Rounding numbers has a surprising amount of detail. It may seem trivial but, as with most things, there is a lot more to consider than is immediately obvious. I expect there have been hundreds if not thousands of pages devoted to rounding in IEEE journals.

An example of the complexity of rounding is what William Kahan called The Tablemaker’s Dilemma: there is no way in general to know in advance how accurately you’ll need to compute a number in order to round it correctly.

Rounding can be subtle in any number system, but there is an alternative number system in which it is a little simpler than in base 10. It’s base 3, but with a twist. Instead of using 0, 1, and 2 as “digits”, we use −1, 0, and 1. This is known as the balanced ternary system: ternary because of base 3, and balanced because the digits are symmetrical about 0.

We need a symbol for −1. A common and convenient choice is to use T. Think of moving the minus sign from in front of a 1 to on top of it. Now we could denote the number of hours in a day as 10T0 because

1 times 3^3 + 0 times 3^2 + (-1)times 3 + 0 = 24

A more formal way of a describing balanced ternary representation of a number x is a set of coefficients tk such that

x = sum_{k=-infty}^infty t_k 3^k

with the restriction that each tk is in the set {−1, 0, 1}.

Balanced ternary representation has many interesting properties. For example, positive and negative numbers can all be represented without a minus sign. See, for example, Brain Hayes’ excellent article Third Base. The property we’re interested in here is that to round a balanced ternary number to the nearest integer, you simply lop off the fractional part. Rounding is the same as truncation. To see this, note that the largest possible fractional part is a sequence of all 1s, which represents ½:

frac{1}{3} + frac{1}{3^2} + frac{1}{3^3} + cdots = frac{1}{2}

Similarly, the most negative possible fractional part is a sequence of all Ts, which represents −½. So unless the fractional part is exactly equal to ½, truncating the fractional part rounds to the nearest integer. If the fractional part is exactly ½ then there is no nearest integer but two integers that are equally near.

Related posts

The post A magical land where rounding equals truncation first appeared on John D. Cook.

Previous Post
Next Post