betaFeedback

WorksMathematics

Counterexamples to the Mathieu Property: From Explicit Quotient Constructions to a General Classification

Abstract

In 1997, Olivier Mathieu conjectured that if every positive power of a function on a compact connected Lie group has average zero, then multiplying those powers by any fixed function should eventually preserve this property. Christopher D. Long disproved this conjecture in 2026 and classified the compact connected groups for which the property holds. This paper extends that classification to a much broader family of algebraic spaces, including nonspherical homogeneous spaces and disconnected groups. We prove a necessary and sufficient condition determining exactly when the Mathieu–Zhao property holds. The proof constructs new explicit counterexamples using the binary octahedral and icosahedral groups, covering cases not addressed by earlier classifications. These counterexamples remain valid under polynomial transformations and give rise to transcendental generating functions. The main result completes the classification of the Mathieu–Zhao property for complex homogeneous spaces whose acting groups and stabilizers have reductive identity components, extending the known results from compact groups and spherical spaces to this general setting.

The problem and the results

Let AA be a commutative complex algebra and let h:A→Ch:A\to\mathbb{C} be a linear functional with h(1)=1h(1)=1. A function P∈AP\in A is called pure if

h(Pm)=0(m≥1).h(P^m)=0 \qquad(m\geq1).

We call PP safe if, for each fixed Q∈AQ\in A, the moments h(QPm)h(QP^m) vanish for all sufficiently large mm. The cutoff is allowed to depend on PP and QQ. Thus ker⁡h\ker h is a Mathieu–Zhao subspace precisely when every pure function is safe. This definition is equivalent to the version using eventual vanishing of the pure moments: take a sufficiently large power of PP and then treat the finitely many residue classes of exponents separately.

For example, let A=C[t,t−1]A=\mathbb{C}[t,t^{-1}] and let h=CTh=\mathrm{CT} extract the constant term. The function P=tP=t is pure and safe: for Q=∑jqjtjQ=\sum_j q_jt^j, one has CT(Qtm)=q−m\mathrm{CT}(Qt^m)=q_{-m}, which is zero once mm exceeds the finite range of negative exponents of QQ. The question is whether this eventual vanishing persists beyond tori.

Mathieu’s Main Conjecture asks whether every pure KK-finite function on a compact connected Lie group KK is safe [11 pp. 264–265]. Here KK-finite means that the translates span a finite-dimensional vector space. Long gave an explicit counterexample on SU(2)\mathrm{SU}(2) [8 Theorem 4.2], then classified the compact connected group case: the Mathieu property holds exactly for tori [9 Theorem 1.1]. Regular functions on a complex reductive group restrict to KK-finite functions on a compact form. Thus the case H={1}H=\{1\} of the classification below is already covered by Long’s compact-group theorem.

We are interested in A=C[G/H]A = \mathbb{C}[G/H], where GG is a complex linear algebraic group, H⊆GH \subseteq G is a closed algebraic subgroup, and both identity components G0G^0 and H0H^0 are reductive. The algebraic Matsushima criterion gives affineness of G0/H0G^0/H^0 [3 Proposition 2.2]. Each component of G/HG/H is a finite quotient of a conjugate of this affine variety, so G/HG/H is affine too. Its normalized Reynolds functional is the unique GG-invariant functional with value one on the constant function one. When GG is connected, it is integration over a compact form, with total mass one.

For connected GG, the space G/HG/H is spherical if a Borel subgroup has an open orbit. For reductive HH in this setting, Long proves that the Mathieu property and equality of the pure set with the nullcone are both equivalent to [G,G]⊆H[G,G] \subseteq H [10 Theorem 1.1]. We extend the Mathieu-property classification to arbitrary reductive homogeneous spaces and disconnected acting groups.

Theorem 1.1. Let GG and HH be as above. Then

ker⁡h⊆C[G/H] is Mathieu–Zhao⟺[G0,G0]⊆H.\ker h \subseteq\mathbb{C}[G/H]\text{ is Mathieu--Zhao} \quad\Longleftrightarrow\quad[G^0,G^0] \subseteq H.

If the containment fails, there are fixed regular functions PP, QQ on G/HG/H such that PP is pure and h(QPm)≠0h(QP^m) \ne0 for infinitely many mm.

The important word in the second assertion is fixed. A different multiplier at each order would not disprove the Mathieu property. Our examples use one polynomial QQ, independent of mm.

A counterexample on SU⁡(2)\operatorname{SU}(2) does not automatically descend to SL⁡2/Γ\operatorname{SL}_2/\Gamma, with Γ⊆SU⁡(2)\Gamma\subseteq\operatorname{SU}(2) finite: both functions must be invariant under right multiplication by Γ\Gamma. Averaging PP enforces invariance, but powers of that average contain mixed products of translates of PP, whose moments are not controlled by h(Pm)=0h(P^m)=0. Our binary-form construction retains all-power vanishing while supplying the required right invariance. The exceptional quotients below have dimension three, while a Borel subgroup of SL⁡2\operatorname{SL}_2 has dimension two; they supply the nonspherical cases left by the rank-one reduction.

Here are the two exceptional examples explicitly. Write

M=(xyzw),xw−yz=1,M = \begin{pmatrix} x & y \\ z & w \end{pmatrix}, \qquad xw-yz=1,

and take one of the following binary forms:

f8(X,Y)=X8+14X4Y4+Y8,f12(X,Y)=XY(X10+11X5Y5−Y10).(1)\begin{aligned} f_8(X,Y) &= X^8 + 14X^4Y^4 + Y^8,\\ f_{12}(X,Y) &= XY(X^{10} + 11X^5Y^5 - Y^{10}). \tag*{(1)} \end{aligned}

The groups 2O2O and 2I2I are the binary octahedral and binary icosahedral subgroups of SU⁡(2)\operatorname{SU}(2). Their standard embeddings fix f8f_8 and f12f_{12}, respectively. An elementary verification is given in Appendix § 9.

For n=8n=8 or 1212, put

b=n2,a=b−1,U=fn(x,y),F=fn(z,w),R=fn(z+Fx,w+Fy)F,P=UaRb,Q=Fa.(2)\begin{aligned} b &= \frac{n}{2}, \qquad a=b-1, \qquad U=f_n(x,y), \qquad F=f_n(z,w),\\ R &= \frac{f_n(z+Fx,w+Fy)}{F}, \qquad P=U^aR^b, \qquad Q=F^a. \tag*{(2)} \end{aligned}

Although RR is written as a quotient, it is a polynomial with integer coefficients. We will prove this before using it.

The parameters used throughout the construction are collected below. The first row also gives the radial example in Section 5. Here c=b(n−1)c=b(n-1) and p=cm−ap=cm-a is the prime index for the exceptional examples; N(T)N(T) denotes the normalizer of the diagonal torus.

Binary formnnaabbccQuotient
X2Y2X^{2}Y^{2}4126SL2/N(T)\mathrm{SL}_{2}/N(T)
f8f_{8}83428SL2/2O\mathrm{SL}_{2}/2O
f12f_{12}125666SL2/2I\mathrm{SL}_{2}/2I

Table 1.

Theorem 1.2. The functions in (1.2) are regular invariants on SL2/2O\mathrm{SL}_{2}/2O when n=8n=8, and on SL2/2I\mathrm{SL}_{2}/2I when n=12n=12. In both cases

h(Pm)=0(m≥1).\mathrm{h}(P^{m})=0 \qquad(m\geq1).

For the same fixed QQ, one has

vp(h(QPm))=−1if n=8 and p=28m−3 is prime,vp(h(QPm))=−1if n=12 and p=66m−5 is prime.(3)\begin{aligned} v_{p}\bigl(\mathrm{h}(QP^{m})\bigr)&=-1 &&\text{if }n=8\text{ and }p=28m-3\text{ is prime},\\ v_{p}\bigl(\mathrm{h}(QP^{m})\bigr)&=-1 &&\text{if }n=12\text{ and }p=66m-5\text{ is prime}. \tag*{(3)} \end{aligned}

Consequently both exceptional quotients fail the Mathieu property.

Here vpv_{p} is the usual valuation on rational numbers, normalized by vp(p)=1v_{p}(p)=1. The moments in this theorem are rational. Dirichlet’s theorem [15 Theorem 18.1] gives infinitely many primes in the progressions 25 (mod 28) and 61 (mod 66), so (1.3) proves nonvanishing at arbitrarily large orders.

The proof has two independent parts. Torus weights and integration by parts give purity. For the marked moments, a congruence reduces Haar integration to one coefficient of a binary form modulo pp. Two explicit Klein identities give a map from the curve defined by the binary form to a nonsingular cubic curve. A differential identity and compatibility of the Cartier operator with pullback transfer the coefficient calculation to that cubic. There it reduces to a binomial coefficient that is nonzero modulo pp.

