Many false proofs in analysis have the same hidden move: two limiting operations are swapped without a theorem that permits the swap.
The final R25 cell is therefore a control layer rather than another catalogue of techniques. It asks when a limit may pass through continuity, an integral, a derivative, an infinite sum, a supremum or a second limit—and what kind of counterexample appears when the required uniformity, domination, monotonicity, compactness or absolute convergence is missing.
The earlier R25 guides supply the objects. Limits, Continuity and Sequences provides epsilon control; Derivatives and Local Approximation provides differentiability; Integration and Accumulation provides Riemann and improper integration; Series and Convergence provides infinite sums; and Measure and Lebesgue Integration provides the modern convergence theorems.
Level: undergraduate real-analysis enrichment. Prerequisites: R25.01–R25.07 as needed, especially pointwise convergence, uniform convergence, differentiation and Lebesgue integration.
Reading route: pointwise versus uniform → sup norm → uniform Cauchy criterion → continuity under uniform limits → failure example xⁿ → interchange of two limits → integration under uniform convergence → moving-spike failure → MCT/Fatou/DCT → differentiation theorem → derivative failure example → series of functions and M-test → power series → rearrangements and absolute convergence → Tonelli/Fubini → suprema and maxima → diagnostic patterns → practice → solutions.
1. Start by naming the two operations
Whenever an argument contains a statement such as
lim_n T(f_n)=T(lim_n f_n),
identify the operator T. Is it evaluation at a point, another limit, differentiation, integration, summation, supremum, expectation or something else?
The equality is not a notation rule. It is a theorem claim whose hypotheses depend on T.
2. Pointwise convergence allows the index threshold to depend on position
A sequence f_n:E→R converges pointwise to f if, for every fixed x∈E and every ε>0, there exists N=N(x,ε) such that n≥N implies
|f_n(x)−f(x)|<ε.
The dependence of N on x is the weakness. Different locations may require radically different stages before the approximation becomes good.
3. Uniform convergence uses one threshold for the whole domain
f_n→f uniformly on E if for every ε>0 there exists N=N(ε) such that for every n≥N and every x∈E,
|f_n(x)−f(x)|<ε.
Now the same N works everywhere. Uniformity turns infinitely many pointwise approximation problems into one global control statement.
4. The sup norm makes uniform convergence visible
When the functions are bounded, define
||g||∞=sup_{x∈E}|g(x)|.
Then f_n→f uniformly exactly when
||f_n−f||∞→0.
This reframes uniform convergence as ordinary metric convergence in a function space, preparing the bridge to R27 functional analysis.
5. Uniform convergence has a Cauchy criterion
If the codomain is complete, a sequence of functions is uniformly convergent exactly when it is uniformly Cauchy:
for every ε>0 there exists N such that m,n≥N implies |f_m(x)−f_n(x)|<ε for every x∈E.
The limit does not need to be known in advance. This is the function-space version of Cauchy completeness developed in R25.06.
6. Uniform limits preserve continuity
If each f_n is continuous on E and f_n→f uniformly, then f is continuous.
The proof uses a three-term estimate:
|f(x)−f(a)|≤|f(x)−f_n(x)|+|f_n(x)−f_n(a)|+|f_n(a)−f(a)|.
Uniform convergence controls the first and third terms with one n across the domain; continuity of that fixed f_n controls the middle term near a.
7. Counterexample: continuous functions can converge pointwise to a discontinuous limit
Let f_n(x)=x^n on [0,1].
For every x<1, x^n→0. At x=1, x^n=1 for every n.
Thus the pointwise limit is
f(x)=0 for 0≤x<1 and f(1)=1, which is discontinuous at 1.
Therefore the convergence cannot be uniform. The missing uniformity appears as a boundary layer near x=1.
8. On smaller compact subintervals xⁿ does converge uniformly
Fix 0≤r<1 and restrict x to [0,r]. Then
sup_{0≤x≤r}x^n=r^n→0.
So x^n→0 uniformly on every [0,r] strictly inside [0,1).
The failure is concentrated near the moving boundary x=1. This local-versus-global distinction recurs throughout analysis.
9. Two limits need not commute
Let f_n(x)=x/(x+1/n) on x≥0.
For each fixed n, lim_{x→0+}f_n(x)=0, so
lim_{n→∞} lim_{x→0+} f_n(x)=0.
For each fixed x>0, f_n(x)→1, so
lim_{x→0+} lim_{n→∞} f_n(x)=1.
The two iterated limits differ. Swapping their order changes the result.
