logo

Proof of Fatou's Lemma 📂Measure Theory

Proof of Fatou's Lemma

Theorem 1

For a sequence {fn}\left\{ f_{n} \right\} of non-negative measurable functions, E(lim infnfn)dmlim infnEfndm \int_{E} \left( \liminf_{n \to \infty} f_{n} \right) dm \le \liminf_{n \to \infty} \int_{E} f_{n} dm


Explanation

A lemma necessary for proving the Monotone Convergence Theorem and the Dominated Convergence Theorem in real analysis. The version of Fatou’s lemma for series without the condition of being measurable functions is as follows.

Series Version: For a sequence {fk:N[0,)}kN\left\{ f_{k} : \mathbb{N} \to [0, \infty) \right\}_{k \in \mathbb{N}} of functions with non-negative values j=1lim infkfk(j)lim infkj=1fk(j),jN \sum_{j=1}^{\infty} \liminf_{k \to \infty} f_{k} (j) \le \liminf_{k \to \infty} \sum_{j=1}^{\infty} f_{k} (j) \qquad , \forall j \in \mathbb{N}

Proof

Strategy: It helps to have an intuition for simple functions. As with all lemmas, they are useful but the proof is quite long and complex, so it is best read when in a clear and healthy state of mind. The proof for the series case is essentially the same.


Part 1.

If we have f:=lim infnfngn:=infknfk f : = \liminf_{n \to \infty} f_{n} \\ \displaystyle g_{n} : = \inf_{k \ge n } f_{k} then f=limngn\displaystyle f = \lim_{n \to \infty} g_{n}. Efdmlim infnEfndm \int_{E} f dm \le \liminf_{n \to \infty} \int_{E} f_{n} dm To show this, it’s sufficient to prove for all simple functions ϕf\phi \le f that Eϕdmlim infnEfndm \int_{E} \phi dm \le \liminf_{n \to \infty} \int_{E} f_{n} dm

Now, let’s define a new simple function for a very small positive number ε>0\varepsilon > 0. Then for a sufficiently large natural number nn, ϕεgnf\phi_{\varepsilon} \le g_{n} \le f will hold. Lastly, if we define Ak:={x  gkϕε}A_{k} : = \left\{ x \ | \ g_{k} \ge \phi_{\varepsilon} \right\}, AkAk+1k=1Ak=R A_{k} \subset A_{k+1} \\ \displaystyle \bigcup_{k=1}^{\infty} A_{k} = \mathbb{R}


Part 2.

Since AnA_{n} and ϕεgn\phi_{\varepsilon} \le g_{n}, AnEϕεdmAnEgndm \int_{A_{n} \cap E} \phi_{\varepsilon} dm \le \int_{A_{n} \cap E} g_{n} dm and since gn=infknfk\displaystyle g_{n} = \inf_{k \ge n } f_{k}, from knk \ge n, we have AnEgndmAnEfkdm \int_{A_{n} \cap E} g_{n} dm \le \int_{A_{n} \cap E} f_{k} dm Meanwhile, since AnEEA_{n} \cap E \subset E, AnEfkdmEfkdm \int_{A_{n} \cap E} f_{k} dm \le \int_{E} f_{k} dm Hence, we obtain the following: AnEϕεdmlim infnEfndm \int_{A_{n} \cap E} \phi_{\varepsilon} dm \le \liminf_{n \to \infty} \int_{E} f_{n} dm


Part 3.

Since ϕϵ\phi_{\epsilon} was a simple function, its image can be represented as a finite set {c1,c2,,cr}\left\{ c_{1} , c_{2} , \cdots , c_{r} \right\}. If we assume Bi:=ϕϵ1({ci})B_{i} := \phi_{\epsilon}^{-1} ( \left\{ c_{i} \right\} ), AnEϕεdm=i=1rcim(AnEBi) \int_{A_{n} \cap E} \phi_{\varepsilon} dm = \sum_{i = 1}^{r} c_{i} m (A_{n} \cap E \cap B_{i})

[7]: AnMA_{n} \in \mathcal{M}, AnAn+1    m(n=1An)=limnm(An)\displaystyle A_{n} \subset A_{n+1} \implies m \left( \bigcup_{n=1}^{\infty} A_{n} \right) = \lim_{n \to \infty} m (A_{n})

From Part 1, since AnAn+1A_{n} \subset A_{n+1} and n=1An=R\displaystyle \bigcup_{n=1}^{\infty} A_{n} = \mathbb{R}, limni=1rcim(AnEBi)=limnAnEϕεdm=Eϕεdm \lim_{n \to \infty} \sum_{i = 1}^{r} c_{i} m (A_{n} \cap E \cap B_{i}) = \lim_{n \to \infty} \int_{A_{n} \cap E} \phi_{\varepsilon} dm = \int_{E} \phi_{\varepsilon} dm Combining with results obtained in Part 2, Eϕεdmlim infkEfkdm \int_{E} \phi_{\varepsilon} dm \le \liminf_{k \to \infty} \int_{E} f_{k} dm


Part 4.

  • Case 1. m({x  ϕ(x)>0})<m( \left\{ x \ | \ \phi (x) >0 \right\} ) < \infty
    Eϕεdm=Eϕdmεm({x  ϕ(x)>0})\displaystyle \int_{E} \phi_{\varepsilon} dm = \int_{E} \phi dm - \varepsilon m( \left\{ x \ | \ \phi (x) >0 \right\} ) Taking the limit as ε0\varepsilon \to 0, yields the following: Eϕεdm=Eϕdm \displaystyle \int_{E} \phi_{\varepsilon} dm = \int_{E} \phi dm

  • Case 2. m({x  ϕ(x)>0})=m( \left\{ x \ | \ \phi (x) >0 \right\} ) = \infty Dn:={x  g(x)12min{ci}i=1n}\displaystyle D_{n} : = \left\{ x \ \left| \ g(x) \ge {{1} \over {2}} \min \left\{ c_{i} \right\}_{i=1}^{n} \right. \right\} Then, DnDn+1n=1Dn=R D_{n} \subset D_{n+1} \\ \displaystyle \bigcup_{n=1}^{\infty} D_{n} = \mathbb{R} Since DnEgndm\displaystyle \int_{D_{n} \cap E} g_{n} dm \to \infty, and from the definition of gng_{n}, we obtain the following: DnEgndmDnEfkdmEfkdm \displaystyle \int_{D_{n} \cap E} g_{n} dm \le \int_{D_{n} \cap E} f_{k} dm \le \int_{E} f_{k} dm

Therefore, in any case, for all simple functions ϕf\phi \le f, the following holds: Eϕdmlim infnEfndm \int_{E} \phi dm \le \liminf_{n \to \infty} \int_{E} f_{n} dm


  1. Capinski. (1999). Measure, Integral and Probability: p82. ↩︎