Theories of Exponential and Logarithmic Functions: Part 3

2 comments
In this concluding post on the theories of exponential and logarithmic functions we will present the most intuitive and obvious approach to define the expression ab directly without going to the number e and the function logx. This approach stems from the fact that an irrational number can be approximated by rational numbers and we can find as good approximations as we want. The idea is that if b is irrational and a>0 then we have many rational approximations b,b, to b and the numbers ab,ab, would be the approximations to the number ab being defined. Inherent in such a procedure is the belief that we can find as good approximations to ab as we want by choosing sufficiently good rational approximations to b. Thus we can see that the numbers 21,21.4,21.41,21.414,21.4142, are approximations to the number 22.

Intuitive Definition of ab

In order to make the definition described in previous paragraph more precise we start with the assumption that a>0 and b is any real number. If {bn} is a sequence of rational numbers such that limnbn=b then we define ab=limnabn While this seems to be the most obvious approach to the theory of exponential and logarithmic functions it has two stumbling blocks:
  • We need to show that the limit in definition (1) exists
  • If there are two sequences {bn},{cn} both tending to b then we must show that limnabn=limnacn
Both these statements are obvious if a=1. In the further development of the theory we can assume that a>1 and the case 0<a<1 can be treated by noting that ab=1/(1/a)b where b is rational and (1/a)>1.

We first show that there is a sequence bn tending to b such that limnabn exists. Clearly we can choose any monotonically increasing sequence bn which tends to b. By any theory of real numbers (Dedekind cuts or Cantor's constructions of reals via Cauchy sequences of rationals) it is possible to find such a sequence bn of rationals. In practice we may choose bn to consist of decimal approximations to b adding each extra digit for each term of the sequence. Note that such a sequence bn is bounded above (because it is increasing and reaches a limit b). This means that there is a rational number B such that bnB for all n. Now we can see that the sequence xn=abn is increasing (because bn is increasing and a>1) and at the same time it is bounded above by aB. It follows that xn tends to a positive limit.

Next we show that if there are two sequences bn,cn of rationals each tending to b then limnabn=limnacn. Clearly if we put dn=bncn then dn is a sequence of rationals tending to 0. We will first show that adn1 as n. Let ϵ>0 be an arbitrary number. We know from previous post that a1/n1 and a1/n1 as n hence there is an integer m such that |a1/n1|<ϵ,|a1/n1|<ϵ whenever nm. Since dn0 it follows that there is a positive integer N such that |dn|<1/m for all nN. This means that a1/m<adn<a1/m for all nN. From equations (2) and (3) we can see that |adn1|<ϵ whenever nN. It follows that adn1 as n. Now we can see that limnabn=limnadn+cn=limnadnlimnacn=limnacn Thus both the stumbling blocks in definition (1) are taken care of and we can proceed further.

We next prove that if b>0 then ab>1. Since b>0 there must be an increasing sequence bn of positive rationals tending to b.  Clearly we have bn>b1>0 for all n>1 and hence it follows that abn>ab1>1 for all n>1. Taking limits as n we see that abab1>1.

We are now in a position to study the function f(x)=ax. As before we will only need to consider the case a>1 and reader can state and prove results for the case 0<a<1 himself. First we see that the law of exponents is valid i.e. axay=ax+y for all x,y. Clearly let xn,yn be sequences of rationals tending to x,y so that (xn+yn)(x+y). Then we have axnayn=axn+yn and taking limits when n we are done. In similar fashion we can prove usual laws of exponents.

We next show that f(x) is strictly increasing for all x. Let x>y so that xy>0 and then axy>1. Multiplying by ay>0 on both sides we get axyay>ay i.e. ax>ay. From this point it is easy to establish most of the common inequalities regarding exponents and we will not go into details. In particular the inequalities related to α,β,r,s in the last post all hold true for all positive real numbers r,s. However there is a caveat. The inequalities may become weak if we try to construct sequences of positive rationals tending to r,s and then letting n. But we don't really need the strict versions of those inequalities.

It will be important now to establish the simple but fundamental limit limx0ax=1 Let ϵ>0 be arbitrary. Then there is a positive integer m such that |a1/n1|<ϵ,|a1/n1|<ϵ for nm. Clearly if we chose x(1/m,1/m) then a1/m<ax<a1/m and hence |ax1|<ϵ. Thus we can choose δ=1/m and then we can see that |ax1|<ϵ whenever 0 < |x| < \delta. Thus a^{x} \to 1 as x \to 0.

As in the last post we had shown that n(x^{1/n} - 1) is decreasing as n increases, we can show that the function g(x) = \dfrac{a^{x} - 1}{x} is increasing for all x > 0. Also the function g(x) satisfies (a - 1) \geq g(x) \geq a^{x - 1} (a - 1) for all a > 1, x > 0. It follows that as x \to 0^{+} the limit \lim_{x \to 0^{+}}\frac{a^{x} - 1}{x} = \phi(a) exists and \phi(a) \geq (a - 1)/a. If x \to 0^{-} then we put x = -y so that y \to 0^{+} and then we have \lim_{x \to 0^{-}}\frac{a^{x} - 1}{x} = \lim_{y \to 0^{+}}\frac{a^{y} - 1}{ya^{y}} = \phi(a)/1 = \phi(a) It follows that the limit \lim_{x \to 0}\frac{a^{x} - 1}{x} = \phi(a)\tag{6} defines a function \phi(a) which is defined for all a > 0.