10. Uniform convergence near the second limit point can restore commutation
A standard theorem says that if f_n→f uniformly in a neighbourhood of a and lim_{x→a}f_n(x)=L_n with L_n→L, then under the appropriate domain conditions one can conclude lim_{x→a}f(x)=L.
The exact theorem can be phrased in several equivalent ways, but the structural message is stable: a common approximation scale prevents the n-limit from changing unpredictably as x approaches a.
11. Uniform convergence permits integration on finite intervals
If f_n are Riemann integrable on [a,b] and f_n→f uniformly, then f is Riemann integrable and
lim_{n→∞}∫_a^b f_n = ∫_a^b f.
The proof is a direct sup-norm estimate:
|∫(f_n−f)|≤(b−a)||f_n−f||∞→0.
Uniform error controls accumulated error because the interval has finite length.
12. Pointwise convergence alone does not permit integral exchange
On [0,1], define f_n(x)=n·1_(0,1/n)(x).
Then f_n(x)→0 pointwise, but
∫₀¹ f_n(x)dx=1
for every n.
The mass concentrates into a narrower spike. Pointwise vision sees the spike leave every fixed x; integration still sees its total mass.
13. Monotone Convergence gives a different route to integral exchange
Uniform convergence is sufficient but not necessary for many Lebesgue-integral limit exchanges.
If 0≤f_n↑f pointwise and the functions are measurable, the Monotone Convergence Theorem gives
∫f_n→∫f.
The control mechanism is monotonicity plus nonnegativity rather than a uniform sup-norm error.
14. Dominated Convergence controls moving behaviour by an integrable envelope
If f_n→f almost everywhere and |f_n|≤g for one integrable g, then the Dominated Convergence Theorem gives
∫f_n→∫f
and even ∫|f_n−f|→0.
Here the common control is not a uniform numerical error but a single integrable envelope that prevents hidden mass from escaping.
15. Fatou’s lemma survives when equality is too much to ask
For nonnegative measurable f_n,
∫liminf f_n≤liminf∫f_n.
Fatou gives a one-sided inequality with very weak hypotheses. It is often the correct tool when no domination or monotonicity is available.
A useful hierarchy is therefore: seek the exact exchange theorem if its hypotheses hold; otherwise ask whether a one-sided inequality still survives.
16. Uniform convergence of functions does not justify differentiating the limit
Differentiation is more sensitive than integration. Even uniform convergence f_n→f can coexist with a nondifferentiable limit.
Consider
f_n(x)=√(x²+1/n)
on a bounded interval containing zero.
These differentiable functions converge uniformly to |x| because 0≤√(x²+1/n)−|x|≤1/√n.
But |x| is not differentiable at zero.
17. The derivative sequence diagnoses the failure
For f_n(x)=√(x²+1/n),
f_n'(x)=x/√(x²+1/n).
For x>0 the derivatives tend to 1; for x<0 they tend to −1; at x=0 they equal 0.
The pointwise derivative limit has a jump. It cannot be a uniform limit of continuous functions on an interval containing zero.
The derivative convergence lacks the uniform control needed to pass differentiability to the limit.
18. A standard termwise-differentiation theorem controls derivatives first
Let f_n be differentiable on [a,b]. Suppose:
- there exists x₀∈[a,b] such that f_n(x₀) converges;
- f_n’ converges uniformly on [a,b] to a function g.
Then f_n converges uniformly to a differentiable f and
f’=g=lim f_n’.
The single-point value anchors the additive constants; uniform derivative convergence controls all differences through the Mean Value Theorem or integration.
19. Convergence of derivatives without an anchor leaves constants undetermined
Take f_n(x)=x+n. Then f_n'(x)=1 for every n, so the derivatives converge uniformly.
But the functions themselves do not converge at any point.
The derivative controls variation, not absolute vertical position. This is why the theorem needs convergence at one base point.
20. Uniform convergence of derivatives is sufficient, not the only possible hypothesis
More advanced theorems permit weaker modes of derivative control—often through Sobolev spaces, weak derivatives, equicontinuity or compactness arguments.
Those belong to later functional-analysis and PDE routes. The R25 rule is narrower: never infer termwise differentiability merely from pointwise or uniform convergence of the original functions.
21. Series of functions are sequences of partial-sum functions
For Σf_n(x), define S_N(x)=Σ_{n=1}^N f_n(x).
Pointwise or uniform convergence of the function series means pointwise or uniform convergence of S_N.
Every theorem about limits of function sequences can therefore be applied to partial sums, provided the hypotheses are translated correctly.
