain conditions. Set a(i)
j,k =xT
kMiyjand
A=X
i
(a(i)
j,k)XiMatn,m [X].
In the ﬁrst step of block Wiedemann we calculate the coeﬃcients of Aup to degree d
m+d
n+O(1).
The goal of the Berlekamp-Massey algorithm is to ﬁnd polynomials of matrices FMatn,n[X]
and GMatn,m[X]satisfying deg(F)d
n+O(1),deg(G)d
n+O(1) and
F A G(mod Xd
m+d
n+O(1)).
Intuitively, we want to produce at least dzero rows in the higher coeﬃcients of F A up to
degree d
m+d
n+O(1). Writing F=PdF
i=0(f(i)
j,k)Xi,dF= deg(F)the jth row of coeﬃcient
dF+bof Gbeing zero corresponds to
(Mbxh)Tvj= 0 for 1hm, 0b < d
m+O(1) where
vj=
n
X
k=1
dF
X
i=0
f(dFi)
j,k ·Miyk.
Coppersmith’s algorithm produces a sequence of matrices (of m+nrows) FtMatm+n,n[X]
and GtMatm+n,m[X]for t=t0,..., d
m+d
n+O(1) (where t0=O(1)) such that
FtAGt(mod Xt)
18
holds and the degrees of Ftand Gtare roughly m
m+nt. In a ﬁrst step t0and Ft0are chosen such
that certain conditions are satisﬁed, in particular we have deg(Ft0) = O(1) and deg(Gt0) =
O(1). To go from tto t+ 1 a polynomial of degree 1 of matrices PtMatm+n,m+n[X]is
constructed and we set Ft+1 =PtFtand Gt+1 =PtGt. This construction is done as follows.
We have FtAGt+EtXt(mod Xt+1)for some matrix Et. Respecting a restriction involving
the degrees of the rows of Gt(essentially we avoid destroying previously constructed zero rows
in the G’s) we perform a Gaussian elimination on Et, i.e., we obtain ˜
Ptsuch that
˜
PtEt=0
1m.
Then we set
Pt=1n0
01mX·˜
Pt.
In this way the degrees of at most mrows are increased when passing from Gtto Gt+1 (due to
the restriction mentioned above ˜
Ptdoes not increase the degrees), so the total number of zero
rows in the coeﬃcients is increased by n. Due to the restriction mentioned above the degrees
of the rows of Ftand of Gtgrow almost uniformly, i.e., they grow on average by m
m+nwhen
going from tto t+ 1.
After t=d
m+d
n+O(1) steps the total number of zero rows in the coeﬃcients of Gtis
m+n
md+O(1) such that we can select mrows that produce at least dzero rows in the coeﬃcients.
These mrows form Fand G.
We now consider unbalanced sequence lengths. Let jbe the length of sequence j, i.e., a(i)
j,k
has been computed for all kand 0ij. Without loss of generality we can assume
12. . . n=. The sum of the lengths of all sequences has to satisfy again Pjj
d·(1 + n
m) + O(1). Moreover we can assume that 1d
m, otherwise we could drop sequence 1
In this setting our goal is to achieve
F A G(mod X+O(1))
with dF= deg(F)d
m,deg(G)d
mand
Xk|F·,k (this denotes the kth column of F).
The latter condition will compensate our ignorance of some rows of the higher coeﬃcients
of A. Indeed, setting for simplicity dF=d
m, the vectors
vj=
n
X
k=1
kd
m
X
i=0
f(dFi)
j,k ·Miyk
satisfy for 1hm, 0b < d
m
(Mbxh)Tvj=
n
X
k=1
kd
m
X
i=0
f(dFi)
j,k a(i+b)
k,h =g(dF+b)
j,h = 0.
19
If i+b > k(thus a(i+b)
k,h not being computed), we have dFi < dF+bkk, so
f(dFi)
j,k = 0 and the sum computes g(dF+b)
j,h .
Our new goal is achieved as before, but we will need steps and the construction of Pthas
to be modiﬁed as follows. In step twe have FtAGt+EtXt(mod Xt+1). Let anbe
maximal such that a1
X
i=1
(m+i)(ni+1 ni)mt
(awill increase during the computation). In the Gaussian elimination of Etwe do not use the
ﬁrst narows for elimination. As a consequence, ˜
Pthas the form
˜
Pt=1na
0.
