Math 5616, Spring 2000: Riemann-Stieltjes integrals: Functions of bounded variation
Page 1 of 2
We will investigate Riemann-Stieltjes integrals that are based on integrators of bounded variation, so we will begin with a discussion of this kind of function. On the interval [0, 1] the graph of the function f (x) = |1 2x| drops from 1 down to zero then bounces back up to one. As we will see, its total variation is 2. Variation measures the up-and-down distance traced out by the point f (x) as x moves through (in this case) the interval [0, 1]. If the function has a jump discontinuity we think of the point as traveling the distance across the jump instantly. Thus the Dirichlet function has innite variation over any subinterval of [0, 1] that has positive length. Denition: A function f (x) dened on an interval [a, b] is a function of bounded variation on [a, b] if there exists a number V such that for every partition of [a, b],
n
|f (xi ) f (xi1 )| V.
i=1
We then dene the variation of f on [a, b] to be the number
n
(1)
Vf [a, b] := sup
[a, b] i=1
|f (xi ) f (xi1 )|,
where [a, b] means is a partition of [a, b]. Our objective is to prove some things about functions of bounded variation. One is that such functions can be written as the dierence between two increasing functions. The path to proving this seems to me to require that we view the sum in (1) as a sum of positive and negative parts of the dierences f (xi ) f (xi1 ). Here is special notation for the sums that appear in the denition of function of bounded variation, and for two pieces of that sum.
n
Vf () :=
i=1 n
|f (xi ) f (xi1 )|, (f (xi ) f (xi1 ))+ ,
i=1 n
(2)
Pf () := Nf () :=
i=1
(f (xi ) f (xi1 )) .
In these denitions, if x is a real number, we write x+ := max(0, x) and x := max(0, x), the positive part and the negative part of x, respectively. Then we have |x| = x+ + x and x = x+ x . We can use these last two formulas to write (3) Vf () = Pf () + Nf () and f (b) f (a) = Pf () Nf ().
The sum with no absolute value signs is telescoping. Since, for each , Vf () Vf [a, b] we have Pf () + Nf () Vf [a, b]. Similarly, (4) where
n n
Vf () Pf [a, b] + Nf [a, b],
(5)
Pf [a, b] := sup
[a, b] i=1
(f (xi ) f (xi1 ))
and Nf [a, b] := sup
[a, b] i=1
(f (xi ) f (xi1 )) .
Math 5616, Spring 2000: Riemann-Stieltjes integrals: Functions of bounded variation The rst of these is called the positive variation, the second the negative variation. Hence (6) We can now state the Theorem we want. Theorem: Let f be a real-valued function dened on [a, b]. Then Vf [a, b] = Pf [a, b] + Nf [a, b]. If one of Vf [a, b], Pf [a, b] and Nf [a, b] is nite, then all three are nite. If f is of bounded variation, f (b) f (a) = Pf [a, b] Nf [a, b]. Vf [a, b] Pf [a, b] + Nf [a, b].
Page 2 of 2
(BV )
Proof: By (6), half of the rst equation is done. To prove that the inequality in the other direction is true, we will use the observation you all made in class, that the second equation in (3) can be solved for one of the two sums in terms of the other. Therefore, for instance, Vf () = Pf () + Nf () = f (b) f (a) + 2Nf (). We can replace the left-hand side by its least upper bound: Vf [a, b] f (b) f (a) + 2Nf (). Now we can take the supremum on the right Vf [a, b] f (b) f (a) + 2Nf [a, b]. Similarly, Vf [a, b] 2Pf [a, b] (f (b) f (a)). On adding these two inequalities and then dividing by 2, we have (7) Hence Vf [a, b] = Pf [a, b] + Nf [a, b]. We never made any assumption that any of the variations was nite. However, we did exploit the niteness of each individual partition-sum. If Vf [a, b] is nite, (7) shows that both of Pf [a, b] and Nf [a, b] are nite. To continue, we will use the second equation in (3), and your observation, twice. Thus, from (3), (8) (9) (10) Nf () = Pf () (f (b) f (a)). Nf [a, b] Pf [a, b] (f (b) f (a)). Nf [a, b] Pf [a, b] (f (b) f (a)). We take the supremum on the right rst and then on the left. This gives Next we return to (8), and this time take the supremum on the left rst and then on the right. This gives The inequalities (9) and (10) show that, if one of Pf [a, b] and Nf [a, b] is nite, so is the other, and then by (6), so is Vf [a, b]. Finally, if f is of bounded variation, (9) and (10) show that the last equation in the Theorem is true. The proof is done. If a function is of bounded variation on [a, b] it is also of bounded variation on each subinterval [a, x] of [a, b]. This observation leads to the Theorem known as the Jordan Decomposition Theorem, which shows that a function of bounded variation can be expressed as the dierence of two increasing functions. Theorem (Jordan Decomposition): If f is a function of bounded variation on [a, b] then f can be written as the dierence of two increasing functions: f (x) = f (a) + Pf [a, x] Nf [a, x]. Conversely, if f (x) = g(x) h(x), where g and h are increasing on [a, b], then f is of bounded variation on [a, b], and Vf [a, b] Vg [a, b] + Vh [a, b] (= |g(b) g(a)| + |h(b) h(a)|). This is a Corollary of Theorem (BV ) that we proved above.. Vf [a, b] Pf [a, b] + Nf [a, b].
