Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 3 — regret of composite mirror descent with time-varying proximal functions, (11)

Proved
AdaGrad.Diag.proposition_3

by mikedeng1 · Oct 5, 2026 · Mathlib 0df444a (Lean v4.33.1)

mirror-descentonline-learningp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1regret

Let η>0\eta>0η>0, let X⊆Rd\mathcal X\subseteq\mathbb R^dX⊆Rd be closed and convex, and let φ\varphiφ and every ftf_tft​ be convex functions Rd→R\mathbb R^d\to\mathbb RRd→R. For each round t≥1t\ge1t≥1 let ht∈Rdh_t\in\mathbb R^dht​∈Rd have nonnegative entries, with proximal function ψt(y)=12∑iht,iyi2\psi_t(y)=\frac12\sum_ih_{t,i}y_i^2ψt​(y)=21​∑i​ht,i​yi2​, Bregman divergence Bψt(y,z)=12∑iht,i(yi−zi)2B_{\psi_t}(y,z)=\frac12\sum_ih_{t,i}(y_i-z_i)^2Bψt​​(y,z)=21​∑i​ht,i​(yi​−zi​)2 and squared dual norm ∥g∥ψt∗2=∑igi2/ht,i\|g\|^2_{\psi_t^*}=\sum_ig_i^2/h_{t,i}∥g∥ψt∗​2​=∑i​gi2​/ht,i​. Let x1,x2,…x_1,x_2,\dotsx1​,x2​,… be a run of the composite mirror descent update (4): gtg_tgt​ is a subgradient of ftf_tft​ at xtx_txt​, with gt,i=0g_{t,i}=0gt,i​=0 whenever ht,i=0h_{t,i}=0ht,i​=0, and xt+1∈Xx_{t+1}\in\mathcal Xxt+1​∈X minimizes η⟨gt,y⟩+ηφ(y)+Bψt(y,xt)\eta\langle g_t,y\rangle+\eta\varphi(y)+B_{\psi_t}(y,x_t)η⟨gt​,y⟩+ηφ(y)+Bψt​​(y,xt​) over X\mathcal XX. Assume x1∈Xx_1\in\mathcal Xx1​∈X minimizes φ\varphiφ over X\mathcal XX. Then for every TTT and every x∗∈Xx^*\in\mathcal Xx∗∈X,

∑t=1T[ft(xt)+φ(xt)−ft(x∗)−φ(x∗)]≤1ηBψ1(x∗,x1)+1η∑t=1T−1[Bψt+1(x∗,xt+1)−Bψt(x∗,xt+1)]+η2∑t=1T∥gt∥ψt∗2.\sum_{t=1}^T\big[f_t(x_t)+\varphi(x_t)-f_t(x^*)-\varphi(x^*)\big]\le\frac1\eta B_{\psi_1}(x^*,x_1)+\frac1\eta\sum_{t=1}^{T-1}\big[B_{\psi_{t+1}}(x^*,x_{t+1})-B_{\psi_t}(x^*,x_{t+1})\big]+\frac\eta2\sum_{t=1}^T\|g_t\|^2_{\psi_t^*}.t=1∑T​[ft​(xt​)+φ(xt​)−ft​(x∗)−φ(x∗)]≤η1​Bψ1​​(x∗,x1​)+η1​t=1∑T−1​[Bψt+1​​(x∗,xt+1​)−Bψt​​(x∗,xt+1​)]+2η​t=1∑T​∥gt​∥ψt∗​2​.

This reduces the regret of composite mirror descent with adaptive proximal functions to two terms: the drift of the Bregman divergences and the dual norms of the subgradients.

