The generalized Markoff solution-count conjecture over finite fields

From papers

Let a11,a22,a33,a12,a13,a23,da_{11},a_{22},a_{33},a_{12},a_{13},a_{23},d be integers, and consider the generalized Markoff equation

a11x2+a22y2+a33z2+a12xy+a13xz+a23yz=dxyz.a_{11}x^2+a_{22}y^2+a_{33}z^2+a_{12}xy+a_{13}xz+a_{23}yz=dxyz.

Let DD and A11,A22,A33A_{11},A_{22},A_{33} denote the quantities defined earlier in the source. Suppose that pp is a prime with pa11a22a33dp\nmid a_{11}a_{22}a_{33}d, let rr be a positive integer, and set q=prq=p^r. Generalized Markoff solution-count conjecture. The number of solutions (x,y,z)Fq3(x,y,z)\in\mathbb{F}_q^3 is as follows: (1) if pDp\nmid D, then p>2p>2 and the number is

q2+((A11q)K+(A22q)K+(A33q)K)q+1;q^2+\left(\genfrac{(}{)}{}{}{ -A_{11}}{q}_K+\genfrac{(}{)}{}{}{ -A_{22}}{q}_K+\genfrac{(}{)}{}{}{ -A_{33}}{q}_K\right)q+1;

(2) if 2<pD2<p\mid D and at least two of A11,A22,A33A_{11},A_{22},A_{33} are divisible by pp, then the number is q2+1q^2+1; (3) if 2<pD2<p\mid D and at most one of A11,A22,A33A_{11},A_{22},A_{33} is divisible by pp, then the number is

q2+((A11q)K+(A22q)K+(A33q)K(Ajjq)K)q+1,q^2+\left(\genfrac{(}{)}{}{}{ -A_{11}}{q}_K+\genfrac{(}{)}{}{}{ -A_{22}}{q}_K+\genfrac{(}{)}{}{}{ -A_{33}}{q}_K-\genfrac{(}{)}{}{}{ -A_{jj}}{q}_K\right)q+1,

where 1j31\leq j\leq3 is any index such that pAjjp\nmid A_{jj}; and (4) if 2=pD2=p\mid D, then the number is

{q2+1if a12a13a230mod2,q2+(1)r2q+1if a12a13a231mod2,q2+(1)rq+1otherwise.\begin{cases} q^2+1 & \text{if }a_{12}\equiv a_{13}\equiv a_{23}\equiv0\bmod 2,\\ q^2+(-1)^r2q+1 & \text{if }a_{12}\equiv a_{13}\equiv a_{23}\equiv1\bmod 2,\\ q^2+(-1)^rq+1 & \text{otherwise.}\end{cases}

The conjecture extends the paper's solution-count theorem from Fp\mathbb{F}_p to Fpr\mathbb{F}_{p^r}. The source motivates it by numerical experiments and notes that fixed equations can be handled directly by Baragar's method, but gives no proof of the general assertion.

Progress summary

Open

The known result covers prime-sized finite fields, while the proposed extension to all prime powers remains unproved.

Xiao-Jie Zhu formulated this conjecture as an extension of his solution-count theorem from prime fields to finite fields of prime-power size. The conjecture gives explicit formulas in the nonsingular, degenerate odd-characteristic, and characteristic-two cases.

Known results

  • Zhu, 2025: proved the corresponding solution-count formulas over Fp\mathbb{F}_p.
  • For a fixed equation, Zhu notes that Baragar's method can determine the count directly, but this does not establish the uniform prime-power assertion.

December 2025 conjectural extension

Zhu proposed the formulas over Fpr\mathbb{F}_{p^r}, motivated by numerical experiments, but supplied no proof of the general statement. No independently verified proof or counterexample was found in the retrieved public record.

Current status (as of August 2026): The prime-field case is settled, but the generalized solution-count conjecture for r>1r>1 remains open.

Sources
Sources & referencesView supporting material

Primary source

Xiao-Jie Zhu, “Quadratic Gauss sums over Z^n/cZ^n: explicit formulas, a duality theorem, and applications to Weil representations and cubic hypersurfaces over F_p”, arXiv:2512.15334 (2025).

