integral_exp_mul_sq_gaussianReal_zero
theoremverifiedimported Lean
Quadratic exponential integral against a centered Gaussian. Let $v > 0$ and write $N(0,v)$ for the centered Gaussian distribution on $\bbr$ with mean $0$ and variance $v$. Then for every real number $t$ with $2tv < 1$, \[ \int_{\bbr} e^{t y^2}\, \mathrm{d}N(0,v)(y) = \frac{1}{\sqrt{1 - 2tv}}. \]
Statement
set_option relaxedAutoImplicit false in set_option autoImplicit false in open MeasureTheory ProbabilityTheory Real NNReal Matrix Finset in variable {d k : ℕ} in lemma integral_exp_mul_sq_gaussianReal_zero (v : ℝ≥0) (hv : v ≠ 0) (t : ℝ) (ht : 2 * t * (v : ℝ) < 1) : ∫ y, Real.exp (t * y ^ 2) ∂(gaussianReal 0 v) = 1 / Real.sqrt (1 - 2 * t * (v : ℝ))
- Source
- TCSlib, `integral_exp_mul_sq_gaussianReal_zero`
- Lean proof
- By Ganesh Sankar, from github.com/Shilun-Allan-Li/tcslib at
c8a02591b1e5(Copyright (c) 2026 Ganesh Sankar. All rights reserved, Apache-2.0). Compiled here against leanprover/lean4:v4.33.1, Mathlib v4.33.1; written for leanprover/lean4:v4.25.0.integral_exp_mul_sq_gaussianReal_zero← integral_exp_mul_sq_gaussianReal_zero (TCSlib/LearningTheory/JohnsonLindenstrauss/ChiSquaredMGF.lean:267): verbatim
- Verified
- 07 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 1
- integral_exp_mul_sq_standardGaussianQuadratic Gaussian integral. Let $Z$ be a standard Gaussian random variable, $Z \sim N(0,1)$, and let $s \in \bbr$ satisfy $2s < 1$. Then…
- Used by 1
- centered_chi_squared_stepCentered chi-squared MGF bound. Let $(\Omega,\mu)$ be a probability space and let $k$ be a positive natural number. Suppose…
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.
set_option relaxedAutoImplicit false inset_option autoImplicit false inopen MeasureTheory ProbabilityTheory Real NNReal Matrix Finset invariable {d k : ℕ} in/-- **Gaussian quadratic MGF closed form** for general variance. For `Y ~ N(0, v)` with `v ≠ 0` and every real `t` with `2tv < 1`, the integral of `exp(t y²)` against the law `N(0, v)` equals `1/√(1 − 2tv)`. This is the scaled form of `E exp(sX²) = (1 − 2s)^{−1/2}` in [DG03, proof of Lemma 2.2]. **Proof sketch.** Step 1: `N(0, v)` is the pushforward of `N(0, 1)` under multiplication by `√v` (Mathlib's `gaussianReal_map_const_mul`). Step 2: change variables in the integral, so the integrand becomes `exp((tv) z²)` against `N(0, 1)`. Step 3: apply the standard-Gaussian closed form `integral_exp_mul_sq_standardGaussian` with `s = tv` (admissible since `2tv < 1`) and match the expression. -/lemma integral_exp_mul_sq_gaussianReal_zero (v : ℝ≥0) (hv : v ≠ 0) (t : ℝ) (ht : 2 * t * (v : ℝ) < 1) : ∫ y, Real.exp (t * y ^ 2) ∂(gaussianReal 0 v) = 1 / Real.sqrt (1 - 2 * t * (v : ℝ)) := by -- Step 1: push forward N(0,1) by multiplication by √v to get N(0,v). have hv_pos : (0 : ℝ) < (v : ℝ) := NNReal.coe_pos.mpr (pos_iff_ne_zero.mpr hv) have hv_nonneg : (0 : ℝ) ≤ (v : ℝ) := hv_pos.le have h_sqrt_sq : Real.sqrt (v : ℝ) ^ 2 = (v : ℝ) := Real.sq_sqrt hv_nonneg -- Identity: (gaussianReal 0 1).map (√v * ·) = gaussianReal 0 v. have h_map : (gaussianReal 0 1).map (fun z => Real.sqrt (v : ℝ) * z) = gaussianReal 0 v := by have := gaussianReal_map_const_mul (μ := 0) (v := (1 : ℝ≥0)) (Real.sqrt (v : ℝ)) simp only [mul_zero] at this rw [this] congr 1 rw [mul_one] apply NNReal.coe_injective simp [h_sqrt_sq] -- Step 2: reduce the integral to one against gaussianReal 0 1. rw [← h_map] rw [integral_map] · -- Now integral is ∫ z, exp(t * (√v * z)²) ∂(gaussianReal 0 1). have h_eq : ∀ z : ℝ, Real.exp (t * (Real.sqrt (v : ℝ) * z) ^ 2) = Real.exp ((t * (v : ℝ)) * z ^ 2) := by intro z congr 1 rw [mul_pow, h_sqrt_sq] ring simp_rw [h_eq] -- Step 3: apply variance-1 lemma with s = t * v. have hs : 2 * (t * (v : ℝ)) < 1 := by have : 2 * t * (v : ℝ) = 2 * (t * (v : ℝ)) := by ring linarith [ht, this] rw [integral_exp_mul_sq_standardGaussian (t * (v : ℝ)) hs] congr 2 ring · exact (measurable_const.mul measurable_id).aemeasurable · -- AEStronglyMeasurable of fun y => exp (t * y^2) under the pushforward apply Measurable.aestronglyMeasurable exact (measurable_const.mul (measurable_id.pow_const _)).exp