TAOCP 4.1 Exercise 28
Let R(e_0,\ldots,e_r) =(1+i)^{e_0}+i(1+i)^{e_1}-(1+i)^{e_2}-i(1+i)^{e_3} +\cdots+i^r(1+i)^{e_r}, \qquad e_0<\cdots<e_r.
Section 4.1: Positional Number Systems
Exercise 28. ▶ [**] $[M24]$ Show that every nonzero complex number of the form $a + bi$ where $a$ and $b$ are integers has a unique "revolving binary representation"
$$(1+i)^{e_0} + i(1+i)^{e_1} - (1+i)^{e_2} - i(1+i)^{e_3} + \cdots + i^r(1+i)^{e_r},$$
where $e_0 < e_1 < \cdots < e_r$. (Compare with exercise 27.)
Verified: yes
Solve time: 9m35s
Let
$$ R(e_0,\ldots,e_r) =(1+i)^{e_0}+i(1+i)^{e_1}-(1+i)^{e_2}-i(1+i)^{e_3} +\cdots+i^r(1+i)^{e_r}, \qquad e_0<\cdots<e_r. $$
We must prove that every nonzero Gaussian integer $z=a+bi$ admits a unique representation of this form.
The proof is most naturally obtained from Exercise 27. Recall that Exercise 27 shows that every positive integer has a unique binary representation
$$ n=2^{e_0}+2^{e_1}+\cdots+2^{e_r}, \qquad e_0<e_1<\cdots<e_r. $$
We shall reduce the present problem to that result.
Existence
Let
$$ z=a+bi\neq0. $$
Choose $m\ge0$ maximal such that $(1+i)^m\mid z$. Then
$$ z=(1+i)^m w, $$
where $w$ is not divisible by $1+i$.
A Gaussian integer is divisible by $1+i$ if and only if its real and imaginary parts have the same parity. Therefore the real and imaginary parts of $w$ have opposite parity. Hence
$$ w\equiv1 \pmod{1+i} \quad\text{or}\quad w\equiv i \pmod{1+i}. $$
Multiplying by a suitable power of $i$, we may write
$$ w=i^s u, \qquad s\in{0,1,2,3}, $$
where
$$ u\equiv1\pmod{1+i}. $$
Thus it suffices to represent every Gaussian integer congruent to $1$ modulo $1+i$.
Consider such a $u$. Since $u-1$ is divisible by $1+i$, write
$$ u=1+(1+i)v . $$
Now repeat the same procedure with $v$. At each step, if the current quotient is nonzero, it is either divisible by $1+i$ or congruent to $1$ or $i$ modulo $1+i$. Recording a digit only when the quotient is not divisible by $1+i$, we obtain
$$ u=\varepsilon_0+\varepsilon_1(1+i)+\varepsilon_2(1+i)^2+\cdots, $$
where each $\varepsilon_j\in{0,1,i}$.
The process terminates because division by $1+i$ decreases the norm by a factor $2$:
$$ N!\left(\frac{x}{1+i}\right)=\frac{N(x)}2. $$
Hence after finitely many steps the quotient becomes $0$, so only finitely many digits are nonzero.
Now let
$$ e_0<e_1<\cdots<e_r $$
be the positions of the nonzero digits. By construction the first nonzero digit is $1$. After a nonzero digit has occurred, the next nonzero digit must be the other residue class modulo $1+i$. Indeed, if
$$ x=\varepsilon +(1+i)y, \qquad \varepsilon\in{1,i}, $$
then $x-\varepsilon$ is divisible by $1+i$; dividing by $1+i$ interchanges the two nonzero residue classes $1$ and $i$ modulo $1+i$. Consequently the successive nonzero digits alternate
$$ 1,\ i,\ 1,\ i,\ \ldots . $$
When these digits are moved back to the original powers of $1+i$, each passage through a factor $1+i$ multiplies the coefficient by $i$. Therefore the coefficients attached to the successive nonzero terms are
$$ 1,\ i,\ i^2,\ i^3,\ldots , $$
that is,
$$ 1,\ i,\ -1,\ -i,\ 1,\ldots . $$
Hence
$$ u=(1+i)^{e_0} +i(1+i)^{e_1} -(1+i)^{e_2} -i(1+i)^{e_3} +\cdots +i^r(1+i)^{e_r}. $$
Multiplying by the initial factor $i^s(1+i)^m$ merely adds the constant $m$ to every exponent and rotates the coefficient cycle by $i^s$. Reindexing the exponents gives again a representation of the required revolving form. Thus every nonzero Gaussian integer possesses such a representation.
Uniqueness
Suppose that
$$ \sum_{k=0}^{r} i^k(1+i)^{e_k} = \sum_{\ell=0}^{t} i^\ell(1+i)^{f_\ell}, $$
with
$$ e_0<\cdots<e_r, \qquad f_0<\cdots<f_t. $$
Subtracting, we obtain a nontrivial relation
$$ \sum_{j\ge0} c_j(1+i)^j=0, $$
where each coefficient satisfies
$$ c_j\in{0,\pm1,\pm i}. $$
Let $m$ be the least index for which $c_m\neq0$. Then
$$ c_m+(1+i)q=0 $$
for some Gaussian integer $q$, hence
$$ 1+i\mid c_m. $$
But none of
$$ \pm1,\ \pm i $$
is divisible by $1+i$, since each has norm $1$ whereas $N(1+i)=2$.
This contradiction shows that no nontrivial relation exists. Therefore the revolving representation is unique.
Hence every nonzero Gaussian integer $a+bi$ has a unique representation
$$ (1+i)^{e_0} +i(1+i)^{e_1} -(1+i)^{e_2} -i(1+i)^{e_3} +\cdots +i^r(1+i)^{e_r}, \qquad e_0<e_1<\cdots<e_r. $$
$\square$