Solutions 1

Proof

Full proof of the generalized Markoff solution-count conjecture

Let

Q(X,Y,Z)=a11X2+a22Y2+a33Z2+a12XY+a13XZ+a23YZ,Q(X,Y,Z) =a_{11}X^2+a_{22}Y^2+a_{33}Z^2 +a_{12}XY+a_{13}XZ+a_{23}YZ,

and suppose that pp is prime, q=prq=p^r with r1r\geq1, and pa11a22a33dp\nmid a_{11}a_{22}a_{33}d. Define

G=(2a11a12a13a122a22a23a13a232a33),D=detG,A11=4a22a33a232,A22=4a11a33a132,A33=4a11a22a122.\begin{aligned} G&= \begin{pmatrix} 2a_{11}&a_{12}&a_{13}\\ a_{12}&2a_{22}&a_{23}\\ a_{13}&a_{23}&2a_{33} \end{pmatrix}, &D&=\det G,\\ A_{11}&=4a_{22}a_{33}-a_{23}^{2}, &A_{22}&=4a_{11}a_{33}-a_{13}^{2}, &A_{33}&=4a_{11}a_{22}-a_{12}^{2}. \end{aligned}

We prove all four cases of Xiao-Jie Zhu's Conjecture 10.1 for every prime power qq. Theorem 1.6 of that source proves the prime-field case r=1r=1; the argument below supplies the conjectured extension to every rr.

1. Projection from the singular origin

Write

C={[X:Y:Z]P2(Fq):Q(X,Y,Z)=0},B={[X:Y:Z]C:XYZ=0},Nq=#{(x,y,z)Fq3:Q(x,y,z)=dxyz}.\begin{aligned} \mathcal C &=\bigl\{[X:Y:Z]\in\mathbb P^2(\mathbb F_q):Q(X,Y,Z)=0\bigr\},\\ \mathcal B &=\bigl\{[X:Y:Z]\in\mathcal C:XYZ=0\bigr\},\\ N_q &=\#\bigl\{(x,y,z)\in\mathbb F_q^3:Q(x,y,z)=dxyz\bigr\}. \end{aligned}

Every nonzero affine point lies on a unique line through the origin with projective direction P=[X:Y:Z]P=[X:Y:Z]. Its nonzero points have the form t(X,Y,Z)t(X,Y,Z), with tFq×t\in\mathbb F_q^\times, and satisfy

t2Q(X,Y,Z)=dt3XYZ.t^2Q(X,Y,Z)=dt^3XYZ.

If XYZ0XYZ\neq0, this line contributes exactly one nonzero solution when Q(P)0Q(P)\neq0, and contributes none when Q(P)=0Q(P)=0. There are (q1)2(q-1)^2 projective directions with XYZ0XYZ\neq0, of which #C#B\#\mathcal C-\#\mathcal B satisfy Q(P)=0Q(P)=0.

If XYZ=0XYZ=0, the line contributes all q1q-1 nonzero points precisely when PBP\in\mathcal B, and none otherwise. Including the origin gives the characteristic-independent identity

Nq=1+(q1)2#C+q#B.(1)\boxed{ N_q=1+(q-1)^2-\#\mathcal C+q\,\#\mathcal B. } \tag{1}

Thus the entire problem reduces to counting points on a projective conic and on its three coordinate sections.

2. Odd characteristic

Suppose that p>2p>2, and let χq\chi_q be the quadratic character of Fq\mathbb F_q, extended by χq(0)=0\chi_q(0)=0. Set

χi=χq(Aii),S=χ1+χ2+χ3.\chi_i=\chi_q(-A_{ii}), \qquad S=\chi_1+\chi_2+\chi_3.

