๐Ÿงฎ Brain Teaser

The Fraction That Eventually Stops Simplifying

Let m0m_0 and n0n_0 be distinct positive integers. For every kโ‰ฅ1k \ge 1, define mkm_k and nkn_k to be the relatively prime positive integers such that

mknk=2mkโˆ’1+12nkโˆ’1+1.\frac{m_k}{n_k} = \frac{2m_{k-1}+1}{2n_{k-1}+1}.

In words: apply xโ†ฆ2x+1x \mapsto 2x+1 to the numerator and the denominator, then reduce to lowest terms.

Prove that 2mk+12m_k+1 and 2nk+12n_k+1 are relatively prime for all but finitely many kk. Equivalently, after finitely many steps, the reduction step never cancels anything again.

Example: 14โ†’39=13โ†’37โ†’715โ†’โ‹ฏ\frac14 \to \frac39 = \frac13 \to \frac37 \to \frac7{15} \to \cdots. Here cancellation happens once, at the first step, and never again.


Field: Number theory / elementary arguments

Why it's interesting: The process looks chaotic, because it is hard to predict when 2m+12m+1 and 2n+12n+1 share a factor. Yet a single well-chosen quantity tames it completely. The problem is a clean example of a monovariant: you find a positive integer that can never increase and strictly decreases whenever cancellation occurs. Infinite descent then finishes the proof.

Difficulty: 2 / 5

Source: Putnam 2025, Problem A1 (solutions by Bhargava, Kedlaya, Ng; materials/putnam/2025s.pdf).

number theorymonovariantinfinite descentgcdodd part

Answer: The Fraction That Eventually Stops Simplifying

Key Idea / Intuition

Track the difference mkโˆ’nkm_k - n_k. Before reduction it simply doubles: (2m+1)โˆ’(2n+1)=2(mโˆ’n)(2m+1)-(2n+1) = 2(m-n). The reduction then divides it by the common factor dd, which is always odd.

So the powers of 22 in the difference are just noise from the doubling. If we strip them off and look at the odd part of โˆฃmkโˆ’nkโˆฃ|m_k - n_k|, each step can only keep it the same or divide it by dd. A positive integer cannot shrink forever, so cancellation (d>1d>1) can happen only finitely often.


Formal Proof / Solution

Setup. Let dk=gcdโก(2mk+1,โ€‰2nk+1)d_k = \gcd(2m_k+1,\, 2n_k+1). Since 2mk+1dk\frac{2m_k+1}{d_k} and 2nk+1dk\frac{2n_k+1}{d_k} are coprime and have the right ratio, we get

mk+1=2mk+1dk,nk+1=2nk+1dk.m_{k+1} = \frac{2m_k+1}{d_k}, \qquad n_{k+1} = \frac{2n_k+1}{d_k}.

For a nonzero integer rr, let Oddโก(r)\operatorname{Odd}(r) denote the largest odd divisor of โˆฃrโˆฃ|r|.

Step 1: The difference never vanishes. The map xโ†ฆ2x+1x \mapsto 2x+1 is injective, and so is division by a common positive number. Hence mkโ‰ nkm_k \ne n_k implies mk+1โ‰ nk+1m_{k+1} \ne n_{k+1}. Since m0โ‰ n0m_0 \ne n_0, induction gives mkโˆ’nkโ‰ 0m_k - n_k \neq 0 for all kk. Therefore Oddโก(mkโˆ’nk)\operatorname{Odd}(m_k - n_k) is a well-defined positive integer.

Step 2: dkd_k is odd and divides mkโˆ’nkm_k - n_k.

  • dkd_k divides the odd number 2mk+12m_k+1, so dkd_k is odd.
  • dkd_k divides (2mk+1)โˆ’(2nk+1)=2(mkโˆ’nk)(2m_k+1) - (2n_k+1) = 2(m_k - n_k).
  • Since gcdโก(dk,2)=1\gcd(d_k, 2) = 1, it follows that dkโˆฃmkโˆ’nkd_k \mid m_k - n_k.

Step 3: The monovariant. We have

mk+1โˆ’nk+1=(2mk+1)โˆ’(2nk+1)dk=2โ‹…mkโˆ’nkdk.m_{k+1} - n_{k+1} = \frac{(2m_k+1)-(2n_k+1)}{d_k} = 2\cdot\frac{m_k - n_k}{d_k}.

Multiplying by 22 does not change the odd part. Dividing by the odd number dkd_k divides the odd part by dkd_k. Hence

Oddโก(mk+1โˆ’nk+1)=Oddโก(mkโˆ’nk)dkโ€…โ€Šโ‰คโ€…โ€ŠOddโก(mkโˆ’nk),\operatorname{Odd}(m_{k+1} - n_{k+1}) = \frac{\operatorname{Odd}(m_k - n_k)}{d_k} \;\le\; \operatorname{Odd}(m_k - n_k),

with equality if and only if dk=1d_k = 1.

Step 4: Descent. The sequence ak=Oddโก(mkโˆ’nk)a_k = \operatorname{Odd}(m_k - n_k) is a nonincreasing sequence of positive integers. It strictly decreases (by a factor of at least 33) exactly at the indices where dk>1d_k > 1.

A nonincreasing sequence of positive integers can strictly decrease only finitely many times; in fact at most logโก3a0\log_3 a_0 times. Hence dk=1d_k = 1 for all but finitely many kk. That is, 2mk+12m_k+1 and 2nk+12n_k+1 are relatively prime for all sufficiently large kk. โ– \blacksquare


Check with the example. Start with 14\frac14: the difference is โˆ’3-3, with odd part 33.

  • Step 1: 39\frac39 has d0=3d_0 = 3, giving 13\frac13. The difference is โˆ’2-2, with odd part 1=3/31 = 3/3. โœ“
  • From now on the odd part is 11, so it cannot drop further. Indeed 37,715,1531,โ€ฆ\frac37, \frac7{15}, \frac{15}{31}, \dots never simplify.

Takeaway. When a process mixes a "growing" operation (doubling) with a "shrinking" one (cancellation), quotient out the growth. What remains, here the odd part, may be a quantity that only shrinks.

Source: Putnam 2025 A1 (materials/putnam/2025s.pdf, solutions by Bhargava, Kedlaya, Ng)

Type: PutnamSource: Putnam 2025 A1 (materials/putnam/2025s.pdf, solutions by Bhargava, Kedlaya, Ng)Edit on GitHub โ†—