OctoInformBeta

OpenAI preprint, September 2026 · §E. Optimization on common log canonical slices

§E (setting): valuations and the common lc slice: Formal expansions turn normalized nonnegative weights into valuations subject to prescribed evaluations

Monomial valuations and finite chart slices.

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

Formal expansions attach weighted orders to functions in a chart algebra.

Let kk be an integral domain, let AA be a kk-algebra, and let σ\sigma be a finite set of coordinates. Write k[[zs:s∈σ]]k[[z_s:s\in\sigma]] for the formal power series ring and cd(f)c_d(f) for the coefficient of zdz^d in a series ff, where d∈ℕσd\in\mathbb N^\sigma. A coordinate weight vector ww for σ\sigma is an element of ℝ≥0σ\mathbb R_{\geq0}^{\sigma}; zero weights and ties are allowed, and the weight of zdz^d is ∑s∈σdsws\sum_{s\in\sigma}d_s w_s.

Fix a finite family of kk-algebra homomorphisms θi:A→k[[zs:s∈σ]]\theta_i:A\longrightarrow k[[z_s:s\in\sigma]], indexed by i∈Ii\in I, and coefficient vectors ai∈ℝ≥0σa_i\in\mathbb R_{\geq0}^{\sigma}. Let h∈Ah\in A, and fix a finite family Fj∈AF_j\in A with thresholds mj∈[0,+∞]m_j\in[0,+\infty], indexed by j∈Jj\in J. All the valuations below are evaluated on chart rings, with values in [0,+∞][0,+\infty].

The Definition

Definition 1. For a nonnegative coordinate weight vector ww, the coefficient weight infimum μw:k[[zs:s∈σ]]→[0,+∞]\mu_w:k[[z_s:s\in\sigma]]\longrightarrow[0,+\infty] is the function specified by (1): μw(f)=infd∈ℕσcd(f)≠0∑s∈σdsws.\mu_w(f)=\inf_{\substack{d\in\mathbb N^\sigma\\c_d(f)\neq0}} \sum_{s\in\sigma}d_s w_s.(1)✦ For a nonnegative coordinate weight vector ww for σ\sigma, the formal monomial valuation vwv_w on k[[zs:s∈σ]]k[[z_s:s\in\sigma]] is the additive valuation specified by (2): vw:k[[zs:s∈σ]]→[0,+∞],f↦μw(f).v_w:k[[z_s:s\in\sigma]]\longrightarrow[0,+\infty], \qquad f\longmapsto\mu_w(f).(2)✦ For a kk-algebra homomorphism θ:A→k[[zs:s∈σ]]\theta:A\longrightarrow k[[z_s:s\in\sigma]], the pulled-back monomial valuation vθ,wv_{\theta,w} on AA is the valuation in (3): vθ,w(f)=vw(θ(f)),f∈A.v_{\theta,w}(f)=v_w\bigl(\theta(f)\bigr),\qquad f\in A.(3)✦ For a∈ℝ≥0σa\in\mathbb R_{\geq0}^{\sigma}, the normalized weight set W(a)W(a) is the subset of the nonnegative weight cone given by (4): W(a)={w∈ℝ≥0σ:∑s∈σasws=1}.W(a)=\left\{w\in\mathbb R_{\geq0}^{\sigma}: \sum_{s\in\sigma}a_s w_s=1\right\}.(4)✦ The evaluation constraint set 𝒞(h,F,m)\mathcal C(h,F,m) is the set of functions from AA to [0,+∞][0,+\infty] satisfying (5): 𝒞(h,F,m)={v:A→[0,+∞]:v(h)=1,v(Fj)≥mjfor every j∈J}.\mathcal C(h,F,m)= \left\{v:A\longrightarrow[0,+\infty]: v(h)=1,\quad v(F_j)\geq m_j\ \text{for every }j\in J\right\}.(5)✦ The finite chart slice Σ(θ,a;h,F,m)\Sigma(\theta,a;h,F,m) is the set of valuations on AA defined by (6): Σ(θ,a;h,F,m)=(⋃i∈I{vθi,w:w∈W(ai)})∩𝒞(h,F,m).\Sigma(\theta,a;h,F,m)= \left(\bigcup_{i\in I}\{v_{\theta_i,w}:w\in W(a_i)\}\right) \cap\mathcal C(h,F,m).(6)✦

In the paper, the result reads: for a normal complex affine germ x∈Sx\in S of positive dimension nn, with SS rationally Gorenstein and Kawamata log terminal, local ring RR and maximal ideal 𝔪\mathfrak m, a nonzero h∈Rh\in R such that (S,div⁡(h))(S,\operatorname{div}(h)) is log canonical, finitely many nonzero Fj∈RF_j\in R and positive integers mjm_j, the common slice Q={v:𝒜(v)=v(h)=1,v(Fj)≥mjfor every j}Q=\{v:\mathcal A(v)=v(h)=1,\ v(F_j)\geq m_j\ \text{for every }j\}, where 𝒜(v)\mathcal A(v) is the log discrepancy, is assumed to contain a centred valuation.

What Is Checked

The definitions are stated over a kk-algebra AA with formal expansion homomorphisms θi\theta_i, nonnegative coordinate weights and evaluation conditions on hh and FjF_j. For formal power series f,gf,g, the coefficient weight infimum satisfies the valuation identities and inequality in (7): μw(0)=+∞,μw(1)=0,μw(f+g)≥min⁡{μw(f),μw(g)},μw(fg)=μw(f)+μw(g).\begin{aligned} \mu_w(0)&=+\infty,&\qquad \mu_w(1)&=0,\\ \mu_w(f+g)&\geq\min\{\mu_w(f),\mu_w(g)\},& \mu_w(fg)&=\mu_w(f)+\mu_w(g). \end{aligned}(7)