For the classification, equivariant pullback carries a bad pair to any homogeneous space mapping onto one of these examples. A simple factor of rank at least two instead produces a two-dimensional SL2\mathrm{SL}_{2}-summand in a representation, to which the radial example applies. In the remaining case, inactive higher-rank factors can be removed, leaving only rank-one factors. A proper stabilizer projection gives a map onto one of the three quotients in the table; otherwise a diagonal coupling again produces such a summand. If [G,G]⊆H[G,G]\subseteq H for connected GG, the quotient is a torus. For disconnected GG, a finite-trace argument separates the moment conditions on its torus components.

Sections 2–4 prove Theorem 1.2. Section 5 gives the radial example, and Sections 6–7 prove Theorem 1.1. Section 7 also distinguishes purity from the invariant-theoretic nullcone. The final section treats polynomial recompositions, ordinary Cauchy transforms and coefficient algebras with nilpotents. We use standard structural facts about reductive groups, the classical finite subgroups of SL2\mathrm{SL}_{2}, the two branching rules stated in Lemma 6.2, and the Duistermaat–van der Kallen theorem in its explicit form (6.2). All special moment identities used in the proof are derived here.

Haar integration and a polynomial construction

On SU(2)\mathrm{SU}(2) the second row is determined by the first:

z=−y‾,w=x‾,∣x∣2+∣y∣2=1.(4)z=-\overline{y},\qquad w=\overline{x},\qquad|x|^{2}+|y|^{2}=1. \tag*{(4)}

Normalized Haar measure is normalized surface measure on this unit three-sphere. The variables ∣x∣2|x|^{2}, arg⁡x\arg x and arg⁡y\arg y have product measure

dt dθ2π dϕ2π,0≤t≤1.\mathrm{d}t\,\frac{\mathrm{d}\theta}{2\pi}\,\frac{\mathrm{d}\phi}{2\pi},\qquad0\leq t\leq1.

One can see this directly by writing x=teiθx=\sqrt{t}e^{i\theta} and y=1−teiϕy=\sqrt{1-t}e^{i\phi} in the surface element. In particular, for nonnegative integers i,ki,k,

h(xiykzkwi)=(−1)k∫01ti(1−t)k dt=(−1)ki!k!(i+k+1)!.(5)\mathrm{h}(x^i y^k z^k w^i)=(-1)^k\int_0^1 t^i(1-t)^k\,\mathrm{d}t=(-1)^k\frac{i!k!}{(i+k+1)!}. \tag*{(5)}

Every monomial whose two phase exponents do not cancel has integral zero.

We shall also use left invariance. The compact diagonal subgroup gives weight +1+1 to x,yx,y and weight −1-1 to z,wz,w; every nonzero weight has zero integral. The complexified infinitesimal left action contains

D=z∂x+w∂y.D=z\partial_x+w\partial_y.

Thus h(DT)=0\mathrm{h}(DT)=0 for every regular polynomial TT, and the product rule gives

h(SDjT)=(−1)jh((DjS)T).(6)\mathrm{h}(SD^jT)=(-1)^j\mathrm{h}((D^jS)T). \tag*{(6)}

These identities involve polynomials on SL2\mathrm{SL}_2 only.

We first work with any homogeneous f∈Z[X,Y]f\in\mathbb{Z}[X,Y] of even degree n=2b≥4n=2b\ge4. Set a=b−1a=b-1 and define U,F,R,P,QU,F,R,P,Q as in (2), with ff in place of fnf_n. Taylor expansion in a new scalar tt gives

f(z+tx,w+ty)=∑k=0nHktk,H0=F.f(z+tx,w+ty)=\sum_{k=0}^{n}H_kt^k,\qquad H_0=F.

Substitution t=Ft=F shows

R=1+∑k=1nFk−1Hk∈Z[x,y,z,w].(7)R=1+\sum_{k=1}^{n}F^{k-1}H_k\in\mathbb{Z}[x,y,z,w]. \tag*{(7)}

If a subgroup fixes ff, its simultaneous right action on the two rows fixes all the polynomials in this formula. Hence the construction respects right invariants.

Proposition 2.1. For every such binary form ff, the polynomial PP is pure.

Proof. Fix m≥1m\ge1, and write A=amA=am and B=bmB=bm. Define

Hk(B)=[tk]f(z+tx,w+ty)B(0≤k≤nB).H_k^{(B)}=[t^k]f(z+tx,w+ty)^B\qquad(0\le k\le nB).

Expanding RBR^B gives

Pm=UA∑k=0nBFk−BHk(B).(8)P^m=U^A\sum_{k=0}^{nB}F^{k-B}H_k^{(B)}. \tag*{(8)}

Every summand is a polynomial, including those with k<Bk<B. Indeed, at most kk of the BB factors contributing to Hk(B)H_k^{(B)} can use a positive power of tt. The remaining factors each contribute FF, so FB−kF^{B-k} divides Hk(B)H_k^{(B)} over Z\mathbb{Z}.

The kkth summand has left weight

nA−(n−2)k=(n−2)(B−k).nA-(n-2)k=(n-2)(B-k).

Only k=Bk=B can contribute to its Haar integral. Homogeneity also gives

Hk(B)=DnB−k(UB)(nB−k)!.(9)H_k^{(B)}=\frac{D^{nB-k}(U^B)}{(nB-k)!}. \tag*{(9)}

To verify this identity, replace tt by t−1t^{-1} in f(x+tz,y+tw)Bf(x+tz,y+tw)^B and multiply by tnBt^{nB}. Consequently, with j=(n−1)Bj=(n-1)B,

h(Pm)=1j!h(UADj(UB)).\mathrm{h}(P^m)=\frac{1}{j!}\mathrm{h}\left(U^A D^j(U^B)\right).

The degree of UAU^A in the top row is nA=(n−2)B<jnA=(n-2)B<j. Equation (6) therefore makes the last expression zero.

Proposition 2.2. Put c=b(n−1)c=b(n-1). For m≥1m\ge1, set A=amA=am, B=bmB=bm and s=b(m−1)s=b(m-1). Then

h(QPm)=h(Tm),Tm=UAF[ts]f(z+tx,w+ty)B.(10)h(QP^m)=h(T_m),\qquad T_m=\frac{U^A}{F}[t^s]f(z+tx,w+ty)^B. \tag*{(10)}

The displayed TmT_m is an integer polynomial, homogeneous of degree

N=cm−b(11)N=cm-b \tag*{(11)}

in each row. In particular its Haar integral has denominator dividing (cm−a)!(cm-a)!.

Proof. Multiplication by Q=FaQ=F^a changes the weight in (2.5) to (n−2)(B−k−b)(n-2)(B-k-b). The zero weight is therefore k=s=B−bk=s=B-b, and its power of FF is −1-1. The same factor-counting argument shows that FbF^b divides the coefficient in (2.7), so division by FF is harmless. The top degree is nA+snA+s and the bottom degree is nB−s−nnB-s-n; both equal cm−bcm-b. Formula (2.2) gives the denominator bound.

A congruence for the marked moments

Lemma 3.1. If T∈Z[x,y,z,w]T\in\mathbb{Z}[x,y,z,w] is homogeneous of degree p−1p-1 in each row, where pp is an odd prime, then ph⁡(T)\operatorname{ph}(T) is pp-integral and

ph⁡(T)≡[Xp−1Yp−1]T(X,Y,X,Y)(modp).(12)\operatorname{ph}(T)\equiv[X^{p-1}Y^{p-1}]T(X,Y,X,Y)\pmod p. \tag*{(12)}

Proof. It suffices to consider xiyp−1−izkwp−1−kx^iy^{p-1-i}z^kw^{p-1-k}. Its integral vanishes unless i+k=p−1i+k=p-1. In that case (2.2) gives (−1)ki!k!/p!(-1)^ki!k!/p!. Wilson’s theorem implies

i!(p−1−i)!≡(−1)i+1(modp),i!(p-1-i)!\equiv(-1)^{i+1}\pmod p,

so multiplication by pp makes this integral congruent to one. The same condition i+k=p−1i+k=p-1 selects precisely the monomials on the right of (3.1).

Choose an index mm for which

p=cm−a is prime.(13)p=cm-a\text{ is prime}. \tag*{(13)}

Then N=p−1N=p-1 in Proposition 2.2. At two equal rows,

[ts]f(X+tX,Y+tY)B=(nBs)f(X,Y)B.[t^s]f(X+tX,Y+tY)^B=\binom{nB}{s}f(X,Y)^B.

It follows that

ph⁡(QPm)≡(nBs)Hp(modp),Hp=[Xp−1Yp−1]f(X,Y)(p−1)/b.(14)\operatorname{ph}(QP^m)\equiv\binom{nB}{s}\mathcal{H}_p\pmod p,\qquad \mathcal{H}_p=[X^{p-1}Y^{p-1}]f(X,Y)^{(p-1)/b}. \tag*{(14)}

We used A+B−1=(p−1)/bA+B-1=(p-1)/b here. Notice also that

nB=p+q,q=bm+a=s+n−1,0≤s≤q<p.nB=p+q,\qquad q=bm+a=s+n-1,\qquad0\le s\le q<p.

