# Discrete logarithms: The past and the odlyzko/doc/ Discrete logarithms: The

date post

24-Jan-2020Category

## Documents

view

0download

0

Embed Size (px)

### Transcript of Discrete logarithms: The past and the odlyzko/doc/ Discrete logarithms: The

Discrete logarithms: The past and the future

Andrew Odlyzko AT&T Labs - Research amo@research.att.com

July 19, 1999

Abstract

The first practical public key cryptosystem to be published, the Diffie-Hellman key exchange algo-

rithm, was based on the assumption that discrete logarithms are hard to compute. This intractability

hypothesis is also the foundation for the presumed security of a variety of other public key schemes.

While there have been substantial advances in discrete log algorithms in the last two decades, in general

the discrete log still appears to be hard, especially for some groups, such as those from elliptic curves.

Unfortunately no proofs of hardness are available in this area, so it is necessary to rely on experience

and intuition in judging what parameters to use for cryptosystems. This paper presents a brief survey

of the current state of the art in discrete logs.

1. Introduction

Many of the popular public key cryptosystems are based on discrete exponentiation. If �

is a group,

such as the multiplicative group of a finite field or the group of points on an elliptic curve, and � is an element of

� , then (writing the group multiplicatively) ��� is the discrete exponentiation of base � to

the power � . This operation shares many properties with ordinary exponentiation, so that, for example, � � ��� ��

� � ��� � ���

The inverse operation is, given � in � , to determine � (if it exists) such that � � � � . The number� , usually taken in the range ��������������� � , where � !"� is the order of H, and ����� is the subgroup generated by � , is called the discrete logarithm of � to base � , since it again shares many properties with the ordinary logarithm. For example, if we use the notation � �$#&%('*)�+ �-, when � � ��� , then, assuming for simplicity that

� is cyclic and is generated by � ,

#&%('�)�+ � �/. ,10 #&%('2)�+ � ,43 #&%('*)�+&. , +657%98 � � �:, � 1

Discrete logs have a long history in number theory. Initially they were used primarily in compu-

tations in finite fields (where they typically appeared in the closely related form of Zech’s logarithm).

However, they were rather obscure, just like integer factorization. Unlike the latter, they could not even

invoke any famous quotes of Gauss (cf. [BachS]) about their fundamental importance in mathematics.

The status of discrete logs started to grow in the 20th century, as more computations were done, and as

more thought went into algorithmic questions. It appears that they started to play an important role in

cryptography already in the 1950s, long before public key systems appeared on the scene, as cryptosys-

tems based on shift-register sequences displaced those based on rotor machines. Discrete logs occur

naturally in that context as tools for finding where in a shift register sequence a particular block occurs.

The main impetus for the intensive current interest in discrete logs, though, came from the invention of

the Diffie-Hellman (DH) method [DiffieH].

The DH key-exchange algorithm was the first practical public key technique to be published, and it

is widely used. The basic approach is that if Alice and Bob wish to create a common secret key, they

agree on a group ����� , and then Alice chooses a random integer ; , while Bob chooses a random integer< . Alice then computes �*= and sends it to Bob over a public channel, while Bob computes �*> and sends

that to Alice. Now Alice and Bob can both compute

� = ? > ��+ � = , > ��+ � > , =A@ while an eavesdropper who happens to have overheard the exchange, and thus knows � , � = , and � > , will hopefully not be able to compute the secret � =B> .

If the discrete log problem for the group ����� is easy, an eavesdropper can compute either ; or < , and can find out what � =C> is. It is an important open question whether determining � =B> knowing just � ,� = , and � > is as hard as the discrete log problem in general. (See [MaurerW] for the latest references on this topic, which will not be covered here. For references on another important subject, namely

that of bit security of the discrete log, which will also not be dealt with here, see [BonehV, HastadN].)

However, a fast discrete log algorithm would definitely destroy the utility of the widely used Diffie-

Hellman protocol. This factor has stimulated an outpouring of research on the complexity of discrete

logs.

This paper is a brief survey of the current state of the art in algorithms for discrete logs. There

are many cryptosystems based on discrete exponentiation other than the DH key exchange algorithm.

Starting especially with the Taher ElGamal proposal [ElGamal], many schemes have been proposed,

including the official U.S. Digital Signature Algorithm (DSA). However, they will not be covered here,

2

and I refer readers to [MenezesVOV, Schneier] for more information about them. Even in the area of

complexity of the discrete log problem there have been several general surveys [Lebedev, McCurley,

Odlyzko1, Odlyzko2, SchirokauerWD], as well as several more recent papers on specialized subfields.

Therefore in this paper I will only give pointers to the latest results, and present some high level obser-

vations about the current status and likely future of the discrete log problem.

2. Why discrete logs?

Almost everything that public key cryptography provides, such as digital signatures and key exchange,

can be accomplished with RSA and its variants. However, cryptosystems based on discrete exponenti-

ation remain of interest for three main reasons:

(a) Patent issues. The Diffie-Hellman patent expired in 1997, while the RSA patent has until the

fall of 2000 to run. Therefore anyone interested in using public key cryptography in the United States

(which is the only place where these patents were applied for and issued) can save money and also

avoid licensing negotiations.

(b) Technical advantages. In many cases where algorithms of comparable functionality exist, say

one over the finite field of integers modulo a prime D , and another using a composite integer � of the same size, breaking the discrete log modulo D appears to be somewhat harder than factoring the integer� . Further, elliptic curve cryptosystems appear to offer the possibility of using much smaller key sizes than would be required by RSA-type cryptosystems of comparable security.

Some other advantages of discrete log cryptosystems come from their limitations. It is widely

believed that the U.S. Digital Signature Algorithm is based on discrete logs because it is harder to use

it for encryption than if it were based on RSA (and thus on integer factorization). This helps enforce

export control regulations on strong encryption without weakening the digital signature methods that

are less stringently controlled. On the other hand, many people like the DH algorithm, since the session

key it generates is evanescent. In the simplest application of RSA to key generation, Alice creates a

session key and transmits it to Bob using Bob’s public key. An eavesdropper who can coerce Bob

afterwards into revealing his private key can then recover the full text of the communication exchanged

by Alice and Bob. In contrast, if Alice and Bob use DH to generate the session key, destroy it after the

session ends, and do not store their communication, then neither coercion nor cryptanalysis will enable

the eavesdropper to find out what information was exchanged.

3

(c) They are different. Cryptographers have learned by bitter experience that it is unwise to put all

eggs in a single basket. It is desirable to have a diversity of cryptosystems, in case one is broken.

It is an unfortunate fact that discrete logs and integer factorization are so close that many algorithms

developed for one problem can be modified to apply to the other. For security, it would be better to have

much more diversity. However, more than two decades after the publication of the first two practical

public key systems, the DH and the RSA algorithms, the only public key cryptosystems that are trusted

and widely deployed are based on the presumed difficulty of the same two problems those schemes

relied upon. Interestingly enough, the earlier discovery of public key cryptography in the classified

community in the early 1970s [GCHQ] also produced essentially the same two algorithms. There have

been many attempts to find public key schemes based on other principles, but so far most have led to

systems that were broken, and the ones that are still standing are often regarded with suspicion.

3. General attacks

This section discusses some general algorithms for discrete logs that assume little knowledge of the

group.

For most commonly encountered cyclic groups � � ����� , there is an efficient method for producing

a unique canonical representation for an element. (There are exceptions, though, such as some class

groups, in which equivalence of two representations is hard to