Processing math: 100%
Skip to main content
Library homepage
 

Text Color

Text Size

 

Margin Size

 

Font Type

Enable Dyslexic Font
Mathematics LibreTexts

5.2: Derivatives of Extended-Real Functions

( \newcommand{\kernel}{\mathrm{null}\,}\)

For a while (in §§2 and 3), we limit ourselves to extended-real functions. Below, f and g are real or extended real (f, g:E1E). We assume, however, that they are not constantly infinite on any interval (a,b),a<b.

Lemma 5.2.1

If f(p)>0 at some pE1, then

x<p<y

implies

f(x)<f(p)<f(y)

for all x,y in a sufficiently small globe Gp(δ)=(pδ,p+δ).

Similarly, if f(p)<0, then x<p<y implies f(x)>f(p)>f(y) for x,y in some Gp(δ).

Proof

If f(p)>0, the "0" case in Definition 1 of §1, is excluded, so

f(p)=limxpΔfΔx>0.

Hence we must also have Δf/Δx>0 for x in some Gp(δ).

It follows that Δf and Δx have the same sign in Gp(δ); i.e.,

f(x)f(p)>0 if x>p and f(x)f(p)<0 if x<p.

(This implies f(p)±. Why? Hence

x<p<yf(x)<f(p)<f(y),

as claimed; similarly in case f(p)<0.

Corollary 5.2.1

If f(p) is the maximum or minimum value of f(x) for x in some Gp(δ), then f(p)=0; i.e., f has a zero derivative, or none at all, at p.

For, by Lemma 1, f(p)0 excludes a maximum or minimum at p. (Why?)

Note 1. Thus f(p)=0 is a necessary condition for a local maximum or minimum at p. It is insufficient, however. For example, if f(x)=x3,f has no maxima or minima at all, yet f(0)=0. For sufficient conditions, see §6.

Figure 22 illustrates these facts at the points p2,p3,,p11. Note that in Figure 22, the isolated points P,Q,R belong to the graph.

Geometrically, f(p)=0 means that the tangent at p is horizontal, or that a two-sided tangent does not exist at p.

Theorem 5.2.1

Let f:E1E be relatively continuous on an interval [a,b], with f0 on (a,b). Then f is strictly monotone on [a,b], and f is signconstant there (possibly 0 at a and b), with f0 if f, and f0 if f.

Proof

By Theorem 2 of Chapter 4, §8, f attains a least value m, and a largest value M, at some points of [a,b]. However, neither can occur at an interior point p(a,b), for, by Corollary 1, this would imply f(p)=0, contrary to our assumption.

Screen Shot 2019-06-26 at 1.54.42 PM.png

Thus M=f(a) or M=f(b); for the moment we assume M=f(b) and m=f(a). We must have m<M, for m=M would make f constant on [a,b], implying f=0. Thus m=f(a)<f(b)=M.

Now let ax<yb. Applying the previous argument to each of the intervals [a,x],[a,y],[x,y], and [x,b] (now using that m=f(a)<f(b)=M), we find that

f(a)f(x)<f(y)f(b). (Why?) 

Thus ax<yb implies f(x)<f(y); i.e., f increases on [a,b]. Hence f cannot be negative at any p[a,b], for, otherwise, by Lemma 1, f would decrease at p. Thus f0 on [a,b].

In the case M=f(a)>f(b)=m, we would obtain f0.

Caution: The function f may increase or decrease at p even if f(p)=0.
See Note 1.

Corollary 5.2.2 (Rolle's theorem)

If : E1E is relatively continuous on [a,b] and if f(a)=f(b), then f(p)=0 for at least one interior point p(a,b).

For, if f0 on all of (a,b), then by Theorem 1, f would be strictly monotone on [a,b], so the equality f(a)=f(b) would be impossible.

Figure 22 illustrates this on the intervals [p2,p4] and [p4,p6], with f(p3)= f(p5)=0. A discontinuity at 0 causes an apparent failure on [0,p2].

Note 2. Theorem 1 and Corollary 2 hold even if f(a) and f(b) are infinite, if continuity is interpreted in the sense of the metric ρ of Problem 5 in Chapter 3, §11. (Weierstrass' Theorem 2 of Chapter 4, §8 applies to (E,ρ), with the same proof.)

Theorem 5.2.2 (Cauchy's law of the mean)

Let the functions f,g:E1E be relatively continuous and finite on [a,b] and have derivatives on (a,b), with f and g never both infinite at the same point p(a,b). Then

g(q)[f(b)f(a)]=f(q)[g(b)g(a)] for at least one q(a,b).

Proof

Let A=f(b)f(a) and B=g(b)g(a). We must show that Ag(q)=Bf(q) for some q(a,b). For this purpose, consider the function h=AgBf. It is relatively continuous and finite on [a,b], as are g and f. Also,

h(a)=f(b)g(a)g(b)f(a)=h(b). (Verify!) 