Taking the coefficient of tst^s in (1+t)p+q≡(1+tp)(1+t)q(1+t)^{p+q}\equiv(1+t^p)(1+t)^q proves

(nBs)≡(qs)≠0(modp).(15)\binom{nB}{s}\equiv\binom{q}{s}\ne0\pmod p. \tag*{(15)}

Thus the entire nonvanishing problem has become the calculation of one binary coefficient Hp\mathcal{H}_p.

The two binary coefficients

For each form in (1.1), define the companion GG by

G8(X,Y)=X2Y2(X4−Y4)2,G12(X,Y)=X20+Y20−228X15Y5+494X10Y10+228X5Y15.(16)\begin{aligned} G_{8}(X,Y)&=X^{2}Y^{2}(X^{4}-Y^{4})^{2},\\ G_{12}(X,Y)&=X^{20}+Y^{20}-228X^{15}Y^{5}+494X^{10}Y^{10}+228X^{5}Y^{15}. \tag*{(16)} \end{aligned}

Let v=deg⁡Gv=\deg G, so v=12v=12 or 2020, and put

J=fXGY−fYGXnv.J=\frac{f_XG_Y-f_YG_X}{nv}.

The polynomial identities we need are

J2=f3G/36−3G3,n=8,J2=12f5+G3/144,n=12.(17)\begin{aligned} J^{2}&=f^{3}G/36-3G^{3},\qquad n=8,\\ J^{2}&=12f^{5}+G^{3}/144,\qquad n=12. \tag*{(17)} \end{aligned}

These identities can be checked by substitution and multiplication; the definitions specify every coefficient. They are a convenient normalization of the classical Klein identities. Appendix C gives explicit verifications in scalar coordinates.

We use a coefficient-extraction rule for differentials in characteristic pp. Its short algebraic proof is included in Appendix B. If LL is a one-variable function field over Fp\mathbb{F}_p and tt is a separating variable, every differential has a unique expression

ω=∑i=0p−1aipti dt.\omega=\sum_{i=0}^{p-1}a_i^p t^i\,dt.

Here “separating” means that LL is finite and separable over Fp(t)\mathbb{F}_p(t). The rule C(ω)=ap−1 dtC(\omega)=a_{p-1}\,dt is independent of the chosen separating variable and commutes with separable field extensions. It is usually called the Cartier operator. In particular,

C(∑jcjtjYp dt)=1Y∑r≥0cpr+p−1tr dt(cj∈Fp).(18)C\left(\frac{\sum_j c_jt^j}{Y^p}\,dt\right)=\frac{1}{Y}\sum_{r\geq0}c_{pr+p-1}t^r\,dt\qquad(c_j\in\mathbb{F}_p). \tag*{(18)}

Work now over Fp\mathbb{F}_p, with p>5p>5 and p≡1(modb)p\equiv1\pmod b, and form the field L=Fp(t,Y)L=\mathbb{F}_p(t,Y) defined by

Yb=f(t,1).Y^b=f(t,1).

The polynomials f(t,1)f(t,1) have simple roots for these primes: for f8f_8 the relevant quadratic discriminant is 142−4=19214^2-4=192, and for f12f_{12} it is 112+4=12511^2+4=125. Together with p∤4⋅5p\nmid4\cdot5, these facts check squarefreeness directly. The polynomial Yb−f(t,1)Y^b-f(t,1) is Eisenstein at any simple root of ff after extending constants; hence it defines a field and tt is separating.

In LL put

g=G(t,1)Yv/2,j=J(t,1)Y(n+v−2)/2.g=\frac{G(t,1)}{Y^{v/2}},\qquad j=\frac{J(t,1)}{Y^{(n+v-2)/2}}.

Equation (4.2) gives

j2=g/36−3g3,n=8,j2=12+g3/144,n=12.(19)\begin{aligned} j^2&=g/36-3g^3,\qquad n=8,\\ j^2&=12+g^3/144,\qquad n=12. \tag*{(19)} \end{aligned}

These are nonsingular cubic equations for p>5p>5. More importantly, differentiation gives the exact identity

dgj=−vdtY.(20)\frac{dg}{j}=-v\frac{dt}{Y}. \tag*{(20)}

Indeed, Euler’s homogeneous formula yields J=f′G/n−fG′/vJ=f'G/n-fG'/v on the chart with second coordinate one, while Y′/Y=f′/(bf)Y'/Y=f'/(bf). Substituting these into the derivative of gg proves (20). Since p>5p>5 and v∈{12,20}v\in\{12,20\}, the scalar −v-v lies in Fp×\mathbb{F}_p^\times. Thus dg≠0dg\ne0. Over the perfect field Fp\mathbb{F}_p, this means that gg is separating: L/Fp(g)L/\mathbb{F}_p(g) is separable. The cubic function field E=Fp(g,j)E=\mathbb{F}_p(g,j) defined by (19) is an intermediate field, so the inclusion E⊆LE\subseteq L is separable as well.

Homogeneity identifies the binary coefficient with the affine one:

Hp=[tp−1]f(t,1)(p−1)/b.\mathcal{H}_p=[t^{p-1}]f(t,1)^{(p-1)/b}.

because f(X,Y)(p−1)/bf(X,Y)^{(p-1)/b} has total degree 2(p−1)2(p-1). Since Yp−1=f(t,1)(p−1)/bY^{p-1}=f(t,1)^{(p-1)/b}, the extraction rule gives

C(dtY)=HpdtY.\mathcal{C}\left(\frac{dt}{Y}\right)=\mathcal{H}_p\frac{dt}{Y}.

The numerator has degree at most 2(p−1)2(p-1), so only the coefficient of tp−1t^{p-1} survives. If the cubic on the right of (19) is denoted by h(g)h(g), the same calculation gives

C(dgj)=[gp−1]h(g)(p−1)/2dgj.\mathcal{C}\left(\frac{dg}{j}\right)=[g^{p-1}]h(g)^{(p-1)/2}\frac{dg}{j}.

Applying C\mathcal{C} to (20) and using separable pullback compares these two eigenvalues. Cartier is Fp\mathbb{F}_p-linear, so the common nonzero factor −v∈Fp×-v\in\mathbb{F}_p^\times cancels. The two coefficients are therefore equal. They are

n=8:Hp=((p−1)/2(p−1)/4)(−1/12)(p−1)/4,n=12:Hp=((p−1)/2(p−1)/3)12(p−1)/6144−(p−1)/3.(21)\begin{aligned} n=8:\quad\mathcal{H}_p&=\binom{(p-1)/2}{(p-1)/4}(-1/12)^{(p-1)/4},\\ n=12:\quad\mathcal{H}_p&=\binom{(p-1)/2}{(p-1)/3}12^{(p-1)/6}144^{-(p-1)/3}. \tag*{(21)} \end{aligned}

Every factor is nonzero in Fp\mathbb{F}_p. For primes in (13), the required congruence p≡1(modb)p\equiv1\pmod b is automatic. Equations (14) and (15) now show that ph⁡(QPm)\operatorname{ph}(Q P^m) is a pp-adic unit. This proves (3). Proposition 2.1, Appendix A and Dirichlet’s theorem complete the proof of Theorem 1.2.

A radial witness from the same construction

Long’s compact-group argument already supplies a radial counterexample for invariant measures on C2\mathbb{C}^2 [9 Theorem 4.1]. We give an alternative pair and its closed marked-moment formula. Take f4(X,Y)=X2Y2f_4(X,Y)=X^2Y^2 in Section 2. The resulting pair has the particularly simple form

P4=x2y2(1+xzw2)4(1+yz2w)4,Q4=z2w2.(22)P_4=x^2y^2(1+xzw^2)^4(1+yz^2w)^4,\qquad Q_4=z^2w^2. \tag*{(22)}

Indeed, the intermediate polynomial RR is (1+xzw2)2(1+yz2w)2(1+xzw^2)^2(1+yz^2w)^2. Both functions in (22) are invariant under the right diagonal torus and under

(0−110).\begin{pmatrix}0&-1\\1&0\end{pmatrix}.

The latter transformation interchanges the two parenthesized factors. Thus they descend to SL2/N(T)\mathrm{SL}_2/N(T), where N(T)N(T) is the normalizer of the diagonal torus.

The first moment illustrates the cancellation. On SU(2)\mathrm{SU}(2), put t=∣x∣2t=|x|^2 and u=x‾yu=\overline{x}y. Then

P4=(xy)2(1−tu)4(1+(1−t)u)4,Q4=u2.P_4=(xy)^2(1-tu)^4(1+(1-t)u)^4,\qquad Q_4=u^2.

Phase integration keeps the coefficient of u2u^2 for P4P_4 and the constant coefficient for Q4P4Q_4P_4. Consequently,

h(P4)=∫01t2(1−t)2(6−28t+28t2) dt=0,h(Q4P4)=∫01t2(1−t)2 dt=130.h(P_4) = \int_0^1 t^2(1-t)^2(6-28t+28t^2)\,\mathrm{d}t = 0,\qquad h(Q_4P_4) = \int_0^1 t^2(1-t)^2\,\mathrm{d}t = \frac{1}{30}.