We can also see that \phi(1) = 0 and (a - 1)/a \leq \phi(a) \leq (a - 1) for a > 1. Thus \phi(a) > 0 for a > 1 and it can be easily seen that \phi(1/a) = -\phi(a) so that \phi(a) < 0 for 0 < a < 1. It can be established that \phi(ab) = \phi(a) + \phi(b) for all positive a, b and \phi(a/b) = \phi(a) - \phi(b). Thus if a > b so that a/b > 1 and we have \phi(a/b) > 0 so that \phi(a) > \phi(b).

It follows that the function \phi(a) is defined for all a > 0 and it is strictly increasing function of a. It satisfies the properties \phi(1) = 0, \phi(1/a) = -\phi(a), \phi(ab) = \phi(a) + \phi(b), \phi(a/b) = \phi(a) - \phi(b) It is now time to identify this function \phi(a) with \log a and thus we define \log x = \lim_{b \to 0}\frac{x^{b} - 1}{b}\tag{7} for all x > 0. Just to repeat the properties of \log function established in terms of \phi we have \begin{align} \log 1 = 0, \log (1/x) = -\log x,\tag{8a}\\ \log(xy) = \log x + \log y,\tag{8b}\\ \log(x/y) = \log x - \log y\tag{8c} \end{align} for all positive x, y. Also \log x is a strictly increasing function of x and \frac{x - 1}{x}\leq \log x \leq (x - 1)\tag{9} for x > 1. From this last equation we can easily show that \lim_{x \to 1}\frac{\log x}{x - 1} = 1\text{ or equivalently }\lim_{x \to 0}\frac{\log(1 + x)}{x} = 1\tag{10} and this limit shows that the derivative of \log x is 1/x (see previous post).

We can also find derivative of f(x) = a^{x}. We have (a^{x})' = \lim_{h \to 0}\frac{a^{x + h} - a^{x}}{h} = a^{x} \lim_{h \to 0}\frac{a^{h} - 1}{h} = a^{x}\log a\tag{11} We next show that \log a^{b} = b\log a\tag{12} for all a > 0 and all b. Clearly from the property \log(ab) = \log a + \log b it is easy to show that (12) holds if b is rational. If b is irrational then we can take a sequence b_{n} of rationals tending to b and then we have \log a^{b_{n}} = b_{n}\log a. Taking limits when n \to \infty and noting that both \log x and a^{x} are continuous we get \log a^{b} = b\log a.

Now \log 2 > 0 and hence given any number M > 0 we can find an integer n such that n \log 2 > M. Let N = 2^{n}. If x > N then \log x > \log N = \log 2^{n} = n\log 2 > M It follows that \log x \to \infty as x \to \infty. Noting that \log (1/x) = -\log x we get \log x \to -\infty as x \to 0^{+}. It follows that \log x is a strictly increasing function from \mathbb{R}^{+} to \mathbb{R} and hence there is a unique number e > 1 such that \log e = 1.

Then for any real x we have \log e^{x} = x\log e = x so that if y = e^{x} then x = \log y. It turns out that the specific power e^{x} is the inverse of the logarithm function. Also we note that (e^{x})' = e^{x}\log e = e^{x} so that e^{x} is its own derivative. We immediately get the formula \lim_{x \to 0}\frac{e^{x} - 1}{x} = 1\tag{13} We will now show that e^{x} = \lim_{t \to \infty}\left(1 + \frac{x}{t}\right)^{t} = \lim_{t \to -\infty}\left(1 + \frac{x}{t}\right)^{t}\tag{14} Note that we can more easily combine these two limits into one and show instead that \lim_{h \to 0}(1 + xh)^{1/h} = e^{x} Clearly we can see that \begin{align} \lim_{h \to 0}\log(1 + xh)^{1/h} &= \lim_{h \to 0}\frac{\log(1 + xh)}{h}\notag\\ &= \lim_{h \to 0}x\cdot\frac{\log(1 + xh)}{xh} = x\notag \end{align} and by continuity of \log x and e^{x} we get \lim_{h \to 0}(1 + xh)^{1/h} = e^{x}. If we replace t in (14) by positive integer n then we immediately get the usual formulas e^{x} = \lim_{n \to \infty}\left(1 + \frac{x}{n}\right)^{n} = \lim_{n \to \infty}\left(1 - \frac{x}{n}\right)^{-n},\,\, e = \lim_{n \to \infty}\left(1 + \frac{1}{n}\right)^{n}\tag{15} Note that the infinite series for e^{x} can either be proved using limits above or using the Taylor's series approach based on derivatives. We have now been able to derive all the common properties of a^{x}, e^{x}, \log x and have established the derivative formulas. The interested reader can carry on from here and establish any property of these functions without any hassle.

Note: This post is inspired by this and this answer on MSE.

Print/PDF Version

2 comments :: Theories of Exponential and Logarithmic Functions: Part 3

Post a Comment

  1. Hello! I did not know about you or this blog until earlier today. I was doing a linguistics project about Indian users of MSE, and I ran across one of your answers. As a bit of a math blogger myself, I of course had to come see what you were doing over here. Well, it was very hard to focus on linguistics after that!

    Hopefully in the three months after finals I will have some free time, and I will definitely be using some of that to read your blog. I think you have a lot to teach me about both mathematics and exposition. I do hope that you'll consider contributing to the upcoming MSE blog because the work you've done over here is fantastic and I think we could benefit greatly from articles written like these.

  2. @proofsareart,
    I have already put my topic for the MSE blog. Let's hope I am able to get it ready for publication by 17th May 2014