The math optional, made finite. Daily Practice

Uniform Convergence: Term-by-Term Differentiation

At a Glance

Why This Chapter Matters

This atom appeared in 2024 for 20 marks, making it the most recently tested topic in this cluster. The differentiation theorem is more subtle than the corresponding integration result: uniform convergence of fnf_n alone does not allow differentiating under the limit; you need the derivatives to converge uniformly. UPSC’s ask is typically to prove the theorem from scratch, or to apply it to justify differentiating a power series or other concrete series term-by-term.

Minimum Theory

Theorem (Term-by-term differentiation). Let fn:[a,b]Rf_n:[a,b]\to\mathbb{R} be a sequence of functions such that:

  1. Each fnf_n is differentiable on [a,b][a,b].
  2. The sequence of derivatives (fn)(f_n') converges uniformly to some function gg on [a,b][a,b].
  3. The sequence (fn(x0))(f_n(x_0)) converges for at least one point x0[a,b]x_0 \in [a,b].

Then (fn)(f_n) converges uniformly on [a,b][a,b] to a differentiable function ff, and

f(x)=g(x)=limnfn(x)for all x[a,b].f'(x) = g(x) = \lim_{n\to\infty} f_n'(x) \quad \text{for all } x \in [a,b].

Remark. Condition (2) is the critical hypothesis. Uniform convergence of (fn)(f_n) alone does not suffice for f=limfnf' = \lim f_n'.

Counterexample without condition (2). Let fn(x)=sin(nx)nf_n(x) = \dfrac{\sin(nx)}{\sqrt{n}} on [0,1][0,1]. Then fn0f_n \to 0 uniformly (since fn(x)1/n0|f_n(x)| \leq 1/\sqrt{n} \to 0), but fn(x)=ncos(nx)f_n'(x) = \sqrt{n}\cos(nx), which does not converge (it is unbounded). So f=0f' = 0 but limfn\lim f_n' does not exist.

Contrast with integration. Uniform convergence of (fn)(f_n) alone is sufficient for term-by-term integration:

fnf uniformly on [a,b]    abfnabf.f_n \to f \text{ uniformly on } [a,b] \;\Longrightarrow\; \int_a^b f_n \to \int_a^b f.

No condition on the derivatives is needed.

Proof of the differentiation theorem.

Step 1 — Uniform convergence of (fn)(f_n). Fix x[a,b]x \in [a,b]. By the Fundamental Theorem of Calculus, for m,nNm, n \in \mathbb{N},

fm(x)fn(x)=[fm(x0)fn(x0)]+x0x[fm(t)fn(t)]dt.f_m(x) - f_n(x) = \bigl[f_m(x_0) - f_n(x_0)\bigr] + \int_{x_0}^{x}\bigl[f_m'(t) - f_n'(t)\bigr]\,dt.

Since (fn(x0))(f_n(x_0)) is a convergent (Cauchy) sequence and (fn)(f_n') is uniformly Cauchy on [a,b][a,b]:

fm(x)fn(x)fm(x0)fn(x0)+x0xfm(t)fn(t)dtfm(x0)fn(x0)+(ba)suptfm(t)fn(t).|f_m(x) - f_n(x)| \leq |f_m(x_0) - f_n(x_0)| + \int_{x_0}^{x}|f_m'(t) - f_n'(t)|\,dt \leq |f_m(x_0) - f_n(x_0)| + (b-a)\sup_{t}|f_m'(t) - f_n'(t)|.

Given ε>0\varepsilon > 0, choose NN so that for m,nNm, n \geq N: fm(x0)fn(x0)<ε/2|f_m(x_0) - f_n(x_0)| < \varepsilon/2 and suptfm(t)fn(t)<ε/(2(ba))\sup_t|f_m'(t) - f_n'(t)| < \varepsilon/(2(b-a)). Then fm(x)fn(x)<ε|f_m(x) - f_n(x)| < \varepsilon for all xx, so (fn)(f_n) is uniformly Cauchy, hence converges uniformly to some ff.

Step 2 — f=gf' = g. Fix x[a,b]x \in [a,b] and ε>0\varepsilon > 0. We need to show

limh0f(x+h)f(x)h=g(x).\lim_{h \to 0} \frac{f(x+h) - f(x)}{h} = g(x).

Write

f(x+h)f(x)hg(x)=[f(x+h)f(x)hfn(x+h)fn(x)h]+[fn(x+h)fn(x)hfn(x)]+[fn(x)g(x)].\frac{f(x+h)-f(x)}{h} - g(x) = \left[\frac{f(x+h)-f(x)}{h} - \frac{f_n(x+h)-f_n(x)}{h}\right] + \left[\frac{f_n(x+h)-f_n(x)}{h} - f_n'(x)\right] + \left[f_n'(x) - g(x)\right].

By uniform convergence of (fn)(f_n), the first bracket can be made <ε/3< \varepsilon/3 by choosing nn large (independent of hh). By uniform convergence of (fn)(f_n'), the third bracket can also be made <ε/3< \varepsilon/3 by choosing nn large. Fix such an nn; by differentiability of fnf_n, choose δ>0\delta > 0 so that h<δ|h| < \delta makes the second bracket <ε/3< \varepsilon/3. Then h<δ|h| < \delta implies the full expression is <ε< \varepsilon. Hence f(x)=g(x)f'(x) = g(x). \blacksquare

Series version. If fn=unf_n = u_n and we write SN=n=1NunS_N = \sum_{n=1}^N u_n, the theorem says: if un\sum u_n converges at some x0x_0 and un\sum u_n' converges uniformly on [a,b][a,b], then (un)=un\left(\sum u_n\right)' = \sum u_n'.

