Discontinuities of monotone functions

Template:Short description Template:Use dmy dates Template:Technical In the mathematical field of analysis, a well-known theorem describes the set of discontinuities of a monotone real-valued function of a real variable; all discontinuities of such a (monotone) function are necessarily jump discontinuities and there are at most countably many of them.

Usually, this theorem appears in literature without a name. It is called Froda's theorem in some recent works; in his 1929 dissertation, Alexandru Froda stated that the result was previously well-known and had provided his own elementary proof for the sake of convenience. Prior work on discontinuities had already been discussed in the 1875 memoir of the French mathematician Jean Gaston Darboux.

Definitions

Denote the limit from the left by $$f\left(x^{-}\right):=\underset{z\nearrow x}{lim}f(z)=\underset{\stackrel{h\to 0}{h>0}}{lim}f(x-h)$$ and denote the limit from the right by $$f\left(x^{+}\right):=\underset{z\searrow x}{lim}f(z)=\underset{\stackrel{h\to 0}{h>0}}{lim}f(x+h).$$

If $f\left(x^{+}\right)$ and $f\left(x^{-}\right)$ exist and are finite then the difference $f\left(x^{+}\right)-f\left(x^{-}\right)$ is called the jumpTemplate:Sfn of $f$ at $x.$

Consider a real-valued function $f$ of real variable $x$ defined in a neighborhood of a point $x.$ If $f$ is discontinuous at the point $x$ then the discontinuity will be a removable discontinuity, or an essential discontinuity, or a jump discontinuity (also called a discontinuity of the first kind).Template:Sfn If the function is continuous at $x$ then the jump at $x$ is zero. Moreover, if $f$ is not continuous at $x,$ the jump can be zero at $x$ if $f\left(x^{+}\right)=f\left(x^{-}\right)\ne f(x).$

Precise statement

Let $f$ be a real-valued monotone function defined on an interval $I.$ Then the set of discontinuities of the first kind is at most countable.

One can proveTemplate:SfnTemplate:Sfn that all points of discontinuity of a monotone real-valued function defined on an interval are jump discontinuities and hence, by our definition, of the first kind. With this remark the theorem takes the stronger form:

Let $f$ be a monotone function defined on an interval $I.$ Then the set of discontinuities is at most countable.

Proofs

This proof starts by proving the special case where the function's domain is a closed and bounded interval $[a,b].$Template:SfnTemplate:Sfn The proof of the general case follows from this special case.

Proof when the domain is closed and bounded

Two proofs of this special case are given.

Proof 1

Let $I:=[a,b]$ be an interval and let $f:I\to ℝ$ be a non-decreasing function (such as an increasing function). Then for any $a<x<b,$ $$f(a)\le f\left(a^{+}\right)\le f\left(x^{-}\right)\le f\left(x^{+}\right)\le f\left(b^{-}\right)\le f(b).$$ Let $\alpha >0$ and let $x_1<x_2<\cdots <x_n$ be $n$ points inside $I$ at which the jump of $f$ is greater or equal to $\alpha$: $$f\left(x_i^{+}\right)-f\left(x_i^{-}\right)\ge \alpha ,i=1,2,\dots ,n$$

For any $i=1,2,\dots ,n,$ $f\left(x_i^{+}\right)\le f\left(x_{i+1}^{-}\right)$ so that $f\left(x_{i+1}^{-}\right)-f\left(x_i^{+}\right)\ge 0.$ Consequently, $$\begin{array}{cc}f(b)-f(a)& \ge f\left(x_n^{+}\right)-f\left(x_1^{-}\right)\\ & =\sum _{i=1}^n[f\left(x_i^{+}\right)-f\left(x_i^{-}\right)]+\sum _{i=1}^{n-1}[f\left(x_{i+1}^{-}\right)-f\left(x_i^{+}\right)]\\ & \ge \sum _{i=1}^n[f\left(x_i^{+}\right)-f\left(x_i^{-}\right)]\\ & \ge n\alpha \end{array}$$ and hence $n\le \frac{f(b)-f(a)}{\alpha }.$

Since $f(b)-f(a)<\mathrm{\infty }$ we have that the number of points at which the jump is greater than $\alpha$ is finite (possibly even zero).