Then we set
Pt=
1naX0 0
01a0
0 0 1mX
·˜
Pt.
Therefore the sum of the degrees of Ftwill be increased by m+naand the number of zero
rows in Gtwill be increased by awhen passing from tto t+1. For a ﬁxed a,(m+a)(na+1na)
m
steps will increase the average degree of the last m+arows from na+1 to na. At
this point awill be increased.
To see why Xk|F·,k holds we have to describe the choice of Ft0(and t0). Let cbe the
number of maximal j, i.e., nc< nc+1 =n. Then Ft0will be of the form
Ft0=1ncXt00
0.
The last m+crows will be chosen such that they are of degree at most t01and such that
the conditions in Coppersmith’s algorithm are satisﬁed. This construction will lead to a value
of t0near m
cinstead of the lower value near m
nin the original algorithm.
Let kbe such that k<  and consider the kth column of Ft. As long as nakthis column
will have the only non-zero entry at row kand this will be Xt. Since nakholds for
tkthis column will be divisible by Xkfor all tk.
For RSA-768 we used the algorithm as described above in the subquadratic version of Thomé.
However, a variant of the algorithm might be useful in certain situations (e.g., if one of the
sequences is much longer than the others) which we will sketch brieﬂy.
Suppose that n1< n. Then for t < (m+1)(nn1)
mwe have a= 1 and Ptis of the form
Pt=1n1X
0.
A product of several of these Ptwill have a similar form, namely an (n1) ×(n1) unit
matrix times a power of Xin the upper left corner and zeros below it.
The basic operations in Thomé’s subquadratic version are building a binary product tree of
these Ptand doing truncated multiplications of intermediate products with Ft0Aor similar
20
polynomials. If we split the computation in this algorithm into two stages, ﬁrst computing
the product of all Ptfor t < (m+1)(nn1)
mand then the remaining product, the matrix
multiplications in the ﬁrst stage become easier due to the special form of the Ptand its
products.
Obviously this can be done in as many stages as there are diﬀerent values among the j.
C Factorizations of p±1and q±1
Referring to the smallest factor of RSA-768 as pand to its cofactor as q, the prime factor-
izations of p±1and q±1are as follows, where “pk denotes a prime number of kdecimal
digits:
p1 = 28·112·13 ·7193 ·160378082551 ·7721565388263419219 ·
111103163449484882484711393053 ·p47,
p+ 1 = 2 ·3·5·31932122749553372262005491861630345183416467 ·p71,
q1 = 22·359 ·p113,
q+ 1 = 2 ·3·23 ·41 ·47 ·239875144072757917901 ·p90.
D Clusters used and block Wiedemann timings
21
Cluster number of CPU type clock speed cores GB RAM interconnect nodes cores seconds per iteration communication
location nodes (GHz) per node per node per job per job stage 1 stage 3
Lausanne 56 2×AMD 2427 2.2 12 16 ib20g 12 144 4.3-4.5 4.8 40%
Tokyo 110 2×Pentium 4 3.0 2 5 eth1g 110 220 5.8, 6.4 7.8 33%, 44%
Grenoble 34 2×Xeon E5420 2.5 8 8 ib20g 24 144 3.7 n/a 30%
Lille 46 2×Xeon E5440 2.8 8 8 mx10g 36 144 3.1 3.3 31%
32 256 3.8 n/a 38%
24 144 4.4 n/a 33%
Nancy 92 2×Xeon L5420 2.5 8 16 ib20g 64 256 2.2 2.4 41%
36 144 3.0 3.2 31%
24 144 3.5 4.2 30%
18 144 n/a 5.0 31%
16 64 n/a 6.5 19%
Orsay 120 2×AMD 250 2.4 2 2 mx10g 98 196 2.8 3.9 32%
Rennes 96 2×Xeon 5148 2.3 4 4 mx10g 64 256 2.5 2.7 37%
49 196 2.9 3.5 33%
Rennes 64 2×Xeon L5420 2.5 8 32 eth1g 49 196 6.2 n/a 67%
24 144 8.4 n/a 67%
18 144 10.0 n/a 68%
8 64 n/a 18.0 56%
Table 2: Data and ﬁrst and third stage block Wiedemann timings for all clusters used.
