subadditive_f_generator_hasDerivAt
For s continuous on (0, ∞) and y > 0, the generator f(x) = α x log x + ∫_1^x (x - t) s(t)/t² dt has derivative f'(y) = α (log y + 1) + ∫_1^y s(t)/t² dt.
Statement
theorem subadditive_f_generator_hasDerivAt (α : ℝ) (s : ℝ → ℝ) (hs : ContinuousOn s (Set.Ioi 0)) {y : ℝ} (hy : 0 < y) : HasDerivAt (subadditive_f_generator α s) (α * (Real.log y + 1) + ∫ t in (1 : ℝ)..y, s t / t ^ 2) y
- Source
- helper for arXiv:2610.21231 (On the subadditivity of mutual f-information), Corollary 1 (equation (4))
- Verified
- 23 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 1
- subadditive_f_generatorThe generator function f associated to parameter α and function s is f(x) = α x log x + ∫_1^x (x - t) (s(t) / t^2) dt.
- Used by 3
- subadditive_f_generator_contDiffOnIf s is twice continuously differentiable on (0, ∞), then the generator f of equation (4) is four times continuously differentiable on (0,…
- subadditive_f_generator_satisfies_weight_boundFirst claim of Corollary 1 of arXiv:2610.21231: let s be twice continuously differentiable and positive on (0, ∞) with s'' < 0, let K_s(x)…
- subadditive_f_generator_deriv2For s continuous on (0, ∞) and y > 0, the derivative of the generator f of equation (4) has derivative α/y + s(y)/y² at y, i.e. f''(y) =…
The proof, explained
To compute the derivative of $f(x) = \alpha x \log x + \int_1^x (x - t) \frac{s(t)}{t^2}\,dt$ at $y > 0$, the proof algebraically rewrites the integrand as $x \frac{s(t)}{t^2} - \frac{t s(t)}{t^2}$, expressing $f(x)$ locally on $(0, \infty)$ as $\alpha x \log x + x \int_1^x \frac{s(t)}{t^2}\,dt - \int_1^x \frac{t s(t)}{t^2}\,dt$. Applying the Fundamental Theorem of Calculus together with the product and difference rules differentiates the integral terms into $\int_1^y \frac{s(t)}{t^2}\,dt$, as the boundary terms $y \frac{s(y)}{y^2}$ cancel. Differentiating $\alpha x \log x$ via the product rule produces $\alpha (\log y + 1)$, and combining the two parts completes the proof.
The set $(0, \infty)$ is open, and both $t \mapsto s(t)/t^2$ and $t \mapsto t s(t)/t^2$ are continuous on $(0, \infty)$.
The interval $(0, \infty)$ is open by `isOpen_Ioi`. Since $s$ is continuous on $(0, \infty)$ and $t^2 > 0$ (and thus $t^2 \neq 0$) for every $t > 0$, the quotient rule for continuity (`ContinuousOn.div`) implies that $t \mapsto s(t)/t^2$ is continuous on $(0, \infty)$. Similarly, multiplying by the identity function preserves continuity, so $t \mapsto t s(t)/t^2$ is also continuous on $(0, \infty)$.
uses
isOpen_Ioi,ContinuousOn.div,Set.mem_Ioi,ne_of_gt,pow_poshave ho : IsOpen (Set.Ioi (0 : ℝ)) := isOpen_Ioi have hcs1 : ContinuousOn (fun t => s t / t ^ 2) (Set.Ioi 0) := by apply ContinuousOn.div hs (continuousOn_id.pow 2) intro t ht; rw [Set.mem_Ioi] at ht; exact ne_of_gt (pow_pos ht 2) have hcs2 : ContinuousOn (fun t => t * s t / t ^ 2) (Set.Ioi 0) := by apply ContinuousOn.div (continuousOn_id.mul hs) (continuousOn_id.pow 2) intro t ht; rw [Set.mem_Ioi] at ht; exact ne_of_gt (pow_pos ht 2)
For any function $f$ continuous on $(0, \infty)$, the integral $u \mapsto \int_1^u f(t)\,dt$ has derivative $f(y)$ at $y$.
