Skip to Content
Math4121Introduction to Lebesgue Integration (Lecture 36)

Math 4121 Lecture 36

Random visit for Lebesgue Integration

Convergence Theorem

Theorem 6.14 Monotone Convergence Theorem

Let {fn}\{f_n\} be a monotone increasing sequence of measurable functions on EE and fnff_n\to f almost everywhere on EE. (fn(x)fn+1(x)f_n(x)\leq f_{n+1}(x) for all xEx\in E and nn)

If there exists A>0A>0 such that EfndmA\left|\int_E f_n \, dm\right|\leq A for all nNn\in \mathbb{N}, then f(x)=limnfn(x)f(x)=\lim_{n\to\infty} f_n(x) exists for almost every xEx\in E and it is integrable on EE and

Efdm=limnEfndm\int_E f \, dm = \lim_{n\to\infty} \int_E f_n \, dm

Proof:

To show the limit exists almost everywhere, let ϵ>0\epsilon>0, set En={xE:fn(x)>Aϵ}E_n=\{x\in E: f_n(x)>\frac{A}{\epsilon}\}. We will show U=n=1EnU=\bigcup_{n=1}^{\infty} E_n has measure <ϵ<\epsilon. fn(x)AϵχEn(x)f_n(x)\geq \frac{A}{\epsilon}\chi_{E_n}(x), so

Aϵm(En)=EAϵχEndmEfndmA\frac{A}{\epsilon}m(E_n)=\int_E \frac{A}{\epsilon}\chi_{E_n}dm\leq \int_E f_n dm\leq A

In particular, m(En)<ϵm(E_n)<\epsilon. Since EnEn+1E_n\subset E_{n+1} for all nn, m(U)=limnm(En)<ϵm(U)=\lim_{n\to\infty} m(E_n)<\epsilon.

limnEfndmEfdmlimn\lim_{n\to\infty} \int_E f_n dm\leq \int_E f dm\leq \lim_{n\to\infty}. To show the reverse inequality, let ϕ\phi be a simple function f\leq f of the form ϕ=i=1kaiχSi\phi=\sum_{i=1}^{k} a_i\chi_{S_i} where SiS_i is sidjoint and i=1kSiE\bigcup_{i=1}^{k} S_i\subseteq E.

Let α(0,1)\alpha\in (0,1) and set An={xS:fn(x)αϕ(x)>0}A_n=\{x\in S:f_n(x)-\alpha\phi(x)>0\}. This ensures that fn(x)αϕ(x)f_n(x)\geq \alpha\phi(x) for all xAnx\in A_n.

Notice that AnAn+1A_n\subset A_{n+1} for all nn and U=n=1AnU=\bigcup_{n=1}^{\infty} A_n. limnm(AnSi)=m(Si)lim_{n\to\infty} m(A_n\cap S_i)=m(S_i) for all ii.

Anϕdm=i=1kaim(SiAn)\int_{A_n} \phi dm=\sum_{i=1}^{k} a_i m(S_i\cap A_n).

As nn\to\infty, m(AnSi)m(Si)m(A_n\cap S_i)\to m(S_i) and i=1kaim(SiAn)Eϕdm\sum_{i=1}^{k} a_i m(S_i\cap A_n)\to \int_E \phi dm.

Let ϵ>0\epsilon>0. There exists n0n_0 large such that Anϕdm>Eϕdmϵ\int_{A_n} \phi dm>\int_E \phi dm-\epsilon for all nn0n\geq n_0.

Then for such nn0n\geq n_0,

EfndmAnfndmAnαϕdm>α(Eϕdmϵ)\int_E f_n dm\geq \int_{A_n} f_n dm\geq \int_{A_n} \alpha \phi dm>\alpha(\int_E \phi dm-\epsilon)

So, limnEfndmα(Eϕdmϵ)\lim_{n\to\infty} \int_E f_n dm\geq \alpha(\int_E \phi dm-\epsilon).

