March 4, 2026
Fractions vs Feelings
Approximation Game
Approximation Game: Math nerds feud over which numbers ‘cheat’
TLDR: A math blog gamifies approximating numbers with fractions without matching exactly. Commenters spotlighted Roth’s theorem—algebraic irrationals have strict limits—sparking debate over whether rationals or irrationals are easier to target and why better-than-allowed approximations hint at transcendental numbers like π.
A math blog just turned fractions into a sport: pick a number r and get as close as possible with a/b—without hitting it exactly. The author sets the rules, shows how rounding up or down pegs a worst-case error near 1/b, and crowns “1-good” winners who beat that. Cue the comments: math drama unlocked.
Top commenter cperciva drops a flex: algebraic irrationals like √2 have strict limits on how close rationals can get—a big deal proven by Roth’s theorem, which helped earn a Fields Medal. The thread’s hot take: if a number allows “too-good” approximations, that’s a neon sign it’s transcendental (think π), not algebraic. Meanwhile, engineers rolled in like, “we just add more decimals,” while meme-makers dubbed continued fractions the “cheat code” for winning.
The core squabble: is the game easier if r is rational or irrational? One camp says rationals are trivial but often forbidden (exact matches aren’t allowed), while the other calls irrationals the “final boss,” daring you to beat the 1/b ceiling. Everyone loved the scorecard vibe with s = error × denominator, turning abstract infinity talk into a leaderboard—and yes, the math flexing was fierce.
Key Points
- •Best low-side approximation for given b: a_low = ceil(r*b) − 1; best high-side: a_high = floor(r*b) + 1.
- •Exact matches to r are disallowed; approximations must be strictly less than or greater than r.
- •The absolute error is ε = |r − a/b|, with a worst-case bound ε ≤ 1/b for these optimal inexact choices.
- •A normalized error score s = ε*b is defined, making the maximum error 1 independent of b.
- •An approximation is called “1-good” if s < 1 (equivalently, ε < 1/b); examples include r = 2 with b = 5 and a rational test case r = 1/4 for analysis.