Thus by Corollary 2, h(q)=0 for some q(a,b). Here, by Theorem 4 of §1, h=(AgBf)=AgBf. (This is legitimate, for, by assumption, f and g never both become infinite, so no indeterminate limits occur.) Thus h(q)=Ag(q)Bf(q)=0, and (1) follows.

Corollary 5.2.3 (Lagrange's law of the mean)

If f:E1E1 is relatively continuous on [a,b] with a derivative on (a,b), then

f(b)f(a)=f(q)(ba) for at least one q(a,b).

Proof

Take g(x)=x in Theorem 2, so g=1 on E1.

Note 3. Geometrically,

f(b)f(a)ba

is the slope of the secant through (a,f(a)) and (b,f(b)), and f(q) is the slope of the tangent line at q. Thus Corollary 3 states that the secant is parallel to the tangent at some intermediate point q; see Figure 23. Theorem 2 states the same for curves given parametrically: x=f(t),y=g(t).

Screen Shot 2019-06-26 at 1.55.27 PM.png

Corollary 5.2.4

Let f be as in Corollary 3. Then

(i) f is constant on [a,b] iff f=0 on (a,b);

(ii) f on [a,b] iff f0 on (a,b); and

(iii) f on [a,b] iff f0 on (a,b).

Proof

Let f=0 on (a,b). If axyb, apply Corollary 3 to the interval [x,y] to obtain

f(y)f(x)=f(q)(yx) for some q(a,b) and f(q)=0.

Thus f(y)f(x)=0 for x,y[a,b], so f is constant.

The rest is left to the reader.

Theorem 5.2.3 (inverse functions)

Let f:E1E1 be relatively continuous and strictly monotone on an interval IE1. Let f(p)0 at some interior point pI. Then the inverse function g=f1 (with f restricted to I) has a derivative at q=f(p), and

g(q)=1f(p).

(If f(p)=±, then g(q)=0.)

Proof

By Theorem 3 of Chapter 4, §9, g=f1 is strictly monotone and relatively continuous on f[I], itself an interval. If p is interior to I, then q=f(p) is interior to f[I]. (Why?)

Now if yf[I], we set

Δg=g(y)g(q),Δy=yq,x=f1(y)=g(y), and f(x)=y

and obtain

ΔgΔy=g(y)g(q)yq=xpf(x)f(p)=ΔxΔf for xp.

Now if yq, the continuity of g at q yields g(y)g(q); i.e., xp. Also, xp iff yq, for f and g are one-to-one functions. Thus we may substitute y=f(x) or x=g(y) to get

g(q)=limyqΔgΔy=limxpΔxΔf=1limxp(Δf/Δx)=1f(p),

where we use the convention 1=0 if f(p)=.

Examples

(A) Let

f(x)=loga|x| with f(0)=0.

Let p>0. Then (x>0)

Δf=f(x)f(p)=logaxlogap=loga(x/p)=logap+(xp)p=loga(1+Δxp).

Thus

ΔfΔx=loga(1+Δxp)1/Δx.

Now let z=Δx/p. (Why is this substitution admissible?) Then using the formula

limz0(1+z)1/z=e (see Chapter 4, §2, Example (C)) 

and the continuity of the log and power functions, we obtain

f(p)=limxpΔfΔx=limz0loga[(1+z)1/z]1/p=logae1/p=1plogae.

The same formula results also if p<0, i.e., |p|=p. At p=0,f has one-sided derivatives (±) only (verify!), so f(0)=0 by Definition 1 in §1.

(B) The inverse of the log a function is the exponential g:E1E1, with

g(y)=ay(a>0,a1).

By Theorem 3, we have

(qE1)g(q)=1f(p),p=g(q)=aq.

Thus

g(q)=11plogae=plogae=aqlogae.

Symbolically,

(loga|x|)=1xlogae(x0);(ax)=axlogae=axlna.

In particular, if a=e, we have logea=1 and logax=lnx; hence

(ln|x|)=1x(x0) and (ex)=ex(xE1).

(C) The power function g:(0,+)E1 is given by

g(x)=xa=exp(alnx) for x>0 and fixed aE1.

By the chain rule (§1, Theorem 3), we obtain

g(x)=exp(alnx)ax=xaax=axa1.

Thus we have the symbolic formula

(xa)=axa1 for x>0 and fixed aE1.

Theorem 5.2.4 (Darboux)

If f:E1E is relatively continuous and has a derivative on an interval I, then f has the Darboux property (Chapter 4, §9 ) on I.

Proof

Let p,qI and f(p)<c<f(q). Put g(x)=f(x)cx. Assume g0 on (p,q) and find a contradiction to Theorem 1. Details are left to the reader.


This page titled 5.2: Derivatives of Extended-Real Functions is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Elias Zakon (The Trilla Group (support by Saylor Foundation)) via source content that was edited to the style and standards of the LibreTexts platform.

Support Center

How can we help?