22. The Weierstrass M-test converts a function series into a numerical majorant
If |f_n(x)|≤M_n for every x∈E and ΣM_n converges, then Σf_n converges uniformly and absolutely on E.
This is an exchange-enabling theorem because uniform convergence can preserve continuity and justify finite-measure integration term by term.
The majorant must be summable. Merely bounding each term by the same constant does not control the infinite tail.
23. Power series have built-in local uniformity inside the radius
If Σc_n(x−a)^n has radius of convergence R, then on every closed subinterval |x−a|≤r<R the series converges uniformly and absolutely.
This local uniformity is why power series may be differentiated and integrated term by term throughout the open interval |x−a|<R.
The endpoints remain separate convergence problems. The radius does not certify them.
24. Conditional numerical series warn against casual reordering
For a conditionally convergent real series, rearrangement can change the sum or destroy convergence.
Riemann’s rearrangement theorem makes the danger exact.
Absolute convergence is the structural control that makes arbitrary rearrangement safe. An infinite sum is not a finite sum with an infinity symbol attached.
25. Double series and iterated sums need the same care
Changing Σ_mΣ_n a_{mn} into Σ_nΣ_m a_{mn} is an order exchange.
For nonnegative terms, Tonelli-type reasoning makes the exchange safe, allowing ∞. For absolutely summable families, Fubini-type reasoning permits order exchange with a finite result.
Conditional cancellation can make different summation orders behave differently.
26. Tonelli and Fubini are exchange theorems
They are often introduced as methods for computing double integrals, but structurally they answer:
When may we exchange the order of integration?
Tonelli uses nonnegativity. Fubini uses integrability, typically ∫|f|<∞. These hypotheses prevent hidden cancellation from making the order significant.
27. Suprema interact cleanly with uniform convergence
For bounded functions f_n,f on E,
|sup_E f_n−sup_E f|≤||f_n−f||∞.
Therefore uniform convergence implies sup f_n→sup f whenever the suprema are finite.
The same holds for infima.
This estimate does not require the suprema to be attained.
28. Maximising points need more care than maximum values
Uniform convergence can control maximum values, but the points at which maxima occur can wander or fail to converge when the limiting function has several maximisers or a flat plateau.
To infer convergence of argmax points, one normally needs compactness plus uniqueness or a quantitative separation condition around the limiting maximiser.
Value convergence and optimiser convergence are different mathematical jobs.
29. Limit and expectation is just limit and integral on a probability space
Expectation is integration with respect to a probability measure.
Therefore Monotone Convergence, Fatou and Dominated Convergence immediately become expectation-limit theorems.
Statements such as E[X_n]→E[X] require hypotheses. Almost-sure or pointwise convergence alone is not enough.
This is a bridge to R29 probability and stochastic processes.
30. Three recurring failure mechanisms
Moving concentration
A spike becomes narrower and taller, so every fixed point eventually sees nothing while an integral still sees finite mass.
Boundary layers
Convergence is excellent away from a boundary but becomes arbitrarily slow near it, as with x^n near x=1.
Conditional cancellation
Positive and negative contributions are individually too large, and only a particular order or pairing creates cancellation. Rearrangement then changes the result.
31. Oscillation is a fourth failure mechanism
A sequence can maintain fixed-size oscillations while approaching no single limit, or can oscillate faster and faster while remaining bounded.
Examples such as sin(nx) or sin(1/x) show why boundedness alone does not provide convergence.
When oscillation is paired with shrinking amplitude, convergence may return; when paired with cancellation inside an integral, weak forms of convergence may appear even without pointwise convergence.
32. A theorem-selection table in words
- Preserve continuity: look for uniform convergence.
- Exchange limit and finite-interval Riemann integral: uniform convergence is a clean sufficient condition.
- Exchange limit and Lebesgue integral for increasing nonnegative functions: Monotone Convergence.
- Control nonnegative sequences without equality: Fatou.
- Exchange limit and Lebesgue integral under a common integrable envelope: Dominated Convergence.
- Differentiate a function-sequence limit: control the derivative sequence uniformly and anchor function values at one point, or invoke another theorem with clearly stated hypotheses.
- Exchange integral order: Tonelli for nonnegative measurable functions; Fubini for integrable signed functions.
- Rearrange an infinite numerical sum safely: absolute convergence is a standard sufficient condition.
33. A dependable exchange-of-limits workflow
Write both operations explicitly before swapping them.
Identify the convergence mode: pointwise, uniform, almost everywhere, in measure, L¹, or another norm.