The following calculation extends this distinction to every order.

Lemma 5.1. For every m≥1m \ge1,

h(P4m)=0,h(Q4P4m)=βm=2m3m+1(2m)!(4m)!(6m)!>0.(23)h(P_4^m)=0,\qquad h(Q_4P_4^m)=\beta_m=\frac{2m}{3m+1}\frac{(2m)!(4m)!}{(6m)!}>0. \tag*{(23)}

Proof. The phase calculation above also proves purity at every order, directly from the binomial theorem:

h(P4m)=1(6m+1)!∑r=02m(−1)r(4mr)(4m2m−r)(2m+r)!(4m−r)!=((4m)!)2(6m+1)!(2m)!∑r=02m(−1)r(2mr)=0.\begin{aligned} h(P_4^m)=\frac{1}{(6m+1)!}\sum_{r=0}^{2m}(-1)^r\binom{4m}{r}\binom{4m}{2m-r}(2m+r)!(4m-r)! \\ &=\frac{((4m)!)^2}{(6m+1)!(2m)!}\sum_{r=0}^{2m}(-1)^r\binom{2m}{r}=0. \end{aligned}

For the marked moment, Proposition 2.2 gives

Tm=(xy)2m(zw)−2[t2m−2](z+tx)4m(w+ty)4m.T_m=(xy)^{2m}(zw)^{-2}[t^{2m-2}](z+tx)^{4m}(w+ty)^{4m}.

Set L=2m−2L=2m-2. Expanding this coefficient and using (5) gives

βm=((4m)!)2(6m−1)!L!∑r=0L(−1)r(Lr)(4m−r)(4m−r−1)(2m+r+1)(2m+r+2).(24)\beta_m=\frac{((4m)!)^2}{(6m-1)!L!}\sum_{r=0}^{L}\frac{(-1)^r\binom{L}{r}}{(4m-r)(4m-r-1)(2m+r+1)(2m+r+2)}. \tag*{(24)}

There is no division by zwzw in the polynomial TmT_m; its displayed coefficient is divisible by (zw)4(zw)^4.

Let SmS_m denote the sum in (24). Use 1/(k(k+1))=∫01uk−1(1−u) du1/(k(k+1))=\int_0^1 u^{k-1}(1-u)\,\mathrm{d}u for each pair of consecutive denominator factors and sum the binomial expansion. This gives

Sm=∫01∫01s2mt2m(1−s)(1−t)(t−s)L ds dt>0.S_m=\int_0^1\int_0^1 s^{2m}t^{2m}(1-s)(1-t)(t-s)^L\,\mathrm{d}s\,\mathrm{d}t>0.

The positivity is immediate because L=2m−2L=2m-2 is even. For its value, use symmetry to integrate over 0<s<t<10<s<t<1 and put s=rts=rt. With

Im=∫01r2m(1−r)L dr=(2m)!L!(4m−1)!,I_m=\int_0^1 r^{2m}(1-r)^L\,\mathrm{d}r=\frac{(2m)!L!}{(4m-1)!},

the tt-integral and then the two rr-integrals give

Sm=2Im(16m(6m+1)−2m+14m(6m+1)(6m+2))=Im6m(6m+2).S_m=2I_m\left(\frac{1}{6m(6m+1)}-\frac{2m+1}{4m(6m+1)(6m+2)}\right)=\frac{I_m}{6m(6m+2)}.

Substitution into (24) gives (23).

The same pair works for much more than surface measure. For v=(v0,v1)∈C2v=(v_0,v_1)\in\mathbb{C}^2, define

M(v)=(v0−v1‾v1v0‾),P(v)=P4(M(v)),Q(v)=Q4(M(v)).M(v)=\begin{pmatrix}v_0&-\overline{v_1}\\v_1&\overline{v_0}\end{pmatrix},\qquad P(v)=P_4(M(v)),\qquad Q(v)=Q_4(M(v)).

These are polynomials in the real and imaginary coordinates of vv. They are unchanged by v↦eiθvv\mapsto e^{i\theta}v, because (22) is right-torus invariant.

Proposition 5.2 (Radial identity). For every compactly supported finite positive SU(2)\mathrm{SU}(2)-invariant measure μ\mu on C2\mathbb{C}^{2} and every m≥1m \ge1,

∫Pm dμ=0,∫QPm dμ=βm∫∥v∥12m−4 dμ.(25)\int\mathcal{P}^{m}\,\mathrm{d}\mu=0,\qquad\int\mathcal{Q}\mathcal{P}^{m}\,\mathrm{d}\mu=\beta_m\int\lVert v\rVert^{12m-4}\,\mathrm{d}\mu. \tag*{(25)}

The second integral is positive if μ\mu is not concentrated at zero.

Proof. The matrix identity M(kv)=kM(v)M(kv)=kM(v) holds for k∈SU(2)k\in\mathrm{SU}(2). For v≠0v\ne0, put ρ=∥v∥\rho=\lVert v\rVert and write M(v)=ρgM(v)=\rho g with g∈SU(2)g\in\mathrm{SU}(2). Using t=∣x∣2t=\lvert x\rvert^{2} and u=xy‾u=x\overline{y} for the unit-radius coordinates of gg, we have

P4(ρg)=ρ4(xy)2(1−ρ4tu)4(1+ρ4(1−t)u)4,Q4(ρg)=ρ4u2.P_{4}(\rho g)=\rho^{4}(xy)^{2}(1-\rho^{4}tu)^{4}(1+\rho^{4}(1-t)u)^{4},\qquad Q_{4}(\rho g)=\rho^{4}u^{2}.

Average the integrands first over kk. Phase integration selects order 2m2m for the pure moment and 2m−22m-2 for the marked moment. The averages are therefore ρ12mh(P4m)=0\rho^{12m}h(P_{4}^{m})=0 and ρ12m−4βm\rho^{12m-4}\beta_m, respectively. The formulas hold at v=0v=0 as well. Integrating these averaged identities against μ\mu proves the result.

Connected reductive groups

We now prove Theorem 1.1 for connected GG. We use compact forms as in [2 Section 3] and the central-isogeny and complete-reducibility facts for reductive groups [4 Propositions 14.2 and 14.10, Sections 24.1 and 24.4]. The root-doublet and subdirect-product arguments follow Long [10, Proposition 3.1 and Lemmas 4.1, 4.2, 6.2]. We give the reductions here and use Proposition 5.2 for the moment input.

We use two elementary facts about normalized Reynolds functionals. If q:X→Yq:X\to Y is a surjective equivariant morphism of homogeneous spaces, then

hX(q∗f)=hY(f).(26)h_X(q^*f)=h_Y(f). \tag*{(26)}

Both sides are normalized invariant functionals on C[Y]\mathbb{C}[Y], and such a functional is unique. The same observation applies when the acting group maps onto the acting group of YY. Also, a central isogeny G~→G\widetilde{G}\to G identifies G~/π−1H\widetilde{G}/\pi^{-1}H with G/HG/H and preserves normalized integration. Thus all witnesses constructed after this change are functions on the original homogeneous space.

We call a submodule a defining doublet if it is isomorphic to the standard two-dimensional SL2\mathrm{SL}_{2}-module.

Proposition 6.1 (A line and a doublet). Suppose an irreducible GG-module VV has a nonzero HH-stable line Cξ\mathbb{C}\xi, and some homomorphism SL2→G\mathrm{SL}_{2}\to G has a defining two-dimensional submodule W⊆VW\subseteq V. Then C[G/H]\mathbb{C}[G/H] contains fixed functions P,QP,Q with

h(Pm)=0,h(QPm)>0(m≥1).h(P^{m})=0,\qquad h(QP^{m})>0\qquad(m\ge1).

Proof. Choose a compact form LL of HH and a compact form KK of GG containing LL. This is possible also for disconnected HH; see [2 Theorems 3.7–3.8]. By conjugating the SL2\mathrm{SL}_{2} homomorphism and WW together, arrange that SU(2)\mathrm{SU}(2) maps into KK. Give VV a KK-invariant Hermitian form. Orthogonal projection ΠW\Pi_W commutes with SU(2)\mathrm{SU}(2), since both WW and W⊥W^{\perp} are stable. The map

Φ:K→W,Φ(k)=ΠW(kξ)\Phi:K\to W,\qquad\Phi(k)=\Pi_W(k\xi)

is SU(2)\mathrm{SU}(2)-equivariant. The pushforward of Haar measure is therefore a compact positive invariant measure on WW. It is not concentrated at zero: continuity and full support of Haar measure would otherwise give Kξ⊆W⊥K\xi\subseteq W^\perp, whereas the span of KξK\xi is all of the irreducible module VV.

