Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Residual comparison between full-solve snapshots

Proved
NearlyOptimalPPR.Computation.recursive_full_solve_comparison

by Shuze Chen · Nov 18, 2026 · Mathlib 0df444a (Lean v4.33.1)

algorithm-analysisgraph-theorypagerank

Let aaa be a permitted full-solve state of open level jjj, and let b≥a+1b\ge a+1b≥a+1 be a later state throughout which that level remains open. Assume the full solve starts with no pending exterior push and no exhausted open budget. Write H=Hj(a)H=H_j(a)H=Hj​(a), F=Hj(b)F=H_j(b)F=Hj​(b), Z=F∖HZ=F\setminus HZ=F∖H, and let hhh be the exact solve on HHH with source rj(a)r_j(a)rj​(a). If yj(a)y_j(a)yj​(a) is the virtual potential, put D={v∈H:h(v)>yj(a,v)}D=\{v\in H:h(v)>y_j(a,v)\}D={v∈H:h(v)>yj​(a,v)} and u=x(b)−x(a+1)u=x(b)-x(a+1)u=x(b)−x(a+1). Then

∑v∈Frj(b,v)≤(1−α)∑v∉Fdvuv+4κ0qjvol⁡(Z)+2κ0qjvol⁡(D)+2∑v∈Drj(a,v).\sum_{v\in F}r_j(b,v)\le (1-\alpha)\sum_{v\notin F}d_vu_v +4\kappa_0q_j\operatorname{vol}(Z) +2\kappa_0q_j\operatorname{vol}(D) +2\sum_{v\in D}r_j(a,v).v∈F∑​rj​(b,v)≤(1−α)v∈/F∑​dv​uv​+4κ0​qj​vol(Z)+2κ0​qj​vol(D)+2v∈D∑​rj​(a,v).

Here qjq_jqj​ is the threshold at the initial snapshot and κ0=ℓ+3\kappa_0=\ell+3κ0​=ℓ+3. The finite undirected input has a nonisolated seed, 0<α<10<\alpha<10<α<1, and positive root threshold.

This is the exact-solve specialization of the comparison behind Lemma 5.7. Its right-hand side separates later exterior flow, newly activated volume, and vertices whose potential doubles.

Preamble
import Definitions.Def_NearlyOptimalPPR_Computation_work_analysis
open Finset Matrix NearlyOptimalPPR.Computation
Formal statement
theorem NearlyOptimalPPR.Computation.recursive_full_solve_comparison (m : SolveModel) (n : ℕ) (g : InputGraph n) (s : Fin n) (α q : ℝ)
    (hα : 0 < α ∧ α < 1) (hq : 0 < q) (hs : 0 < g.graph.degree s)
    (ℓ : ℕ) (schedule : BlockSchedule) (a b : ℕ) (hab : a+1 ≤ b) :
    let run := recursiveRun m g s α q ℓ schedule
    let st := run a
    let next := run b
    let j := st.top
    let H := (st.frames j).active
    let F := (next.frames j).active
    let Z := F \ H
    let h := solveOn g α H (st.frames j).residual
    let u := next.x-(run (a+1)).x
    let D := doubledSet H h (virtualPotential st j)
    (∀ k, a ≤ k → k < b → j ≤ (run k).top) → j ≤ next.top →
    st.done = false → ¬componentComplete st → (st.frames j).pc = .fullSolve →
      ¬anyExhausted st → pendingPush st = none →
      (∑ v ∈ F, (next.frames j).residual v) ≤
        (1-α)*(∑ v ∈ Fᶜ, LocalPush.deg g.graph v*u v)+
        4*(kappa ℓ 0 : ℝ)*(st.frames j).q*(volumeNat g Z : ℝ)+
        2*(kappa ℓ 0 : ℝ)*(st.frames j).q*(volumeNat g D : ℝ)+
        2*∑ v ∈ D, (st.frames j).residual v := by sorry
Source
Li, Xu, Yang, Zhou, Nearly optimal Personalized PageRank computation, STOC 2027 submission #884, https://github.com/sigact-stoc/stoc2027/releases/download/papers-1/884.pdf, Lemmas 5.7–5.8 and their proof pp.26–28; Lemma 4.4 p.18 and finite-prefix counts pp.28–29.

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me