Search for the theorem whose conclusion is exactly the desired interchange. Check every hypothesis rather than matching only the theorem name.
If a hypothesis is missing, do not immediately conclude the exchange fails. Instead, either prove the result by another method or build a counterexample targeting the missing control.
Typical diagnostic questions are: Can mass concentrate? Can convergence slow near a moving boundary? Can derivatives become unbounded or nonuniform? Is cancellation conditional? Does an exceptional set remain significant?
34. Common misconceptions
- Pointwise convergence preserves continuity. False.
- Uniform convergence preserves differentiability. False.
- Pointwise convergence permits integral exchange. False.
- If two iterated limits exist, they must agree. False.
- Convergent series can always be rearranged. False for conditional convergence.
- Fubini allows any double integral to change order. False without its hypotheses.
- A theorem’s sufficient condition is always necessary. False; several different control mechanisms may justify the same exchange.
35. Independent practice: twenty questions
- Define pointwise convergence.
- Define uniform convergence.
- Express uniform convergence using the sup norm.
- State the uniform Cauchy criterion.
- What property of continuous functions is preserved by uniform limits?
- Find the pointwise limit of x^n on [0,1].
- Why does x^n fail to converge uniformly on [0,1]?
- Why does x^n converge uniformly on [0,r] when r<1?
- For f_n(x)=x/(x+1/n), compare the two iterated limits as n→∞ and x→0+.
- Give a sufficient condition for passing a uniform limit through a Riemann integral on [a,b].
- Explain the moving-spike counterexample n1_(0,1/n).
- State the Monotone Convergence Theorem.
- State the Dominated Convergence Theorem.
- Why does uniform convergence of f_n not imply differentiability of the limit?
- For √(x²+1/n), identify the uniform limit.
- State one standard theorem permitting differentiation of a sequence limit.
- Why is convergence of f_n at one point needed in that theorem?
- State the Weierstrass M-test.
- When may an absolutely convergent series be rearranged?
- Distinguish the typical hypotheses of Tonelli and Fubini.
36. Worked solutions and checks
1. For every fixed x and ε>0, there is N depending on x and ε so n≥N gives |f_n(x)−f(x)|<ε.
2. For every ε>0, there is one N depending only on ε that works for every x in the domain.
3. ||f_n−f||∞→0.
4. For every ε>0, sufficiently late f_m and f_n differ by less than ε uniformly over the domain.
5. Continuity.
6. The limit is 0 on [0,1) and 1 at x=1.
7. The pointwise limit is discontinuous while each x^n is continuous; a uniform limit of continuous functions would be continuous.
8. sup_{[0,r]}x^n=r^n→0.
9. lim_n lim_{x→0+}f_n=0, while lim_{x→0+}lim_n f_n=1.
10. If f_n are Riemann integrable and converge uniformly to f on the finite interval, then ∫f_n→∫f.
11. f_n→0 pointwise but every integral equals 1 because mass concentrates into a shrinking interval.
12. If 0≤f_n↑f measurably, then ∫f_n→∫f.
13. If f_n→f almost everywhere and |f_n|≤g for one integrable g, then ∫f_n→∫f.
14. Differentiation amplifies small-scale variation; uniformly convergent differentiable functions can approach a nondifferentiable limit.
15. |x|.
16. One standard version assumes differentiability on [a,b], convergence f_n(x₀) at one point, and uniform convergence of f_n’; then f_n converges uniformly to differentiable f with f’=lim f_n’.
17. Derivatives determine functions only up to additive constants; the base-point convergence fixes that freedom.
18. If |f_n(x)|≤M_n everywhere and ΣM_n converges, then Σf_n converges uniformly and absolutely.
19. Arbitrary rearrangement preserves the sum under absolute convergence.
20. Tonelli uses nonnegative measurability and may allow ∞; Fubini uses integrability of the signed function, typically ∫|f|<∞.
37. R25 return path
The eight R25 cells now form one connected analysis route: limits and continuity → derivatives and local approximation → Riemann integration → series and power series → multivariable calculus → completeness → measure and Lebesgue integration → theorem-controlled exchange of limiting operations.
Publication of all eight cells is coverage, not final certification of the region. The Atlas still requires a separate audit of bridges, mathematical review, sources and reader return paths before any “Complete for v1” status would be justified.
Sources and further study
For a rigorous university route through sequences and series of functions, uniform convergence and termwise operations, see MIT OpenCourseWare 18.100A Introduction to Analysis. For measure-theoretic limit exchange, continue with MIT OpenCourseWare 18.125 Measure and Integration.