Measure with a given density with respect to another measure #
For a measure μ
on α
and a function f : α → ℝ≥0∞
, we define a new measure μ.withDensity f
.
On a measurable set s
, that measure has value ∫⁻ a in s, f a ∂μ
.
An important result about withDensity
is the Radon-Nikodym theorem. It states that, given measures
μ, ν
, if HaveLebesgueDecomposition μ ν
then μ
is absolutely continuous with respect to
ν
if and only if there exists a measurable function f : α → ℝ≥0∞
such that
μ = ν.withDensity f
.
See MeasureTheory.Measure.absolutelyContinuous_iff_withDensity_rnDeriv_eq
.
Given a measure μ : Measure α
and a function f : α → ℝ≥0∞
, μ.withDensity f
is the
measure such that for a measurable set s
we have μ.withDensity f s = ∫⁻ a in s, f a ∂μ
.
Equations
- MeasureTheory.Measure.withDensity μ f = MeasureTheory.Measure.ofMeasurable (fun (s : Set α) (x : MeasurableSet s) => ∫⁻ (a : α) in s, f a ∂μ) ⋯ ⋯
Instances For
In the next theorem, the s-finiteness assumption is necessary. Here is a counterexample
without this assumption. Let α
be an uncountable space, let x₀
be some fixed point, and consider
the σ-algebra made of those sets which are countable and do not contain x₀
, and of their
complements. This is the σ-algebra generated by the sets {x}
for x ≠ x₀
. Define a measure equal
to +∞
on nonempty sets. Let s = {x₀}
and f
the indicator of sᶜ
. Then
∫⁻ a in s, f a ∂μ = 0
. Indeed, consider a simple functiong ≤ f
. It vanishes ons
. Then∫⁻ a in s, g a ∂μ = 0
. Taking the supremum overg
gives the claim.μ.withDensity f s = +∞
. Indeed, this is the infimum ofμ.withDensity f t
over measurable setst
containings
. Ass
is not measurable, such a sett
contains a pointx ≠ x₀
. Thenμ.withDensity f t ≥ μ.withDensity f {x} = ∫⁻ a in {x}, f a ∂μ = μ {x} = +∞
. One checks thatμ.withDensity f = μ
, whileμ.restrict s
gives zero mass to sets not containingx₀
, and infinite mass to those that contain it.
This is Exercise 1.2.1 from [tao2010]. It allows you to express integration of a measurable
function with respect to (μ.withDensity f)
as an integral with respect to μ
, called the base
measure. μ
is often the Lebesgue measure, and in this circumstance f
is the probability density
function, and (μ.withDensity f)
represents any continuous random variable as a
probability measure, such as the uniform distribution between 0 and 1, the Gaussian distribution,
the exponential distribution, the Beta distribution, or the Cauchy distribution (see Section 2.4
of [wasserman2004]). Thus, this method shows how to one can calculate expectations, variances,
and other moments as a function of the probability density function.
The Lebesgue integral of g
with respect to the measure μ.withDensity f
coincides with
the integral of f * g
. This version assumes that g
is almost everywhere measurable. For a
version without conditions on g
but requiring that f
is almost everywhere finite, see
lintegral_withDensity_eq_lintegral_mul_non_measurable
If f
is almost everywhere positive and finite, then μ ≪ μ.withDensity f
. See also
withDensity_absolutelyContinuous
for the reverse direction, which always holds.
A sigma-finite measure is absolutely continuous with respect to some finite measure.