Methods to solve Ax2 + Bxy + Cy2 + Dx + Ey + F = 0

by Dario Alejandro Alpern

The purpose of this article is to show how to solve the Diophantine Equation Ax2 + Bxy + Cy2 + Dx + Ey + F = 0. The term Diophantine Equation means that the solutions (x, y) should be integer numbers. For example, the equation 4y2 - 25y + 10 = 0 has solutions given by the horizontal line y = 2.5, but since 2.5 is not an integer number, we will say that the equation has no solutions.

There are several cases that depend on the values of A, B and C. The names are taken from the figures represented by the equation in the plane xy: a line, an ellipse, a parabola or a hyperbola (or two lines). These figures are the set of real solutions. In our situation, the set of solutions are represented by isolated point/s in the plane xy.


Contents


Linear case: A = B = C = 0

The equation is now: Dx + Ey + F = 0. There are several cases:

Example 1: Solve 10x + 84y + 16 = 0.

gcd(D, E) = gcd(10, 84) = 2. Since the constant term is also multiple of 2, we will divide the equation by this gcd.

The equation is now: 5x + 42y + 8 = 0.

Now we must apply the Generalized Euclidean algorithm:

Multiplying the last equation by -F = -8 we obtain:
(-136) * 5 + 16 * 42 = -8

Adding and subtracting de t = 5 * 42 t we obtain:
(-136 + 42 t) * 5 + (16 - 5 t) * 42 = -8

So, the solution is given by the set:

x = -136 + 42 t
y = 16 - 5 t

where t is any integer number.


Simple Hyperbolic case A = C = 0; B != 0

Since A = C = 0 the original equation is reduced to Bxy + Dx + Ey + F = 0, so:

Bxy + Dx + Ey + F=0
Bxy + Dx + Ey=-F
B2xy + BDx + BEy=-BF
B2xy + BDx + BEy + DE=DE - BF
(Bx + E) (By + D)=DE - BF

There are two cases: DE - BF = 0 (two lines parallel to x and y axes respectively) and DE - BF != 0 (a hyperbola whose asymptotes are parallel to x and y axes).

In the first case a necessary condition to have solutions occurs when one of the parentheses equal zero, i.e., Bx + E = 0 or By + D = 0. Since B != 0, we have solutions for:

x = -E
B
 ,   y = any integer  (if E is multiple of B)
x = any integer ,   y = -D
B
    (if D is multiple of B)

In the second case the values of x and y are found by finding all divisors of DE - BF. Let d1, d2, ..., dn be the set of divisors of DE - BF.So,

Bx + E=di     By + D=(DE - BF) / di
Bx=di - E     By=(DE - BF) / di - D
x=di - E
B
y=(DE - BF) / di - D
B

Example 2: Solve 2xy + 5x + 56y + 7 = 0.

In this case the divisors of DE - BF = 5*56 - 2*7 = 266 are: ±1, ±2, ±7, ±14, ±19, ±38, ±133, ±266.

Since (2x + 56) (2y + 5) = 266 we obtain:

The only 8 solutions to the requested equations are marked above in red.


Elliptical case B2 - 4AC < 0

Since the ellipse is a closed figure, the number of solutions will be finite.

Operating with the original quadratic equation:

Ax2 + Bxy + Cy2 + Dx + Ey + F = 0

Cy2 + (Bx + E)y + (Ax2 + Dx + F) = 0

y =-(Bx + E) ± sqrt[(Bx + E)2 - 4C(Ax2 + Dx + F)]
2C
    (*)

For any value of x there will be two values of y except at the left and right extremes of the ellipse. In this case there will be only one value of y. To determine the location of the left and right extremes we should equal the square root to zero, so the previous expression returns only one value of y.

(Bx + E)2 - 4C(Ax2 + Dx + F) = 0

(B2 - 4AC)x2 + 2(BE - 2CD)x + (E2 - 4CF) = 0

So the values of x should be between the roots of this equation. If the roots are not real, there will be no solutions to the original equation, else, all integer values of x should be replaced in equation (*) in order to find an integer value of y.

Example 3: Solve 42x2 + 8xy + 15y2 + 23x + 17y - 4915 = 0.

Since B2 - 4AC = 82 - 4*42*15 = -2456 < 0 the equation is elliptical.

The values of x should be between the roots of (B2 - 4AC)x2 + 2(BE - 2CD)x + (E2 - 4CF) = -2456x2 - 1108x + 295189 = 0. The roots equal -11.19... and 10.74..., so we have to check the values of x from -11 to 10.

The only value of x that replaced in (*) makes y integer occurs for x = -11, where y = -1, therefore this is the only solution to this problem.


Parabolic case B2 - 4AC = 0

Let g = gcd(A,C), a = A/g >= 0, b = B/g, c = C/g >= 0.

Since b2 = 4ac is positive, we can choose g with the same sign of A. In this way a and c will be positive (or one of them zero).

The expression b2 - 4ac = 0 implies that b2/4 = ac. Since gcd(a,c) = 1, both a and c are perfect squares.

