The algorithm was published by René Schoof in 1985 and it was a theoretical breakthrough, as it was the first deterministic polynomial time algorithm for counting points on elliptic curves. Before Schoof's algorithm, approaches to counting points on elliptic curves such as the naive and baby-step giant-step algorithms were, for the most part, tedious and had an exponential running time.
This article explains Schoof's approach, laying emphasis on the mathematical ideas underlying the structure of the algorithm.
Hasse's theorem states that if is an elliptic curve over the finite field , then satisfies
This powerful result, given by Hasse in 1934, simplifies our problem by narrowing down to a finite (albeit large) set of possibilities. Defining to be , and making use of this result, we now have that computing the cardinality of modulo where , is sufficient for determining , and thus . While there is no efficient way to compute directly for general , it is possible to compute for a small prime, rather efficiently. We choose to be a set of distinct primes such that . Given for all , the Chinese remainder theorem allows us to compute .
In order to compute for a prime , we make use of the theory of the Frobenius endomorphism and division polynomials. Note that considering primes is no loss since we can always pick a bigger prime to take its place to ensure the product is big enough. In any case Schoof's algorithm is most frequently used in addressing the case since there are more efficient, so called adic algorithms for small-characteristic fields.
Given the elliptic curve defined over we consider points on over , the algebraic closure of ; i.e. we allow points with coordinates in . The Frobenius endomorphism of over extends to the elliptic curve by .
This map is the identity on and one can extend it to the point at infinity , making it a group morphism from to itself.
The Frobenius endomorphism satisfies a quadratic polynomial which is linked to the cardinality of by the following theorem:
Theorem: The Frobenius endomorphism given by satisfies the characteristic equation
Thus we have for all that , where + denotes addition on the elliptic curve and and denote scalar multiplication of by and of by .
One could try to symbolically compute these points , and as functions in the coordinate ring of and then search for a value of which satisfies the equation. However, the degrees get very large and this approach is impractical.
Schoof's idea was to carry out this computation restricted to points of order for various small primes . Fixing an odd prime , we now move on to solving the problem of determining , defined as , for a given prime . If a point is in the -torsion subgroup, then where is the unique integer such that and . Note that and that for any integer we have . Thus will have the same order as . Thus for belonging to , we also have if . Hence we have reduced our problem to solving the equation
The lth division polynomial is such that its roots are precisely the x coordinates of points of order l. Thus, to restrict the computation of to the l-torsion points means computing these expressions as functions in the coordinate ring of Eand modulo the lth division polynomial. I.e. we are working in . This means in particular that the degree of X and Y defined via is at most 1 in y and at most in x.
The scalar multiplication can be done either by double-and-add methods or by using the th division polynomial. The latter approach gives:
where is the nth division polynomial. Note that is a function in x only and denote it by .
We must split the problem into two cases: the case in which , and the case in which . Note that these equalities are checked modulo .
Note that this computation fails in case the assumption of inequality was wrong.
We are now able to use the x-coordinate to narrow down the choice of to two possibilities, namely the positive and negative case. Using the y-coordinate one later determines which of the two cases holds.
We first show that X is a function in x alone. Consider . Since is even, by replacing by , we rewrite the expression as
and have that
Now if for one then satisfies
for all l-torsion points P.
As mentioned earlier, using Y and we are now able to determine which of the two values of ( or ) works. This gives the value of . Schoof's algorithm stores the values of in a variable for each prime l considered.
We begin with the assumption that . Since l is an odd prime it cannot be that and thus . The characteristic equation yields that . And consequently that . This implies that q is a square modulo l. Let . Compute in and check whether . If so, is depending on the y-coordinate.
If q turns out not to be a square modulo l or if the equation does not hold for any of w and , our assumption that is false, thus . The characteristic equation gives .
If you recall, our initial considerations omit the case of . Since we assume q to be odd, and in particular, if and only if has an element of order 2. By definition of addition in the group, any element of order 2 must be of the form . Thus if and only if the polynomial has a root in , if and only if .
1. An elliptic curve .
2. An integer q for a finite field with .
The number of points of E over .
Choose a set of odd primes S not containing psuch that Putif, else.Compute the division polynomial .
All computations in the loop below are performed in the ring Fordo:Let be the unique integersuch that and .Compute , and .ifthenCompute .fordo:ifthenifthen;else.else ifq is a square modulo lthencompute w with compute ifthenelse ifthenelseelse
Use the Chinese Remainder Theorem to compute t modulo N
from the equations , where .
Most of the computation is taken by the evaluation of and , for each prime , that is computing , , , for each prime . This involves exponentiation in the ring and requires multiplications. Since the degree of is , each element in the ring is a polynomial of degree . By the prime number theorem, there are around primes of size , giving that is and we obtain that . Thus each multiplication in the ring requires multiplications in which in turn requires bit operations. In total, the number of bit operations for each prime is . Given that this computation needs to be carried out for each of the primes, the total complexity of Schoof's algorithm turns out to be . Using fast polynomial and integer arithmetic reduces this to .
In the 1990s, Noam Elkies, followed by A. O. L. Atkin, devised improvements to Schoof's basic algorithm by restricting the set of primes considered before to primes of a certain kind. These came to be called Elkies primes and Atkin primes respectively. A prime is called an Elkies prime if the characteristic equation: splits over , while an Atkin prime is a prime that is not an Elkies prime. Atkin showed how to combine information obtained from the Atkin primes with the information obtained from Elkies primes to produce an efficient algorithm, which came to be known as the Schoof–Elkies–Atkin algorithm. The first problem to address is to determine whether a given prime is Elkies or Atkin. In order to do so, we make use of modular polynomials, which come from the study of modular forms and an interpretation of elliptic curves over the complex numbers as lattices. Once we have determined which case we are in, instead of using division polynomials, we are able to work with a polynomial that has lower degree than the corresponding division polynomial: rather than . For efficient implementation, probabilistic root-finding algorithms are used, which makes this a Las Vegas algorithm rather than a deterministic algorithm. Under the heuristic assumption that approximately half of the primes up to an bound are Elkies primes, this yields an algorithm that is more efficient than Schoof's, with an expected running time of using naive arithmetic, and using fast arithmetic. It should be noted that while this heuristic assumption is known to hold for most elliptic curves, it is not known to hold in every case, even under the GRH.
Several algorithms were implemented in C++ by Mike Scott and are available with source code. The implementations are free (no terms, no conditions), and make use of the MIRACL library which is distributed under the AGPLv3.