🧮 Brain Teaser

The Sequence That Must Have a Convergent Subsequence... or Does It?

Let fn:[0,1]→Rf_n : [0,1] \to \mathbb{R} be a sequence of functions defined by

fn(x)=sin⁡(nx).f_n(x) = \sin(nx).

Does (fn)(f_n) have a subsequence that converges pointwise almost everywhere on [0,1][0,1]?

Surprisingly: yes. But can you find a subsequence that converges pointwise everywhere on [0,1][0,1]?

More precisely: prove that no subsequence fnk(x)=sin⁡(nkx)f_{n_k}(x) = \sin(n_k x) converges at every point of [0,1][0,1].

Riemann-Lebesgueweak convergencepointwise convergenceL2Weyl equidistributiondominated convergence

Answer: The Sequence That Must Have a Convergent Subsequence... or Does It?

Key Idea / Intuition

The sequence sin⁡(nx)\sin(nx) is bounded in L2[0,1]L^2[0,1], and by the Riemann–Lebesgue lemma its L2L^2-norm doesn't vanish — it stays at 1/21/\sqrt{2}. By a soft functional-analysis argument, every subsequence converges weakly to zero in L2L^2, but any pointwise-everywhere convergent subsequence would have to converge to a measurable limit, and then by dominated convergence the L2L^2 norms would have to be 1/21/\sqrt{2} while the limit is zero — giving a contradiction. A more elementary route uses the Weyl equidistribution theorem: for any fixed irrational x/πx/\pi, the values sin⁡(nkx)\sin(n_k x) are equidistributed and can't converge. One cannot avoid such irrationalities simultaneously for all xx.


Formal Proof / Solution

Step 1: Weak convergence to zero.

For any g∈L2[0,1]g \in L^2[0,1], by the Riemann–Lebesgue lemma: ∫01sin⁡(nx) g(x) dx→0as n→∞.\int_0^1 \sin(nx)\, g(x)\, dx \to 0 \quad \text{as } n \to \infty.

So fn⇀0f_n \rightharpoonup 0 weakly in L2[0,1]L^2[0,1], and the same holds for any subsequence.

Step 2: L2L^2-norm stays bounded away from zero.

∥fn∥L22=∫01sin⁡2(nx) dx=12−sin⁡(2n)4n→12.\|f_n\|_{L^2}^2 = \int_0^1 \sin^2(nx)\, dx = \frac{1}{2} - \frac{\sin(2n)}{4n} \to \frac{1}{2}.

So ∥fnk∥L22→1/2≠0\|f_{n_k}\|_{L^2}^2 \to 1/2 \neq 0 for any subsequence.

Step 3: Pointwise a.e. convergence implies L2L^2 convergence via DCT.

Suppose, for contradiction, that some subsequence sin⁡(nkx)→h(x)\sin(n_k x) \to h(x) pointwise everywhere on [0,1][0,1].

Since ∣sin⁡(nkx)∣≤1|\sin(n_k x)| \leq 1, dominated convergence gives: ∫01h(x)2 dx=lim⁡k→∞∫01sin⁡2(nkx) dx=12.\int_0^1 h(x)^2\, dx = \lim_{k\to\infty} \int_0^1 \sin^2(n_k x)\, dx = \frac{1}{2}.

But weak convergence forces h=0h = 0 a.e.: for any measurable set AA, ∫Ah(x) dx=lim⁡k→∞∫01sin⁡(nkx) 1A(x) dx=0\int_A h(x)\, dx = \lim_{k\to\infty} \int_0^1 \sin(n_k x)\, \mathbf{1}_A(x)\, dx = 0 by Riemann–Lebesgue (since 1A∈L2⊂L1\mathbf{1}_A \in L^2 \subset L^1).

If ∫Ah dx=0\int_A h\, dx = 0 for all measurable AA, then h=0h = 0 a.e., contradicting ∥h∥L22=1/2>0\|h\|_{L^2}^2 = 1/2 > 0.

Conclusion.

No subsequence sin⁡(nkx)\sin(n_k x) can converge pointwise everywhere on [0,1][0,1]. (It can converge almost everywhere — that follows from the fact that sin⁡(nkx)⇀0\sin(n_k x) \rightharpoonup 0 in L2L^2 and by Banach–Saks or Komlós-type arguments a subsequence of Cesàro means converges a.e., but the full sequence itself already fails everywhere off a measure-zero set by Weyl equidistribution.)

Bonus elegant version via Weyl.

Fix any xx with x/πx/\pi irrational (a full-measure set). The sequence {nkx/π(mod1)}\{n_k x / \pi \pmod{1}\} is equidistributed mod 1 for any lacunary subsequence, so sin⁡(nkx)\sin(n_k x) visits both positive and negative values infinitely often — hence diverges. Since irrationals have full measure, no subsequence can converge everywhere.

Type: analysisEdit on GitHub ↗