Fall 2013 - Problem 12

bounded variation

Let f ⁣:[0,1]R\func{f}{\br{0,1}}{\R} be a continuous function that is absolutely continuous on each interval [ε,1]\br{\epsilon, 1} with 0<ε10 < \epsilon \leq 1.

  1. Show that ff is not necessarily absolutely continuous on [0,1]\br{0, 1}.
  2. Show that if ff is of bounded variation on [0,1]\br{0, 1}, then ff is absolutely continuous on [0,1]\br{0, 1}.
Solution.
  1. Let f ⁣:[0,1]R\func{f}{\br{0,1}}{\R} be given by f(x)=xsin(1x)f\p{x} = x\sin\p{\frac{1}{x}} when x0x \neq 0 and f(0)=0f\p{0} = 0. Observe that if ε>0\epsilon > 0, then on [ε,1]\br{\epsilon, 1},

    f(x)=sin(1x)1xcos(1x)    f(x)1+1ε,f'\p{x} = \sin\p{\frac{1}{x}} - \frac{1}{x} \cos\p{\frac{1}{x}} \implies \abs{f'\p{x}} \leq 1 + \frac{1}{\epsilon},

    so ff is Lipschitz on [ε,1]\br{\epsilon, 1}, hence absolutely continuous. To show that ff is not absolutely continuous on [0,1]\br{0, 1}, let

    an=1π2+2πnandbn=1π2+2πn,a_n = \frac{1}{\frac{\pi}{2} + 2\pi n} \quad\text{and}\quad b_n = \frac{1}{-\frac{\pi}{2} + 2\pi n},

    i.e., so an<bna_n < b_n and these are essentially points of local maximum and minimum, respectively. Observe that

    bnan=π4π2n2π241n2f(bn)f(an)=an+bn=4πn4πn2π24.\begin{gathered} \abs{b_n - a_n} = \frac{\pi}{4\pi^2 n^2 - \frac{\pi^2}{4}} \leq \frac{1}{n^2} \\ \abs{f\p{b_n} - f\p{a_n}} = a_n + b_n = \frac{4\pi n}{4\pi n^2 - \frac{\pi^2}{4}}. \end{gathered}

    Thus, for any δ>0\delta > 0, we can pick NNN \in \N large enough so that

    n=Nbnan<δ.\sum_{n=N}^\infty \abs{b_n - a_n} < \delta.

    But

    n=Nf(bn)f(an)=,\sum_{n=N}^\infty \abs{f\p{b_n} - f\p{a_n}} = \infty,

    so we can pick M>NM > N so large that

    n=NMbnan<δ,butn=NMf(bn)f(an)1,\sum_{n=N}^M \abs{b_n - a_n} < \delta, \quad\text{but}\quad \sum_{n=N}^M \abs{f\p{b_n} - f\p{a_n}} \geq 1,

    so ff cannot be absolutely continuous.

  2. We denote the variation of ff on [a,b]\br{a, b} via

    Vf([a,b])=supa=t1<<tn=b;nNi=1n1f(ti+1f(ti)).V_f\p{\br{a,b}} = \sup_{a = t_1 < \cdots < t_n = b; n \in \N} \sum_{i=1}^{n-1} \abs{f\p{t_{i+1} - f\p{t_i}}}.

    First, we will show that xVf([0,x])x \mapsto V_f\p{\br{0,x}} is continuous. Let ε>0\epsilon > 0 and {0=t0<t1<<tn=1}\set{0 = t_0 < t_1 < \cdots < t_n = 1} be a partition such that

    i=1n1f(ti+1f(ti))Vf([0,1])ε.\sum_{i=1}^{n-1} \abs{f\p{t_{i+1} - f\p{t_i}}} \geq V_f\p{\br{0,1}} - \epsilon.

    Since ff is continuous at 00, we can pick δ(t0,t1)\delta \in \p{t_0, t_1} small enough so that f(δ)f(0)<ε\abs{f\p{\delta} - f\p{0}} < \epsilon. On the other hand, δ<t1<<tn=1\delta < t_1 < \cdots < t_n = 1 is a partition of [δ,1]\br{\delta, 1}, and so

    Vf([0,1])εf(δ)f(0)+f(t1)f(δ)+i=2n1f(ti+1f(ti))ε+Vf([δ,1]).V_f\p{\br{0,1}} - \epsilon \leq \abs{f\p{\delta} - f\p{0}} + \abs{f\p{t_1} - f\p{\delta}} + \sum_{i=2}^{n-1} \abs{f\p{t_{i+1} - f\p{t_i}}} \leq \epsilon + V_f\p{\br{\delta, 1}}.

    Rearranging, we see Vf([0,δ])=Vf([0,1])Vf([δ,1])<2εV_f\p{\br{0, \delta}} = V_f\p{\br{0, 1}} - V_f\p{\br{\delta, 1}} < 2\epsilon, which proves the claim.

    To show that ff is absolutely continuous on [0,1]\br{0, 1}, let ε>0\epsilon > 0 and let h>0h > 0 be such that Vf([0,h])εV_f\p{\br{0, h}} \leq \epsilon. Since ff is absolutely continuous on [0,1]\br{0, 1}, let δ>0\delta > 0 come from the definition of absolute continuity.

    Suppose a1<b1a2<b2an<bna_1 < b_1 \leq a_2 < b_2 \leq \cdots \leq a_n < b_n is such that i=1nbiai<δ\sum_{i=1}^n \abs{b_i - a_i} < \delta. Then shrinking hh if necessary, we may assume without loss of generality that h[a1,b1]h \in \br{a_1, b_1} and so

    i=1nf(bi)f(ai)f(a1)f(0)+f(h)f(a1)+f(b1)f(h)+i=2nf(bi)f(ai)Vf([0,h])+f(b1)f(h)+i=2nf(bi)f(ai)2ε.\begin{aligned} \sum_{i=1}^n \abs{f\p{b_i} - f\p{a_i}} &\leq \abs{f\p{a_1} - f\p{0}} + \abs{f\p{h} - f\p{a_1}} + \abs{f\p{b_1} - f\p{h}} + \sum_{i=2}^n \abs{f\p{b_i} - f\p{a_i}} \\ &\leq V_f\p{\br{0, h}} + \abs{f\p{b_1} - f\p{h}} + \sum_{i=2}^n \abs{f\p{b_i} - f\p{a_i}} \\ &\leq 2\epsilon. \end{aligned}

    Indeed, {0,a1,h}\set{0, a_1, h} gives a partition of [0,h]\br{0, h}, and {h,b1,a2,b2,,an,bn}\set{h, b_1, a_2, b_2, \ldots, a_n, b_n} gives a disjoint collection of intervals in [h,1]\br{h, 1} with total length less than δ\delta, and this completes the proof.