Define the following sets: $$S_1:=\{x:x\in I,f\left(x^{+}\right)-f\left(x^{-}\right)\ge 1\},$$ $$S_n:=\{x:x\in I,\frac{1}{n}\le f\left(x^{+}\right)-f\left(x^{-}\right)<\frac{1}{n-1}\},n\ge 2.$$

Each set $S_n$ is finite or the empty set. The union $S=\bigcup _{n=1}^{\mathrm{\infty }}{S}_n$ contains all points at which the jump is positive and hence contains all points of discontinuity. Since every $S_i,i=1,2,\dots$ is at most countable, their union $S$ is also at most countable.

If $f$ is non-increasing (or decreasing) then the proof is similar. This completes the proof of the special case where the function's domain is a closed and bounded interval. $◼$

Proof 2

For a monotone function $f$, let $f\nearrow$ mean that $f$ is monotonically non-decreasing and let $f\searrow$ mean that $f$ is monotonically non-increasing. Let $f:[a,b]\to ℝ$ is a monotone function and let $D$ denote the set of all points $d\in [a,b]$ in the domain of $f$ at which $f$ is discontinuous (which is necessarily a jump discontinuity).

Because $f$ has a jump discontinuity at $d\in D,$ $f\left(d^{-}\right)\ne f\left(d^{+}\right)$ so there exists some rational number $y_d\in \mathbb{Q}$ that lies strictly in between $f\left(d^{-}\right)\text{ and }f\left(d^{+}\right)$ (specifically, if $f\nearrow$ then pick $y_d\in \mathbb{Q}$ so that $f\left(d^{-}\right)<y_d<f\left(d^{+}\right)$ while if $f\searrow$ then pick $y_d\in \mathbb{Q}$ so that $f\left(d^{-}\right)>y_d>f\left(d^{+}\right)$ holds).

It will now be shown that if $d,e\in D$ are distinct, say with $d<e,$ then $y_d\ne y_e.$ If $f\nearrow$ then $d<e$ implies $f\left(d^{+}\right)\le f\left(e^{-}\right)$ so that $y_d<f\left(d^{+}\right)\le f\left(e^{-}\right)<y_e.$ If on the other hand $f\searrow$ then $d<e$ implies $f\left(d^{+}\right)\ge f\left(e^{-}\right)$ so that $y_d>f\left(d^{+}\right)\ge f\left(e^{-}\right)>y_e.$ Either way, $y_d\ne y_e.$

Thus every $d\in D$ is associated with a unique rational number (said differently, the map $D\to \mathbb{Q}$ defined by $d\mapsto y_d$ is injective). Since $\mathbb{Q}$ is countable, the same must be true of $D.$ $◼$

Proof of general case

Suppose that the domain of $f$ (a monotone real-valued function) is equal to a union of countably many closed and bounded intervals; say its domain is $\bigcup _n[a_n,b_n]$ (no requirements are placed on these closed and bounded intervalsTemplate:Efn). It follows from the special case proved above that for every index $n,$ the restriction $f{|}_{[a_n,b_n]}:[a_n,b_n]\to ℝ$ of $f$ to the interval $[a_n,b_n]$ has at most countably many discontinuities; denote this (countable) set of discontinuities by $D_n.$ If $f$ has a discontinuity at a point $x_0\in \bigcup _n[a_n,b_n]$ in its domain then either $x_0$ is equal to an endpoint of one of these intervals (that is, $x_0\in \{a_1,b_1,a_2,b_2,\dots \}$) or else there exists some index $n$ such that $a_n<x_0<b_n,$ in which case $x_0$ must be a point of discontinuity for $f{|}_{[a_n,b_n]}$ (that is, $x_0\in D_n$). Thus the set $D$ of all points of at which $f$ is discontinuous is a subset of $\{a_1,b_1,a_2,b_2,\dots \}\cup \bigcup _n{D}_n,$ which is a countable set (because it is a union of countably many countable sets) so that its subset $D$ must also be countable (because every subset of a countable set is countable).

