Convex smooth gradients are cocoercive
AutoSamplingTheory.TechnicalLemmas.Analysis.ConvexSmoothGradient.gradient_cocoercive · theorem · Teaching coverage
Statement
For a nonnegative real β and a C¹ convex real-valued function on a complete real inner-product space satisfying the global β quadratic upper model, every ordered pair x,y satisfies the division-free cocoercivity inequality.
All objects and hypotheses
- E is a complete real inner-product space, f:E→ℝ, and all points range over the entire space E.
- f is C¹ (ContDiff ℝ 1 f) and convex on E (ConvexOn ℝ Set.univ f). No second derivative is assumed.
- The genuine gradient is the Riesz representative of the Fréchet derivative. The displayed quadratic upper model holds for every x,y.
- β is a nonnegative real number, including zero.
Mathematical proof
1. Use a strictly positive relaxed modulus
Fix x,y and set d=∇f(y)−∇f(x). For ε>0, the β upper model implies the (β+ε) model because squared norm is nonnegative. Thus the positive-modulus Bregman bound applies even when β=0.
Corresponding Lean step
hu' follows by real arithmetic from hu and sq_nonneg; two calls of gradient_gap_sq_le_bregman.
2. Exchange endpoints and add
The values f(x),f(y) cancel. The two Bregman gaps sum to the gradient-displacement pairing, while reversing the gradient difference preserves its squared norm.
Corresponding Lean step
hxy and hyx; norm_sub_rev, inner_neg_right, inner_sub_left and nlinarith.
3. Include the zero boundary by a real limit
Let ε decrease to zero. The right side is affine and continuous in ε, and order is closed in ℝ. This proves the scaled bound for all β≥0. At β=0 the norm square must be zero, so the gradient is constant; no inverse at zero is used.
Corresponding Lean step
Tendsto.add/mul and tendsto_id.mono_left nhdsWithin_le_nhds; ge_of_tendsto with self_mem_nhdsWithin.
Lean statement · gradient_cocoercive
For a nonnegative real β and a C¹ convex real-valued function on a complete real inner-product space satisfying the global β quadratic upper model, every ordered pair x,y satisfies the division-free cocoercivity inequality.
Braces mark parameters Lean can infer; square brackets request structures such as a measurable space or probability measure. Named hypotheses are mathematical premises, not facts established by this declaration. Section parameters are described in the mathematical hypotheses above; the module link retains their exact source context.
theorem gradient_cocoercive {f : E → ℝ} {β : ℝ}
(hf : ContDiff ℝ 1 f) (hc : ConvexOn ℝ univ f) (hβ : 0 ≤ β)
(hu : ∀ x y, f y ≤ f x + inner ℝ (gradient f x) (y - x) + β / 2 * ‖y - x‖ ^ 2)
(x y : E) :
‖gradient f y - gradient f x‖ ^ 2 ≤
β * inner ℝ (gradient f y - gradient f x) (y - x)Lean proof · gradient_cocoercive
Fix x,y and set d=∇f(y)−∇f(x). For ε>0, the β upper model implies the (β+ε) model because squared norm is nonnegative. Thus the positive-modulus Bregman bound applies even when β=0. The values f(x),f(y) cancel. The two Bregman gaps sum to the gradient-displacement pairing, while reversing the gradient difference preserves its squared norm. Let ε decrease to zero. The right side is affine and continuous in ε, and order is closed in ℝ. This proves the scaled bound for all β≥0. At β=0 the norm square must be zero, so the gradient is constant; no inverse at zero is used.
Braces mark parameters Lean can infer; square brackets request structures such as a measurable space or probability measure. Named hypotheses are mathematical premises, not facts established by this declaration. Section parameters are described in the mathematical hypotheses above; the module link retains their exact source context.
theorem gradient_cocoercive {f : E → ℝ} {β : ℝ}
(hf : ContDiff ℝ 1 f) (hc : ConvexOn ℝ univ f) (hβ : 0 ≤ β)
(hu : ∀ x y, f y ≤ f x + inner ℝ (gradient f x) (y - x) + β / 2 * ‖y - x‖ ^ 2)
(x y : E) :
‖gradient f y - gradient f x‖ ^ 2 ≤
β * inner ℝ (gradient f y - gradient f x) (y - x) := by
have he (ε : ℝ) (hε : 0 < ε) :
‖gradient f y - gradient f x‖ ^ 2 ≤
(β + ε) * inner ℝ (gradient f y - gradient f x) (y - x) := by
have hu' (a b : E) : f b ≤ f a + inner ℝ (gradient f a) (b - a) +
(β + ε) / 2 * ‖b - a‖ ^ 2 := by
have := hu a b
nlinarith [sq_nonneg ‖b - a‖]
have hxy := gradient_gap_sq_le_bregman hf hc (add_pos_of_nonneg_of_pos hβ hε) hu' x y
have hyx := gradient_gap_sq_le_bregman hf hc (add_pos_of_nonneg_of_pos hβ hε) hu' y x
rw [norm_sub_rev (gradient f x), show x - y = -(y - x) by abel, inner_neg_right] at hyx
rw [inner_sub_left]
nlinarith
have ht : Tendsto (fun ε : ℝ => (β + ε) *
inner ℝ (gradient f y - gradient f x) (y - x)) (𝓝[>] 0)
(𝓝 (β * inner ℝ (gradient f y - gradient f x) (y - x))) := by
simpa using (tendsto_const_nhds.add (tendsto_id.mono_left nhdsWithin_le_nhds :
Tendsto (fun ε : ℝ => ε) (𝓝[>] 0) (𝓝 0))).mul tendsto_const_nhds
apply ge_of_tendsto ht
filter_upwards [self_mem_nhdsWithin] with ε hε
exact he ε hε
/-- A C¹ convex function with the global quadratic upper model has a
`β`-Lipschitz gradient, including `β = 0`. -/Scope and omitted-condition boundaries
- The complete real inner-product space explicitly generalizes the source Euclidean space. A constrained set is not covered: the trial gradient step must remain in the domain. No Riemannian, stochastic, or companion-paper theorem is claimed.
- Definition1.12 permits β≥0, but the reciprocal in (3.4) and the suggested step 1/β require β>0; the source does not state that restriction explicitly. The first declaration covers this positive domain. The scaled (3.5) and Lipschitz conclusions are separately established at β=0.
Source and reuse
ASTIS parents called
Mathlib API called (external library)
- norm_sub_rev; inner_neg_right; inner_sub_left
- Tendsto.add; Tendsto.mul; tendsto_id; nhdsWithin_le_nhds; ge_of_tendsto; self_mem_nhdsWithin
Mathematical sources
- Sinho Chewi, Exercise3.1 (3.4)-(3.5) — ASTIS-authored explanation of the tangent-plane descent route and explicit zero-modulus extension.
- Pinned Optlib convex_to_lower / lower_to_lipschitz — External mathematical reuse provenance, Apache-2.0; not an imported compiled dependency.
- Reciprocal formula, zero modulus and nonexpansive gradient-step consumers — Compiled use of the actual declarations.
ASTIS prose is not a quotation or a source-equivalence certificate. Definitions and aliases are explained as constructions, not counted as new mathematical proofs.