Paper 2009/318

The Fermat factorization method revisited

Robert ERRA and Christophe GRENIER

Abstract

We consider the well known Fermat factorization method ({\it FFM}) when it is applied on a balanced RSA modulus $N=pq>0$, with primes $p$ and $q$ supposed of equal length. We call the {\it Fermat factorization equation} the equation (and all the possible variants) solved by the FFM like $\mathcal{P}(x,y)=(x+2R{)}^2-y^2-4N=0$ (where $R=\lceil N^{1/2}\rceil$). These equations are bivariate integer polynomial equations and we propose to solve them directly using Coppersmith's methods for bivariate integer polynomials. As we use them as a black box, our proofs will be brief. We show first that, using Coppersmith's methods, we can factor $$ in a polynomial time if $$, with $$ and, using the fact that the Newton polygon of $$ is a lower triangle we show a better result: we can indeed factor $$ in a polynomial time if $$. Unfortunately this shows that using Coppersmith's methods for bivariate integer polynomials is no better than the FFM, because in that case the FFM is immediate. This is confirmed by numerical experiments. We then propose another method: solving the {\it modular} Fermat factorization equation $$. Since Coppersmith's methods for {\it modular} multivariate integer polynomial equations are {\it empirical}, there relies on the the famous {\it "resultant heuristic"}, we get only an empirical method that can factor $$ in a polynomial time if $$. We conclude with proposals for future works.