Let $f$ be continuous on $(0, \infty)$. Since $y > 0$, the unordered closed interval between $1$ and $y$ is entirely contained in $(0, \infty)$, so $f$ is continuous on it and therefore interval integrable. Because $(0, \infty)$ is open, continuity on $(0, \infty)$ implies continuity at each of its points, in particular at $y$, and also implies strong measurability of $f$ near $y$ (`ContinuousAt.stronglyMeasurableAtFilter`). The Fundamental Theorem of Calculus (`intervalIntegral.integral_hasDerivAt_right`) then guarantees that the derivative of $u \mapsto \int_1^u f(t)\,dt$ at $y$ is $f(y)$.
uses
Set.mem_Ioi,Set.mem_uIcc,ContinuousAt.stronglyMeasurableAtFilter,intervalIntegral.integral_hasDerivAt_righthave h_deriv_integral : ∀ (f : ℝ → ℝ), ContinuousOn f (Set.Ioi 0) → HasDerivAt (fun u => ∫ t in (1 : ℝ)..u, f t) (f y) y := by intro f hf have hsub : Set.uIcc 1 y ⊆ Set.Ioi 0 := by intro t ht rw [Set.mem_Ioi] rw [Set.mem_uIcc] at ht rcases ht with ⟨h1, h2⟩ | ⟨h1, h2⟩ <;> linarith have hint : IntervalIntegrable f MeasureTheory.volume 1 y := (hf.mono hsub).intervalIntegrable have hfc : ∀ x ∈ Set.Ioi (0 : ℝ), ContinuousAt f x := fun x hx => hf.continuousAt (ho.mem_nhds hx) have hmeas : StronglyMeasurableAtFilter f (nhds y) MeasureTheory.volume := ContinuousAt.stronglyMeasurableAtFilter ho hfc y hy have hconty : ContinuousAt f y := hfc y hy exact intervalIntegral.integral_hasDerivAt_right hint hmeas hconty
The functions $x \mapsto \int_1^x s(t)/t^2\,dt$ and $x \mapsto \int_1^x t s(t)/t^2\,dt$ have derivatives $s(y)/y^2$ and $y s(y)/y^2$ at $y$, respectively.
This specializes the Fundamental Theorem of Calculus established in the previous step to the two integrands $t \mapsto s(t)/t^2$ and $t \mapsto t s(t)/t^2$, which are both continuous on $(0, \infty)$.
have hf1 : HasDerivAt (fun x => ∫ t in (1 : ℝ)..x, s t / t ^ 2) (s y / y ^ 2) y := h_deriv_integral (fun t => s t / t ^ 2) hcs1 have hf2 : HasDerivAt (fun x => ∫ t in (1 : ℝ)..x, t * s t / t ^ 2) (y * s y / y ^ 2) y := h_deriv_integral (fun t => t * s t / t ^ 2) hcs2
The function $x \mapsto \alpha x \log x$ has derivative $\alpha (\log y + 1)$ at $y$.
Because $y > 0$, $y \neq 0$. Differentiating $\log x$ gives $y^{-1}$ (`Real.hasDerivAt_log`), and differentiating the identity gives $1$ (`hasDerivAt_id`). By the product rule, the derivative of $x \log x$ is $1 \cdot \log y + y \cdot y^{-1} = \log y + 1$, where $y \cdot y^{-1} = 1$ by `mul_inv_cancel₀`. Scaling by the constant $\alpha$ yields derivative $\alpha (\log y + 1)$.
uses
ne_of_gt,Real.hasDerivAt_log,hasDerivAt_id,mul_inv_cancel₀,one_mul,mul_assochave hy0 : y ≠ 0 := ne_of_gt hy have hlog : HasDerivAt Real.log y⁻¹ y := Real.hasDerivAt_log hy0 have hid : HasDerivAt (fun x => x) 1 y := hasDerivAt_id y have hmul : HasDerivAt (fun x => x * Real.log x) (1 * Real.log y + y * y⁻¹) y := hid.mul hlog rw [mul_inv_cancel₀ hy0, one_mul] at hmul have hα : HasDerivAt (fun x => α * x * Real.log x) (α * (Real.log y + 1)) y := by simpa only [mul_assoc] using hmul.const_mul α
The combination $x \mapsto x \int_1^x \frac{s(t)}{t^2}\,dt - \int_1^x \frac{t s(t)}{t^2}\,dt$ has derivative $\int_1^y \frac{s(t)}{t^2}\,dt$ at $y$.
