Convolution of Gaussians:
ProvedStatUnlearn.gaussianReal_conv_gaussianRealFor real means and variances , the convolution of the two Gaussians on the line is Gaussian with the parameters added:
In other words, the Gaussian family is closed under convolution: adding independent Gaussian noise to a Gaussian observation yields another Gaussian, with the variances adding. The proof identifies both sides by their characteristic functions — , the characteristic function of a convolution is the product of the characteristic functions, and two finite Borel measures on the line with the same characteristic function coincide — after which the identity reduces to the exponential law .
This is the algebraic fact that makes "add independent noise" a Markov kernel on the Gaussian baselines. It is what converts the data-processing inequality for trade-off functions into monotonicity in the shift parameter: from and one can reach a pair of Gaussians with a smaller separation, never a larger one.
import Mathlib.Probability.Distributions.Gaussian.Real import Mathlib.MeasureTheory.Measure.CharacteristicFunction open MeasureTheory ProbabilityTheory
namespace StatUnlearn
theorem gaussianReal_conv_gaussianReal (μ ν : ℝ) (v w : NNReal) :
(gaussianReal μ v).conv (gaussianReal ν w) = gaussianReal (μ + ν) (v + w) := by
sorry
end StatUnlearn