When does the Riemann Integral-like characterization of Lebesgue Integrals fail?

Question

Wheeden and Zygmund's book "Measure and Integral" gives an interesting characterization of the Lebesgue Integral that is reminiscent of the Riemann Integral. If $E$ be a Lebesgue measurable set, then then the Lebesgue integral of $f$ on $E$ is equal to $\sup\Sigma_{k=1}^n (\inf_{E_K}f)\lambda(E_k)$, where the supremum is taken over all finite partitions $E_1,...,E_n$ of $E$ into Lebesgue measurable sets. (This remains true for arbitrary measure spaces.)

But Wheeden and Zygmund say that if you switch the supremum and infimum, this need not hold true. So my question is, what is an example of a Lebesgue integrable function $f$ such that the Lebesgue integral of $f$ on $E$ is not equal to $\inf\Sigma_{k=1}^n (\sup_{E_K}f)\lambda(E_k)$?

Also, is there a subset of the set of Lebesgue integrable functions for which the switched version does hold true?

Answer 1

Let $E = (0,1)$ and $f(x) = \frac{1}{\sqrt{x}}$. Then $\int f < \infty$ but if $E = \sqcup_{k=1}^n E_k$ is a finite partition, then for some $k$, we have $\sup_{E_k} f = \infty$ and $\lambda(E_k) > 0$.

Answer 2

Let $E=[0,1]$ and let $f$ be the characteristic function of $E \cap \mathbb Q$.