None of the three coordinate vertices lies on C\mathcal C, since each aiia_{ii} is nonzero. The coordinate sides therefore contribute disjointly to B\mathcal B. For example, on Z=0Z=0 the defining equation is

a11X2+a12XY+a22Y2=0,a_{11}X^2+a_{12}XY+a_{22}Y^2=0,

which has exactly 1+χq(a1224a11a22)=1+χ31+\chi_q(a_{12}^2-4a_{11}a_{22})=1+\chi_3 projective roots. The other two sides give the analogous expressions, so

#B=3+S.(2)\boxed{\#\mathcal B=3+S.} \tag{2}

Nonsingular conic

If pDp\nmid D, the conic C\mathcal C is nonsingular. Every nonsingular projective plane conic over a finite field has exactly q+1q+1 rational points: diagonalization and the intersection of two subsets of Fq\mathbb F_q having (q+1)/2(q+1)/2 elements give a rational point, and projection from that point parametrizes the conic by P1(Fq)\mathbb P^1(\mathbb F_q). Hence

#C=q+1.\#\mathcal C=q+1.

Substitution into (1) and (2) gives

Nq=q2+q(χ1+χ2+χ3)+1.(3)\boxed{ N_q=q^2+q(\chi_1+\chi_2+\chi_3)+1. } \tag{3}

Moreover, pDp\nmid D automatically implies p>2p>2: modulo 22, the matrix GG is alternating of odd order and therefore singular.

Degenerate conic

Suppose now that pDp\mid D. Since all diagonal coefficients of QQ are nonzero, the rank of GG is either 11 or 22.

If rankG=1\operatorname{rank}G=1, then QQ is a nonzero scalar multiple of the square of a linear form. Consequently,

A11=A22=A33=0,#C=q+1,Nq=q2+1.(4)A_{11}=A_{22}=A_{33}=0, \qquad \#\mathcal C=q+1, \qquad N_q=q^2+1. \tag{4}

Suppose instead that rankG=2\operatorname{rank}G=2. Its adjugate has rank one and is symmetric, so

adj(G)=cvvT(5)\operatorname{adj}(G)=c\,vv^{\mathsf T} \tag{5}

for some cFq×c\in\mathbb F_q^\times and nonzero vFq3v\in\mathbb F_q^3. In particular, at least one diagonal cofactor Ajj=cvj2A_{jj}=cv_j^2 is nonzero, and every nonzero diagonal cofactor has the same quadratic character:

χq(Aii)=χq(c)whenever Aii0.(6)\chi_q(-A_{ii})=\chi_q(-c) \qquad\text{whenever }A_{ii}\neq0. \tag{6}

In fact at least two diagonal cofactors are nonzero. Otherwise vv would be supported on a single coordinate jj, while Gadj(G)=0G\operatorname{adj}(G)=0 implies Gv=0Gv=0; its jjth coordinate would give 2ajjvj=02a_{jj}v_j=0, contradicting p2ajjp\nmid 2a_{jj}.

Fix an index jj with Ajj0A_{jj}\neq0. The restriction of QQ to the coordinate line Xj=0X_j=0 is a nondegenerate binary quadratic form with discriminant Ajj-A_{jj}. Its projective zero set therefore consists of 1+χj1+\chi_j points. The radical of GG is spanned by vv and does not lie on this coordinate line, because vj0v_j\neq0. Accordingly, C\mathcal C is the union of the lines joining this radical point to the 1+χj1+\chi_j points of the binary section. Therefore

#C=1+q(1+χj)=q+1+qχj.(7)\#\mathcal C =1+q(1+\chi_j) =q+1+q\chi_j. \tag{7}

Equations (1), (2), and (7) yield

Nq=q2+q(χ1+χ2+χ3χj)+1.(8)\boxed{ N_q=q^2+q\left(\chi_1+\chi_2+\chi_3-\chi_j\right)+1. } \tag{8}

If at least two diagonal cofactors vanish modulo pp, the preceding observation excludes rank two. Thus GG has rank one, all three diagonal cofactors vanish, and (4) gives

Nq=q2+1.(9)\boxed{N_q=q^2+1.} \tag{9}

If at most one diagonal cofactor vanishes, (8) is exactly the third conjectured formula. Its value is independent of the admissible index jj by (6).

