|
|
package r1csqap
import ( "bytes" "fmt" "math/big"
"github.com/mottla/go-snark/fields" )
// Transpose transposes the *big.Int matrix
func Transpose(matrix [][]*big.Int) [][]*big.Int { var r [][]*big.Int for i := 0; i < len(matrix[0]); i++ { var row []*big.Int for j := 0; j < len(matrix); j++ { row = append(row, matrix[j][i]) } r = append(r, row) } return r }
// ArrayOfBigZeros creates a *big.Int array with n elements to zero
func ArrayOfBigZeros(num int) []*big.Int { bigZero := big.NewInt(int64(0)) var r = make([]*big.Int, num, num) for i := 0; i < num; i++ { r[i] = bigZero } return r } func BigArraysEqual(a, b []*big.Int) bool { if len(a) != len(b) { return false } for i := 0; i < len(a); i++ { if !bytes.Equal(a[i].Bytes(), b[i].Bytes()) { return false } } return true }
// PolynomialField is the Polynomial over a Finite Field where the polynomial operations are performed
type PolynomialField struct { F fields.Fq }
// NewPolynomialField creates a new PolynomialField with the given FiniteField
func NewPolynomialField(f fields.Fq) PolynomialField { return PolynomialField{ f, } }
// Mul multiplies two polinomials over the Finite Field
func (pf PolynomialField) Mul(a, b []*big.Int) []*big.Int { r := ArrayOfBigZeros(len(a) + len(b) - 1) for i := 0; i < len(a); i++ { for j := 0; j < len(b); j++ { r[i+j] = pf.F.Add( r[i+j], pf.F.Mul(a[i], b[j])) } } return r }
// Div divides two polinomials over the Finite Field, returning the result and the remainder
func (pf PolynomialField) Div(a, b []*big.Int) ([]*big.Int, []*big.Int) { // https://en.wikipedia.org/wiki/Division_algorithm
r := ArrayOfBigZeros(len(a) - len(b) + 1) rem := a for len(rem) >= len(b) { l := pf.F.Div(rem[len(rem)-1], b[len(b)-1]) pos := len(rem) - len(b) r[pos] = l aux := ArrayOfBigZeros(pos) aux1 := append(aux, l) aux2 := pf.Sub(rem, pf.Mul(b, aux1)) rem = aux2[:len(aux2)-1] } return r, rem }
func max(a, b int) int { if a > b { return a } return b }
// Add adds two polinomials over the Finite Field
func (pf PolynomialField) Add(a, b []*big.Int) []*big.Int { r := ArrayOfBigZeros(max(len(a), len(b))) for i := 0; i < len(a); i++ { r[i] = pf.F.Add(r[i], a[i]) } for i := 0; i < len(b); i++ { r[i] = pf.F.Add(r[i], b[i]) } return r }
// Sub subtracts two polinomials over the Finite Field
func (pf PolynomialField) Sub(a, b []*big.Int) []*big.Int { r := ArrayOfBigZeros(max(len(a), len(b))) for i := 0; i < len(a); i++ { r[i] = pf.F.Add(r[i], a[i]) } for i := 0; i < len(b); i++ { r[i] = pf.F.Sub(r[i], b[i]) } return r }
// Eval evaluates the polinomial over the Finite Field at the given value x
func (pf PolynomialField) Eval(v []*big.Int, x *big.Int) *big.Int { r := big.NewInt(int64(0)) for i := 0; i < len(v); i++ { xi := pf.F.Exp(x, big.NewInt(int64(i))) elem := pf.F.Mul(v[i], xi) r = pf.F.Add(r, elem) } return r }
// NewPolZeroAt generates a new polynomial that has value zero at the given value
func (pf PolynomialField) NewPolZeroAt(pointPos, totalPoints int, height *big.Int) []*big.Int { //todo note that this will blow up. big may be necessary
fac := 1 //(xj-x0)(xj-x1)..(xj-x_j-1)(xj-x_j+1)..(x_j-x_k)
for i := 1; i < totalPoints+1; i++ { if i != pointPos { fac = fac * (pointPos - i) } }
facBig := big.NewInt(int64(fac)) hf := pf.F.Div(height, facBig) r := []*big.Int{hf} for i := 1; i < totalPoints+1; i++ { if i != pointPos { ineg := big.NewInt(int64(-i)) //is b1 necessary?
b1 := big.NewInt(int64(1)) r = pf.Mul(r, []*big.Int{ineg, b1}) } } return r }
// LagrangeInterpolation performs the Lagrange Interpolation / Lagrange Polynomials operation
func (pf PolynomialField) LagrangeInterpolation(v []*big.Int) []*big.Int { // https://en.wikipedia.org/wiki/Lagrange_polynomial
var r []*big.Int for i := 0; i < len(v); i++ { r = pf.Add(r, pf.NewPolZeroAt(i+1, len(v), v[i])) //r = pf.Mul(v[i], pf.NewPolZeroAt(i+1, len(v), v[i]))
} //
return r }
// R1CSToQAP converts the R1CS values to the QAP values
//it uses Lagrange interpolation to to fit a polynomial through each slice. The x coordinate
//is simply a linear increment starting at 1
//within this process, the polynomial is evaluated at position 0
//so an alpha/beta/gamma value is the polynomial evaluated at 0
// the domain polynomial therefor is (-1+x)(-2+x)...(-n+x)
func (pf PolynomialField) R1CSToQAP(a, b, c [][]*big.Int) (alphas [][]*big.Int, betas [][]*big.Int, gammas [][]*big.Int, domain []*big.Int) { aT := Transpose(a) bT := Transpose(b) cT := Transpose(c) fmt.Println(aT) fmt.Println(bT) fmt.Println(cT)
for i := 0; i < len(aT); i++ { alphas = append(alphas, pf.LagrangeInterpolation(aT[i])) }
for i := 0; i < len(bT); i++ { betas = append(betas, pf.LagrangeInterpolation(bT[i])) }
for i := 0; i < len(cT); i++ { gammas = append(gammas, pf.LagrangeInterpolation(cT[i])) } //it used to range till len(alphas)-1, but this was wrong.
z := []*big.Int{big.NewInt(int64(1))} for i := 1; i < len(a); i++ { z = pf.Mul( z, []*big.Int{ pf.F.Neg( big.NewInt(int64(i))), big.NewInt(int64(1)), }) } return alphas, betas, gammas, z }
// CombinePolynomials combine the given polynomials arrays into one, also returns the P(x)
func (pf PolynomialField) CombinePolynomials(r []*big.Int, ap, bp, cp [][]*big.Int) ([]*big.Int, []*big.Int, []*big.Int, []*big.Int) { var ax []*big.Int for i := 0; i < len(r); i++ { m := pf.Mul([]*big.Int{r[i]}, ap[i]) ax = pf.Add(ax, m) } var bx []*big.Int for i := 0; i < len(r); i++ { m := pf.Mul([]*big.Int{r[i]}, bp[i]) bx = pf.Add(bx, m) } var cx []*big.Int for i := 0; i < len(r); i++ { m := pf.Mul([]*big.Int{r[i]}, cp[i]) cx = pf.Add(cx, m) }
px := pf.Sub(pf.Mul(ax, bx), cx) return ax, bx, cx, px }
// DivisorPolynomial returns the divisor polynomial given two polynomials
func (pf PolynomialField) DivisorPolynomial(px, z []*big.Int) []*big.Int { quo, _ := pf.Div(px, z) return quo }
|