Back to index

In the bracketed representations we use on this website, you may notice how the first two numbers have always the same parity. For example,

10 = \lbrace 5,1 \rbrace &= \frac{2^5 - 2^1}{3}
           & \mbox{(with 5 and 1 being both odd)} \\
      11 = \lbrace 10,6,3,1,0 \rbrace &= \frac{2^{10} - (2^6 + 3 \cdot 2^3 + 9 \cdot 2^1 + 27)}{81}
           & \mbox{(with 10 and 6 being both even)}

Let m > 0.

If 2^{a_m} - \displaystyle\sum_{i=0}^{m-1} 2^{a_i} 3^{m-1-i} is divisible by 3^m, with a_m > a_{m-1}, then a_m \equiv a_{m-1} \pmod 2.

We require m > 0, so that there are at least two numbers in the representation.

The converse is not true; for a counterexample, note that 2^7 - (2^3 + 3 \cdot 2^1) is not divisible by 9, so \lbrace 7,3,1 \rbrace is not a valid representation.

2^{a_m} - \displaystyle\sum_{i=0}^{m-1} 2^{a_i} 3^{m-1-i}
        = 2^{a_m} - 2^{a_{m-1}} - 3 \displaystyle\sum_{i=0}^{m-2} 2^{a_i} 3^{m-2-i}

If the above is divisible by 3^m it follows that it is also divisible by 3 and, as a result, so is 2^{a_m} - 2^{a_{m-1}}. Therefore

2^{a_m} - 2^{a_{m-1}} = 2^{a_{m-1}} \left( 2^{a_m-a_{m-1}} - 1 \right)

is divisible by 3. By Euclid's Lemma, 2^{a_m-a_{m-1}} - 1 is divisible by 3, and

2^{a_m-a_{m-1}} \equiv 1 \pmod 3

It should be evident at this point that only an even power of 2 could be congruent to 1 modulo 3, but we will give this a detailed treatment.

By the Division Algorithm, let a_m-a_{m-1} = 2q+r for some integers q and r, where r is either 0 or 1.

Our goal is to show that r=0. So far we have

2^{a_m-a_{m-1}} = 2^{2q+r} = (2^2)^q \cdot 2^r \equiv 1 \pmod 3

But 2^2 = 4 \equiv 1 \pmod 3. By the rules of modular arithmetic, we then have

1^q \cdot 2^r = 2^r \equiv 1 \pmod 3

But we know that r can only be 0 or 1, and only r=0 will satisfy the congruence above.

Therefore a_m-a_{m-1} = 2q, and

a_m \equiv a_{m-1} \pmod 2

\square