Applying the product rule to $x \cdot \int_1^x \frac{s(t)}{t^2}\,dt$ gives derivative $1 \cdot \int_1^y \frac{s(t)}{t^2}\,dt + y \frac{s(y)}{y^2}$. Subtracting the derivative of $\int_1^x \frac{t s(t)}{t^2}\,dt$, which is $y \frac{s(y)}{y^2}$, the terms $y \frac{s(y)}{y^2}$ cancel out by elementary ring algebra, leaving $\int_1^y \frac{s(t)}{t^2}\,dt$.
have hprod : HasDerivAt (fun x => x * (∫ t in (1 : ℝ)..x, s t / t ^ 2)) (1 * (∫ t in (1 : ℝ)..y, s t / t ^ 2) + y * (s y / y ^ 2)) y := hid.mul hf1 have hsub_deriv : HasDerivAt (fun x => x * (∫ t in (1 : ℝ)..x, s t / t ^ 2) - ∫ t in (1 : ℝ)..x, t * s t / t ^ 2) (∫ t in (1 : ℝ)..y, s t / t ^ 2) y := by have h := hprod.sub hf2 have hr : (1 * (∫ t in (1 : ℝ)..y, s t / t ^ 2) + y * (s y / y ^ 2)) - (y * s y / y ^ 2) = ∫ t in (1 : ℝ)..y, s t / t ^ 2 := by ring rw [hr] at h exact h
The sum $x \mapsto \alpha x \log x + \big(x \int_1^x \frac{s(t)}{t^2}\,dt - \int_1^x \frac{t s(t)}{t^2}\,dt\big)$ has derivative $\alpha (\log y + 1) + \int_1^y \frac{s(t)}{t^2}\,dt$ at $y$.
Adding the derivatives obtained in the previous two steps yields the derivative of the sum.
have htot : HasDerivAt (fun x => α * x * Real.log x + (x * (∫ t in (1 : ℝ)..x, s t / t ^ 2) - ∫ t in (1 : ℝ)..x, t * s t / t ^ 2)) (α * (Real.log y + 1) + ∫ t in (1 : ℝ)..y, s t / t ^ 2) y := hα.add hsub_deriv
Near $y$, the generator function $\text{subadditive\_f\_generator } \alpha\ s$ is identical to $x \mapsto \alpha x \log x + x \int_1^x \frac{s(t)}{t^2}\,dt - \int_1^x \frac{t s(t)}{t^2}\,dt$.
Since $(0, \infty)$ is a neighborhood of $y$, it suffices to show equality for all $x > 0$. The $\alpha x \log x$ terms match by definition. For the integral term, since $x > 0$, the interval between $1$ and $x$ lies in $(0, \infty)$, making both $t \mapsto x s(t)/t^2$ and $t \mapsto t s(t)/t^2$ interval integrable. Expanding $(x - t) s(t)/t^2 = x s(t)/t^2 - t s(t)/t^2$, splitting the integral (`intervalIntegral.integral_sub`), and factoring out the constant $x$ (`intervalIntegral.integral_const_mul`) shows the two expressions agree on $(0, \infty)$.
uses
Filter.eventuallyEq_iff_exists_mem.mpr,Set.mem_Ioi,Set.mem_uIcc,intervalIntegral.integral_sub,intervalIntegral.integral_const_mulhave heq : subadditive_f_generator α s =ᶠ[nhds y] fun x => α * x * Real.log x + (x * (∫ t in (1 : ℝ)..x, s t / t ^ 2) - ∫ t in (1 : ℝ)..x, t * s t / t ^ 2) := by apply Filter.eventuallyEq_iff_exists_mem.mpr refine ⟨Set.Ioi 0, ho.mem_nhds hy, fun x hx => ?_⟩ dsimp [subadditive_f_generator] congr 1 have hxpos : 0 < x := hx have hsubx : Set.uIcc 1 x ⊆ Set.Ioi 0 := by intro t ht rw [Set.mem_Ioi] rw [Set.mem_uIcc] at ht rcases ht with ⟨h1, h2⟩ | ⟨h1, h2⟩ <;> linarith have h1 : IntervalIntegrable (fun t => x * (s t / t ^ 2)) MeasureTheory.volume 1 x := ((continuousOn_const.mul hcs1).mono hsubx).intervalIntegrable have h2 : IntervalIntegrable (fun t => t * s t / t ^ 2) MeasureTheory.volume 1 x := (hcs2.mono hsubx).intervalIntegrable have h_integrand : (fun t => (x - t) * s t / t ^ 2) = fun t => x * (s t / t ^ 2) - t * s t / t ^ 2 := by ext t; ring rw [h_integrand] rw [intervalIntegral.integral_sub h1 h2] rw [intervalIntegral.integral_const_mul]
The generator $\text{subadditive\_f\_generator } \alpha\ s$ has derivative $\alpha (\log y + 1) + \int_1^y \frac{s(t)}{t^2}\,dt$ at $y$.
