The usual dominated convergence theorem asks for one integrable function dominating every term. In applications the natural estimate often changes with the approximation. A bound for each term is not enough: the bounds must themselves avoid losing mass. Convergence of the bounds almost everywhere together with convergence of their integrals provides exactly the control used below.
Let be any measure space; no finiteness or -finiteness is assumed. All functions in this unit, including the limits, are assumed real-valued and measurable (an almost-everywhere limit need not be measurable on a noncomplete space without a suitable representative), and statements involving almost-everywhere inequalities are understood outside measurable null sets. We use Fatou's lemma, monotone convergence and the standard linearity of the finite integral.
The varying-bound theorem
Theorem 1.7.1
(Dominated convergence with varying bounds)
Suppose and almost everywhere, where , almost everywhere, , and
Then and
Proof
(Fatou applied to the slack in the bounds)
Combine the countably many exceptional null sets. Outside their union, for every and taking limits yields . Consequently all the functions in (2) are integrable. Put and
The inequality follows from . Moreover almost everywhere. After setting to zero on a common measurable null set if needed, without changing any integral or the almost-everywhere limit, Fatou gives
All displayed integrals are finite, and the sequence is bounded by , so the subtraction is legitimate. Since these error integrals are nonnegative, (4) forces their limit to be zero. The final assertion follows from .
Proof X-ray
Where the integral condition is spent
Fatou only ever gives lower bounds, so the proof applies it to the nonnegative slack rather than to the error . Pointwise convergence of and identifies . The key move is the middle equality in (4): it splits into , and the hypothesis is used exactly there, to replace by . Finiteness of every integral licenses the subtraction, and comparison with closes the argument. Without the integral hypothesis, mass escaping from would appear as an unknown positive defect and the inequality would say nothing about .
What the hypothesis says about the bounds
Proposition 1.7.2
(The integral condition is convergence of the bounds)
Let , and almost everywhere. Then if and only if .
Proof
(Minimum with the limit)
Indeed almost everywhere and is dominated by . Dominated convergence and the identity
therefore give when . Conversely, if in , then .
The theorem above proves its conclusion directly from Fatou; this observation is a second way to understand the assumption.
An example without one common dominator
Worked example 1.7.3
(Moving spikes with converging integrals)
On with Lebesgue measure, for integers let
The spike intervals are disjoint, since , and move to infinity. Hence pointwise, while . The theorem gives ; here direct calculation gives the exact error .
Worked example 1.7.4
(No integrable common dominator exists)
Suppose a single integrable measurable dominated all of the previous example almost everywhere. The countable union of the individual exceptional null sets would still be null, and almost everywhere on because it dominates a nonnegative . On each spike interval almost everywhere. Thus for each ,
which tends to infinity. The usual fixed-dominator theorem therefore does not apply, while the varying-bound theorem does.
Common mistake
(Using the supremum of the bounds as a dominator)
The natural candidate is measurable and dominates every , but in the spike example it equals , whose integral is . Integrability of each bound does not make their supremum integrable; that is precisely the gap the varying-bound theorem fills.
Worked example 1.7.5
(Signed terms below varying bounds)
On let and for . Then , and pointwise, and . The theorem gives . Directly, , confirming it. A common dominator would satisfy on and so .
A missing integral condition
The integral hypothesis in (1) cannot be dropped.
Counterexample mode
Bounds that converge pointwise but lose mass
Claim: it suffices that with each integrable and almost everywhere. On , let . Both converge pointwise to zero, since each lies outside once , and . But for every while , and also does not tend to . The repaired statement is the theorem: the mass of the bounds must converge as well.
Practice with solutions
Quick check
(Why does hold almost everywhere under the theorem's hypotheses?)
Solution · Answer
Outside the countable union of exceptional sets, both and hold and for every ; take limits.
Quick check
(If , almost everywhere and , what can be concluded for ?)
Solution · Answer
The bound also gives almost everywhere by squeezing. Directly, gives convergence and convergence of integrals without needing Fatou.
Quick check
(In the moving-spike example, why does an almost-everywhere dominator still have infinite integral?)
Solution · Answer
The exceptional sets have null union; disjointness makes the first spikes contribute at least , which diverges.
Quick check
(Does the theorem require ?)
Solution · Answer
No. The theorem and proof apply to arbitrary and never require finite total measure; the example uses .