Aspects of Index Calculus Algorithms for Discrete Logarithm and Class Group Computations

Date of Submission

September 2021

Date of Award

9-1-2022

Institute Name (Publisher)

Indian Statistical Institute

Document Type

Doctoral Thesis

Degree Name

Doctor of Philosophy

Subject Name

Computer Science

Department

Applied Statistics Unit (ASU-Kolkata)

Supervisor

Sarkar, Palash (ASU-Kolkata; ISI)

Abstract (Summary of the Work)

The thesis focuses on two problems: The discrete logarithm problem(DLP) and the class group com- putation problem. Besides the inherent mathematical appeal, both of these problems have strong cryp- tographical interest. Discrete Logarithm Problem The computation of discrete logarithms is known to be a hard problem in general. The intractability of the discrete logarithm problem has lead to schemes that revolutionized cryptography. The known algo- rithms for DLP can be categorised into two major categories, namely, generic algorithms which work in any group and index calculus algorithms which work in a special class of groups. It is known that Pollard's rho algorithm is the best generic algorithm to compute discrete logarithm. The complete discrete logarithm in a nite eld is found by applying Pollard's rho in subproblems that may arise when index calculus algorithms are applied. The tag tracing variant of Pollard's rho is a highly e cient algorithm. We have intertwined the technique of Montgomery multiplication with tag tracing in elds of prime order. As a result, the expensive modular reductions are completely replaced by divisions by a suitable power of two. The implementation of these divisions is a cakewalk as it is just right shift operations. In doing this, we do not compromise with any advantage that tag tracing o ered. The essential di erence is instead of doing eld multiplication after a certain number of steps we just do a Montgomery multiplication after the same number of steps. The net result of our work is that the speedup is huge without any additional costs of memory or time. We next focus on general medium prime elds. We perform a record discrete log computation for a eld with 22-bit characteristic and 1051-bit size. It successfully combats the challenge to perform a larger discrete logarithm computation for a medium prime case eld than what had been reported earlier with- out losing generality. The work also reports progress in discrete logarithm computation for the general medium prime case using the function eld sieve algorithm. Fields considered earlier with large sizes en- joyed the advantage o ered by the Kummer extension property. This made the factor base size small and 2 �� 1 descent which is one of the last steps of the index calculus technique easier. In our case, the factor base is larger than what has been considered earlier. The di culty of 2��1 descent is also clearly visible. An increase in the size of the eld makes various steps of the function eld sieve more complicated. Our study delves into these di culties. The linear algebra step along with descent form the main stumbling blocks. Some previously known techniques have been analysed and implemented e ciently. The manner in which the various methods are used in our work is equally applicable for any general medium prime eld for performing future record computations, provided suitable computational resources are present. We next consider the problem of computing discrete logarithms with function eld sieve in Fpn for a small characteristic p and composite n. The last phase of the function eld sieve, namely the individual logarithm step itself again consist of two sub-steps of initial splitting and descent. The focus in this work is on the initial splitting phase. We have proposed a new algorithm for initial splitting in small characteristic elds of composite extension degree. It has been shown to be better than the other existing algorithms like Waterloo algorithm and Guillevic's initial splitting algorithm. Implementation of the new algorithm has also been done. Our algorithm reduces the cost of generation of polynomials which are to be tested for smoothness. Additionally, our algorithm is completely parallelizable compared to some sort of semi-parallelism possible for Guillevic's algorithm. The usage of this algorithm is appropriate when attempting record discrete logarithm computations over such elds. Class Group Computation The class group computation problem is quite important and relevant as it provides information about the structure of multiplication in the eld. However, computationally it remains a di cult problem to date. Class group computation is one of the four major problems in computational algebraic number theory postulated by Zassenhaus (the others being computation of unit group, ring of integers, Galois group). Class groups are important both mathematically as well as cryptographically. We have introduced a technique to obtain practical speed up for relation collection in class group computations. We have done Magma implementations of both the new algorithm as well as the previous one given by G elin. Timing results con rm that there is indeed a substantial practical speed-up in relation collection by the new algorithm over G elin's algorithm

Comments

ProQuest Collection ID: https://www.proquest.com/pqdtlocal1010185/dissertations/fromDatabasesLayer?accountid=27563

Control Number

ISILib-TH515

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

DOI

http://dspace.isical.ac.in:8080/jspui/handle/10263/2146

Link to Full Text

Share

COinS