In particular, because every interval (including open intervals and half open/closed intervals) of real numbers can be written as a countable union of closed and bounded intervals, it follows that any monotone real-valued function defined on an interval has at most countable many discontinuities.

To make this argument more concrete, suppose that the domain of $f$ is an interval $I$ that is not closed and bounded (and hence by Heine–Borel theorem not compact). Then the interval can be written as a countable union of closed and bounded intervals $I_n$ with the property that any two consecutive intervals have an endpoint in common: $I={\cup }_{n=1}^{\mathrm{\infty }}{I}_n.$ If $I=(a,b]\text{ with }a\ge -\mathrm{\infty }$ then $I_1=[{\alpha }_1,b],I_2=[{\alpha }_2,{\alpha }_1],\dots ,I_n=[{\alpha }_n,{\alpha }_{n-1}],\dots$ where ${\left({\alpha }_n\right)}_{n=1}^{\mathrm{\infty }}$ is a strictly decreasing sequence such that ${\alpha }_n\to a.$ In a similar way if $I=[a,b),\text{ with }b\le +\mathrm{\infty }$ or if $I=(a,b)\text{ with }-\mathrm{\infty }\le a<b\le \mathrm{\infty }.$ In any interval $I_n,$ there are at most countable many points of discontinuity, and since a countable union of at most countable sets is at most countable, it follows that the set of all discontinuities is at most countable. $◼$

Jump functions

Examples. Let Template:Mvar₁ < Template:Mvar₂ < Template:Mvar₃ < ⋅⋅⋅ be a countable subset of the compact interval [[[:Template:Mvar]],Template:Mvar] and let μ₁, μ₂, μ₃, ... be a positive sequence with finite sum. Set

$f(x)=\sum _{n=1}^{\mathrm{\infty }}{\mu }_n{\chi }_{[x_n,b]}(x)$

where χ_A denotes the characteristic function of a compact interval Template:Mvar. Then Template:Mvar is a non-decreasing function on [[[:Template:Mvar]],Template:Mvar], which is continuous except for jump discontinuities at Template:Mvar_{Template:Mvar} for Template:Mvar ≥ 1. In the case of finitely many jump discontinuities, Template:Mvar is a step function. The examples above are generalised step functions; they are very special cases of what are called jump functions or saltus-functions.Template:SfnTemplate:Sfn

More generally, the analysis of monotone functions has been studied by many mathematicians, starting from Abel, Jordan and Darboux. Following Template:Harvtxt, replacing a function by its negative if necessary, only the case of non-negative non-decreasing functions has to be considered. The domain [[[:Template:Mvar]],Template:Mvar] can be finite or have ∞ or −∞ as endpoints.

The main task is to construct monotone functions — generalising step functions — with discontinuities at a given denumerable set of points and with prescribed left and right discontinuities at each of these points. Let Template:Mvar (Template:Mvar ≥ 1) lie in (Template:Mvar, Template:Mvar) and take λ₁, λ₂, λ₃, ... and μ₁, μ₂, μ₃, ... non-negative with finite sum and with λ_{Template:Mvar} + μ_{Template:Mvar} > 0 for each Template:Mvar. Define

$f_n(x)=0$ for $x<x_n,f_n(x_n)={\lambda }_n,f_n(x)={\lambda }_n+{\mu }_n$ for $x>x_n.$

Then the jump function, or saltus-function, defined by

$f(x)=\sum _{n=1}^{\mathrm{\infty }}{f}_n(x)=\sum _{x_n\le x}{\lambda }_n+\sum _{x_n<x}{\mu }_n,$

is non-decreasing on [[[:Template:Mvar]], Template:Mvar] and is continuous except for jump discontinuities at Template:Mvar for Template:Mvar ≥ 1.Template:SfnTemplate:SfnTemplate:Sfn

