Calculus I · Textbook draft · 20 minute reading

Exponentials, Logarithms, and Inverses

Differentiate inverse functions and derive logarithmic, exponential, inverse-trigonometric, and real-power rules on their domains.

Before you begin

  • Chain rule, inverse functions, logarithm laws, and trigonometric derivatives.

Learning objectives

  • Apply the inverse derivative theorem with a nonzero derivative.
  • Differentiate exponential, logarithmic, and inverse-trigonometric functions.
  • Use logarithmic differentiation without losing sign or domain restrictions.

Exponential growth describes a quantity whose instantaneous rate is proportional to its current size. Before using that model, we need the derivative of an exponential and a precise account of the inverse relationship between exponentials and logarithms.

We take the natural exponential and logarithm as elementary functions with their usual continuous inverse relationship and the normalization lim⁡h→0(eh−1)/h=1\lim_{h\to0}(e^h-1)/h=1. This characterization of the base ee is an input here, not a consequence of the differentiation rules already proved. One can instead construct log⁡x\log x by an integral and derive exponentials after the fundamental theorem; that alternative ordering leads to the same functions.

Exponentials and inverse rates

Using ex+h=exehe^{x+h}=e^xe^h in the difference quotient gives (ex)′=ex(e^x)'=e^x. The chain rule gives (eu(x))′=eu(x)u′(x)(e^{u(x)})'=e^{u(x)}u'(x). For a fixed base b>0b>0, write bx=exlog⁡bb^x=e^{x\log b}, obtaining (bx)′=bxlog⁡b(b^x)'=b^x\log b. If b=1b=1, the formula correctly returns zero.

If y=f(x)y=f(x) has a differentiable inverse locally, the identity f(f−1(y))=yf(f^{-1}(y))=y and the chain rule imply

(f−1)′(y)=1f′(f−1(y)),(f^{-1})'(y)=\frac1{f'(f^{-1}(y))},

provided the denominator is nonzero. A standard inverse derivative theorem ensures differentiability of the inverse when ff is continuously differentiable near the point with nonzero derivative. The requirement is substantive: x3x^3 has an inverse, but at zero its derivative is zero and the inverse cube root lacks a finite derivative there.

Applying the theorem to exe^x gives (log⁡x)′=1/x(\log x)'=1/x for x>0x>0. For x≠0x\ne0, (log⁡∣x∣)′=1/x(\log|x|)'=1/x on each half-line. A logarithm of a negative argument remains undefined over the reals; the absolute value supplies a different, well-defined function.

Variable arguments and real powers

For a differentiable nonzero uu, differentiation of log⁡∣u∣\log|u| gives u′/uu'/u wherever uu does not vanish. Thus (log⁡(x2+1))′=2x/(x2+1)(\log(x^2+1))'=2x/(x^2+1) on all real inputs. For log⁡(3x−2)\log(3x-2) we additionally require x>2/3x>2/3.

For a real constant rr and x>0x>0, define xr=erlog⁡xx^r=e^{r\log x}. Differentiation yields rxr−1rx^{r-1}. Some rational powers extend to negative arguments and require separate domain care; this proof covers the positive domain uniformly.

When both base and exponent vary, an ordinary power rule is insufficient. For y=xxy=x^x on x>0x>0, take logarithms:

log⁡y=xlog⁡x,y′y=log⁡x+1,y′=xx(log⁡x+1).\log y=x\log x,\qquad \frac{y'}y=\log x+1, \qquad y'=x^x(\log x+1).

The function is positive on the stated domain, so taking the logarithm is legitimate. For y=(x2+1)sin⁡xy=(x^2+1)^{\sin x}, the same method gives

y′=(x2+1)sin⁡x[cos⁡xlog⁡(x2+1)+2xsin⁡xx2+1].y'=(x^2+1)^{\sin x}\left[\cos x\log(x^2+1)+\frac{2x\sin x}{x^2+1}\right].

Inverse trigonometric functions

Restrict sine to [−π/2,π/2][-\pi/2,\pi/2] to define arcsin⁡\arcsin. At interior inputs ∣x∣<1|x|<1, set y=arcsin⁡xy=\arcsin x. Then sin⁡y=x\sin y=x, so cos⁡y y′=1\cos y\,y'=1. On that branch cosine is positive, giving

(arcsin⁡x)′=11−x2.(\arcsin x)'=\frac1{\sqrt{1-x^2}}.

Likewise (arccos⁡x)′=−1/1−x2(\arccos x)'=-1/\sqrt{1-x^2} for ∣x∣<1|x|<1. For y=arctan⁡xy=\arctan x in (−π/2,π/2)(-\pi/2,\pi/2), sec⁡2y y′=1\sec^2y\,y'=1, hence (arctan⁡x)′=1/(1+x2)(\arctan x)'=1/(1+x^2) for every real xx. The inverse branch determines the sign of a square root; it cannot be ignored.

A growth model

For P(t)=P0ektP(t)=P_0e^{kt} with P0>0P_0>0, P′=kPP'=kP. The parameter kk has units inverse time. Doubling time for k>0k>0 solves ekt=2e^{kt}=2, giving log⁡2/k\log2/k. A proportional growth model assumes a constant rate parameter and no limiting resources or changing conditions. Differentiating the formula does not establish those modeling assumptions.

Exercises

  1. Differentiate e−3x2e^{-3x^2} and 25x2^{5x}.
  2. Differentiate log⁡∣x2−4∣\log|x^2-4| and state the domain.
  3. Differentiate arctan⁡(2x)\arctan(2x) and arcsin⁡(x/2)\arcsin(x/2) on their derivative domains.
  4. Differentiate xsin⁡xx^{\sin x} on x>0x>0.
  5. A positive quantity grows as P0e0.04tP_0e^{0.04t} with tt in years. Find its doubling time and its instantaneous relative rate.
  6. Why can the inverse derivative formula not give a finite derivative for the cube root at zero?
Worked solutions
  1. −6xe−3x2-6xe^{-3x^2} and 5log⁡2 25x5\log2\,2^{5x}.
  2. 2x/(x2−4)2x/(x^2-4), with x≠±2x\ne\pm2.
  3. 2/(1+4x2)2/(1+4x^2) for all real xx; 1/[21−x2/4]1/[2\sqrt{1-x^2/4}] for ∣x∣<2|x|<2.
  4. xsin⁡x[cos⁡xlog⁡x+(sin⁡x)/x]x^{\sin x}[\cos x\log x+(\sin x)/x].
  5. log⁡2/0.04≈17.329\log2/0.04\approx17.329 years, and P′/P=0.04P'/P=0.04 per year.
  6. The original cube function has zero derivative at zero, so the theorem’s nonzero-denominator hypothesis fails. Direct quotients confirm the inverse has unbounded slope.