Riemann Integral vs. Lebesgue Integral [dark version]

Lebesgue integration replaces the Riemann integral’s “rectangle-and-partition” machinery with a partition based on function values, letting integration work cleanly in higher dimensions and for functions with many discontinuities. The core shift is that Lebesgue integration partitions the real line (the range of the function) and then measures how much of the domain maps into each value band—so the integral depends on a notion of “volume” (a measure) rather than on geometric partitions of the domain.

Riemann integration starts with the idea of area under a graph. On an interval [a,b], it approximates the area using upper and lower sums built from partitions of the x-axis: the upper sum uses suprema of f on each subinterval, the lower sum uses infima. When these bounds can be squeezed together arbitrarily closely, the common value defines the Riemann integral. That framework is intuitive in one dimension, but it strains in two ways.

First, extending Riemann integration to higher dimensions becomes laborious because partitions of the domain must match the geometry of the region. In R², rectangles become cuboids in R³, and for irregular domains (like a circle), even the “right” partition is awkward—requiring more technical limit arguments to approximate the region.

Second, Riemann integration is sensitive to discontinuities. While continuous functions behave well, Riemann’s definition effectively relies on controlling how function values oscillate across subintervals. If discontinuities are too wild—especially if they occur infinitely often in a way that prevents the upper and lower sums from converging—Riemann integrability can fail.

A third, deeper issue is the relationship between integrals and limits. For sequences of functions f_n, swapping a limit with an integral is not automatic. For the Riemann integral, pulling lim_{n→∞} inside the integral typically requires strong conditions such as uniform convergence. When uniform convergence fails, the interchange can break, even if the “final” answer might still exist.

Lebesgue integration addresses all three problems by changing what gets partitioned and what gets measured. Instead of slicing the domain into x-intervals, it slices the range into value intervals C_i. For each C_i, it looks at the preimage f^{-1}(C_i): the set of points in the domain whose function values land in that band. The “height” of the corresponding generalized rectangle is the value band, and the “width” is the measure of that preimage set. With a measure μ on the domain (length in R, area in R², volume in R³, or an abstract measure space), Lebesgue integration can be defined for functions on very general spaces.

Because the definition uses measure, Lebesgue integration avoids the geometric partition headaches of higher dimensions. It also weakens the continuity requirement: even if a function has infinitely many discontinuities, it can still be integrable as long as the set of problematic points has measure zero. Finally, Lebesgue’s framework comes with stronger limit theorems, making it far more reliable for exchanging limits and integrals than the Riemann approach.

In short: Riemann integration is the classical, geometry-first method; Lebesgue integration is the measure-first method that generalizes smoothly across dimensions, tolerates discontinuities better, and supports more powerful limit operations.