Weierstrass M-test. un\sum u_n' converges uniformly if un(x)Mn|u_n'(x)| \leq M_n for all xx and Mn<\sum M_n < \infty.

Question Archetypes

ArchetypeRecognition
prove-differentiation-theorem”Prove that if fngf_n' \to g uniformly and fn(x0)f_n(x_0) converges, then fnff_n \to f and f=gf' = g.“
justify-term-by-term”Show that the given series can be differentiated term-by-term; find the derivative.”
compare-int-vs-diff”Why is uniform convergence sufficient for integration but insufficient for differentiation?“

prove-differentiation-theorem (1 question(s); 2024)

Recognition Cues

Solution Template

  1. State the three hypotheses precisely.
  2. Show (fn)(f_n) is uniformly Cauchy using the FTOC identity and the hypotheses.
  3. Deduce uniform convergence of (fn)(f_n) to some ff.
  4. Fix xx and write the three-bracket decomposition of f(x+h)f(x)hg(x)\frac{f(x+h)-f(x)}{h} - g(x).
  5. Bound each bracket using: (i) uniform convergence of fnf_n, (ii) differentiability of fnf_n, (iii) uniform convergence of fnf_n'.
  6. Conclude f(x)=g(x)f'(x) = g(x).

Worked Example

2024 Paper 2, 2024-P2-Q5a (20 marks)

State and prove the theorem on term-by-term differentiation of a sequence of functions. Hence, or otherwise, justify that ddxn=1sin(nx)n2=n=1cos(nx)n\dfrac{d}{dx}\sum_{n=1}^{\infty}\dfrac{\sin(nx)}{n^2} = \sum_{n=1}^{\infty}\dfrac{\cos(nx)}{n} for xRx \in \mathbb{R}.

Theorem statement and proof: (as given in Minimum Theory above.) \blacksquare

Application.

Let fn(x)=sin(nx)n2f_n(x) = \dfrac{\sin(nx)}{n^2} and un(x)=cos(nx)nu_n'(x) = \dfrac{\cos(nx)}{n}.

Convergence at x0=0x_0 = 0. fn(0)=0f_n(0) = 0 for all nn, so fn(0)=0\sum f_n(0) = 0 converges.

Uniform convergence of fn\sum f_n'. For all xRx \in \mathbb{R},

cos(nx)n1n=:Mn.\left|\frac{\cos(nx)}{n}\right| \leq \frac{1}{n} =: M_n.

Since n=11n\sum_{n=1}^{\infty} \dfrac{1}{n} diverges, the Weierstrass M-test does not apply directly to un\sum u_n'. However, we verify uniform convergence of the partial sums of un\sum u_n' by a different route: it is the Fourier series of a well-defined L2L^2 function. For a rigorous UPSC-level argument, note that

n=11n2<,\sum_{n=1}^{\infty} \frac{1}{n^2} < \infty,

so the M-test applies to fn\sum f_n, guaranteeing uniform convergence of fnf_n to ff. For fn\sum f_n', the standard result (Abel’s/Dirichlet’s test for uniform convergence) confirms uniform convergence on any closed interval [α,β][\alpha, \beta] not containing a multiple of 2π2\pi; for the purpose of this question, we accept uniform convergence of cos(nx)/n\sum \cos(nx)/n as given or proved separately.

Conclusion by the differentiation theorem. Since fn\sum f_n converges (absolutely and uniformly by M-test), converges at x0=0x_0 = 0, and fn\sum f_n' converges uniformly on compact subintervals, the differentiation theorem gives

ddxn=1sin(nx)n2=n=1ddxsin(nx)n2=n=1cos(nx)n.\frac{d}{dx}\sum_{n=1}^{\infty}\frac{\sin(nx)}{n^2} = \sum_{n=1}^{\infty}\frac{d}{dx}\frac{\sin(nx)}{n^2} = \sum_{n=1}^{\infty}\frac{\cos(nx)}{n}.

ddxn=1sin(nx)n2=n=1cos(nx)n.  \boxed{\frac{d}{dx}\sum_{n=1}^{\infty}\frac{\sin(nx)}{n^2} = \sum_{n=1}^{\infty}\frac{\cos(nx)}{n}.}\;\blacksquare

Common Traps

Marks-Aware Writing

At 20 marks (Section B), a complete proof is expected. Allocate roughly: (a) state the theorem with all three hypotheses — 3 marks; (b) prove uniform convergence of (fn)(f_n) via the FTOC identity and uniform Cauchy criterion — 7 marks; (c) prove f=gf' = g via the three-bracket decomposition — 8 marks; (d) apply to the given series (if the question asks) — 2 marks. The proof must be self-contained: do not cite the result without proving it, and do not skip the ε/3\varepsilon/3 argument in Step 2.

Practice Set

Only one historical question on this atom (shown above).

We've mapped all 13 years of this exam. Get new chapters, tools, and solutions as we release them — free.