Identify WW with C2\mathbb{C}^2 by an SU(2)\mathrm{SU}(2)-equivariant unitary isomorphism and apply Proposition 5.2 to this measure. The polynomials P=P∘ΦP=\mathcal{P}\circ\Phi and Q=Q∘ΦQ=\mathcal{Q}\circ\Phi are regular on GG: their holomorphic factors are matrix coefficients. In an orthonormal basis, ρ(k)‾=(ρ(k)−1)T\overline{\rho(k)}=(\rho(k)^{-1})^{\mathrm{T}} for k∈Kk\in K, so the conjugate factors extend as matrix coefficients of the dual representation. Moreover Φ(kl)=χ(l)Φ(k)\Phi(kl)=\chi(l)\Phi(k) for the character χ\chi of the HH-stable line. On LL that character is unitary. The phase invariance of P,QP,Q therefore makes P,QP,Q right LL-invariant. Zariski density of LL in HH makes their regular extensions right HH-invariant. Thus they belong to C[G/H]\mathbb{C}[G/H] and (25) proves the claim. □

Lemma 6.2. Every nontrivial finite-dimensional rational representation of a connected simple complex group of rank at least two contains a defining doublet for some root SL2\mathrm{SL}_2. Proof. For type A2A_2, an irreducible module has highest weight (a,b)(a,b), a,b≥0a,b\geq0. The GL3↓GL2\mathrm{GL}_3\downarrow\mathrm{GL}_2 interlacing rule says that the polynomial highest weight (a+b,b,0)(a+b,b,0) contains every pair satisfying

a+b≥μ1≥b≥μ2≥0.a+b\geq\mu_1\geq b\geq\mu_2\geq0.

If b>0b>0, use (μ1,μ2)=(b,b−1)(\mu_1,\mu_2)=(b,b-1); otherwise use (1,0)(1,0). In the latter case nontriviality gives a≥1a\geq1. The difference is one, so the root SL2\mathrm{SL}_2 has a defining summand.

For B2B_2, write a highest weight as (a,b)(a,b) with a≥b≥0a\geq b\geq0, both integral or both half-integral. The Spin⁡5↓Spin⁡4\operatorname{Spin}_5\downarrow\operatorname{Spin}_4 rule contains all (c,d)(c,d) of the same parity with

a≥c≥b≥∣d∣.a\geq c\geq b\geq\lvert d\rvert.

The two SL2\mathrm{SL}_2 highest weights are c−d,c+dc-d,c+d. Use (c,d)=(b,b−1)(c,d)=(b,b-1) when b>0b>0, and (1,0)(1,0) otherwise. The first choice is valid even when b=12b=\frac{1}{2}, since ∣b−1∣≤b\lvert b-1\rvert\leq b; the second uses a≥1a\geq1. These are the classical branching rules; see [12 Theorem 2.1 and equation (4.4)].

Every simple root system of rank at least two contains a root-generated subalgebra of type A2A_2 or B2B_2: take adjacent simple roots, using the long-root A2A_2 inside G2G_2 if necessary. A nontrivial representation of a simple Lie algebra is faithful, so its restriction to this subalgebra is nontrivial. A nontrivial irreducible constituent and the preceding branching arguments give the doublet. Root homomorphisms integrate it to the group, also after composing a simply connected cover. □

Call a normal simple factor active if it acts nontrivially on G/HG/H. Since G/HG/H is affine, an active factor acts nontrivially on some irreducible submodule U⊆C[G/H]U\subseteq\mathbb{C}[G/H]. Evaluation at eHeH is a nonzero HH-fixed vector of U∗U^*. It is nonzero because a GG-stable space of functions vanishing at eHeH would vanish at every point. Thus Lemma 6.2 and Proposition 6.1 handle every active factor of rank at least two.

Lemma 6.3. Let G=Z×SL2G=Z\times\mathrm{SL}_2, with ZZ a torus, and let HH be reductive. If every projection H→SL2H\to\mathrm{SL}_2 is surjective, then either [G,G]⊆H[G,G]\subseteq H or the hypothesis of Proposition 6.1 holds.

Proof. The projections of H∘H^\circ are surjective as well, since their images have finite index in the connected target. The center of Lie⁡(H∘)\operatorname{Lie}(H^\circ) maps to zero in each sl2\mathfrak{sl}_2. Hence its derived algebra projects onto every factor. Decompose that semisimple algebra into simple ideals. For each coordinate, exactly one ideal maps nontrivially: its image is an ideal of sl2\mathfrak{sl}_2, and two nonzero such images could not commute. That ideal maps isomorphically onto sl2\mathfrak{sl}_2. Every simple ideal appears in at least one coordinate, since it cannot map nontrivially to the torus.

Thus the coordinates partition into blocks on which the derived algebra is diagonal, up to inner automorphisms. After conjugating individual factors,

[H0,H0]=∏BΔBSL2.[H^{0},H^{0}]=\prod_{B}\Delta_{B}\mathrm{SL}_{2}.

The equality holds for groups because connected algebraic subgroups in characteristic zero are determined by their Lie algebras. The connected center of H0H^{0} lies in ZZ. If every block is a singleton, [G,G]⊆H[G,G]\subseteq H. Otherwise choose two indices i,ji,j in one block. On the irreducible module V=Ci2⊗Cj2V=\mathbb{C}_{i}^{2}\otimes\mathbb{C}_{j}^{2}, that block acts diagonally and the connected center acts trivially. Hence the entire fixed subspace is

VH0=C(ei,1⊗ej,2−ei,2⊗ej,1),V^{H^{0}}=\mathbb{C}(e_{i,1}\otimes e_{j,2}-e_{i,2}\otimes e_{j,1}),

where the bases correspond under the chosen diagonal identifications. This is the alternating summand in C2⊗C2=Sym⁡2C2⊕⋀2C2\mathbb{C}^{2}\otimes\mathbb{C}^{2}=\operatorname{Sym}^{2}\mathbb{C}^{2}\oplus\bigwedge^{2}\mathbb{C}^{2}. Normality of H0H^{0} in HH makes this whole fixed subspace, and thus its unique line, HH-stable. The iith factor has the defining submodule Ci2⊗Ce1\mathbb{C}_{i}^{2}\otimes\mathbb{C}e_{1} in VV.

Proof of Theorem 1.1 for connected GG. If [G,G]⊆H[G,G]\subseteq H, then G/HG/H is a torus and its normalized functional is constant term: HH is the inverse image of a closed subgroup of the torus G/[G,G]G/[G,G], hence is normal with torus quotient. We use the Duistermaat–van der Kallen theorem [5 pp. 222–224] in the following form:

CT⁡(fm)=0 (m≥1),f≠0⟹0∉conv⁡(Supp⁡f).(27)\operatorname{CT}(f^{m})=0\ (m\geq1),\quad f\ne0\quad\Longrightarrow\quad0\notin\operatorname{conv}(\operatorname{Supp}f). \tag*{(27)}

A strictly positive linear functional on the finite support of ff eventually dominates the weights of any fixed multiplier qq. Therefore CT⁡(qfm)=0\operatorname{CT}(qf^{m})=0 for all large mm.

Now assume [G,G]⊈H[G,G]\not\subseteq H. Pass to a finite central cover Z×∏Si→GZ\times\prod S_{i}\to G. If an active factor has rank at least two, the preceding argument supplies the required pair. Otherwise quotient out the higher-rank factors contained in the stabilizer. They are normal and act trivially, so this leaves an isomorphic homogeneous space with acting group Z×SL2rZ\times\mathrm{SL}_{2}^{r}.

Consider the image HiH_{i} of the stabilizer in one SL2\mathrm{SL}_{2} factor. This image is closed and reductive. If it is proper and positive-dimensional, its identity component is a torus; the full image normalizes that torus and so lies in a conjugate of N(T)N(T). If finite, the classical finite-subgroup classification [6 p. 215] puts it in one of the following groups:

N(T),2O,2I.N(T),\qquad2O,\qquad2I.

Here cyclic and binary dihedral groups lie in N(T)N(T), while 2T2T lies in 2O2O. In each case projection gives a genuine surjection of the original homogeneous space onto the corresponding quotient of SL2\mathrm{SL}_{2}. Pull back the pair from Lemma 5.1 or Theorem 1.2, using (6.1). Finally, if every projection is surjective, apply Lemma 6.3 and Proposition 6.1. The alternative [G,G]⊆H[G,G]\subseteq H is excluded. This proves the connected classification and its fixed-multiplier assertion.

Disconnected groups and moment nullcones

We first give the finite-trace fact needed for disconnected spaces. It does not rely on the signs of complex moments.

Lemma 7.1 (Finite-trace separation). Let f1,…,frf_{1},\ldots,f_{r} be Laurent polynomials in finitely many variables over a characteristic-zero field. If

∑i=1rCT⁡(fiN)=0(N≥1),\sum_{i=1}^{r}\operatorname{CT}(f_{i}^{N})=0\qquad(N\geq1),

then CT⁡(fim)=0\operatorname{CT}(f_i^m)=0 for every ii and m≥1m \ge1.