Multiplying the original equation by sqrt a:

sqrt ag(ax2 + bxy + cy2) + sqrt aDx + sqrt aEy + sqrt aF = 0

sqrt ag(sqrt ax + sqrt cy)2 + sqrt aDx + sqrt aEy + sqrt aF = 0

where for sqrt c the sign of B/A is taken.

Adding and subtracting sqrt cDy:

sqrt ag(sqrt ax + sqrt cy)2 + D(sqrt ax + sqrt cy) - sqrt cDy + sqrt aEy + sqrt cF = 0

Let u = sqrt ax + sqrt cy:    (i)

sqrt agu2 + Du + (sqrt aE - sqrt cD)y + sqrt aF = 0

(sqrt cD - sqrt aE)y = sqrt agu2 + Du + sqrt aF (ii)

There are two cases: sqrt cD - sqrt aE = 0 (two parallel lines) or sqrt cD - sqrt aE != 0 (a parabola).

In the first case, sqrt cD - sqrt aE = 0.

From (ii): sqrt agu2 + Du + sqrt aF = 0

Since x and y should be integer numbers, the equation (i) implies that the number u (the root of the above equation) should be also integer. Let u1 and u2 be the roots of the above equation.

From (i) we have: sqrt ax + sqrt cy - u1 = 0 and sqrt ax + sqrt cy - u2 = 0 which can be solved with the methods for the linear equation.

In the second case, sqrt agu2 + Du + sqrt aF should be multiple of sqrt cD - sqrt aE.

Let u0, u1,... the values of u in the range 0 <= u < |sqrt cD - sqrt aE| for which the above condition holds.

So u = ui + (sqrt cD - sqrt aE)t, where t is any integer number.   (iii)

Replacing (iii) in (ii):

(sqrt cD - sqrt aE)y = sqrt ag[ui + (sqrt cD - sqrt aE)t]2 + D[ui + (sqrt cD - sqrt aE)t] + sqrt aF

y = sqrt ag(sqrt cD - sqrt aE)t2 + (D + 2sqrt agui)t + sqrt agui2 + Dui + sqrt aF
sqrt cD - sqrt aE

From (i) and (iii):

u = sqrt ax + sqrt cy = ui + (sqrt cD - sqrt aE)t

sqrt ax=sqrt asqrt cg(sqrt aE - sqrt cD)t2 + (sqrt cD - sqrt aE - 2sqrt asqrt cgui - sqrt cD)t +
+
ui - sqrt csqrt agui2 + Dui + sqrt aF
sqrt cD - sqrt aE

sqrt ax=sqrt asqrt cg(sqrt aE - sqrt cD)t2 + (- sqrt aE - 2sqrt asqrt cgui)t +
+
ui(sqrt cD - sqrt aE) - sqrt asqrt cgui2 + sqrt cDui + sqrt asqrt cF
sqrt cD - sqrt aE

sqrt ax=sqrt asqrt cg(sqrt aE - sqrt cD)t2 + (- sqrt aE - 2sqrt asqrt cgui)t -
-
sqrt asqrt cgui2 + sqrt aEui + sqrt asqrt cF
sqrt cD - sqrt aE

x = sqrt cg(sqrt aE - sqrt cD)t2 - (E + 2sqrt cgui)t - sqrt cgui2 + Eui + sqrt cF
sqrt cD - sqrt aE

y = sqrt ag(sqrt cD - sqrt aE)t2 + (D + 2sqrt agui)t + sqrt agui2 + Dui + sqrt aF
sqrt cD - sqrt aE

Example 4: Find the solutions for 8 x2 - 24 xy + 18 y2 + 5x + 7y + 16 = 0

We have to calculate the values g, a, c, sqrt a, sqrt c, sqrt cD - sqrt aE and sqrt agu2 + Du + sqrt aF.

g = gcd(8, 18) = 2
a = 8/2 = 4
c = 18/2 = 9
sqrt a = 2
sqrt c = -3 (since B/A = -24/8 < 0)
sqrt cD - sqrt aE = -3 * 5 - 2 * 7 = -29 (second case)
sqrt agu2 + Du + sqrt aF = 4u2 + 5u + 32

We have to determine the values of u in the range 0 <= u < 29 for which 4u2 + 5u + 32 is multiple of 29.

The values of u are: u0 = 2 and u1 = 4.

For u0 = 2:

x = -174 t2 - 17 t - 2
y = -116 t2 - 21 t - 2

For u0 = 4:

x = -174 t2 - 41 t - 4
y = -116 t2 - 37 t - 4


Hyperbolic case B2 - 4AC > 0

Contents


Find solutions of the homogeneous equation Ax2 + Bxy + Cy2 + F = 0

If F = 0 we have the trivial solution x = 0 and y = 0. Now we will investigate if there are more solutions.