To prove this, note that sup |Template:Mvar_{Template:Mvar}| = λ_{Template:Mvar} + μ_{Template:Mvar}, so that Σ Template:Mvar_{Template:Mvar} converges uniformly to Template:Mvar. Passing to the limit, it follows that

$f(x_n)-f(x_n-0)={\lambda }_n,f(x_n+0)-f(x_n)={\mu }_n,$ and $f(x\pm 0)=f(x)$

if Template:Mvar is not one of the Template:Mvar_{Template:Mvar}'s.

Conversely, by a differentiation theorem of Lebesgue, the jump function Template:Mvar is uniquely determined by the properties: (1) being non-decreasing and non-positive; (2) having given jump data at its points of discontinuity Template:Mvar_{Template:Mvar}; (3) satisfying the boundary condition Template:Mvar(Template:Mvar) = 0; and (4) having zero derivative almost everywhere.

Template:Collapse top Property (4) can be checked following Template:Harvtxt, Template:Harvtxt and Template:Harvtxt. Without loss of generality, it can be assumed that Template:Mvar is a non-negative jump function defined on the compact [[[:Template:Mvar]],Template:Mvar], with discontinuities only in (Template:Mvar,Template:Mvar).

Note that an open set Template:Mvar of (Template:Mvar,Template:Mvar) is canonically the disjoint union of at most countably many open intervals Template:Mvar_{Template:Mvar}; that allows the total length to be computed ℓ(Template:Mvar)= Σ ℓ(Template:Mvar_{Template:Mvar}). Recall that a null set Template:Mvar is a subset such that, for any arbitrarily small ε' > 0, there is an open Template:Mvar containing Template:Mvar with ℓ(Template:Mvar) < ε'. A crucial property of length is that, if Template:Mvar and Template:Mvar are open in (Template:Mvar,Template:Mvar), then ℓ(Template:Mvar) + ℓ(Template:Mvar) = ℓ(Template:Mvar ∪ Template:Mvar) + ℓ(Template:Mvar ∩ Template:Mvar).Template:Sfn It implies immediately that the union of two null sets is null; and that a finite or countable set is null.

Proposition 1. For Template:Mvar > 0 and a normalised non-negative jump function Template:Mvar, let Template:Mvar_{Template:Mvar}(Template:Mvar) be the set of points Template:Mvar such that

$\frac{f(t)-f(s)}{t-s}>c$

for some Template:Mvar, Template:Mvar with Template:Mvar < Template:Mvar < Template:Mvar. Then Template:Mvar_{Template:Mvar}(Template:Mvar) is open and has total length ℓ(Template:Mvar_{Template:Mvar}(Template:Mvar)) ≤ 4 Template:Mvar⁻¹ (Template:Mvar(Template:Mvar) – Template:Mvar(Template:Mvar)).

Note that Template:Mvar_{Template:Mvar}(Template:Mvar) consists the points Template:Mvar where the slope of Template:Mvar is greater that Template:Mvar near Template:Mvar. By definition Template:Mvar_{Template:Mvar}(Template:Mvar) is an open subset of (Template:Mvar, Template:Mvar), so can be written as a disjoint union of at most countably many open intervals Template:Mvar_{Template:Mvar} = (Template:Mvar_{Template:Mvar}, Template:Mvar_{Template:Mvar}). Let Template:Mvar_{Template:Mvar} be an interval with closure in Template:Mvar_{Template:Mvar} and ℓ(Template:Mvar_{Template:Mvar}) = ℓ(Template:Mvar_{Template:Mvar})/2. By compactness, there are finitely many open intervals of the form (Template:Mvar,Template:Mvar) covering the closure of Template:Mvar_{Template:Mvar}. On the other hand, it is elementary that, if three fixed bounded open intervals have a common point of intersection, then their union contains one of the three intervals: indeed just take the supremum and infimum points to identify the endpoints. As a result, the finite cover can be taken as adjacent open intervals (Template:Mvar_{Template:Mvar},Template:Mvar_{Template:Mvar}), (Template:Mvar_{Template:Mvar},Template:Mvar_{Template:Mvar}), ... only intersecting at consecutive intervals. Hence

