Differentiation of measures

2010 Mathematics Subject Classification: Primary: 28A15 Secondary: 49Q15 [MSN][ZBL]

Some authors use this name for the outcome of the Radon-Nikodym theorem or for the density of the Radon-Nykodim decomposition (see for instance Section 32 of [Ha]).

Other authors use the name for the following theorem which gives an explicit characterization of the Radon-Nykodim decomposition for locally finite Radon measures on the euclidean space. This theorem is used often in Geometric measure theory and credited to Besicovitch.

Theorem (cp. with Theorem 2.12 of [Ma] and Theorem 2 in Section 1.6 of [EG]) Let μμ and νν be two locally finite Radon measures on RnRn. Then,

  • the limit

f(x):=limr↓0ν(Br(x))μ(Br(x))f(x):=limr↓0ν(Br(x))μ(Br(x))exists at μμ-a.e. xx and defines a μμ-measurable map;

  • the set

S:={x:limr↓0ν(Br(x))μ(Br(x))=∞}(1)(1)S:={x:limr↓0ν(Br(x))μ(Br(x))=∞}is νν-measurable and a μμ-null set;

  • νν can be decomposed as νa+νsνa+νs, where

νa(E)=∫Efdμνa(E)=∫Efdμandνs(E)=ν(S∩E).νs(E)=ν(S∩E).Moreover, for μμ-a.e. xx we have:limr↓01μ(Br(x))∫Br(x)|f(y)−f(x)|dμ(y)=0andlimr↓0νs(Br(x))μ(Br(x))=0.(2)(2)limr↓01μ(Br(x))∫Br(x)|f(y)−f(x)|dμ(y)=0andlimr↓0νs(Br(x))μ(Br(x))=0.


The first identity in 22 relates to the concept of Lebesgue point.

The theorem can be generalized to signed measures νν and measures taking values in a finite-dimensional Banach space VV. In that case:

  • ∥ν(Br(x))∥V‖ν(Br(x))‖V substitutes ν(Br(x))ν(Br(x)) in 11;
  • ∥f(y)−f(x)∥V‖f(y)−f(x)‖V substitutes the integrand |f(y)−f(x)||f(y)−f(x)| in 22;
  • |ν|(Br(x))|ν|(Br(x)) substitutes ν(Br(x))ν(Br(x)) in 22, where |ν||ν| denotes the total variation of νν (see Signed measure for the relevant definition).

The theorem does not hold in general metric spaces. It holds provided the metric space satisfies some properties about covering of sets with balls,


“Rectifiable sets, densities and tangent measures
Order Now on customessaymasters.com