1
$\begingroup$

How are logarithms defined explicitly?

One way I can come up with is the following, first start by using McLaurin series representation of exponent function:

$$e^x = \sum_{k=o}^{\infty}\frac{x^k}{k!} = 1 + x + \frac{x^2}{2!} + \frac{x^3}{3!}...$$

Now, simply invert the function (solving for x), cutting it off at a certain point:

$$y=1 + x + \frac{x^2}{2!} + \frac{x^3}{3!}$$ $$ln(y)=x(y)$$

But as can be seen, the function is really difficult to solve as each added term increases the order of the polynomial. My question is, is there a simpler explicit definition for logarithms, and how to define logarithms explicitly accurately at all?

(Edit: previously there was an unnecessary bit regarding how calculators accomplish the calculation).

$\endgroup$
3
$\begingroup$

It's hard to say there's "a" correct definition. But what many texts do is not define $\ln$ as the inverse of $\exp$, but define each individually then show their relation. In particular, $\ln x = \int_{1}^{x} \frac{1}{t} \mathrm{d}t$ makes sense as it's the unique function with derivative $1/x$ and with value $0$ at $1$. From this, you can show $\ln (a b) = \ln(a) + \ln(b)$ and all the rest.

$\endgroup$
1
$\begingroup$

Calculators don't actually need to define logarithms -- it is enough that they can somehow calculate a number that is within a desired tolerance from the true value of the logarithm, and if a method achieves that, then it doesn't matter whether it's based on a principled definition of the logarithm function or not.

Calculators that are based on binary floating point typically calculate $\log_2 x$ natively and then scale that to get logarithms to other bases.

First, multiply or divide by an appropriate integer power of $2$ such that the number is between $1$ and $2$. (This is an extremely cheap operation to do with binary floating-point representation). The integer power becomes the integral part of $\log_2 x$, and then one only needs to be able to approximate $\log_2 x$ on the interval $[1,2)$ to find the fractional part.

Methods to do this approximation vary. Sometimes this is done with a polynomial approximation -- which might be a Taylor approximation of $\log_2(x+1)$, or a hand-chosen polynomial that minimizes the error over the entire interval $[1,2)$ with fewer terms than a Taylor polynomial would need.

Some implementations uses several approximating polynomials, chosen for subintervals of $[1,2)$ by table lookup.

There are also CORDIC-like methods, where you try express $x$ as a product of factors of the form $1\pm2^{-n}$ (multiplying by such a factor is cheap!), and then add together the logarithms of the factors you used -- which are themselves looked up in a table baked into the calculator during manufacture.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.