OctoInformBeta

OpenAI preprint, September 2026 · §4. A uniquely represented function with intermediate growth

Heights, the scalar m_* and the sequence W_j: With hˆ\hat h the canonical height on y2=x3−25xy^2=x^3-25x, log⁡Wj\log W_j lies within cc of 2m2hˆ(P)j22m^2\hat h(P)j^2 and v=Wm(Q)v=W_m(Q) is single-fold on point codes

Canonical heights and a height sequence of quadratic logarithmic growth.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let EE be the curve y2=x3−25xy^2=x^3-25x over ℚ\mathbb Q, with point at infinity OO. Its group of rational points is E(ℚ)=ℤP⊕TE(\mathbb Q)=\mathbb ZP\oplus T, with PP a point of infinite order and TT the finite torsion subgroup, so that every rational point is nP+tnP+t with n∈ℤn\in\mathbb Z and t∈Tt\in T. A point is measured by the height of its xx-coordinate, a natural number. Two questions concern the multiples of a point: how fast the height of [j]P[j]P grows with jj, and whether, for a fixed mm, the relation between a point QQ and the height vv of [m]Q[m]Q has a single-fold Diophantine representation once points are written as tuples of natural numbers.

The library states both answers for this curve, along with the basic properties of the canonical height and the existence of an even mm and of constants cc and κ\kappa bound by a set of inequalities.

Heights on EE

The height H(r)H(r) of a rational number r=a/dr=a/d in lowest terms, d≥1d\ge1, is max⁡(|a|,d)\max(|a|,d). The height Hx(Q)H_x(Q) of a point QQ of E(ℚ)E(\mathbb Q) is the natural number H(x(Q))H(x(Q)) when Q≠OQ\ne O, and Hx(O)=1H_x(O)=1.✦ The naive height hx(Q)h_x(Q) of a point QQ of E(ℚ)E(\mathbb Q) is the real number hx(Q)=log⁡H(x(Q))(Q≠O),hx(O)=0.h_x(Q)=\log H(x(Q))\quad(Q\ne O),\qquad h_x(O)=0.(1)✦

The canonical height hˆ(Q)\hat h(Q) of a point QQ of E(ℚ)E(\mathbb Q) is the real number equal to one half of the limit, as N→∞N\to\infty, of the scaled naive heights 4−Nhx([2N]Q)4^{-N}h_x([2^N]Q).✦

Multiples of PP

The decomposition of E(ℚ)E(\mathbb Q) is given by an isomorphism ff of E(ℚ)E(\mathbb Q) onto ℤ×T\mathbb Z\times T: the point PP is f−1(1,0)f^{-1}(1,0)✦, the point nP+tnP+t is f−1(n,t)f^{-1}(n,t) for n∈ℤn\in\mathbb Z and t∈Tt\in T✦, and the index of a point QQ is the integer nn with f(Q)=(n,t)f(Q)=(n,t) for some t∈Tt\in T✦.

For a natural number mm, the multiple height Wm(Q)W_m(Q) of a point QQ of E(ℚ)E(\mathbb Q) is the natural number Wm(Q)=Hx([m]Q),W_m(Q)=H_x([m]Q),(2)✦ and the height sequence of mm is the sequence of natural numbers Wj=Wm([j]P),j∈ℕ.W_j=W_m([j]P),\qquad j\in\mathbb N.(3)✦

Point Codes

A rational code is a sextuple (p,n,d,t,s,q)(p,n,d,t,s,q) of natural numbers✦; its value is the rational number (p−n)/d(p-n)/d✦. A rational code is valid when pn=0,d=1+t+s,(p+n)t+d=1+qd.pn=0,\qquad d=1+t+s,\qquad (p+n)t+d=1+qd.(4)✦ The code of a rational number rr is the valid rational code of value rr.✦

A point code is a flag o∈ℕo\in\mathbb N together with two rational codes, its xx-code and its yy-code: thirteen natural numbers in all.✦ The height of a point code is the natural number max⁡(p+n,d)\max(p+n,d) formed from the first three entries pp, nn, dd of its xx-code.✦ The code of a point Q≠OQ\ne O of E(ℚ)E(\mathbb Q) is the point code with flag 00, xx-code the code of x(Q)x(Q) and yy-code the code of y(Q)y(Q); the code of OO has flag 11 and the code of 00 as both its xx-code and its yy-code.✦

For a natural number mm, the multiplication relation of mm holds between point codes AA and BB when AA is the code of a point QQ of E(ℚ)E(\mathbb Q) and BB is the code of [m]Q[m]Q.✦ The relation 𝒲m\mathcal W_m between point codes and natural numbers holds of (A,v)(A,v) when AA stands in the multiplication relation of mm with some point code of height vv.✦