Ax2 + Bxy + Cy2 = -F

Multiplying by 4A:
4A2x2 + 4ABxy + 4ACy2 = -4AF
4A2x2 + 4ABxy + B2y2 - B2y2 + 4ACy2 = -4AF
(2Ax + By)2 - (B2 - 4AC)y2 = -4AF

This can be interpreted as a difference of squares:

(2Ax + By + sqrt(B2 - 4AC) y) (2Ax + By - sqrt(B2 - 4AC) y) = -4AF
(2Ax + (B + sqrt(B2 - 4AC))y) (2Ax + (B - sqrt(B2 - 4AC))y) = -4AF

Since -4AF = 0, the condition to have more solutions is that B2 - 4AC should be a perfect square.

Now the same method used for the linear equation (since the equation are represented by two lines in the plane xy intersecting at the point (0, 0)) can be used in order to find the solutions.

If F != 0 and B2 - 4AC = k2 for some integer k, the parentheses in the equation above should be factors of -4AF.

Let u1, u2,... be the positive and negative divisors of -4AF.

Then we have the following set of two linear equations in two unknowns:

2Ax + (B+k)y = ui
2Ax + (B-k)y = -4AF/ui

So we have:

y = (ui + 4AF/ui) / (2k)
x = (ui - (B+k)y)

We should discard the values of ui that makes x or y non-integer.


Let's consider now the case F != 0 and B2 - 4AC not a perfect square.

If F is not a multiple of gcd(A, B, C), the equation has no solutions, otherwise we can divide all coefficients of the equation by this gcd.

If 4 F2 < B2 - 4AC, the solutions of the equation will be amongst the convergents of the continued fraction of the roots of the equation At2 + Bt + C = 0.

The continued fraction expansion of a quadratic irrationality is periodic. Since B2 - 4AC is not a perfect square the number of solutions will be infinite or none.

In the other hand, if 4 F2 >= B2 - 4AC solutions can be obtained as follows:

Let G = gcd(x,y), x = Gu and y = Gv.

The original equation is then: AG2u2 + BG2uv + CG2v2 + F = 0, so F will be multiple of G2.

Dividing the equation by G2:

Au2 + Buv + Cv2 + F/G2 = 0    (1).

Once the values of u and v are found, we can easily determine x = Gu and y = Gv.

So we can assume that gcd(x,y) = 1.

Let x = sy - Fz    (2).

Replacing in the original equation:

A(sy - Fz)2 + B(sy - Fz)y + Cy2 + F = 0

As2y2 - 2AFsyz + AF2z2 + Bsy2 - BFyz + Cy2 = -F

(As2 + Bs + C) y2 + (-2As - B)Fyz + AF2z2 = - F

Dividing by -F:

-(As2 + Bs + C) y2 / F + (2As + B)yz - AFz2 = 1    (3)

Now we must determine the values of s between 0 and F - 1 such that As2 + Bs + C = 0 (mod F). Once the values of y and z are found using continued fraction expansions of the roots of -(As2 + Bs + C) t2 / F + (2As + B)t - AF = 0, the value of x is found by (2). If no solutions are found amongst the convergents, there will be no solutions to (1).

If the original equation has solutions, there should be a solution to the previous congruence, except when gcd(A,B,F) > 1. In this case, if gcd(B,C,F) = 1 we should make the substitution y = sx - Fz    (4), so replacing in the original equation:

Ax2 + Bx(sx - Fz) + C(sx - Fz)2 + F = 0

Ax2 + Bsx2 - BFxz + Cs2x2 - 2CFsxz + CF2z2 = -F

(Cs2 + Bs + A) x2 + (-2Cs - B)Fxz + CF2z2 = - F

Dividing by -F:

-(Cs2 + Bs + A) x2 / F + (2Cs + B)xz - CFz2 = 1    (5).

Now we must determine the values of s between 0 and F - 1 such that Cs2 + Bs + A = 0 (mod F). Once the values of x and z are found using continued fraction expansions of the roots of -(Cs2 + Bs + A) t2 / F + (2Cs + B)t - CF = 0, the value of y is found by (4). If no solutions are found amongst the convergents, there will be no solutions to (1).

The equations (4) and (5) have no solutions when both gcd(A,B,F) and gcd(B,C,F) are greater than 1. In this case we will use the following approach:

Let i, j, m and n be four integers such that in - jm = 1    (6).

If x = iX + jY and y = mX + nY    (7) we obtain X = nx - jy and Y = -mx + iy    (8).

Since the transformation is reversible, we can convert any (x,y) to (X,Y) and vice versa. So we will work with (X,Y) and with these solutions will can compute the values of (x,y) that satisfies the original equation.

Ax2 + Bxy + Cy2 =
= A(iX+jY)2 + B(iX+jY)(mX+nY) + C(mX+nY)2 =
= aX2 + bXY + cY2

where:

a = Ai2 + Bim + Cm2    (9)
b = 2Aij + Bin + Bjm + 2Cmn    (10)
c = Aj2 + Bjn + Cn2    (11)

So we have to find the values of i and m such that a = Ai2 + Bim + Cm2 is relatively prime to F.