Finally, because all coefficients come from the prime subfield, for every integer bb one has

χq(b)=(bp)r=(bpr)K.(10)\chi_q(b) =\left(\frac{b}{p}\right)^{r} =\left(\frac{b}{p^r}\right)_{K}. \tag{10}

Thus (3), (9), and (8) agree exactly with the Kronecker-symbol expressions in the first three parts of Conjecture 10.1.

3. Characteristic two

Now let p=2p=2 and q=2rq=2^r. The source assumes integer coefficients with 2a11a22a33d2\nmid a_{11}a_{22}a_{33}d, so reduction to Fq\mathbb F_q gives

a11=a22=a33=d=1.a_{11}=a_{22}=a_{33}=d=1.

Write

e=a12mod2,f=a13mod2,g=a23mod2,k=e+f+g{0,1,2,3},ε=(1)r.e=a_{12}\bmod2, \qquad f=a_{13}\bmod2, \qquad g=a_{23}\bmod2, \qquad k=e+f+g\in\{0,1,2,3\}, \qquad \varepsilon=(-1)^r.

Here kk is the ordinary integer count of nonzero mixed coefficients. On a coordinate side, a zero mixed coefficient produces the equation

U2+V2=(U+V)2=0,U^2+V^2=(U+V)^2=0,

with one projective root. A nonzero mixed coefficient produces

U2+UV+V2=0.U^2+UV+V^2=0.

This has two projective roots precisely when Fq\mathbb F_q contains F4\mathbb F_4, equivalently when rr is even, and has no projective roots when rr is odd. Its number of roots is therefore 1+ε1+\varepsilon. Since the coordinate vertices remain disjoint from C\mathcal C, we obtain

#B=3+kε.(11)\boxed{\#\mathcal B=3+k\varepsilon.} \tag{11}

The conic itself has four elementary forms, up to permutation of coordinates.

If k=0k=0, then

Q(X,Y,Z)=(X+Y+Z)2,#C=q+1.(12)Q(X,Y,Z)=(X+Y+Z)^2, \qquad \#\mathcal C=q+1. \tag{12}

If k=1k=1, take the mixed term to be XYXY. Then

Q(X,Y,Z)=X2+Y2+Z2+XY.Q(X,Y,Z)=X^2+Y^2+Z^2+XY.

Its partial derivatives are YY, XX, and 00; their unique common projective zero is [0:0:1][0:0:1], which is not on the conic. Thus the conic is nonsingular. The rational point [1:0:1][1:0:1] and projection from that point give

#C=q+1.(13)\#\mathcal C=q+1. \tag{13}

If k=2k=2, take the mixed terms to be XYXY and XZXZ. With U=Y+ZU=Y+Z,

Q(X,Y,Z)=X2+XU+U2.(14)Q(X,Y,Z)=X^2+XU+U^2. \tag{14}

If k=3k=3, set U=X+YU=X+Y and V=X+ZV=X+Z. Then

Q(X,Y,Z)=U2+UV+V2.(15)Q(X,Y,Z)=U^2+UV+V^2. \tag{15}

In either (14) or (15), the conic is the union of two distinct rational lines when rr is even and has only its rational vertex when rr is odd. Consequently,

#C=q+1+qεfor k{2,3}.(16)\#\mathcal C=q+1+q\varepsilon \qquad\text{for }k\in\{2,3\}. \tag{16}

Combining (1), (11), and (12)–(16), we find

Nq={q2+1,k=0,q2+(1)rq+1,k{1,2},q2+(1)r2q+1,k=3.(17)\boxed{ N_q= \begin{cases} q^2+1,&k=0,\\ q^2+(-1)^r q+1,&k\in\{1,2\},\\ q^2+(-1)^r 2q+1,&k=3. \end{cases} } \tag{17}

These are precisely the three characteristic-two alternatives in part (4) of Conjecture 10.1. Together, (3), (9), (8), and (17) prove the complete conjecture over every finite field Fpr\mathbb F_{p^r}.

0 endorsements
Shivam Patel ·