“n/a” means that the job conﬁguration was not used.
: ﬁgure per iteration per sequence when two sequences are processed in parallel, in which case a part of the communication time is hidden in the local
computation time (the communication number shows the pure communication percentage),
for all other ﬁgures just a single sequence is processed.
22
... Many techniques have already been suggested to protect highvalued digital assets; but unable to provide tight security against the attackers. In 2010, Kleinjung et al. break 768-bits RSA security by factorization [14]. Approximately a decade later, E. Thomé [15] break the security of 795-bits RSA in December 2019. ...
... For encryption, the plaintext M is converted into integer value m and the m is encrypted using Equation (14). ...
Conference Paper
Recent developments suggest RSA suffers from integer factorization issues because it depends on two prime numbers. In this paper, we analyze the issues of integer factorization and suggest possible future solutions for RSA for conventional computers. Recent developments in Quantum Computing make integer factorization easier than ever. Therefore, RSA cryptography will obsolete soon after the launching of the complete quantum computer. Still, quantum computers are designed especially for particular purposes, but not for general purposes. Therefore, we analyze the enhancement mechanism of RSA till the arrival of general-purpose quantum computers. We analyze the linear RSA, multiple RSA, recursive RSA, and multiple recursive RSA in this paper. We describe these variants of RSA mathematically and analyze the feasibility of these RSA variants for the future. However, we found that reRSA is the most suitable variant of RSA.
... A trusted third party distributes the public key, and the private key is kept secret. The most famous public-key cryptography is RSA [25]; however, it has factorization issue [13,28,32]. Recent literature suggests that RSA can be broken in a few hours, which causes serious concerns to the cryptography community [10]. ...
... There are currently several encryption algorithms that use prime factorization, but historically the most important is RSA [3]. One of the first standards was RSA-768 [4], which represents a 232-digit number. The standard has been broken by scientists by bringing together hundreds of computers that have been working for a period representing 2000 years of one computer work. ...
Article
Full-text available
Skewes’ number was discovered in 1933 by South African mathematician Stanley Skewes as upper bound for the first sign change of the difference π ( x ) − li( x ). Whether a Skewes’ number is an integer is an open problem of Number Theory. Assuming Schanuel’s conjecture, it can be shown that Skewes’ number is transcendental. In our paper we have chosen a different approach to prove Skewes’ number is an integer, using lattice points and tangent line. In the paper we acquaint the reader also with prime numbers and their use in RSA coding, we present the primary algorithms Lehmann test and Rabin-Miller test for determining the prime numbers, we introduce the Prime Number Theorem and define the prime-counting function and logarithmic integral function and show their relation.
... Latest technological advancements and communications, on the other hand, have made it easier to achieve the computational effort capable of breaking specific asymmetric schemes, increasing the recommended minimum key size. For example, since 768-bit and 1024-bit RSA implementations were broken around 2010 [21,22] the minimum recommended RSA key size is now between 2048 and 4096 bits (depending on the type of information to be protected). Key size expansion is a temporary solution until technology catches up and provides the necessary computational effort. ...
Conference Paper
Full-text available
The Internet of Things (IoT) is shaping the world of every economy and client. By 2023, clients, companies, and governments will be able to install a total of 40 billion IoT devices worldwide. The 2021's challenges simply confirmed the necessity to secure all four elements of the IoT Model: Sensors, Networks, Cloud, and Applications. Therefore, it is important to be pragmatic regarding the extreme threats and high worth of attacks on secrecy in IoT systems. Although the concept of convergence of IoT and Quantum Computing is not a standby topic, it had been mentioned in several works of literature and coated by many researchers, however, nothing is getting ready for sensible applications thus far. Quantum Computing is not prepared nevertheless, it is so far from preparation on a billboard scale. Quantum computing resolves most of the issues in IoT however it occurs yet below the analysis phase and can take over a couple of years to induce in public obtainable use. During this research, the author will start through a broad introduction to the internet of things and Quantum Computing and carry on with issues that can be featured throughout the convergence of Quantum Computing and IoT. To solve these integration issues in terms of cryptography this research will also provide the answer. At long last, this research will call attention to the hybrid of both fields to existing IoT network design.