Proof. It is enough to consider a number field of coefficients. Indeed, if one selected moment were nonzero, specialize the finitely generated Q\mathbb{Q}-algebra of coefficients after inverting that moment. A closed point gives a number-field specialization preserving it and every assumed identity.

Fix m≥1m \ge1 and 1≤k≤r1 \le k \le r. There are arbitrarily large primes pp splitting completely in the coefficient field and avoiding its coefficient denominators. For completeness, this infinitude needs no density theorem. Take an integral primitive element of the Galois closure. Schur’s elementary argument gives infinitely many prime divisors of values of its minimal polynomial. Outside its discriminant and index, such a divisor gives a degree-one prime in the Galois field; all residue degrees are equal, so the prime splits completely. Schur’s argument itself is Euclid’s: after translating a nonconstant integer polynomial to have nonzero constant term aa, if all prime divisors of its nonzero integer values belonged to a finite set with product BB, the polynomial f(aBt)/af(aBt)/a would be integral, nonconstant and congruent to one modulo BB. A sufficiently large value has a prime divisor outside that set.

Choose such pp larger than the absolute value of every exponent coordinate in every fimf_i^m, and put N=m(1+p+⋯+pk−1)N=m(1+p+\cdots+p^{k-1}). In the residue field,

fiN=∏j=0k−1fi(zpj)m.f_i^N=\prod_{j=0}^{k-1}f_i(z^{p^j})^m.

An exponent equation α0+pα1+⋯+pk−1αk−1=0\alpha_0+p\alpha_1+\cdots+p^{k-1}\alpha_{k-1}=0 forces α0=0\alpha_0=0 by reduction modulo pp and the exponent bound. Repeating gives every αj=0\alpha_j=0. Hence, writing ai=CT⁡(fim)a_i=\operatorname{CT}(f_i^m),

CT⁡(fiN)≡aik(modp).\operatorname{CT}(f_i^N)\equiv a_i^k \pmod p.

The assumed trace identity makes ∑iaik\sum_i a_i^k divisible by infinitely many primes, so it is zero in the number field. We have proved this for k=1,…,rk=1,\ldots,r. Newton’s identities now give ∏i(T−ai)=Tr\prod_i(T-a_i)=T^r, so every ai=0a_i=0. This also contradicts the nonzero moment preserved in the initial specialization, if one existed.

Let K=G0K=G^0 and choose g1,…,grg_1,\ldots,g_r representing G/(KH)G/(KH). The connected components of XX are the affine, open-and-closed orbits

Xi=K/Hi,Hi=K∩giHgi−1.X_i=K/H_i,\qquad H_i=K\cap g_iHg_i^{-1}.

Consequently C[X]=∏iC[Xi]\mathbb{C}[X]=\prod_i\mathbb{C}[X_i]. Transitivity of GG on these components and normalization imply

h((fi))=1r∑ihi(fi).(28)h((f_i))=\frac{1}{r}\sum_i h_i(f_i). \tag*{(28)}

If [K,K]⊈H[K,K]\not\subseteq H, use the connected counterexample on K/(K∩H)K/(K\cap H) and extend both functions by zero. This is a regular operation on an open-and-closed component. Equation (28) preserves its pure moments and its nonzero marked orders, up to 1/r1/r.

If D=[K,K]⊆HD=[K,K]\subseteq H, then DD lies in every HiH_i, because it is normal in GG. Each component is a torus. Embed its character lattice Mi=X∗(K/Hi)M_i=X^*(K/H_i) into M=X∗(K/D)M=X^*(K/D). For a pure tuple (fi)(f_i), Lemma 7.1 gives purity on every component. Apply (6.2) and separate each component support. The maximum of the finitely many multiplier cutoffs proves safety in (7.1). This completes Theorem 1.1.

We can also distinguish the Mathieu property from the moment-nullcone statement. Let ZXZ_X be the pure set and NG\mathcal{N}_G the GG-nullcone: the functions whose orbit closure contains zero in a finite-dimensional rational submodule containing the function. Corollary 7.2. One has ZX=NGZ_X = \mathcal{N}_G if and only if D⊆HD \subseteq H and

θ:⨁iMi⟶M,(χi)⟼∑iχi(29)\theta:\bigoplus_i M_i \longrightarrow M,\qquad(\chi_i)\longmapsto\sum_i\chi_i \tag*{(29)}

is injective. The lattices in (29) are defined after imposing D⊆HD \subseteq H.

Proof. The GG-nullcone equals the KK-nullcone, since a GG-orbit closure is a finite union of translates of the KK-orbit closure, where K=G0K = G^0. For this connected group, the Hilbert–Mumford criterion gives a one-parameter subgroup with strictly positive weights on any unstable function [7 Introduction and Section 2]. These weights eventually dominate those of any fixed multiplier, so unstable functions are safe. The non-toral fixed bad pairs therefore lie outside the nullcone.

In the toral case, finite-trace separation and (6.2) give a separate positive functional on every component support. If θ\theta is injective, these extend to one functional on M⊗QM \otimes\mathbb{Q}, positive on the entire union of supports. A multiple lifts to a cocharacter of KK, since the central torus maps onto K/DK/D with finite kernel. Its inverse contracts the tuple to zero, with the translation action on functions. Conversely, a nonzero lattice relation ∑iχi=0\sum_i \chi_i = 0 gives a pure tuple with component χi\chi_i when χi≠0\chi_i \ne0 and zero otherwise. To exclude zero from its orbit closure, let I={i:χi≠0}I = \{i : \chi_i \ne0\} and let vi∈C[X]v_i \in\mathbb{C}[X] equal χi\chi_i on XiX_i and zero elsewhere. The tuple is v=∑i∈Iviv = \sum_{i\in I}v_i in the finite-dimensional rational KK-module W=⨁i∈ICviW = \bigoplus_{i\in I}\mathbb{C}v_i. If ziz_i are the coordinate functions on WW, then ∏i∈Izi∈C[W]K\prod_{i\in I}z_i \in\mathbb{C}[W]^K because ∑iχi=0\sum_i\chi_i = 0. This polynomial has value one at vv and zero at the origin. Thus vv is outside the nullcone. The product here is a polynomial on WW, rather than a product of component-supported functions in C[X]\mathbb{C}[X]. Finally, all nullcone elements are pure because the moments are invariant homogeneous polynomials vanishing at zero.

Recompositions and ordinary Cauchy transforms

Return to either exceptional pair P,QP,Q in Theorem 1.2, with Γ=2O\Gamma= 2O or 2I2I, respectively. Write

(c,a)=(28,3) or (66,5),bj=h(QPj).(c,a)=(28,3)\text{ or }(66,5),\qquad b_j=h(QP^j).

The facts needed in this section are

b0=0,den⁡(bj)∣(cj−a)! (j≥1),vp(bj)=−1 when p=cj−a is prime.(30)b_0=0,\qquad\operatorname{den}(b_j)\mid(cj-a)!\ (j\geq1),\qquad v_p(b_j)=-1\text{ when }p=cj-a\text{ is prime}. \tag*{(30)}

The first is a weight calculation, the second is Proposition 2.2, and the third is Theorem 1.2. Here den⁡(bj)\operatorname{den}(b_j) is the positive denominator of the reduced rational number bjb_j.

For regular S,VS,V on SL2/Γ\mathrm{SL}_2/\Gamma, evaluate the Haar integral on the compact quotient SU(2)/Γ\mathrm{SU}(2)/\Gamma. The ordinary marked Cauchy transform is the convergent germ

CVS(z)=h(Vz−S)=∑m≥0h(VSm)zm+1for ∣z∣>max⁡∣S∣.C^S_V(z)=h\left(\frac{V}{z-S}\right)=\sum_{m\geq0}\frac{h(VS^m)}{z^{m+1}}\qquad\text{for }|z|>\max|S|.

We use this ordinary series throughout this section.

Theorem 8.1. For each d≥1d\geq1, choose an integer ee satisfying

0≤e<d,gcd⁡(cd,ce−a)=1(31)0\leq e<d,\qquad\gcd(cd,ce-a)=1 \tag*{(31)}

and set Qd=QPeQ_d=QP^e. Such an ee always exists. For every complex polynomial HH of degree dd, the same multiplier QdQ_d satisfies

h(QdH(P)m)≠0for infinitely many m,CQdH(P)(z) is transcendental over C(z).h(Q_dH(P)^m)\ne0\quad\text{for infinitely many }m,\qquad C^{H(P)}_{Q_d}(z)\text{ is transcendental over }\mathbb{C}(z).

Furthermore

h(H(P)m)=H(0)m,C1H(P)(z)=1z−H(0).(32)h\left(H(P)^m\right)=H(0)^m,\qquad C_1^{H(P)}(z)=\frac{1}{z-H(0)}. \tag*{(32)}

In particular, every nonzero polynomial recomposition with H(0)=0H(0)=0 is pure and unsafe.