Single-fold Relations

A single-fold representation of a relation RR on tuples of natural numbers indexed by a set α\alpha is a polynomial FF with integer coefficients in input variables a=(ai)i∈αa=(a_i)_{i\in\alpha} and finitely many auxiliary variables ww such that a∈Ra\in R exactly when F(a,w)=0F(a,w)=0 for some tuple ww of natural numbers, and such that any two tuples ww and w′w' with F(a,w)=0F(a,w)=0 and F(a,w′)=0F(a,w')=0 are equal.✦ A relation RR on tuples of natural numbers indexed by a set α\alpha is single-fold when it has a single-fold representation.✦

The Claim

Theorem 1. Let QQ range over the rational points of EE. The scaled naive heights of QQ converge: 124−Nhx([2N]Q)→hˆ(Q)(N→∞).\tfrac12\,4^{-N}h_x([2^N]Q)\longrightarrow\hat h(Q)\qquad(N\to\infty).(5)✦ For every integer nn, hˆ([n]Q)=n2hˆ(Q).\hat h([n]Q)=n^2\,\hat h(Q).(6)✦ If QQ has infinite order, then hˆ(Q)>0.\hat h(Q)>0.(7)✦ There is a real number c>0c>0 such that, for every QQ, |hx(Q)−2hˆ(Q)|≤c.|h_x(Q)-2\hat h(Q)|\le c.(8)✦ For every QQ, log⁡Hx(Q)=hx(Q).\log H_x(Q)=h_x(Q).(9)✦ If mm is an even natural number and (Wj)(W_j) its height sequence, then for every n∈ℤn\in\mathbb Z and t∈Tt\in T, Wm(nP+t)=W|n|.W_m(nP+t)=W_{|n|}.(10)✦ If mm is a natural number and cc a real number with |hx(Q)−2hˆ(Q)|≤c|h_x(Q)-2\hat h(Q)|\le c for every QQ, then the height sequence of mm satisfies, for every j∈ℕj\in\mathbb N, |log⁡Wj−2m2hˆ(P)j2|≤c.\bigl|\log W_j-2m^2\hat h(P)\,j^2\bigr|\le c.(11)✦ There are a real number c>0c>0 and positive integers κ\kappa and mm, with mm even and |hx(Q)−2hˆ(Q)|≤c|h_x(Q)-2\hat h(Q)|\le c for every QQ, such that 2c<2m2hˆ(P),6c<log⁡κ<4m2hˆ(P)−6c.2c<2m^2\hat h(P),\qquad 6c<\log\kappa<4m^2\hat h(P)-6c.(12)✦ Finally, let mm be a natural number, α\alpha a set, σ\sigma a map from the thirteen entries of a point code into α\alpha, and u∈αu\in\alpha. With a∘σa\circ\sigma the point code whose entries are those of aa at the places σ\sigma gives, {a∈ℕα:𝒲m(a∘σ,au)}is single-fold.\bigl\{a\in\mathbb N^{\alpha} : \mathcal W_m(a\circ\sigma,\,a_u)\bigr\}\ \text{is single-fold}.(13)✦

In the paper, the result reads: H(a/d)=max⁡(|a|,d)H(a/d)=\max(|a|,d) for reduced a/da/d, hx(Q)=log⁡H(x(Q))h_x(Q)=\log H(x(Q)) with hx(O)=0h_x(O)=0, and hˆ(Q)=12lim⁡N→∞4−Nhx([2N]Q)\hat h(Q)=\frac12\lim_{N\to\infty}4^{-N}h_x([2^N]Q); hˆ([j]Q)=j2hˆ(Q)\hat h([j]Q)=j^2\hat h(Q), hˆ(Q)>0\hat h(Q)>0 for nontorsion QQ, and |log⁡H(x(Q))−2hˆ(Q)|≤cE|\log H(x(Q))-2\hat h(Q)|\le c_E; with κ>e6cE\kappa>e^{6c_E} and an even m*m_* making aE=2m*2hˆ(P)a_E=2m_*^2\hat h(P) satisfy aE>2cEa_E>2c_E and 2aE−6cE>log⁡κ2a_E-6c_E>\log\kappa, W(Q)=H(x([m*]Q))W(Q)=H(x([m_*]Q)) with W(O)=1W(O)=1, Wj=W(nP+t)W_j=W(nP+t) for |n|=j|n|=j, and log⁡Wj=aEj2+εj\log W_j=a_Ej^2+\varepsilon_j with |εj|≤cE|\varepsilon_j|\le c_E, its cEc_E and m*m_* being the cc and mm above.

What Is Checked

The claim is stated over the curve EE and its group of rational points with the isomorphism ff onto ℤ×T\mathbb Z\times T, TT finite, the point PP and the index it gives, the heights HxH_x, hxh_x and hˆ\hat h, the multiple heights WmW_m and their height sequences; and (13) is stated over rational codes with their value and validity, point codes and their height, the codes of rational numbers and of points, the multiplication relation, the relation 𝒲m\mathcal W_m and single-fold representations.

About these constructions the library proves that every rational number rr has exactly one valid rational code of value rr.✦