... Another scenario of those attacks may be an improper generation of the random primes, i.e., unstable PRNGs usage in the implementation. Even research was done in which a researchers group tried to estimate the factor of an RSA-232-bit number, and they estimated that it would take more than 1500 years to crack the cipher [105], but recently in the year 2020, RSA-232 was factored by [117], [118]. It can be said from the prior summaries and discussion that there has been numerous research that has already been done for the RSA security development (Fig 6), and for many upcoming years, it can be said that, in terms of security enhancements via mathematical modifications and complexity increment, we have a possibility to rely on these already proposed RSA variants after testing and comparing their authenticity on some real-world applications. ...
Article
Full-text available
The interconnected digital world is generating enormous data that must be secured from unauthorized access. Advancement in technologies and new innovative methods applied by attackers play an instrumental role in breaching data security. Public key Cryptography provides a set of cryptographic algorithms in achieving data security through confidentiality, integrity and authentication. Among all cryptographic algorithms in general and public key cryptography in particular, RSA is one of the most widely used and applied algorithms. Since its inception, it is commonly being adopted in securing data across different domains such as cloud, image and others. Despite its importance and wide applications, no such systematic and extensive survey exists in the literature. A systematic and thorough study of RSA based cryptography is presented in this work covering several domains. All the available works in this direction are divided into 11 different categories, viz, Hybrid, Parallel, Cloud, Image, Multiple-Keys, Chinese-Remainder-Theorem-based, Digital-Signatures, K-Nearest-Theorem-based, Batch, Wireless, and Core-Modifications. This study methodically explores RSA-based cryptosystems, either modifications in core RSA or applications of enhanced RSA across different domains, systematically categorizing in various categories and eventually providing findings and indications. The current study compares RSA methods based on parameters such as key generation, encryption schemes, decryption schemes, key features and enhancements, and also finds the leading areas where modified RSA has been applied in the recent past. As a result, this study will guide researchers and practitioners in understanding the past and present status of RSA cryptography along with the possibility of its applications in other domains.
... Keys in asymmetric cryptography are also more vulnerable to brute force attacks than in secret-key cryptography. There exist algorithms for publickey cryptography that allow attackers to crack private keys faster than a brute force method would require [6]. One of the main strength of public-key cryptography remains in the difficulty of factorization to the large primes. ...
Article
We propose an optical physical layer encryption method based on real-time optical signal temporal scrambling and decoy-state quantum key distribution. The real-time optical signal processing is operating with an array of tunable Fabry-Perot cavities by using the updating and synchronization keys. Based on this scheme, an experimental demonstration is given to realize secure communication for optical physical layer in a standard 10 Gbps data transmission with quantum key distribution in real time. An on-line transmitted data over 100 km of a standard single mode fiber is realized with power penalty of as low as 3.45 dB, where the average key generation rate is better than 800 bps. The experimental results indicate that, the proposed encryption scheme of dynamic scrambling with quantum keys exhibits significantly potential for improving the confidentiality of optical physical layer.
Chapter
With the recent shift to post-quantum algorithms it becomes increasingly important to provide precise bit-security estimates for code-based cryptography such as McEliece and quasi-cyclic schemes like BIKE and HQC. While there has been significant progress on information set decoding (ISD) algorithms within the last decade, it is still unclear to which extent this affects current cryptographic security estimates. We provide the first concrete implementations for representation-based ISD, such as May-Meurer-Thomae (MMT) or Becker-Joux-May-Meurer (BJMM), that are parameter-optimized for the McEliece and quasi-cyclic setting. Although MMT and BJMM consume more memory than naive ISD algorithms like Prange, we demonstrate that these algorithms lead to significant speedups for practical cryptanalysis on medium-sized instances (around 60 bit). More concretely, we provide data for the record computations of McEliece-1223 and McEliece-1284 (old record: 1161), and for the quasi-cyclic setting up to code length 2918 (before: 1938). Based on our record computations we extrapolate to the bit-security level of the proposed BIKE, HQC and McEliece parameters in NIST’s standardization process. For BIKE/HQC, we also show how to transfer the Decoding-One-Out-of-Many (DOOM) technique to MMT/BJMM. Although we achieve significant DOOM speedups, our estimates confirm the bit-security levels of BIKE and HQC. For the proposed McEliece round-3 192 bit and two out of three 256 bit parameter sets, however, our extrapolation indicates a security level overestimate by roughly 20 and 10 bits, respectively, i.e., the high-security McEliece instantiations may be a bit less secure than desired.
