The Simplest Math Problem No One Can Solve - Collatz Conjecture

V
Veritasium
Education3 min read23 min video
Jul 30, 2021|45,229,679 views|938,036|81,059
Save to Pod

Key Moments

TL;DR

All positive integers reach 1 under 3x+1; patterns, randomness, and open questions.

Key Insights

1

The 3x+1 rules generate highly variable “hailstone” sequences, where numbers bounce up and down before stabilizing on 1.

2

Some seeds (like 27) peak extremely high and take many steps to reach 1, illustrating the chaotic paths possible.

3

Benford’s law appears in the distribution of leading digits of hailstone sequences, a sign of underlying statistical structure rather than a proof.

4

Terence Tao and others showed strong probabilistic results: almost all sequences dip below their starting value and obey even stronger slowly growing bounds.

5

There are genuine surprises on the negative side of the number line, including multiple independent loops, challenging intuition about positivity.

6

Despite massive brute-force checks (up to 2^68), no counterexample exists yet, but the problem remains unproven and possibly undecidable in principle.

THE SIMPLE RULES BEHIND THE CONJECTURE

At its core, the Collatz conjecture is built on two simple rules: if a number n is odd, replace it with 3n + 1; if it is even, replace it with n/2. Repeatedly applying these rules from any positive seed generates a sequence, commonly called a hailstone path, that rises and falls in often dramatic fashion. For example, starting with 7 produces a cascade ending at 1, after which the sequence cycles 1-4-2-1. Despite the elegance of the rules, the long-term behavior of all seeds remains mysterious, inviting deep questions about inevitability and convergence.

HAILSTONES AND PATHS: HOW SEQUENCES DANCE

The trajectories of hailstone sequences vary wildly from seed to seed. The famous case of 27 climbs to a peak of 9,232 before descending back to 1, taking 111 steps in total. Even seeds close to each other can have vastly different journeys. Visualizations portray these paths as directed graphs or coral-like spirals, illustrating how countless streams funnel toward the 4-2-1 loop. The existence of occasional peaks and long wandering phases shows how deceptively simple rules can yield intricate, almost organic structures.

LEADING DIGITS AND BENFORD'S LAW: A SURPRISING PATTERN

When you histogram the leading digits of numbers encountered along many hailstone sequences, you observe a striking, stable distribution: most numbers begin with 1, then decreasing frequencies for higher digits, exactly the kind of pattern Benford’s law predicts. This statistical regularity demonstrates an underlying order amid chaotic movement, and it helps explain why large numerical datasets born from iterative processes often resist naive intuition about randomness.

SHRINKING VERSUS GROWING: WHAT THE ODDS TELL US

Although odd steps briefly raise the value (roughly tripling and adding one), the subsequent halving steps typically pull the number down. On average, you tend to move from one odd term to the next by a factor around 3/4, a geometric mean less than one. This favors eventual descent toward smaller numbers, even as occasional seeds surge high. An intuitive picture emerges: the process is biased toward shrinking, which helps explain the frequent convergence toward 1 in many observed paths.

EVIDENCE, LIMITS, AND NEAR-MISS PROGRESS

Vast computational searches have tested seeds up to astronomical sizes—in particular up to 2^68—without finding a counterexample. Beyond brute force, progress is probabilistic and subtle: partial results show that almost all trajectories dip below their starting value, and stronger bounds hold for all seeds, albeit in asymptotic senses. The landscape includes dramatic examples like 27 and a few negative seeds, which reveal unexpected structure. Yet this evidence does not constitute a proof, leaving the conjecture tantalizingly unresolved.

COULD IT BE UNDECIDABLE OR FALSE? PROLOGUES OF A DEEPER LIMIT

A key twist is not just whether the conjecture is true, but whether it can be decided at all. John Conway’s fractal-like generalization of 3x+1 led to a touring-complete machine, highlighting that some questions about dynamical systems can inherit undecidability aspects. The possibility remains that the Collatz problem could be true but unprovable within current axioms, or even false due to deeper, hidden counterexamples. This perspective reframes the quest as exploring the limits of mathematical proof itself rather than merely hunting a specific solution.

3x+1 Quick Start Cheat Sheet

Practical takeaways from this episode

Do This

Choose a seed and apply: if n is odd, compute 3n+1; if even, replace n with n/2.
Track the sequence; many seeds descend to 1 and enter the 4-2-1 loop.
Understand that the problem is not yet proven; use it as a learning/curiosity exercise.

Avoid This

Don’t claim a proof of the conjecture from brute-force checks up to large bounds.
Don’t assume convergence for all seeds without a formal argument.

Benford-like leading-digit distribution described

Data extracted from this episode

Leading digitObserved percentage
130%
217%
312%
49%
58%
67%
76%
85%
94%

Common Questions

Pick any positive integer. If it’s odd, multiply by 3 and add 1; if it’s even, divide by 2. Repeat. The conjecture claims every seed eventually reaches the cycle 1-4-2-1, though this remains unproven.

Topics

Mentioned in this video

study2^68

the enormous bound up to which seeds have been brute-force tested for convergence; used as evidence but not a proof.

personAlex Kovich

co-investigator with Sinai examining the patterns in hailstone paths.

studyBenford's Law

distribution of leading digits observed in many real-world datasets; used to spot irregularities or fraud.

toolBrilliant.org

online learning platform sponsor mentioned for interactive math courses and problem solving.

toolFract Tran

John Conway's generalized machine based on Collatz-like rules; Turing complete and linked to the halting problem.

personGeorge Paa

original proposer of a counterexample-style perspective in the discussion of number theory (referenced as George paa).

personJeffrey Lagarias

one of the world's authorities on 3x+1.

personJohn Conway

created the fract Tran generalization of 3x+1 and proved it is Turing complete.

personLther Collets

German mathematician after whom the conjecture is commonly named (as the Collets/Collatz conjecture).

personPaul Erdős

famous mathematician quoted as saying mathematics is not yet ripe for such questions.

personRho Teras

proved in 1976 that almost all Collatz sequences reach below their starting value (later refinements followed).

studySyracuse problem

alternative label for aspects of the Collatz-like questions; referenced as part of origin stories.

personTerry Tao

one of the world's most prominent mathematicians; proved almost all numbers will eventually be smaller than any given growing function f(x).

personYakov Sinai

collaborator who studied the paths of hailstone numbers with Alex Kovich.

More from Veritasium

View all 12 summaries

Found this useful? Build your knowledge library

Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.

Try Summify free