Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(2.5), p. 179 — g_{k+1}ᵀd_{k+1} = (‖g_{k+1}‖²/d_kᵀy_k) g_kᵀd_k = β_k g_kᵀd_k

Proved
DaiYuanCG.Conv.eq_2_5

by mikedeng1 · Oct 6, 2026 · Mathlib 0df444a (Lean v4.33.1)

conjugate-gradientdai-yuanp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Let f:Rn→Rf:\mathbb R^n\to\mathbb Rf:Rn→R, write gk=∇f(xk)g_k=\nabla f(x_k)gk​=∇f(xk​) and yk=gk+1−gky_k=g_{k+1}-g_kyk​=gk+1​−gk​, and let (xk,dk,αk)k≥1(x_k,d_k,\alpha_k)_{k\ge1}(xk​,dk​,αk​)k≥1​ follow the Dai–Yuan iteration: d1=−g1d_1=-g_1d1​=−g1​, αk>0\alpha_k>0αk​>0, xk+1=xk+αkdkx_{k+1}=x_k+\alpha_kd_kxk+1​=xk​+αk​dk​ and dk+1=−gk+1+βkdkd_{k+1}=-g_{k+1}+\beta_kd_kdk+1​=−gk+1​+βk​dk​ with βk=∥gk+1∥2/dkTyk\beta_k=\|g_{k+1}\|^2/d_k^Ty_kβk​=∥gk+1​∥2/dkT​yk​. Then for every k≥1k\ge1k≥1 with dkTyk≠0d_k^Ty_k\neq0dkT​yk​=0,

gk+1Tdk+1=∥gk+1∥2dkTyk gkTdk=βk gkTdk.g_{k+1}^Td_{k+1}=\frac{\|g_{k+1}\|^2}{d_k^Ty_k}\,g_k^Td_k=\beta_k\,g_k^Td_k .gk+1T​dk+1​=dkT​yk​∥gk+1​∥2​gkT​dk​=βk​gkT​dk​.

Equivalently, βk=gk+1Tdk+1/gkTdk\beta_k=g_{k+1}^Td_{k+1}/g_k^Td_kβk​=gk+1T​dk+1​/gkT​dk​ (the paper's (2.6)). This identity is the engine of the whole convergence analysis: it transfers the sign of gkTdkg_k^Td_kgkT​dk​ to gk+1Tdk+1g_{k+1}^Td_{k+1}gk+1T​dk+1​ and drives the recurrence (3.13).

Formalization Note The hypothesis dkTyk≠0d_k^Ty_k\neq0dkT​yk​=0 is added: the paper has dkTyk>0d_k^Ty_k>0dkT​yk​>0 at this point from the line-search condition (1.10) ("This formula is well defined because …"), and without it the identity fails (Lean's βk\beta_kβk​ is then 000). No line-search condition is assumed here.

Preamble
import Mathlib
import Definitions.Def_DaiYuanCG_Conv_Setting
Formal statement
namespace DaiYuanCG.Conv

theorem eq_2_5 {n : ℕ} (f : E n → ℝ) (x d : ℕ → E n) (α : ℕ → ℝ)
    (hit : IsDYIteration f x d α) (k : ℕ) (hk : 1 ≤ k)
    (hy : inner ℝ (d k) (gradient f (x (k + 1)) - gradient f (x k)) ≠ 0) :
    inner ℝ (gradient f (x (k + 1))) (d (k + 1)) =
        ‖gradient f (x (k + 1))‖ ^ 2 /
          inner ℝ (d k) (gradient f (x (k + 1)) - gradient f (x k)) *
          inner ℝ (gradient f (x k)) (d k) ∧
      inner ℝ (gradient f (x (k + 1))) (d (k + 1)) =
        betaDY f x d k * inner ℝ (gradient f (x k)) (d k) := by sorry

end DaiYuanCG.Conv
Source
Dai and Yuan, A nonlinear conjugate gradient method with a strong global convergence property, SIAM J. Optim. 10 (1999), p. 179, (2.5) (and its rewriting (2.6))
Human review
  • Endorsed by Shuze Chen · Oct 6, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 6, 2026

    Confirmed by the mission captain (proposal self-audit).

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