Since gcd(C, F) > 1 we have gcd(Ai2 + Bim + Cm2, C) = 1, so gcd(i, C) = 1 and gcd(Ai+Bm, C) = 1.

Since gcd(A, F) > 1 we have gcd(Ai2 + Bim + Cm2, A) = 1, so gcd(m, A) = 1 and gcd(Bi+Cm, A) = 1.

From (6), gcd(i, m) = 1.

If F = 0 (mod p) (p prime):

ABCi, mExamples
A = 0B = 0C = 0Not applicable (gcd(A, B, C) = 1)
A = 0B = 0C != 0m != 0i = 0, m = 1
A = 0B != 0C = 0i != 0, m != 0i = 1, m = 1
A = 0B != 0C != 0m != 0, i != -Cm/Bi = 1-C, m = B
A != 0B = 0C = 0i != 0i = 1, m = 0
A != 0B = 0C != 0i != 0 or m != 0i = 1, m = 1
A != 0B != 0C = 0i != 0, m != -Ai/Bi = B, m = 1-A
A != 0B != 0C != 0i != 0 or m != 0i = 1, m = 1

While it is possible to generate the values of i and m from their values modulo different primes, it is very tedious and it is not necessary, because from the table above, almost all values of i and m can be used. So it is better to use the following pseudocode in order to find both values:

    for i=0 to |F|-1
       for m=0 to i+1
          if gcd(i, m) = 1 and gcd(Ai2 + Bim + Cm2, F) = 1, end.
       next m
    next i

With the values of i and m just found, we can compute the values of j and n from (6) using the methods for the linear equation. Then we have to compute a, b and c using (9), (10) and (11), from which the set of solutions (X,Y) can be found. With the formula (7) we can find the set of solutions (x,y).

Credits: This method was e-mailed to me by Iain Davidson.

Example 5: Find some solutions for 18 x2 + 41 xy + 19 y2 - 24 = 0

First of all we must determine the gcd of all coefficients but the constant term, that is: gcd(18, 41, 19) = 1.

Dividing the equation by the greatest common divisor we obtain:
18 x2 + 41 xy + 19 y2 - 24 = 0

Let x = sy - fz, so [-(as2 + bs + c)/f]y2 + (2sa + b)yz - afz2 = 1.

So 18 s2 + 41 s + 19 should be multiple of 24.

This holds for s = 19.

Since 2 * 2 is a divisor of the constant term (-24), the solutions should be 2 times the solutions of 18 u2 + 41 uv + 19 v2 - 6 = 0.

We have to find the continued fraction expansion of the roots of 18 t2 + 41 t + 19 = 0, that is, (sqrt(313) - 41) / 38

The continued fraction expansion is:
-1 + //2, 1, 5, 8, 1, 2, 17, 2, 1, 8, 5, 1, 3, 1, 1, 2, 2, 1, 1, 3//

where the periodic part is marked in bold (the period has 19 coefficients).

The following table shows how the values of U0 and V0 are found (the third column are the values for 18 u2 + 41 uv + 19 v2):

cnunvn
 10
-1-11-4
2-1212
1-23-3
5-11172
8-90139-11
1-1011566
2-292451-1
17-506578236
2-1042216097-11
1-15487239202
8-134318207457-3
5-6870771 06120512
1-8213951 268662-4
3-3 1512624 8671919
1-3 9726576 135853-8
1-7 12391911 0030446
2-18 22049528 141941-6
2-43 56490967 2869268
1-61 78540495 428867-9
1-105 350313162 7157934
3-377 836343583 576246-12
1-483 186656746 2920393
5-2793 7696234315 036441-2
8-22833 34364035266 58356711
1-25627 11326339581 620008-6
2-74087 570166114429 8235831
17-1 285115 8060851 984888 620919-6
2-2 644319 1823364 084207 06542111
1-3 929434 9884216 069095 686340-2
8-34 079799 08970452 636972 5561413
5-174 328430 436941269 253958 467045-12
1-208 408229 526645321 890931 0231864
3-799 553119 0168761234 926751 536603-9
1-1007 961348 5435211556 817682 5597898
1-1807 514467 5603972791 744434 096392-6
2-4622 990283 6643157140 306550 7525736
2-11053 495034 88902717072 357535 601538-8
1-15676 485318 55334224212 664086 3541119
1-26729 980353 44236941285 021621 955649-4
3-95866 426378 880449148067 728952 22105812

As explained above, x = 2u and y = 2v, so:

X0 = -202
Y0 = 312

X0 = 202
Y0 = -312

X0 = -10130
Y0 = 15646

X0 = 10130
Y0 = -15646

X0 = -14 247838 (8 digits)
Y0 = 22 006088 (8 digits)

X0 = 14 247838 (8 digits)
Y0 = -22 006088 (8 digits)

X0 = -9245 980567 328630 (16 digits)
Y0 = 14280 613101 505146 (17 digits)

X0 = 9245 980567 328630 (16 digits)
Y0 = -14280 613101 505146 (17 digits)

The other root of the equation 18 t2 + 41 t + 19 = 0 is (-sqrt(313) - 41) / 38

Its continued fraction expansion is:
-2 + //2, 1, 2, 2, 1, 1, 3, 1, 5, 8, 1, 2, 17, 2, 1, 8, 5, 1, 3, 1//

where the periodic part is marked in bold (the period has 19 coefficients).

