Real Analysis

Uniform Convergence

f_n → f uniformly iff sup_x |f_n(x) − f(x)| → 0 — preserves continuity, integrability, and limits of integrals

A sequence of functions f_n: D → ℝ converges uniformly to f if for every ε > 0 there exists N such that |f_n(x) − f(x)| < ε for all x ∈ D and all n ≥ N. The "uniform" means N depends only on ε, not on x — distinguishing this from pointwise convergence where N may depend on each x. Crucial because uniform limits preserve continuity, Riemann integrability, and allow swapping ∫ and lim. Famous counterexample with pointwise: f_n(x) = x^n on [0, 1] converges pointwise to a discontinuous function. Weierstrass M-test: if |aₙ(x)| ≤ Mₙ and Σ Mₙ < ∞, then Σ aₙ(x) converges uniformly — workhorse for proving Fourier and power series convergence.

  • Definitionsup |f_n − f| → 0
  • Nindependent of x
  • Preservescontinuity, Riemann
  • Allows∫ ↔ lim swap
  • WorkhorseWeierstrass M-test
  • Counterexamplex^n on [0,1]

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Why uniform convergence matters

  • Fourier series. Sufficient (via Weierstrass M-test or Dirichlet conditions) to write a periodic function as a uniformly convergent trigonometric series — basis of signal processing and quantum mechanics.
  • Power series. Converge uniformly on every closed sub-disk strictly inside the radius of convergence — gives term-by-term differentiation and integration of analytic functions.
  • Riemann integration. The interchange ∫ lim = lim ∫ holds for uniformly convergent sequences of integrable functions — bedrock of integration of series.
  • Continuity preservation. A uniform limit of continuous functions is continuous; the function space C(K) is closed under uniform limits, hence complete (a Banach space under sup norm).
  • ODE theory (Picard). Successive approximations form a uniformly Cauchy sequence in C([a, b]); their uniform limit is the unique solution.
  • Approximation theory. Weierstrass approximation theorem: any continuous function on [a, b] is the uniform limit of polynomials — supports numerical analysis and machine learning function representations.
  • Functional analysis. The sup-norm topology on C(K) is uniform convergence; convergence in this Banach space is exactly uniform convergence.

Common misconceptions

  • Uniform = pointwise. Uniform is strictly stronger; pointwise allows N to depend on x, uniform doesn't. The classic x^n on [0, 1] separates them.
  • Uniform implies pointwise. Yes — but pointwise does NOT imply uniform. Pointwise without bounded sup gives non-uniform convergence.
  • Uniform allows swapping derivatives. No — derivatives need uniform convergence of the f_n′, not just of f_n. f_n(x) = sin(n²x)/n is uniformly small but f_n′ blows up.
  • Series converge uniformly automatically. Need a test like Weierstrass M; pointwise convergence of series doesn't imply uniform.
  • Uniform on (0, 1) ⇒ uniform on [0, 1]. Open vs closed matters: x^n is uniform on [0, 1 − ε] but never on [0, 1].
  • Uniform Cauchy ≠ uniform convergent. Equivalent in C(K) (complete), but in general spaces the limit must exist; uniform Cauchy is the stand-in when the limit isn't known a priori.

Frequently asked questions

What's the difference between uniform and pointwise convergence?

Pointwise: for every x and every ε > 0, there exists N(x, ε) such that |f_n(x) − f(x)| < ε for n ≥ N. Uniform: for every ε > 0, there exists N(ε) (no x-dependence) such that |f_n(x) − f(x)| < ε for all x and n ≥ N. Equivalently, sup_x |f_n(x) − f(x)| → 0. Uniform convergence requires the same N to work for every x, which is strictly stronger than pointwise. Standard counterexample: f_n(x) = x^n on [0, 1] converges pointwise to f(x) = 0 for x < 1, f(1) = 1 — but the convergence is not uniform because near x = 1 you need huge n to get f_n(x) close to 0.

Why does uniform preserve continuity but pointwise doesn't?

Continuous limit theorem: a uniform limit of continuous functions is continuous. Proof uses the triangle inequality |f(x) − f(c)| ≤ |f(x) − f_n(x)| + |f_n(x) − f_n(c)| + |f_n(c) − f(c)| — pick n large so the first and third are < ε/3 uniformly in x and c, then continuity of f_n gives the middle term < ε/3. The pointwise N(x, ε) doesn't allow this — the first and third terms can't be controlled simultaneously near c. Counterexample: f_n(x) = x^n on [0, 1] are all continuous; pointwise limit is discontinuous at x = 1.

What is the Weierstrass M-test?

If |aₙ(x)| ≤ Mₙ for all x ∈ D and Σ Mₙ converges, then Σ aₙ(x) converges uniformly on D. The bound Mₙ controls each term independently of x — that's what makes the convergence uniform. The M-test is the standard way to prove power series, Fourier series, and many integrals converge uniformly. Example: Σ sin(nx)/n² is uniformly convergent on ℝ because |sin(nx)/n²| ≤ 1/n² and Σ 1/n² = π²/6 converges.

How does uniform convergence let you swap limit and integral?

If f_n → f uniformly on [a, b] and each f_n is Riemann-integrable, then f is Riemann-integrable and ∫_a^b f_n → ∫_a^b f. Proof: |∫_a^b (f_n − f)| ≤ ∫_a^b |f_n − f| ≤ (b − a) · sup |f_n − f| → 0. Pointwise convergence is not enough; the limit may even fail to be integrable. The Lebesgue dominated convergence theorem replaces uniform convergence with a domination by an integrable function — a much weaker hypothesis but requires Lebesgue, not Riemann, integration.

Does differentiation commute with uniform limit?

Not from uniform convergence alone. Counterexample: f_n(x) = sin(n²x)/n converges uniformly to 0 on ℝ, but f_n′(x) = n cos(n²x) does not converge to 0 (the derivatives blow up). The correct theorem requires uniform convergence of the derivatives: if f_n → f pointwise and f_n′ → g uniformly on a closed bounded interval, then f is differentiable, f′ = g, and f_n converges uniformly. So differentiation requires more — uniform convergence of f_n′, not just of f_n.

What about uniform convergence on compact subsets?

A weaker but very practical condition: f_n → f uniformly on every compact subset K ⊂ D. Often called 'locally uniform' or 'compact-open' convergence. Power series converge uniformly on every compact subset inside the open disk of convergence (but not on the whole disk). Locally uniform is enough to inherit continuity and to swap limit with integral on compact intervals. The right notion in complex analysis (Weierstrass theorem on holomorphic limits requires only locally uniform convergence) and in PDE.