Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Cube-free outputs and cubic multiplicity


Status

The elementary statements below are proved and author-recorded. They concern F(x,y,z)=x3+2y3+4z3F(x,y,z)=x^3+2y^3+4z^3. A positive proportion of input triples have cube-free output; this is not a positive-density theorem for distinct outputs. The remaining issue is still multiplicity.

A cube-free sieve at the critical scale

Theorem. Put

ρp=p−9#{u mod p3:F(u)≡0(modp3)}.\rho_p=p^{-9}\#\{\mathbf u\bmod p^3:F(\mathbf u)\equiv0\pmod{p^3}\}.

Then

#{x∈[1,N]3∩Z3:F(x) is cube-free}=δFN3+o(N3),δF=∏p(1−ρp)>0.(1)\#\{\mathbf x\in[1,N]^3\cap\mathbb Z^3:F(\mathbf x)\text{ is cube-free}\} =\delta_F N^3+o(N^3),\qquad \delta_F=\prod_p(1-\rho_p)>0. \tag{1}

Every triple counted in (1) is primitive.

For p≥5p\ge5, first count the triples with all coordinates divisible by pp: there are at most (N/p)3(N/p)^3. Otherwise choose a unit coordinate. Its coefficient and the derivative of its cube term are units modulo pp. For fixed other coordinates there are at most three possible unit residues modulo p3p^3, each lifting uniquely from a root modulo pp. Union over the three possible choices of coordinate gives

Bp(N):=#{x∈[1,N]3:p3∣F(x)}≤10N3p−3+9N2.(2)B_p(N):=\#\{\mathbf x\in[1,N]^3:p^3\mid F(\mathbf x)\} \le 10N^3p^{-3}+9N^2. \tag{2}

Since 0<F(x)≤7N30<F(\mathbf x)\le7N^3, only primes p≤71/3Np\le7^{1/3}N occur. The elementary bound π(t)≪t/log⁡t\pi(t)\ll t/\log t consequently gives, for P≥3P\ge3,

#{x:∃p>P, p3∣F(x)}≪N3P−2+N3log⁡N.(3)\#\{\mathbf x:\exists p>P,\ p^3\mid F(\mathbf x)\} \ll N^3P^{-2}+\frac{N^3}{\log N}. \tag{3}

The same unit-coordinate argument in a complete residue box gives ρp≤10p−3\rho_p\le10p^{-3} for p≥5p\ge5. Every factor 1−ρp1-\rho_p is positive, because F(1,0,0)=1F(1,0,0)=1. Thus the product in (1) converges to a positive number.

For fixed PP, CRT and elementary box counting give

#{x∈[1,N]3:p3∤F(x) (p≤P)}=N3∏p≤P(1−ρp)+OP(N2).\#\{\mathbf x\in[1,N]^3:p^3\nmid F(\mathbf x)\ (p\le P)\} =N^3\prod_{p\le P}(1-\rho_p)+O_P(N^2).

Use (3), let N→∞N\to\infty, and then let P→∞P\to\infty. This proves (1). If a prime divided all coordinates, its cube would divide the output, proving primitivity. At p=2p=2, more explicitly, 8∣F(x)8\mid F(\mathbf x) if and only if all three coordinates are even: the valuations of the nonzero terms are distinct modulo three. Hence ρ2=1/8\rho_2=1/8.

Bounded average logarithm of the cube part

For a positive integer nn, define

D(n)=∏pp⌊vp(n)/3⌋.D(n)=\prod_p p^{\lfloor v_p(n)/3\rfloor}.

Thus D(n)3D(n)^3 is the largest perfect cube dividing nn.

Theorem. Uniformly for N≥2N\ge2,

∑x∈[1,N]3log⁡D(F(x))≪N3.(4)\sum_{\mathbf x\in[1,N]^3}\log D(F(\mathbf x))\ll N^3. \tag{4}

First restrict to primitive triples. At every prime, at least one coordinate is a unit. For p≥5p\ge5, the corresponding one-variable congruence modulo pkp^k has at most three unit roots. This remains true at p=3p=3, since cubing on (Z/3kZ)×(\mathbb Z/3^k\mathbb Z)^\times has kernel of order at most three. The coefficients of FF are units there. At p=2p=2, a primitive triple has v2(F(x))≤2v_2(F(\mathbf x))\le2, by the distinct-valuation argument above, so it contributes no cube factor. Consequently for every odd prime and k≥1k\ge1,

#{x∈[1,N]3:gcd⁡(x)=1, pk∣F(x)}≪N3p−k+N2.(5)\#\{\mathbf x\in[1,N]^3:\gcd(\mathbf x)=1,\ p^k\mid F(\mathbf x)\} \ll N^3p^{-k}+N^2. \tag{5}

The absolute constant in (5) does not depend on p,k,Np,k,N.