The multiplier depends on the degree, not on the coefficients of HH. For example, the smallest possible ee in (31) gives an unambiguous choice. We first record two elementary algebraic facts used to allow arbitrary complex coefficients.

Lemma 8.2. Let LL be a finitely generated field over Q\mathbb{Q}, and let S⊂L∗S\subset L^* be finite. For all but finitely many rational primes pp, there is a valuation ww on LL extending vpv_p, normalized by w(p)=1w(p)=1, for which every member of SS is a unit.

Proof. Choose a transcendence basis tt with LL finite over Q(t)\mathbb{Q}(t). On Q(t)\mathbb{Q}(t) use the pp-adic Gauss valuation: on a polynomial it is the minimum coefficient valuation, and on a quotient it is the difference. Choose monic equations over Q(t)\mathbb{Q}(t) for each α∈S\alpha\in S and its inverse. All coefficients of these finitely many equations are integral for all but finitely many Gauss valuations. Extend the valuation to LL; existence follows from the valuation-ring domination lemma [14 Lemma 10.50.2]. The monic equations force both w(α)≥0w(\alpha)\geq0 and w(α−1)≥0w(\alpha^{-1})\geq0, so w(α)=0w(\alpha)=0. Values in an algebraic extension are rational multiples of w(p)w(p): in an algebraic equation the minimum valuation must occur at least twice, which gives an integral multiple of the element’s value in the original value group. This justifies the normalization.

Lemma 8.3 (Finite-generation form of Eisenstein’s argument). If LL has characteristic zero and A(t)∈L[[t]]A(t)\in L[[t]] is algebraic over L(t)L(t), all coefficients of AA belong to one finitely generated Z\mathbb{Z}-subalgebra of LL.

Proof. Choose F(t,Y)∈L[t,Y]F(t,Y)\in L[t,Y] with F(t,A)=0F(t,A)=0 and FY(t,A)≠0F_Y(t,A)\ne0; characteristic zero guarantees such a relation. Put r=ord⁡tFY(t,A)r=\operatorname{ord}_t F_Y(t,A), choose N>rN>r, and write A=a+tNBA=a+t^N B, where a=A mod tNa=A\bmod t^N. Taylor expansion gives

G(t,Y)=t−N−rF(t,a+tNY)∈L[t,Y].G(t,Y)=t^{-N-r}F(t,a+t^N Y)\in L[t,Y].

For the constant term, divisibility follows by comparing with F(t,A)=0F(t,A)=0; all terms of order at least two have tt-order at least 2N>N+r2N>N+r. The coefficient of YY has a nonzero constant term uu, and all higher powers of YY have zero constant term. Thus G(0,Y)G(0,Y) is linear. Successive coefficient comparison in G(t,B)=0G(t,B)=0 uses only addition, multiplication and division by uu. It puts all coefficients of BB in the ring generated by the coefficients of GG and u−1u^{-1}. Adjoining the finitely many coefficients of aa proves the lemma. This is the part of Eisenstein’s theorem needed here; see [13] for its classical and more general forms.

Proof of Theorem 8.1. Since gcd⁡(c,a)=1\gcd(c,a)=1, the choice in (31) is possible. For every prime ℓ∣d\ell\mid d with ℓ∤c\ell\nmid c, choose e≠a/c(modℓ)e\ne a/c\pmod{\ell}. A prime dividing cc cannot divide aa, so imposes no condition. The Chinese remainder theorem combines the choices modulo rad⁡(d)\operatorname{rad}(d), and a representative lies below dd.

Write H(T)=∑j=0dhjTjH(T)=\sum_{j=0}^{d}h_jT^j, and let LL be the field generated by its coefficients over Q\mathbb{Q}. For a sufficiently large prime

p=c(dm+e)−a,(33)p=c(dm+e)-a, \tag*{(33)}

Lemma 8.2 gives a ww making every nonzero hjh_j a unit. Expansion of H(T)mH(T)^m yields

h(QdH(P)m)=hdmbdm+e+∑j<dm[Tj]H(T)mbj+e.h\left(Q_dH(P)^m\right)=h_d^m b_{dm+e}+\sum_{j<dm}[T^j]H(T)^m b_{j+e}.

The first summand has valuation −1-1 by (30). Every other summand is integral: its coefficients are integral, and c(j+e)−a<pc(j+e)-a<p makes its factorial denominator prime to pp. The term b0b_0 is zero. A unique term of negative valuation cannot cancel. The progression (33) is coprime by (31); Dirichlet’s theorem supplies infinitely many such primes after any finite exclusions. This proves the fixed-multiplier nonvanishing assertion.

Suppose the ordinary series were algebraic over C(z)\mathbb{C}(z). It would be algebraic over L(z)L(z) as well. To see this descent, a polynomial relation with fixed degree bounds is a linear system over LL in finitely many unknown coefficients. Extending scalars does not change its rank. Even though the system has infinitely many rows, its row space has a finite basis, so a nonzero complex solution gives a nonzero LL solution.

Apply Lemma 8.3 to A(t)=∑m≥0h(QdH(P)m)tmA(t)=\sum_{m\geq0}h(Q_dH(P)^m)t^m. There is a finite set SA⊂L∗S_A\subset L^* whose Z\mathbb{Z}-algebra contains every coefficient of AA. For a sufficiently large prime in (33), apply Lemma 8.2 simultaneously to SAS_A and the nonzero coefficients of HH. It makes every coefficient of AA integral, whereas the preceding expansion gives valuation −1-1 for its mmth coefficient. This contradiction proves transcendence. Finally, all positive powers of PP integrate to zero, so expanding H(P)mH(P)^m leaves only H(0)mH(0)^m. Summing its geometric series proves (32).

Thus the unit Cauchy transform is rational while the marked transform is transcendental for the specified fixed polynomial multiplier.

Coefficient algebras

There is also an exact statement with nilpotents retained. Let RR be any commutative unital C\mathbb{C}-algebra, and extend hh RR-linearly to hR:R⊗CC[SL2/Γ]→Rh_R:R\otimes_{\mathbb{C}}\mathbb{C}[\mathrm{SL}_2/\Gamma]\to R. An element is eventually pure if its positive moments vanish for all sufficiently large orders. Safety in the following statement is against every multiplier in this entire algebra.

Corollary 8.4. For H(T)=∑j=0dcjTj∈R[T]H(T)=\sum_{j=0}^{d}c_jT^j\in R[T],

H(P) is pure ⟺c0=0,H(P)\text{ is pure }\Longleftrightarrow c_0=0,
H(P) is eventually pure ⟺c0 is nilpotent,H(P)\text{ is eventually pure }\Longleftrightarrow c_0\text{ is nilpotent},
H(P) is safe ⟺ every cj is nilpotent ⟺H(P) is nilpotent.H(P)\text{ is safe }\Longleftrightarrow\text{ every }c_j\text{ is nilpotent }\Longleftrightarrow H(P)\text{ is nilpotent}.

Proof. The formula hR(H(P)m)=c0mh_R(H(P)^m)=c_0^m proves the first two assertions. If every coefficient is nilpotent, their finitely generated ideal is nilpotent. Explicitly, cjrj=0c_j^{r_j}=0 makes its 1+∑j(rj−1)1+\sum_j(r_j-1) power zero. Thus H(P)H(P) is nilpotent and safe.

Conversely, a nonnilpotent coefficient survives in a prime quotient of RR. Pass to the fraction field of that quotient. If the image of HH has positive degree, the valuation argument in Theorem 8.1 applies to the finitely generated subfield over Q\mathbb{Q} generated by its coefficients. It gives a fixed multiplier QPeQP^e with infinitely many nonzero moments. Those moments were therefore nonzero in RR. If the image of HH is a nonzero constant, the multiplier one already detects failure. Hence safety forces every coefficient to be nilpotent. Since any nilpotent function is safe, the final equivalence follows too.

Invariance of the exceptional binary forms

We give a verification using the root sets; it also fixes the embeddings of the binary groups. Identify P1\mathbb{P}^1 with the unit sphere by

q⟼(2ℜq,2ℑq,∣q∣2−1)1+∣q∣2.q\longmapsto\frac{(2\Re q,2\Im q,|q|^2-1)}{1+|q|^2}.

The projective action of SU(2)\mathrm{SU}(2) is the rotation action.

For f8f_8, the finite roots satisfy q4=−7+43q^4=-7+4\sqrt{3}. They have arguments (2k+1)π/4(2k+1)\pi/4 and squared moduli 2±32\pm\sqrt{3}. Their sphere coordinates are exactly (±1,±1,±1)/3(\pm1,\pm1,\pm1)/\sqrt{3}, the eight vertices of a cube. Its rotation group is S4S_4, and its inverse image in SU(2)\mathrm{SU}(2) is 2O2O. The root set makes the line Cf8\mathbb{C}f_8 stable. Its scalar character factors through S4S_4, because the central element −I-I acts trivially in even degree. The matrix diag⁡(eπi/4,e−πi/4)\operatorname{diag}(e^{\pi i/4},e^{-\pi i/4}) fixes f8f_8 directly and projects to a quarter-turn, a four-cycle on the body diagonals. Every character of S4S_4 has the same value ϵ∈{1,−1}\epsilon\in\{1,-1\} on its transpositions and value ϵ3=ϵ\epsilon^3=\epsilon on a four-cycle. Thus this character is trivial and f8f_8 itself is invariant.