Because `subadditive_f_generator` coincides with the split expression on a neighborhood of $y$, its derivative at $y$ is equal to the derivative computed for that expression.
exact htot.congr_of_eventuallyEq heq
isOpen_Ioi- The interval $(a, \infty)$ is open in the order topology.
ContinuousOn.div- The quotient of two functions continuous on a set $s$ is continuous on $s$, provided the denominator does not vanish on $s$.
Set.mem_Ioi- An element $x$ belongs to $(b, \infty)$ if and only if $b < x$.
ne_of_gt- If $a > b$, then $a \neq b$.
pow_pos- If $a > 0$, then every positive integer power $a^n > 0$.
Set.mem_uIcc- An element lies in the unordered interval $[a, b]^*$ if and only if it lies between $a$ and $b$ (either $a \le x \le b$ or $b \le x \le a$).
ContinuousAt.stronglyMeasurableAtFilter- A function continuous on an open set $s$ is strongly measurable at the neighborhood filter of every point in $s$.
intervalIntegral.integral_hasDerivAt_right- Fundamental Theorem of Calculus: if $f$ is interval integrable on $[a, b]$, strongly measurable near $b$, and continuous at $b$, then $u \mapsto \int_a^u f(t)\,dt$ has derivative $f(b)$ at $u = b$.
Real.hasDerivAt_log- The natural logarithm function has derivative $x^{-1}$ at any non-zero real number $x$.
hasDerivAt_id- The identity function has derivative $1$ everywhere.
mul_inv_cancel₀- For any non-zero element $a$, $a \cdot a^{-1} = 1$.
one_mul- Multiplying by $1$ on the left leaves an element unchanged.
mul_assoc- Multiplication is associative: $(a \cdot b) \cdot c = a \cdot (b \cdot c)$.
Filter.eventuallyEq_iff_exists_mem.mpr- Two functions are eventually equal along the neighborhood filter of $y$ if they agree on some neighborhood of $y$.
intervalIntegral.integral_sub- The integral of the difference of two integrable functions is the difference of their integrals.
intervalIntegral.integral_const_mul- A constant factor can be pulled out of an interval integral.
- The splitting of the integral $\int_1^x (x-t)\frac{s(t)}{t^2}\,dt = x \int_1^x \frac{s(t)}{t^2}\,dt - \int_1^x \frac{t s(t)}{t^2}\,dt$ is proved only on the open neighborhood $(0, \infty)$ of $y$ to avoid dividing by zero. The derivative is then transferred to `subadditive_f_generator` via local equality near $y$ (`congr_of_eventuallyEq`).
- Applying the Fundamental Theorem of Calculus in Mathlib (`intervalIntegral.integral_hasDerivAt_right`) requires checking strong measurability at the filter `nhds y`, which Lean derives from continuity on the open set $(0, \infty)$.
Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
theorem subadditive_f_generator_hasDerivAt (α : ℝ) (s : ℝ → ℝ) (hs : ContinuousOn s (Set.Ioi 0)) {y : ℝ} (hy : 0 < y) : HasDerivAt (subadditive_f_generator α s) (α * (Real.log y + 1) + ∫ t in (1 : ℝ)..y, s t / t ^ 2) y := by have ho : IsOpen (Set.Ioi (0 : ℝ)) := isOpen_Ioi have hcs1 : ContinuousOn (fun t => s t / t ^ 2) (Set.Ioi 0) := by apply ContinuousOn.div hs (continuousOn_id.pow 2) intro t ht; rw [Set.mem_Ioi] at ht; exact ne_of_gt (pow_pos ht 2) have hcs2 : ContinuousOn (fun t => t * s t / t ^ 2) (Set.Ioi 0) := by apply ContinuousOn.div (continuousOn_id.mul hs) (continuousOn_id.pow 