Write C=71/3C=7^{1/3} and Jp=⌊log⁡(7N3)/(3log⁡p)⌋J_p=\lfloor\log(7N^3)/(3\log p)\rfloor. Expanding log⁡D(n)=∑p∑j≥1(log⁡p)1p3j∣n\log D(n)=\sum_p\sum_{j\ge1}(\log p)\mathbf1_{p^{3j}\mid n}, equation (5) bounds the primitive contribution to (4) by a constant times

N3∑plog⁡pp3−1+N2∑p≤CNJplog⁡p.N^3\sum_p\frac{\log p}{p^3-1} +N^2\sum_{p\le CN}J_p\log p.

The first series converges. The second prime sum is at most 13log⁡(7N3)π(CN)≪N\tfrac13\log(7N^3)\pi(CN)\ll N. The primitive contribution is therefore O(N3)O(N^3).

For an arbitrary triple write x=gy\mathbf x=g\mathbf y, with y\mathbf y primitive. Homogeneity gives D(F(x))=gD(F(y))D(F(\mathbf x))=gD(F(\mathbf y)). Sum the primitive estimate at N/gN/g, together with the term log⁡g\log g, and use ∑gg−3(1+log⁡g)<∞\sum_g g^{-3}(1+\log g)<\infty. This proves (4). In particular,

#{x∈[1,N]3:D(F(x))>H}≪N3log⁡H(H>1).(6)\#\{\mathbf x\in[1,N]^3:D(F(\mathbf x))>H\} \ll\frac{N^3}{\log H}\qquad(H>1). \tag{6}

This controls the output's cube part, not the logarithm of its representation multiplicity.

Primitive collisions at cube-free outputs remain

The split identity already proved in the energy note is

F(27m+92s,212m+480s,171m+515s)=F(27m+12s,212m+642s,171m+391s).(7)\begin{aligned} F(27m+92s,212m+480s,171m+515s)\\ =F(27m+12s,212m+642s,171m+391s). \end{aligned} \tag{7}

Its common value is

G(m,s)=39076783m3+310349304m2s+837996084ms2+768326188s3.(8)G(m,s)=39076783m^3+310349304m^2s +837996084ms^2+768326188s^3. \tag{8}

There are ≫T2\gg T^2 pairs 1≤m,s≤T1\le m,s\le T for which (8) is cube-free. Here is a sieve proof, including the local conditions.

The binary cubic GG has no rational projective zero: the rank-two linear map on either side of (7) would turn such a zero into a nonzero rational zero of FF, contrary to its distinct 22-adic valuations. Thus GG is irreducible over Q\mathbb Q, and its discriminant is nonzero. Also

gcd⁡(G(1,0),G(0,1))=gcd⁡(39076783,768326188)=1.\gcd(G(1,0),G(0,1))=\gcd(39076783,768326188)=1.

For each prime, at least one of these two values is a unit, so GG has no fixed prime-cube divisor.

Outside finitely many primes dividing its discriminant, every nonzero zero of GG modulo pp has a nonzero partial derivative. Split off the pairs with p∣m,sp\mid m,s; these contribute at most (T/p)2(T/p)^2. For the other pairs, choose a coordinate with nonzero derivative and fix the other coordinate. There are at most three simple roots modulo p3p^3 for either choice. Hence

#{1≤m,s≤T:p3∣G(m,s)}≤T2p−2+6T(Tp−3+1).(9)\#\{1\le m,s\le T:p^3\mid G(m,s)\} \le T^2p^{-2}+6T(Tp^{-3}+1). \tag{9}

Only primes p≪GTp\ll_G T can occur. Once PP contains the finitely many bad primes, (9) gives the tail bound

#{(m,s):∃p>P, p3∣G(m,s)}≪GT2P−1+T2log⁡T.\#\{(m,s):\exists p>P,\ p^3\mid G(m,s)\} \ll_G T^2P^{-1}+\frac{T^2}{\log T}.

In complete residue boxes the local excluded proportions are O(p−2)O(p^{-2}). Their product of complements is positive by local nonobstruction. Finite CRT, followed by this tail estimate, proves the asserted positive input density of cube-free GG-values.

For these parameters the two triples in (7) are positive, primitive and distinct. They lie in [1,854T]3[1,854T]^3. The map from parameters to collision pairs is injective: their difference is (80s,−162s,124s)(80s,-162s,124s), which determines ss, and then the first coordinate determines mm. Thus this fixed split plane supplies ≫N2\gg N^2 primitive collisions with cube-free output in a box of side NN.

Consequence for the density route

The restriction to cube-free outputs retains ≫N3\gg N^3 input triples, and 8F(x)∈S38F(\mathbf x)\in S_3. A bound O(N3)O(N^3) for their restricted collision count would therefore give positive lower density of S3S_3. No such bound, nor the weaker bounded-fiber positive-mass criterion, has been established. Equations (4) and (7) show why merely deleting common dilations does not complete the argument: primitive split collisions survive, and the critical average over large directions is still uncontrolled.

The primitive-fiber analysis now separates the bounded entropy of common dilations from primitive multiplicity. It also shows that every curve defined over Q\mathbb Q on the special cubic's output surface has degree divisible by three, excluding rational lines and conics even without cube-freeness. No uniform estimate for its higher-degree curves or primitive fibers was obtained.