The following table shows how the values of U0 and V0 are found (the third column are the values for 18 u2 + 41 uv + 19 v2):

cnunvn
 10
-2-219
2-32-8
1-536
2-138-6
2-31198
1-4427-9
1-75464
3-269165-12
1-3442113
5-19891220-2
8-16256997111
1-1824511191-6
2-52746323531
17-914927561192-6
2-1 8826001 15473711
1-2 7975271 715929-2
8-24 26281614 8821693
5-124 11160776 126774-12
1-148 37442391 0089434
3-569 234876349 153603-9
1-717 609299440 1625468
1-1286 844175789 316149-6
2-3291 2976492018 7948446
2-7869 4394734826 905837-8
1-11160 7371226845 7006819
1-19030 17659511672 606518-4
3-68251 26690741863 52023512
1-87281 44350253536 126753-3
5-504658 484417309544 1540002
8-4 124549 3188382 529889 358753-11
1-4 629207 8032552 839433 5127536
2-13 382964 9253488 208756 384259-1
17-232 139611 534171142 388292 0451566
2-477 662187 993690292 985340 474571-11
1-709 801799 527861435 373632 5197272
8-6156 076584 2165783775 974400 632387-3
5-31490 184720 61075119315 245635 68166212
1-37646 261304 82732923091 220036 314049-4
3-144428 968635 09273888588 905744 6238099
1-182075 229939 920067111680 125780 937858-8

As explained above, x = 2u and y = 2v, so:

X0 = 10
Y0 = -6

X0 = -10
Y0 = 6

X0 = 6582 595298 (10 digits)
Y0 = -4037 589688 (10 digits)

X0 = -6582 595298 (10 digits)
Y0 = 4037 589688 (10 digits)

X0 = 9 258415 606510 (13 digits)
Y0 = -5 678867 025506 (13 digits)

X0 = -9 258415 606510 (13 digits)
Y0 = 5 678867 025506 (13 digits)

X0 = 464 279223 068342 (15 digits)
Y0 = -284 776584 090312 (15 digits)

X0 = -464 279223 068342 (15 digits)
Y0 = 284 776584 090312 (15 digits)


Find recurrences among the solutions of the homogeneous equation

Now that some solutions of the original equation were found, we will find other solutions, in fact, a family of infinite solutions, where:

Xn+1 = P Xn + Q Yn
Yn+1 = R Xn + S Yn

where P, Q, R and S should be determined.

Let M(x, y) = Ax2 + Bxy + Cy2 = M and N(u, v) = u2 + Buv + ACv2 = N.

M(p, q) = Ap2 + Bpq + Cq2

M(p, q)/A = p2 + (B/A)pq + (C/A)q2

M(p, q)/A = p2 + Dpq + Eq2

M(p/q, 1)/A = (p/q)2 + D(p/q) + E    (12)

The roots of M(p/q, 1)/A = (p/q - J) (p/q - J') = 0  (13) are:

J =-B + sqrt(B2 - 4AC)
2A
   and   J' =-B - sqrt(B2 - 4AC)
2A

It can be easily shown by equating (12) and (13) that:

J2 = -DJ - E    (14)
J'2 = -DJ' - E    (15)
J + J' = -D    (16)
JJ' = E    (17)

The roots of N(p/q, 1) = (p/q - K) (p/q - K') = 0 are:

K =-B + sqrt(B2 - 4AC)
2
   and   K' =-B - sqrt(B2 - 4AC)
2

so K = AJ, K' = AJ'    (18)

M(p, q)/A = (p - Jq)(p - J'q) = M    (19)

N(r, s) = (r - Ks)(r - K's) = N    (20)

From (18) we obtain:

(p - Jq)(r - Ks) = (p - Jq)(r - AJs) = (pr - AJps - Jqr + AJ2qs)

From (14) we obtain:

[pr - AJps - Jqr + A(-DJ - E)qs] = (pr - AEqs) - (Aps + qr + AEqs)J = (pr - Cqs) - (Aps + qr + Bqs)J    (21)

From (18) we obtain:

(p - J'q)(r - K's) = (p - J'q)(r - AJ's) = (pr - AJ'ps - J'qr + AJ'2qs)

From (15) we obtain:

[pr - AJ'ps - J'qr + A(-DJ' - E)qs] = (pr - AEqs) - (Aps + qr + AEqs)J' = (pr - Cqs) - (Aps + qr + Bqs)J'    (22)

Let X = pr - Cqs and Y = Aps + qr + Bqs    (23).

