The Most Irrational Number

Pose the driving question: not all irrationals are equally irrational. Introduce the idea of measuring irrationality by how badly rationals approximate them.

• Irrational means not expressible as a ratio of integers.
• But every irrational can be approximated by rationals to arbitrary precision.
• The question is which one is hardest to pin down with fractions.

Learner tries to write fractions that get close to π, √2, and φ, then sees how small the error must be for each.

• Guess best rational approximations within a denominator limit.
• Compare how close you got for each constant.
• Notice which one feels hardest to pin down.

Introduce continued fractions as a way to read off the best rational approximations. Show the first several terms of √2, e, π, and φ.

• A continued fraction writes a number as a + 1/(b + 1/(c + ...)).
• Truncating the continued fraction gives the best rational approximation at that depth.
• Big partial quotients mean the approximation jumps a long way.

An interactive plot showing the approximation error versus truncation depth for √2, e, π, and φ. The learner sees how rapidly each curve descends.

• Truncate the continued fraction at increasing depth.
• Plot the resulting error on a log scale.
• Identify which curve falls slowest — that is the most irrational.

Explain the theorem: among all irrationals, the one whose continued fraction partial quotients are all 1 is the hardest to approximate by rationals. Equivalently, any irrational has a convergent at least as good as φ's.

• Hurwitz's theorem: infinitely many rationals p/q satisfy |α - p/q| < 1/(√5 q²).
• The constant √5 is sharp, and equality holds for φ.
• So φ is the unique worst case: every other irrational is approximated strictly better.

A new constant, the plastic number ρ ≈ 1.3247 with continued fraction [1; 1, 1, 1, 1, ...]. The learner tests whether ρ might be 'more irrational' than φ.

• Truncate the plastic number's continued fraction and read off convergents.
• Compare the error curve to φ's.
• Conclude that ρ is actually easier to approximate, despite its all-1 pattern ending.

Discuss what the result does and does not claim: irrationality is a property; 'most irrational' is a quantitative competition that depends on the chosen measure.

• Liouville numbers exist that defeat φ in specific approximations, but only finitely often.
• Lebesgue-almost-every real is approximated about as well as φ — φ is the worst typical case.
• There is no 'most transcendental' or 'most algebraic' in the same clean sense.

Recap the driving question and its resolution. The golden ratio, via its continued fraction of all 1s, is the hardest real number for rationals to approximate.

• Continued fraction partial quotients measure approximation difficulty.
• φ = [1; 1, 1, 1, ...] has the smallest possible partial quotients forever.
• Hurwitz's theorem with constant √5 certifies φ as the unique worst case.