For f12f_{12}, let φ=(1+5)/2\varphi=(1+\sqrt{5})/2. Besides 00 and ∞\infty, its roots are

q=φ−1e2πik/5,q=φe(2k+1)πi/5(0≤k<5).q=\varphi^{-1}e^{2\pi ik/5},\qquad q=\varphi e^{(2k+1)\pi i/5}\quad(0\le k<5).

Their sphere images are two poles and two regular pentagons at heights −1/5-1/\sqrt{5} and 1/51/\sqrt{5}, of radius 2/52/\sqrt{5}, offset by π/5\pi/5. These are the vertices of a regular icosahedron. Its rotation group is A5A_5 and its inverse image is 2I2I. Again the line spanned by the binary form is stable, and its scalar character factors through A5A_5. A three-cycle is conjugate to its inverse in A5A_5, so its character value has both square and cube equal to one. It is therefore one. Three-cycles generate A5A_5, proving invariance.

The differential extraction rule

Here we prove the coordinate independence used in Section 4. Let LL be a one-variable function field over Fp\mathbb{F}_p, and choose a separating element tt. Then L=Lp(t)L=L^p(t) and 1,t,…,tp−11,t,\ldots,t^{p-1} is a basis over LpL^p. Indeed, L/Lp(t)L/L^p(t) is purely inseparable but is also separable, as a subextension of the separable extension L/Fp(t)L/\mathbb{F}_p(t). It is therefore trivial; and t∉Lpt\notin L^p, since dt≠0\mathrm{d}t\ne0.

Every differential is consequently ∑i=0p−1aiti dt\sum_{i=0}^{p-1}a_i t^i\,\mathrm{d}t. Modulo exact differentials, only the last term remains: differentiating ∑biti\sum b_i t^i shows that the exact differentials have the other p−1p-1 basis terms. Thus ΩL/dL\Omega_L/\mathrm{d}L is one-dimensional over LpL^p, with basis [tp−1 dt][t^{p-1}\,\mathrm{d}t].

For u∈Lu\in L put γ(u)=[up−1 du]\gamma(u)=[u^{p-1}\,\mathrm{d}u] in this quotient. It satisfies

γ(uv)=upγ(v)+vpγ(u),γ(u+v)=γ(u)+γ(v).\gamma(uv)=u^p\gamma(v)+v^p\gamma(u),\qquad\gamma(u+v)=\gamma(u)+\gamma(v).

The product formula is literal. For additivity, differentiate the integer polynomial ((U+V)p−Up−Vp)/p((U+V)^p-U^p-V^p)/p and reduce modulo pp; its differential is exactly the difference of the two sides before taking classes. These rules also imply γ(up)=0\gamma(u^p)=0.

Writing u=∑aitiu=\sum a_i t^i now gives

[up−1 du]=(dudt)p[tp−1 dt].[u^{p-1}\,\mathrm{d}u]=\left(\frac{\mathrm{d}u}{\mathrm{d}t}\right)^p[t^{p-1}\,\mathrm{d}t].

If uu is another separating coordinate, the class of a differential written as apup−1 dua^p u^{p-1}\,\mathrm{d}u is therefore (a du/dt)p[tp−1 dt](a\,\mathrm{d}u/\mathrm{d}t)^p[t^{p-1}\,\mathrm{d}t]. The extracted differential is a dua\,\mathrm{d}u in either coordinate. This proves that the rule defining C\mathcal{C} is well defined. For a separable extension of function fields, the same tt is separating in both and the same expansion works in both, proving compatibility with pullback. Formula (18) follows by grouping powers whose exponents are congruent to p−1p-1. See also [1 Section 2.2] for the usual Cartier–Manin terminology.

Checking the polynomial identities

Here are explicit forms of the algebra in (17), so that the proof does not require a table of invariant-theoretic normalizations. For n=8n = 8, put

B=XY(X4−Y4),T=X12−33X8Y4−33X4Y8+Y12.B = XY(X^4 - Y^4), \qquad T = X^{12} - 33X^8Y^4 - 33X^4Y^8 + Y^{12}.

Differentiation gives G=B2G = B^2 and J=BT/6J = BT/6. To check T2=f83−108B4T^2 = f_8^3 - 108B^4, set s=X4+Y4s = X^4 + Y^4 and u=X4Y4u = X^4Y^4. Then f8=s2+12uf_8 = s^2 + 12u, T=s(s2−36u)T = s(s^2 - 36u) and B4=u(s2−4u)2B^4 = u(s^2 - 4u)^2. The identity follows by expanding these three short expressions.

For n=12n = 12, differentiation gives 12J=−T12J = -T, where

T=X30+Y30+522(X25Y5−X5Y25)−10005(X20Y10+X10Y20).T = X^{30} + Y^{30} + 522(X^{25}Y^5 - X^5Y^{25}) - 10005(X^{20}Y^{10} + X^{10}Y^{20}).

With t=X5/Y5t = X^5/Y^5, the required identity becomes the following one-variable polynomial identity:

(t6+522t5−10005t4−10005t2−522t+1)2=(t4−228t3+494t2+228t+1)3+1728t(t2+11t−1)5.\begin{aligned} (t^6 + 522t^5 - 10005t^4 - 10005t^2 - 522t + 1)^2 \\ = (t^4 - 228t^3 + 494t^2 + 228t + 1)^3 + 1728t(t^2 + 11t - 1)^5. \end{aligned}

Multiplying the displayed polynomials verifies it. Homogenization then gives T2=G3+1728f125T^2 = G^3 + 1728f_{12}^5, hence J2=G3/144+12f125J^2 = G^3/144 + 12f_{12}^5 as asserted.

AI assistance and review. ChatGPT (OpenAI) assisted with mathematical arguments, literature checks, exposition, proof review and verification code.

References

  1. [1]J. D. Achter and E. W. Howe, Hasse–Witt and Cartier–Manin matrices: a warning and a request, arXiv:1710.10726v5, corrected preprint, 2020.DOI
  2. [2]J. Adams and O. Taïbi, Galois and Cartan cohomology of real groups, preprint, 2018, Section 3. Author’s manuscript.DOI
  3. [3]I. V. Arzhantsev, Invariant ideals and Matsushima’s criterion, arXiv:math/0506430v1, 2005.
  4. [4]A. Borel, Linear Algebraic Groups, second edition, Graduate Texts in Mathematics 126, Springer, 1991.
  5. [5]J. J. Duistermaat and W. van der Kallen, Constant terms in powers of a Laurent polynomial, Indag. Math. (N.S.) 9 (1998), 221–231. doi:10.1016/S0019-3577(98)80020-7.DOI
  6. [6]Y. Z. Flicker, Finite subgroups of SL(2,F) and automorphy, Manuscripta Math. 167 (2022), 213–243. doi:10.1007/s00229-021-01274-x.DOI
  7. [7]G. Kempf, Instability in invariant theory, original preprint, transcription by I. Morrison, arXiv:1807.02890v1, 2018.arxiv.org/abs/1807.02890
  8. [8]C. D. Long, Counterexamples to the xz-Conjecture and the Mathieu Conjecture for SU(2), arXiv:2607.19012v1, 2026.arxiv.org/abs/2607.19012
  9. [9]C. D. Long, The Mathieu Property for Compact Connected Lie Groups, preprint, 2026, arXiv:2609.16178v1.arxiv.org/abs/2609.16178
  10. [10]C. D. Long, A Classification of Mathieu’s Second Conjecture for Reductive Spherical Pairs, September 2026. Versioned GitHub source, revision e9edac2.
  11. [11]O. Mathieu, Some conjectures about invariant theory and their applications, Séminaires et Congrès 2 (1997), 263–279. Publisher’s excerpt, pp. 263–267.
  12. [12]A. I. Molev, Gelfand–Tsetlin bases for classical Lie algebras, in Handbook of Algebra, vol. 4, Elsevier, 2006, 109–170. arXiv:math/0211289v2.
  13. [13]G. Rond, About Eisenstein’s theorem, arXiv:2412.11738v2, 2024.arxiv.org/abs/2412.11738
  14. [14]The Stacks Project Authors, The Stacks Project, Tag 0018, Lemma 10.50.2.
  15. [15]A. V. Sutherland, Dirichlet L-functions, primes in arithmetic progressions, MIT 18.785, Lecture 18, 2021, Theorem 18.1. Lecture notes.

Paper details

vlnc:W_gfpy6rax
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

OpenAI Codex

View statement
Looking for related papers…
Contents