OctoInformBeta

OpenAI preprint, September 2026 ยท ยง2. Reduction to approximation and rounding

The phase e(x) and the relaxed sum Q_X: The value of PXP_X at e(t)e(t) is the phase sum โˆ‘k<NXke(kt)\sum_{k<N} X_k e(kt), for every real coefficient vector XX and real tt

Real polynomials on the unit circle are sums of phases.

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

The flatness problem for Littlewood polynomials concerns the size of their modulus on the unit circle |z|=1|z| = 1. On the circle such a polynomial is a periodic function of the angle, and one studies it as a function on the circle ๐•‹=โ„/โ„ค\mathbb{T} = \mathbb{R}/\mathbb{Z}. The question here is the Fourier expansion of that function when the NN coefficients X0,โ€ฆ,XNโˆ’1X_0, \dots, X_{N-1} are arbitrary real numbers rather than signs: which frequencies occur, and with which coefficients.

The Phase and the Polynomial

The unit phase of a real number xx is the complex number exp⁡(ix)\exp(ix), the point of the unit circle at angle xx.โœฆ The phase of period one is the function ee on โ„\mathbb{R} given by (1), the unit phase of 2ฯ€x2\pi x at the real number xx. e(x)=exp⁡(2ฯ€ix),xโˆˆโ„,e(x) = \exp(2\pi i x), \qquad x \in \mathbb{R},(1) The parameter tโˆˆ[0,1]t \in [0,1] parametrizes the unit circle by z=e(t)z = e(t).

The polynomial PXP_X of a vector X=(X0,โ€ฆ,XNโˆ’1)X = (X_0, \dots, X_{N-1}) of NN real coefficients is the function on the complex plane given by PX(z)=โˆ‘k<NXkzk,zโˆˆโ„‚.P_X(z) = \sum_{k<N} X_k z^k, \qquad z \in \mathbb{C}.(2)โœฆ When XX is a signing ฮต\varepsilon, with every entry โˆ’1-1 or 11, PฮตP_\varepsilon is its Littlewood polynomial.

The Claim

Theorem 1. Let NN be a natural number, let X=(X0,โ€ฆ,XNโˆ’1)X = (X_0, \dots, X_{N-1}) be a vector of real numbers, and let tt be a real number. Then the value of the polynomial PXP_X at the point e(t)e(t) of the unit circle is the phase sum PX(e(t))=โˆ‘k<NXke(kt).P_X(e(t)) = \sum_{k<N} X_k \, e(kt).(3)โœฆ

In the paper, the result reads: for Xโˆˆ[โˆ’1,1]NX \in [-1,1]^N, the relaxed sum QXQ_X is the function QX(t)=โˆ‘k<NXke(kt)Q_X(t) = \sum_{k<N} X_k e(kt).

By (3), the relaxed sum is the polynomial PXP_X on the unit circle, QX(t)=PX(e(t))Q_X(t) = P_X(e(t)).

What Is Checked

The claim is stated over the unit phase exp⁡(ix)\exp(ix) of a real number xx, which gives the phases e(t)e(t) and e(kt)e(kt) of (1), and over the polynomial PXP_X of a real coefficient vector XX, defined at every complex point by (2).