Multiplying (21) by (22) we obtain:

(M(p, q)/A) N(r, s) = (X - YJ) (X - YJ') = X2 - (J + J')XY + JJ'Y2

Multiplying equations (16) and (17) we obtain: (M(p, q)/A) N(r, s) = X2 + DY + EY2

Multiplying by A we get (from (19) and (20)):

AX2 + BXY + CY2 = MN

Letting M = -F and N = 1 we can see that X and Y are also solutions of the original equation.

Let r and s be a solution to N(r, s) = r2 + Brs + ACs2 = 1,
Xn = p, Yn = q, Xn+1 = X and Yn+1 = Y (since the last two pairs of numbers are solutions to the original equation).

From (23) we obtain:

Xn+1 = rXn - CsYn+1
Yn+1 = AsXn + rYn+1 + BsYn+1

This means that:

Xn+1 = P Xn + Q Yn
Yn+1 = R Xn + S Yn
P = r    (24)
Q = -Cs    (25)
R = As    (26)
S = r + Bs    (27)
where
r2 + Brs + ACs2 = 1    (28)

Credits: This method was e-mailed to me by Iain Davidson. I've made some modifications.


Find solutions of the general quadratic equation

Ax2 + Bxy + Cy2 + Dx + Ey + F = 0

Multiplying the equation by 4A:

4A2x2 + 4ABxy + 4ACy2 + 4ADx + 4AEy + 4AF = 0
(2Ax + By + D)2 - (By + D)2 + 4ACy2 + 4AEy + 4AF = 0
(2Ax + By + D)2 + (4AC - B2)y2 + (4AE - 2BD)y + (4AF - D2) = 0

Let x1 = 2Ax + By + D
and g = gcd(4AC - B2, 2AE - BD).

Multiplying by (4AC - B2)/g:

4AC - B2
g
x12 +(4AC - B2)2
g
y2 +2(4AC - B2) (2AE - BD)
g
y +(4AC - B2) (4AF - D2)
g
= 0

4AC - B2
g
x12 + g y12 +(4AC - B2) (4AF - D2) - (2AE - BD)2
g
= 0

4AC - B2
g
x12 + g y12 +4A(4ACF - AE2 - B2F + BDE - CD2)
g
= 0

where:

y1 =4AC - B2
g
y + 2AE - BD
g


Find recurrences among the solutions of the general quadratic equation

We will assume that the solutions will have the form:

Xn+1 = P Xn + Q Yn + K
Yn+1 = R Xn + S Yn + L

Replacing in the original equation x by Px + Qy + K and y by Rx + Sy + L:

A(Px + Qy + K)2 + B(Px + Qy + K) (Rx + Sy + L) + C(Rx + Sy + L)2 + D(Px + Qy + K) + E(Rx + Sy + L) + F = 0

(AP2 + BPR + CR2)x2 + (2APQ + B(PS+QR) + 2CRS)xy + (AQ2 + BQS + CS2)y2 + (2ABP + B(KR+LP) + 2CLR + DP + ER)x + (2AKQ + B(KS+LQ) + 2CLS + DQ + ES)y + (AK2 + BKL + CL2 + DK + EL + F) = 0    (29)

Now we will investigate the values inside the parentheses.

This means that 2AKP + B(KR+LP) + 2CLR + DP + ER = D and 2AKQ + B(KS+LQ) + 2CLS + DQ + ES = E.

These two equations are equivalent to:

(2AP+BR)K + (BP+2CR)L = -D(P-1) - ER
and (2AQ+BS)K + (BQ+2CS)L = -DQ - E(S-1)

Solving the equation system for K and L:

K =D[BQ - 2C(PS-QR-S)] + E[B(PS-RQ-P) - 2CR]
4AC (PS - QR) + B2 (QR - PS)

L =D[B(PS-RQ-S) - 2AQ] + E[BR - 2A(PS-RQ-P)]
4AC (PS - QR) + B2 (QR - PS)

Since PS - QR = r(r+Bs) - (-Cs)As = r2 + Brs + ACs2 = 1, these equations can be simplified to:

K =D[BQ - 2C(1-S)] + E[B(1-P) - 2CR]
4AC - B2

L =D[B(1-S) - 2AQ] + E[BR - 2A(1-P)]
4AC - B2

Now we must show that the expression inside the right parentheses of (29) equals F. This means that we have to prove that the values of K and L just found verify the equation Z = AK2 + BKL + CL2 + DK + EL = 0    (33).

The expansion is very complicated and will not be reproduced here, but fortunately it is a multiple of 4AC-B2, so it cancels the square in the denominator, since it is (4AC-B2)2.

This means that Z(4AC-B2) is an integer number and it is equal to:

AD2Q2 - 2ADEPQ + AE2P2 - AE2 + BD2QS - BDEPS - BDEQR + BDE + BE2PR + CD2S2 - CD2 - 2CDERS + CE2R2

Reordering terms:

AD2Q2 + BD2QS + CD2S2 - CD2 - 2ADEPQ - BDEPS - BDEQR - 2CDERS + BDE + AE2P2 + BE2PR + CE2R2 - AE2

D2(AQ2 + BQS + CS2) - CD2 - DE(2APQ + BPS + BQR + 2CRS) + BDE + E2(AP2 + BPR + CR2) - AE2

From (30), (31) and (32):

Z(4AC-B2) = CD2 - CD2 - BDE + BDE + AE2 - AE2 = 0

This means that Z = 0, so (33) holds, then (29) holds too.