Conference Paper
After going through past research, it is observed that the security of data depends on five important key features such as authentication, data Integrity, mutual authentication between two parties, non-repudiation, and key exchange algorithm. Even if one of these features fail then it may breach security in real-world scenarios. The Diffie-Hellman Key Exchange (DHKE) algorithm allows securing one feature called key exchange in the cryptography field. However, this key exchange algorithm is unable to provide authentication. As a result, this key exchange scheme is prone to a hostile attack named as Man-In-The-Middle Attack (MITM). To solve this type of attack, a novel algorithm has been proposed called modification of the DHKE algorithm. In this algorithm, they claim that the MITM attack on classical DHKE can be prevented by applying the necessary functionalities like encryption and Zero-Knowledge Proof (ZKP). Although, after going through result analysis and working techniques, it is observed that the modified DHKE algorithm is still suffering from MITM attack. In this paper, the main objective is to develop an algorithm for finding out this type of security flaw in the existing modified DH algorithm which includes the concept of required authentication and encryption functions and able to perform the MITM attack on the existing modified DHKE algorithm.
Article
The continued fraction method for factoring integers, which was introduced by D. H. Lehmer and R. E. Powers, is discussed along with its computer implementation. The power of the method is demonstrated by the factorization of the seventh Fermat number F7 and other large numbers of interest.
Article
In this paper we exhibit the full prime factorization of the ninth Fermât number Fg = 2512 + 1. It is the product of three prime factors that have 7, 49, and 99 decimal digits. We found the two largest prime factors by means of the number field sieve, which is a factoring algorithm that depends on arithmetic in an algebraic number field. In the present case, the number field used was Q(v^2). The calculations were done on approximately 700 workstations scattered around the world, and in one of the final stages a supercomputer was used. The entire factorization took four months.
Article
This chapter discusses algorithms that solve two basic problems in computational number theory—factoring integers into prime factors and finding discrete logarithms. In the factoring problem, one is given an integer n1 and is asked to find the decomposition of n into prime factors. It is common to split this problem into two parts. The first is called primality testing: given n, it is determined whether n is prime or composite. The second is called factorization: if n is composite, a nontrivial divisor of n is to be calculated. In the discrete logarithm problem, one is given a prime number p, and two elements h, y of the multiplicative group F*p of the field of integers modulo p. The algorithms and their analyses depend on many different parts of number theory. Number theory is considered the purest of all sciences, and within number theory the hunt for large primes and for factors of large numbers has always been remote from applications, even to other questions of a number-theoretic nature.
Article
Factorization of large numbers amounts to trial and retrial. The quadratic and algebraic number sieves produce sophisticated trials which are efficient for numbers up to 150 digits.
Article
The paper describes a "probabilistic algorithm" for finding a factor of any large composite integer $n$ (the required input is the integer $n$ together with an auxiliary sequence of random numbers). It is proved that the expected number of operations which will be required is $O(\exp\{\beta(\ln n \ln \ln n)^{1/2}\})$ for some constant $\beta > 0$. Asymptotically, this algorithm is much faster than any previously analyzed algorithm for factoring integers; earlier algorithms have all required $O(n^\alpha)$ operations where $\alpha > 1/5$.
Article
We propose a method of solving large sparse systems of homogeneous linear equations over GF(2), the field with two elements. We modify an algorithm due to Wiedemann. A block version of the algorithm allows us to perform 32 matrix-vector operations for the cost of one. The resulting algorithm is competitive with structured Gaussian elimination in terms of time and has much lower space requirements. It may be useful in the last stage of integer factorization.
Article
The continued fraction method for factoring integers, which was introduced by D. H. Lehmer and R. E. Powers, is discussed along with its computer implementation. The power of the method is demonstrated by the factorization of the seventh Fermat number $F_7$ and other large numbers of interest.