The following is an implementation of elliptical curve Point Multiplication, but it's not working as expected (using recent Chrome / Node with BigInt for illustration):
const bi0 = BigInt(0)
const bi1 = BigInt(1)
const bi2 = BigInt(2)
const bi3 = BigInt(3)
const absMod = (n, p) => n < bi0 ? (n % p) + p : n % p
export function pointAdd (xp, yp, xq, yq, p) {
const lambda = (yq - yp) / (xq - xp)
const x = absMod(lambda ** bi2 - xp - xq, p)
const y = absMod(lambda * (xp - x) - yp, p)
return { x, y }
}
export function pointDouble (xp, yp, a, p) {
const numer = bi3 * xp ** bi2 + a
const denom = (bi2 * yp) ** (p - bi2)
const lambda = (numer * denom) % p
const x = absMod(lambda ** bi2 - bi2 * xp, p)
const y = absMod(lambda * (xp - x) - yp, p)
return { x, y }
}
export function pointMultiply (d, xp, yp, a, p) {
const add = (xp, yp, { x, y }) => pointAdd(xp, yp, x, y, p)
const double = (x, y) => pointDouble(x, y, a, p)
const recur = ({ x, y }, n) => {
if (n === bi0) { return { x: bi0, y: bi0 } }
if (n === bi1) { return { x, y } }
if (n % bi2 === bi1) { return add(x, y, recur({ x, y }, n - bi1)) }
return recur(double(x, y), n / bi2)
}
return recur({ x: xp, y: yp }, d)
}
Given a known curve with properties, the above succeeds for points P2 - P5, but starts failing at P6 onwards:
const p = BigInt('17')
const a = BigInt('2')
const p1 = { x: BigInt(5), y: BigInt(1) }
const d = BigInt(6)
const p6 = pointMultiply(d, p1.x, p1.y, a, p)
p6.x === BigInt(16) // incorrect value of 8 was returned
p6.y === BigInt(13) // incorrect value of 14 was returned
The known curve has values:
P X Y
——————————
1 5 1
2 6 3
3 10 6
4 3 1
5 9 16
6 16 13
7 0 6
8 13 7
9 7 6
10 7 11
I'm not sure if I misunderstand the algorithm or I've made an error in the implementation.
I don't know javascript very well, so the code confuses me. But ...
In function
pointAdd
, and everywhere else, "division" must be done mod p. You're evidently doing it correctly inpointDouble
, but not inpointAdd
. InpointAdd
, use the same pattern: Instead ofdo
though I would think it would much clearer and less error prone to simply have a modular inverse function instead of computing Xp-2 everywhere you need it.