$\ell (J_k)\le \sum _m(t_{k,m}-s_{k,m})\le \sum _m{c}^{-1}(f(t_{k,m})-f(s_{k,m}))\le 2c^{-1}(f(b_k)-f(a_k)).$

Finally sum both sides over Template:Mvar.

Proposition 2. If Template:Mvar is a jump function, then Template:Mvar '(Template:Mvar) = 0 almost everywhere.

To prove this, define

$Df(x)=\underset{s,t\to x,s<x<t}{lim\; sup}\frac{f(t)-f(s)}{t-s},$

a variant of the Dini derivative of Template:Mvar. It will suffice to prove that for any fixed Template:Mvar > 0, the Dini derivative satisfies Template:MvarTemplate:Mvar(Template:Mvar) ≤ Template:Mvar almost everywhere, i.e. on a null set.

Choose ε > 0, arbitrarily small. Starting from the definition of the jump function Template:Mvar = Σ Template:Mvar_{Template:Mvar}, write Template:Mvar = Template:Mvar + Template:Mvar with Template:Mvar = Σ_{Template:Mvar≤Template:Mvar} Template:Mvar_{Template:Mvar} and Template:Var = Σ_{Template:Mvar>Template:Mvar} Template:Mvar_{Template:Mvar} where Template:Mvar ≥ 1. Thus Template:Mvar is a step function having only finitely many discontinuities at Template:Mvar_{Template:Mvar} for Template:Mvar ≤ Template:Mvar and Template:Mvar is a non-negative jump function. It follows that Template:MvarTemplate:Mvar = Template:Mvar' +Template:MvarTemplate:Mvar = Template:MvarTemplate:Mvar except at the Template:Mvar points of discontinuity of Template:Mvar. Choosing Template:Mvar sufficiently large so that Σ_{Template:Mvar>Template:Mvar} λ_{Template:Mvar} + μ_{Template:Mvar} < ε, it follows that Template:Mvar is a jump function such that Template:Mvar(Template:Mvar) − Template:Mvar(Template:Mvar) < ε and Template:Mvar ≤ Template:Mvar off an open set with length less than 4ε/Template:Mvar.

By construction Template:Mvar ≤ Template:Mvar off an open set with length less than 4ε/Template:Mvar. Now set ε' = 4ε/Template:Mvar — then ε' and Template:Mvar are arbitrarily small and Template:Mvar ≤ Template:Mvar off an open set of length less than ε'. Thus Template:Mvar ≤ Template:Mvar almost everywhere. Since Template:Mvar could be taken arbitrarily small, Template:Mvar and hence also Template:Mvar ' must vanish almost everywhere. Template:Collapse bottom As explained in Template:Harvtxt, every non-decreasing non-negative function Template:Mvar can be decomposed uniquely as a sum of a jump function Template:Mvar and a continuous monotone function Template:Mvar: the jump function Template:Mvar is constructed by using the jump data of the original monotone function Template:Mvar and it is easy to check that Template:Mvar = Template:Mvar − Template:Mvar is continuous and monotone.

See also

• Template:Annotated link
• Template:Annotated link
• Monotone function

Bibliography

• Template:Cite book
• Template:Cite journal
• Template:Cite book (subscription required)
• Template:Cite book
• Template:Cite book
• Template:Citation; reprinted by Dover, 2003
• Template:Cite book
• Template:Cite book
• Template:Cite journal
• Template:Cite book
• Template:Citation
• Template:Citation
• Template:Citation
• Template:Cite book Reprint of the 1955 original.
• Template:Cite book
• Template:Cite journal
• Template:Citation
• Template:Cite book
• Template:Cite journal