Since α,ϵ\alpha,\epsilon are arbitrary, set α1\alpha\to 1 and ϵ0\epsilon\to 0 to get limnEfndmEϕdm\lim_{n\to\infty} \int_E f_n dm\geq \int_E \phi dm.

For any simple function ϕf\phi\leq f, taking sup over all simple functions ϕf\phi\leq f gives limnEfndmEfdm\lim_{n\to\infty} \int_E f_n dm\geq \int_E f dm.

QED

Lemma Absolute Integrability

ff is integrable on EE if and only if f|f| is integrable on EE and EfdmEfdm\left|\int_E f \, dm\right|\leq \int_E |f| \, dm.

Proof:

If f+f^+ and ff^- are integrable and f=f+f|f|=f^+-f^-. So setting E1={xE:f(x)0}E_1=\{x\in E: f(x)\geq 0\} and E2={xE:f(x)<0}E_2=\{x\in E: f(x)<0\}, these are disjoint and E=E1E2E=E_1\cup E_2.

Efdm=E1f+dm+E2fdm\int_E |f| dm=\int_{E_1} f^+ dm+\int_{E_2} f^- dm

For the reverse inequality, note that

Ef+dmEfdm\int_E f^+ dm\leq \int_E |f| dm

and

EfdmEfdm\int_E f^- dm\leq \int_E |f| dm

QED

Corollary Properties of Integrals

Let ff and gg be integrable on EE, and cRc\in \mathbb{R}.

  1. E(cf)dm=cEfdm\int_E (cf) dm=c\int_E f dm
  2. E(f+g)dm=Efdm+Egdm\int_E (f+g) dm=\int_E f dm+\int_E g dm

Proof:

First we prove it for f,gf,g nonnegative and c0c\geq 0.

Take simple functions ϕnf\phi_n\to f and ψng\psi_n\to g pointwise. Then cϕncfc\phi_n\to cf and ϕn+ψnf+g\phi_n+\psi_n\to f+g pointwise.

By Monotone Convergence Theorem,

Ecfdm=limnEcϕndm=climnEϕndm=cEfdm\int_E cf dm=\lim_{n\to\infty} \int_E c\phi_n dm=c\lim_{n\to\infty} \int_E \phi_n dm=c\int_E f dm

Second part leave as homework.

QED

Theorem 6.8

Riemann integrable functions are Lebesgue integrable and the values of the integrals are the same.

Proof:

Say ff is Riemann integrable on [a,b][a,b]. mf(x)Mm\leq f(x)\leq M for all x[a,b]x\in [a,b].

We can find a partition PnPn+1P_n\subseteq P_{n+1} of [a,b][a,b] such that L(Pn,f)abfdxL(P_n,f)\nearrow \int_a^b f dx and U(Pn,f)abfdxU(P_n,f)\searrow \int_a^b f dx.

Let ϕn=i=1kmiχIi\phi_n=\sum_{i=1}^{k} m_i \chi_{I_i} and ψn=i=1kMiχIi\psi_n=\sum_{i=1}^{k} M_i \chi_{I_i} where IiI_i is an interval in PnP_n.

So abϕndm=L(Pn,f)\int_a^b \phi_n dm=L(P_n,f) and abψndm=U(Pn,f)\int_a^b \psi_n dm=U(P_n,f).

mϕnfψnMm\leq \phi_n\leq f\leq \psi_n\leq M for all nn. almost everywhere.

By Monotone Convergence Theorem, to ϕnm\phi_{n-m} we have g(x)=limnϕn(x)g(x)=\lim_{n\to\infty} \phi_n(x), h(x)=limnψn(x)h(x)=\lim_{n\to\infty} \psi_n(x) exists for almost every x[a,b]x\in [a,b].

g(x)f(x)h(x)g(x)\leq f(x)\leq h(x) almost everywhere.

So

abgdm=abfdm=abhdm=abfdx\int_a^b g dm= \int_a^b f dm= \int_a^b h dm =\int_a^b f dx

So h(x)=f(x)=g(x)h(x)=f(x)=g(x) almost everywhere.

QED

Last updated on