Suppose $(\Omega, \mathcal{F}, P)$ is a probability space (or finite measure space, I doubt this matters) and that $\mathcal{F}$ is generated by the algebra $\mathcal{B}$.

It is well-known that for all $\epsilon > 0$ and all $A \in \mathcal{F}$, there exists $B \in \mathcal{B}$ such that $$P(A\Delta B) < \epsilon.$$

Note that this implies $|P(A) - P(B)| < \epsilon.$

Can this kind of approximation be extended to integrals?

Is it the case that for all $\epsilon > 0$, all bounded measurable $f$, and all $A \in \mathcal{F}$, there exists $B \in \mathcal{B}$ such that $$\Big |\int_A f dP - \int_B f dP \Big | < \epsilon?$$

It is clearly true for $f = 1$ in view of the approximation result mentioned above. It's also true for indicator functions $\mathbf{1}_F$, using the result above and the fact that $$P(A \Delta B) < \epsilon \implies |P(A \cap F) - P(B \cap F)| < \epsilon,$$ which in turn follows from $F \cap(A \Delta B) = (F \cap A) \Delta (F \cap B)$.

My idea now was to finish by using the monotone class theorem for functions. I need to show that the set of functions for which the desired property holds is a vector space and closed under monotone limits. I'm stuck here.

For instance, suppose $f$ and $g$ have the desired property. We want to show that $f+g$ has the desired property. By the triangle inequality, $$\Big|\int_A (f+g) dP - \int_B (f+g) dP \Big| \leq \Big| \int_A fdP - \int_B fdP \Big| + \Big| \int_A gdP - \int_B gdP \Big|.$$

So to conclude, I need a $B \in \mathcal{B}$ that works for both $f$ and $g$, and I can't convince myself that such a $B$ exists. I reckon I'll face the same sort of problem for the monotone limit case.

Does the result hold? If so, can the proof be completed along the lines I suggest? If not, does anyone know of a similar approximation result for integrals?


We can actually prove the following:

for all $\epsilon > 0$, all integrable function $f$, and all $A \in \mathcal{F}$, there exists $B \in \mathcal{B}$ such that $$\Big |\int_A f dP - \int_B f dP \Big | < \epsilon.$$

Indeed, for all integrable function $f$ and all $A \in \mathcal{F}$, $$\Big |\int_A f dP - \int_B f dP \Big | =\Big |\int f\left(\mathbf 1_A-\mathbf 1_B\right) dP \Big |\leqslant \int\left| f\right|\left|\mathbf 1_A-\mathbf 1_B\right| dP =\int_{A\Delta B}\left| f\right| \mathbb dP.$$

Now, we use the fact that for any integrable function $f$ and any $\varepsilon\gt 0$, there exists $\delta\gt 0$ such that if $C$ has a measure smaller than $\delta$, then $\int_C\left| f\right|\mathrm dP\lt\varepsilon$.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.