Let K =KDD + KEE
4AC - B2
   and   L =LDD + LEE
4AC - B2
    (34)

To continue simplifying the expressions we should note the following:

KD = BQ - 2C(1 - S)
KD = B(-Cs) - 2C(1 - r - Bs)
KD = -BCs - 2C + 2Cr + 2BCs
KD = C(-2 + 2r + Bs)    (35)
KD = C(P + S - 2)

LE = BR - 2A(1 - P)
LE = ABs - 2A + 2Ar
LE = A(-2 + 2r + Bs)
LE = A(P + S - 2)

KE = B(1 - P) - 2CR
KE = B(1 - r) - 2ACs
KE = B - Br - 2ACs    (36)

LD = B(1 - S) - 2AQ
LD = B(1 - r - Bs) + 2ACs
LD = B - Br - B2s + 2ACs

LD - KE = (4AC - B2)s    (37)

So:

K =CD(P+S-2) + E(B-Br-2ACs)
4AC - B2

L =D(B-Br-2ACs) + AE(P+S-2)
4AC - B2
+ Ds

Generally the numerators will not be multiple of 4AC - B2, so using this formula we cannot find a recurrence for all values of D and E.

For some values of D and E there will be solutions, as shown below. Using equations (24) - (27):

KDLE - KELD = 4ACr2 + 4ABCrs + 4A2C2s2 - B2r2 - B3rs - AB2Cs2 - 4ABCs - B3s + 4AC - B2 - 8ACr + 2B2r =

= (4AC - B2) (r2 + Brs + ACs2) - (4AC - B2)Bs + (4AC - B2) - (4AC - B2)2r =

= (4AC - B2) (2 - 2r - Bs)

The equal signs shown below mean congruence mod 4AC - B2.

KDLE - KELD = 0 => KD/KE = LD/LE    (38)

Since K and L must be integers they should be (from (34)):

KDD + KEE = 0 => E = (-KD/KE)D    (39)

LDD + LEE = 0 => E = (-LD/LE)D

These equations are consistent because of equation (38).

In some cases (see example 6) we can find a recurrence by using the solutions -r and -s since (-r)2 + B(-r)(-s) + AC(-s)2 = r2 + Brs + ACs2 = 1.

If no solutions were found (as in example 7), we should use the next pair of solutions (r1, s1) of r2 + Brs + ACs2 = 1 because there will always be solutions as shown below.

First we should find r1 and s1 from r and s. To do that we use the formulas (24) - (28).

r1 = r r + (-ACs)s = r2 - ACs2
s1 = s r + (r + Bs)s = 2rs + Bs2

Now the values of r and s should be replaced by r1 and s1.

From (24): P1 = r1 = r2 - ACs2
From (25): Q1 = -Cs1 = -C(2rs + Bs2)
From (26): R1 = As1 = A(2rs + Bs2)
From (27): S1 = r1 + Bs1 = r2 + 2Brs + (B2 - AC)s2

From (35):

K1D = C(-2 + 2r1 + Bs1)
K1D = C[-2 + 2(r2 - ACs2) + B(2rs + Bs2)]
K1D = C[-2 + 2(r2 + Brs + ACs2) - 4ACs2 + B2s2]
K1D = C[-2 + 2 + (B2 - 4AC)s2]
K1D = (B2 - 4AC)Cs2

From (36):

K1E = B - Br1 - 2ACs1
K1E = B - Br2 + ABCr2 - 4ACrs - 2ABCr2
K1E = B - B(r2 + ACs2) - 4ACrs
K1E = B - B(r2 + ACs2 + Brs - Brs) - 4ACrs
K1E = B - B(1 - Brs) - 4ACrs
K1E = (B2 - 4AC)rs

From (37):

L1D - K1E = (4AC - B2)s1
L1D = (B2 - 4AC) (rs - 2rs - Bs2)
L1D = (B2 - 4AC) (-rs - Bs2)

Therefore:

K1 =K1DD + K1EE
4AC - B2
= -CDs2 - Ers

L1 =L1DD + L1EE
4AC - B2
= Ds(r + Bs) - AEs2

So, finally:

Xn+1 = (r2 - ACs2)Xn - Cs(2r+Bs)Yn - CDs2 - Ers    (40)

Yn+1 = As(2r+Bs)Xn + [r2 + 2Brs + (B2-AC)s2]Yn + Ds(r+Bs) - AEs2    (41)

Notice that in this case, in order to find the solutions using the continued fraction method, we will need to compute two entire periods if the period length is even and four if it is odd.

Example 6: 3x2 + 13xy + 5y2 + Dx + Ey + F = 0

The first solution of r2 + Brs + ACs2 = r2 + 13 rs + 15 s2 = 1 using the continued fraction method is r = -8351 and s = 6525.