2) intro t ht; rw [Set.mem_Ioi] at ht; exact ne_of_gt (pow_pos ht 2) have h_deriv_integral : ∀ (f : ℝ → ℝ), ContinuousOn f (Set.Ioi 0) → HasDerivAt (fun u => ∫ t in (1 : ℝ)..u, f t) (f y) y := by intro f hf have hsub : Set.uIcc 1 y ⊆ Set.Ioi 0 := by intro t ht rw [Set.mem_Ioi] rw [Set.mem_uIcc] at ht rcases ht with ⟨h1, h2⟩ | ⟨h1, h2⟩ <;> linarith have hint : IntervalIntegrable f MeasureTheory.volume 1 y := (hf.mono hsub).intervalIntegrable have hfc : ∀ x ∈ Set.Ioi (0 : ℝ), ContinuousAt f x := fun x hx => hf.continuousAt (ho.mem_nhds hx) have hmeas : StronglyMeasurableAtFilter f (nhds y) MeasureTheory.volume := ContinuousAt.stronglyMeasurableAtFilter ho hfc y hy have hconty : ContinuousAt f y := hfc y hy exact intervalIntegral.integral_hasDerivAt_right hint hmeas hconty have hf1 : HasDerivAt (fun x => ∫ t in (1 : ℝ)..x, s t / t ^ 2) (s y / y ^ 2) y := h_deriv_integral (fun t => s t / t ^ 2) hcs1 have hf2 : HasDerivAt (fun x => ∫ t in (1 : ℝ)..x, t * s t / t ^ 2) (y * s y / y ^ 2) y := h_deriv_integral (fun t => t * s t / t ^ 2) hcs2 have hy0 : y ≠ 0 := ne_of_gt hy have hlog : HasDerivAt Real.log y⁻¹ y := Real.hasDerivAt_log hy0 have hid : HasDerivAt (fun x => x) 1 y := hasDerivAt_id y have hmul : HasDerivAt (fun x => x * Real.log x) (1 * Real.log y + y * y⁻¹) y := hid.mul hlog rw [mul_inv_cancel₀ hy0, one_mul] at hmul have hα : HasDerivAt (fun x => α * x * Real.log x) (α * (Real.log y + 1)) y := by simpa only [mul_assoc] using hmul.const_mul α have hprod : HasDerivAt (fun x => x * (∫ t in (1 : ℝ)..x, s t / t ^ 2)) (1 * (∫ t in (1 : ℝ)..y, s t / t ^ 2) + y * (s y / y ^ 2)) y := hid.mul hf1 have hsub_deriv : HasDerivAt (fun x => x * (∫ t in (1 : ℝ)..x, s t / t ^ 2) - ∫ t in (1 : ℝ)..x, t * s t / t ^ 2) (∫ t in (1 : ℝ)..y, s t / t ^ 2) y := by have h := hprod.sub hf2 have hr : (1 * (∫ t in (1 : ℝ)..y, s t / t ^ 2) + y * (s y / y ^ 2)) - (y * s y / y ^ 2) = ∫ t in (1 : ℝ)..y, s t / t ^ 2 := by ring rw [hr] at h exact h have htot : HasDerivAt (fun x => α * x * Real.log x + (x * (∫ t in (1 : ℝ)..x, s t / t ^ 2) - ∫ t in (1 : ℝ)..x, t * s t / t ^ 2)) (α * (Real.log y + 1) + ∫ t in (1 : ℝ)..y, s t / t ^ 2) y := hα.add hsub_deriv have heq : subadditive_f_generator α s =ᶠ[nhds y] fun x => α * x * Real.log x + (x * (∫ t in (1 : ℝ)..x, s t / t ^ 2) - ∫ t in (1 : ℝ)..x, t * s t / t ^ 2) := by apply Filter.eventuallyEq_iff_exists_mem.mpr refine ⟨Set.Ioi 0, ho.mem_nhds hy, fun x hx => ?_⟩ dsimp [subadditive_f_generator] congr 1 have hxpos : 0 < x := hx have hsubx : Set.uIcc 1 x ⊆ Set.Ioi 0 := by intro t ht rw [Set.mem_Ioi] rw [Set.mem_uIcc] at ht rcases ht with ⟨h1, h2⟩ | ⟨h1, h2⟩ <;> linarith have h1 : IntervalIntegrable (fun t => x * (s t / t ^ 2)) MeasureTheory.volume 1 x := ((continuousOn_const.mul hcs1).mono hsubx).intervalIntegrable have h2 : IntervalIntegrable (fun t => t * s t / t ^ 2) MeasureTheory.volume 1 x := (hcs2.mono hsubx).intervalIntegrable have h_integrand : (fun t => (x - t) * s t / t ^ 2) = fun t => x * (s t / t ^ 2) - t * s t / t ^ 2 := by ext t; ring rw [h_integrand] rw [intervalIntegral.integral_sub h1 h2] rw [intervalIntegral.integral_const_mul] exact htot.congr_of_eventuallyEq heq