Formalization Note Specialised to diagonal quadratic proximal functions; their strong convexity with respect to ∥⋅∥ψt\|\cdot\|_{\psi_t}∥⋅∥ψt​​ is implied by the quadratic form, and no monotonicity of hth_tht​ is needed. The paper's "assume w.l.o.g. that φ(x1)=0\varphi(x_1)=0φ(x1​)=0" is used in its proof as φ(x1)≤φ(xT+1)\varphi(x_1)\le\varphi(x_{T+1})φ(x1​)≤φ(xT+1​); here it is the hypothesis that x1x_1x1​ minimizes φ\varphiφ over X\mathcal XX (without it the inequality fails). The condition "ht,i=0⇒gt,i=0h_{t,i}=0\Rightarrow g_{t,i}=0ht,i​=0⇒gt,i​=0" makes Lean's a/0=0a/0=0a/0=0 agree with the paper's dual norm.

Preamble
import Mathlib
import Definitions.Def_AdaGrad_Diag_Setup
Formal statement
namespace AdaGrad.Diag

/-- Duchi, Hazan, Singer, JMLR 12 (2011), Proposition 3, (11), p. 2129, specialised to diagonal
quadratic proximal functions `ψ_t(y) = ½ ∑_i h_{t,i} y_i²` with weights `h_{t,i} ≥ 0`. Let the
`f_t` and `ϕ` be convex, `X` convex, and let `x_1, x_2, …` be a run of the composite mirror descent
update (4) with subgradients `g_t` of `f_t` at `x_t`, where `h_{t,i} = 0` forces `g_{t,i} = 0`.
Assume `x_1 ∈ X` minimizes `ϕ` over `X` (the paper's "w.l.o.g. `ϕ(x_1) = 0`", see the
Formalization Note). Then for every `T` and every `x* ∈ X`,
`∑_{t=1}^T [f_t(x_t) + ϕ(x_t) − f_t(x*) − ϕ(x*)]
   ≤ (1/η) B_{ψ_1}(x*, x_1) + (1/η) ∑_{t=1}^{T−1} [B_{ψ_{t+1}}(x*, x_{t+1}) − B_{ψ_t}(x*, x_{t+1})]
     + (η/2) ∑_{t=1}^T ‖g_t‖²_{ψ_t*}`. -/
theorem proposition_3 {d : ℕ} (η : ℝ) (hη : 0 < η) (X : Set (EuclideanSpace ℝ (Fin d)))
    (hXc : Convex ℝ X) (hXcl : IsClosed X)
    (ϕ : EuclideanSpace ℝ (Fin d) → ℝ) (hϕ : ConvexOn ℝ Set.univ ϕ)
    (f : ℕ → EuclideanSpace ℝ (Fin d) → ℝ) (hf : ∀ t, ConvexOn ℝ Set.univ (f t))
    (h : ℕ → Fin d → ℝ) (x g : ℕ → EuclideanSpace ℝ (Fin d))
    (hh : ∀ t, 1 ≤ t → ∀ i, 0 ≤ h t i)
    (hhg : ∀ t, 1 ≤ t → ∀ i, h t i = 0 → g t i = 0)
    (hrun : IsCompositeMDRun η X ϕ f h x g)
    (hx1 : x 1 ∈ X) (hϕx1 : ∀ y ∈ X, ϕ (x 1) ≤ ϕ y)
    (T : ℕ) (xstar : EuclideanSpace ℝ (Fin d)) (hxstar : xstar ∈ X) :
    regret f ϕ x xstar T
      ≤ 1 / η * bregman (h 1) xstar (x 1)
        + 1 / η * ∑ t ∈ Finset.Ico 1 T,
            (bregman (h (t + 1)) xstar (x (t + 1)) - bregman (h t) xstar (x (t + 1)))
        + η / 2 * ∑ t ∈ Finset.Icc 1 T, dualNormSq (h t) (g t) := by sorry

end AdaGrad.Diag
Source
Duchi, Hazan, Singer, Adaptive Subgradient Methods for Online Learning and Stochastic Optimization, JMLR 12 (2011), p. 2129, Proposition 3, (11); proof p. 2154
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