4

I've noticed that $a \bmod b$ (with 'mod' in the operational sense) can also be represented using various tricks of formulation. For instance, that

$$a \bmod b = a - b\left\lfloor\frac{a}{b}\right\rfloor .$$

There are several ways to do it by (ab)using various functions like ceiling, max, abs, and the like. However, I realized yesterday that

$$a \bmod 2 = \frac{1-(-1)^a}{2}.$$

I find this interesting as I consider basic exponentiation to be a purer operation in some sense than something like floor, perhaps in that it doesn't have a built-in sense of conditionality or knowledge of fractional parts. Furthermore, you can use substitution to reach arbitrarily high powers of $2$, as in

$$a \bmod 4 = \frac{1-(-1)^a}{2}+1-(-1)^{\frac{a-\frac{1-(-1)^a}{2}}{2}},$$

which amounts to right-shifting $a$ by one spot and repeating the process to get the second bit you need for the $\bmod 4$. I realize that's not pretty, but I'm interested that it's possible. The motivation here is identifying situations where formulae may implicitly support a richness of computational complexity one wouldn't expect, which parity detection and manipulation goes a long way towards. Which leads me to my question:

Is there some way using exponentiation or other basic operations to find a comparable expression for $a \bmod 3$, or ideally, $a \bmod b$?

Trevor
  • 6,144
  • 16
  • 39

2 Answers2

4

Any function $f(n)$ on the integers that is periodic with period $m$ can be written in terms of powers of $\zeta=e^{2\pi i/m}$, a primitive $m$th root of unity: $$ f(n) = \sum_{k=0}^{m-1} \hat f(k) \zeta^{nk}, \quad\text{where } \hat f(k) = \frac1m \sum_{j=0}^{m-1} f(j) \zeta^{-jk}. $$ Notice that when $m=2$, we have $\zeta=-1$ and this formula becomes $$ f(n) = \hat f(0) + \hat f(1)(-1)^n, \quad\text{where } \hat f(0) = \frac12(f(0)+f(1)) \text{ and } \hat f(1) = \frac12(f(0)-f(1)); $$ this is the formula you included when $f(n) = n\mod 2$.

When $m=3$, we have $\zeta=e^{2\pi i/3} = \frac12(-1+i\sqrt3)$ and $\zeta^2=\bar\zeta=\frac12(-1-i\sqrt3)$, and $$ f(n) = \hat f(0) + \hat f(1) \zeta^n + \hat f(2) \bar\zeta^n. $$ In the specific case $f(n)=n\mod 3$, the coefficients are \begin{align*} \hat f(0) &= \frac13(0+1+2) = 1, \\ \hat f(1) &= \frac13(0+1\bar\zeta+2\zeta) = -\frac12+\frac i{2\sqrt3}, \\ \hat f(2) &= \frac13(0+1\zeta+2\bar\zeta) = -\frac12-\frac i{2\sqrt3}. \end{align*} Putting it all together, $$ n\mod3 = 1 + \bigg({-}\frac12+\frac i{2\sqrt3}\bigg)\bigg(\frac{-1+i\sqrt3}2\bigg)^n + \bigg({-}\frac12-\frac i{2\sqrt3}\bigg)\bigg(\frac{-1-i\sqrt3}2\bigg)^n. $$

Greg Martin
  • 92,241
  • 1
    Thanks for this, although I'm getting hung up on the fact that the two functions refer to each other. Maybe that recursion always cancels out, but it's not immediately clear to me why it would. Am I missing something / is there some way to restate things to avoid that? – Trevor Jul 26 '20 at 23:13
  • 2
    If you're talking about the first pair of equalities, you can read it as: "Given a function $f$, define $\hat f(k)=\dots$. With that definition, it is a true fact that $f(n) = \dots$." – Greg Martin Jul 27 '20 at 04:44
  • 1
    Note to self: this probably has a nice representation in the Eisenstein integers. – Trevor Jan 30 '21 at 22:56
2

Note that $\sin(2\pi n/3)=0, \dfrac{\sqrt3}2, $ or $-\dfrac{\sqrt3}2$, according as $n\equiv0, 1, $ or $2\bmod3$, respectively.

Furthermore, $f(x)=\dfrac{x\left(x-\dfrac{\sqrt3}2\right)2}{-\dfrac{\sqrt3}2\left(\dfrac{\sqrt3}2-\dfrac{\sqrt3}2\right)}+\dfrac{x\left(x+\dfrac{\sqrt3}2\right)1}{\dfrac{\sqrt3}2\left(\dfrac{\sqrt3}2+\dfrac{\sqrt3}2\right)}=\dfrac{3x^2-\dfrac{\sqrt3}2x}{\dfrac 32}=2x^2-\dfrac x{\sqrt3}$

has the property that $f(0)=0, f\left(\dfrac{\sqrt3}2\right)=1, $ and $f\left(-\dfrac{\sqrt3}2\right)=2$.

Therefore, $f(\sin(2\pi n/3))=2\sin^2(2\pi n/3)-\dfrac {\sin(2\pi n/3)}{\sqrt3}=n \bmod 3.$

J. W. Tanner
  • 63,683
  • 4
  • 43
  • 88