The Weierstrass M-Test Is Tight: A Conditionally Convergent Series of Functions
Consider the series
f(x)=โn=1โโn(โ1)nโsin(nx),xโR.
(a) Show that for each fixed xโ(0,2ฯ), the series converges.
(b) Show that the convergence is not uniform on (0,2ฯ).
(Hint for (b): Think about what happens near x=0.)
uniform convergenceFourier seriesDirichlet testpointwise convergencesawtooth function
Answer: Conditionally Convergent Series of Functions: Pointwise but Not Uniform
Key Idea / Intuition
The series converges pointwise by the Dirichlet test: the partial sums of sin(nx) are bounded (they satisfy a telescoping geometric sum estimate) and 1/nโ0. But uniform convergence fails because near x=0 the terms sin(nx)/n don't die uniformly โ you can always find an x small enough that the partial sums are large. In short: pointwise convergence can hold everywhere while uniform convergence fails because the "trouble spot" chases xโ0.
Formal Proof / Solution
Part (a): Pointwise Convergence
Fix xโ(0,2ฯ). We apply the Dirichlet test for series: if bnโ=1/nโ0 and the partial sums ANโ(x)=โn=1Nโ(โ1)nsin(nx) are bounded in N, then the series converges.
We bound the partial sums using the identity
โn=1Nโ(โ1)nsin(nx)=Im(โn=1Nโ(โeix)n)=Im(1+eixโeix(1โ(โeix)N)โ).
Since xโ(0,2ฯ), we have eix๎ =โ1, so โฃ1+eixโฃ=2โฃcos(x/2)โฃ>0. Therefore
โฃANโ(x)โฃโคโฃ1+eixโฃ2โ=โฃcos(x/2)โฃ1โ<โ.
This bound is finite for each fixed xโ(0,2ฯ), so by the Dirichlet test, the series converges pointwise. โ
Part (b): Convergence Is Not Uniform
Suppose for contradiction the convergence were uniform on (0,2ฯ). Then the partial sums
SNโ(x)=โn=1Nโn(โ1)nโsin(nx)
satisfy supxโ(0,2ฯ)โโฃf(x)โSNโ(x)โฃโ0.
In particular, the tailRNโ(x)=f(x)โSNโ(x) would go to zero uniformly, so in particular SNโ(x) would be uniformly Cauchy, meaning
Now consider the single term remainder: take M=N+1, so we need
supxโ(0,2ฯ)โโN+1(โ1)N+1โsin((N+1)x)โ=N+11โsupxโโฃsin((N+1)x)โฃ=N+11โโ0.
This part is fine. The real issue is more subtle โ let us use a necessary condition for uniform convergence.
Key test: If โfnโ converges uniformly on (0,2ฯ), then supxโโฃfnโ(x)โฃโ0. Here fnโ(x)=n(โ1)nโsin(nx), and
supxโ(0,2ฯ)โโnsin(nx)โโ=n1โโ0.
This is consistent, so this test does not immediately give a contradiction. We need a sharper argument.
Better approach โ look at SNโ near x=0.
It is a classical fact (Fourier series of a sawtooth) that the series โn=1โโn(โ1)n+1โsin(nx) represents x/2 on (โฯ,ฯ). Rearranging signs,
So f(x)=โx/2, which is continuous on (0,2ฯ) but has a jump discontinuity at 0 and 2ฯ.
Now, each partial sum SNโ(x) is a finite sum of continuous functions, hence continuous and bounded on [0,2ฯ] with SNโ(0)=0 for all N.
If SNโโf=โx/2 uniformly on (0,2ฯ), then f would extend to a continuous function on [0,2ฯ] (as the uniform limit of continuous functions). But f(x)=โx/2โ0 as xโ0+ while we also have SNโ(0)=0 for all N. So far no contradiction.
The actual non-uniformity is near x=2ฯ: As xโ2ฯโ, f(x)=โx/2โโฯ, but SNโ(2ฯ)=0 for all N (since sin(2ฯn)=0). So
This shows supโฃfโSNโโฃโฅฯ>0 for all N, so convergence is not uniform. โ
Summary
| | What holds |
|---|---|
| Pointwise convergence on (0,2ฯ) | โ (Dirichlet test) |
| Uniform convergence on (0,2ฯ) | โ (limit function has a boundary jump) |
The moral: the Weierstrass M-test is sufficient but not necessary for uniform convergence. Conditional convergence (via Dirichlet) gives pointwise results, but the boundary behavior can destroy uniformity.