P = r = -8351
Q = -Cs = -32625
R = As = 19575
S = r + Bs = 76474

K =CD(P+S-2) + E(B-Br-2ACs)
4AC-B2
=-340605
109
D +87174
109
E

L =D(B-Br-2ACs) + AE(P+S-2)
4AC - B2
+ Ds =798399
109
D -204363
109
E

The numerator of K (or L) is not a multiple of the denominator (4AC - B2 = -109), so there is no recurrence with the values of P, Q, R, S shown above, except for special cases (according to (39), when E = 93 D (mod 109)).

Using the solution r = 8351, s = -6525 we get:

P = r = 8351
Q = -Cs = 32625
R = As = -19575
S = r + Bs = -76474

K =CD(P+S-2) + E(B-Br-2ACs)
4AC-B2
=3125 D - 800 E

L =D(B-Br-2ACs) + AE(P+S-2)
4AC - B2
+ Ds = -7325 D + 1875 E

So, the recursive relation between solutions is:

Xn+1 = 8351 Xn - 32625 Yn + (3125 D - 800 E)

Yn+1 = -19575 Xn - 76474 Yn + (-7325 D + 1875 E)

Check: Knowing that x = 2, y = 3 is a solution of 3x2 + 13xy + 5y2 - 11x - 7y - 92 = 0, find other two solutions.

Replacing D = -11 and E = -7 in the previous equations:

Xn+1 = 8351 Xn - 32625 Yn - 28775
Yn+1 = -19575 Xn + 76474 Yn + 67450

So, replacing here X0 = 2 and Y0 = 3, we find X1 = 85802 and Y1 = -201122.

and replacing X1 = 85802 and Y1 = -201122, we find X2 = -5845 101523 and Y2 = 13701 097128.

Replacing these values in the original equation we can check that these values are correct.

Example 7: 3x2 + 14xy + 6y2 + Dx + Ey + F = 0

The first solution of r2 + Brs + ACs2 = r2 + 14 rs + 18 s2 = 1 using the continued fraction method is r = -391 and s = 273.

P = r = -391
Q = -Cs = -1638
R = As = 819
S = r + Bs = 3431

K =CD(P+S-2) + E(B-Br-2ACs)
4AC-B2
=-147 D + 35 E

L =D(B-Br-2ACs) + AE(P+S-2)
4AC - B2
+ Ds = 308 D -147
2
E

The numerator of L is not a multiple of the denominator (4AC - B2 = -124), so there is no recurrence with the values of P, Q, R, S shown above, except for special cases (when E is even).

Using the solution r = 391, s = -273 we get:

P = r = 391
Q = -Cs = 1638
R = As = -819
S = r + Bs = -3431

K =CD(P+S-2) + E(B-Br-2ACs)
4AC-B2
=-4563
31
D -1092
31
E

L =D(B-Br-2ACs) + AE(P+S-2)
4AC - B2
+ Ds =4563
62
D -9555
31
E

The numerator of K (or L) is not a multiple of the denominator (4AC - B2 = -124), so there is no recurrence with the values of P, Q, R, S shown above, except for special cases.

Using (40) and (41):

P1 = r2 - ACs2 = -1188641
Q1 = -Cs(2r+Bs) = -4979520
R1 = As(2r+Bs) = 2489760
S1 = r2 + 2Brs + (B2-AC)s2 = 10430239
K1 = -CDs2 - Ers = -106743 D + 447174 E
L1 = Ds(r+Bs) - AEs2 = 936663 D - 223587 E

So, the recursive relation between solutions is:

Xn+1 = -1188641 Xn - 4979520 Yn + (106743 D - 447174 E)

Yn+1 = 2489760 Xn + 10430239 Yn + (936663 D - 223587 E)

Check: Knowing that x = 4, y = 7 is a solution of 3x2 + 14xy + 6y2 - 17x - 23y - 505 = 0, find other two solutions.

Replacing D = -17 and E = -23 in the previous equations:

Xn+1 = -1188641 Xn - 4979520 Yn + 5146869

Yn+1 = 2489760 Xn + 10430239 Yn - 10780770

So, replacing here X0 = 4 and Y0 = 7, we find X1 = -34 464335 and Y1 = 72 189943.

and replacing X1 = -34 464335 and Y1 = 72 189943, we find X2 = -318 505538 201756 and Y2 = 667 150425 396007.

Replacing these values in the original equation we can check that these values are correct.


Programs that use these methods

I've written two programs that use these methods:

Press here to run a JavaScript program compatible with Internet Explorer 3.0 or Netscape Navigator 3.0 or later (not finished yet).

Press here to run a Java program compatible with Internet Explorer 4.0 or Netscape Navigator 4.0 or later. It is about 50 times faster than the previous program.

Both programs run in two modes: solution only and step-by-step. In the last mode, the program explains how the results are found.

Click here to send me your comments.

Nedstat Counter Last modification: December 17th, 2000.