ExplainerScience

What Is a Mathematical Proof and Why Does Every Step Need Checking?

5 min read / 2026-08-05

A mathematical proof is a step-by-step argument that shows a claim is true in every possible case, which is why mathematicians double-check AI-generated proofs line by line.

1,000+Open problems Paul Erdős posed in his lifetime

What it means

A mathematical proof is a chain of logical steps that starts from accepted facts, called axioms, and ends at a conclusion that must be true no matter what. Unlike a science experiment, which can only show something is likely true based on evidence, a proof aims for certainty. Once proven, a mathematical statement stays true forever, even if new tools or computers appear later. This is different from checking your bank balance on an app, where you trust the number shown; in math, you must be able to trace every step yourself.

How it works

A proof usually starts with a clear claim, like 'every even number greater than 2 can be written as a sum of two primes.' The mathematician then builds a logical path, using definitions and previously proven results, until the claim is fully justified with no gaps. Each step must follow strictly from the one before it. If even one step is wrong or unjustified, the whole proof can collapse, even if the final answer happens to be correct. This is why reviewers, called referees, read proofs carefully before they are accepted by the math community.

A simple example

Think of proving that the sum of two even numbers is always even. You could say: an even number can be written as 2 times some whole number, so two even numbers are 2a and 2b. Adding them gives 2a plus 2b, which equals 2 times (a plus b), and that is even by definition. This short proof covers every possible pair of even numbers at once, not just one example, which is what makes it different from just testing a few number pairs on a calculator.

Why people talk about it

AI systems can now search through huge numbers of possible proof strategies far faster than a person, which is how they have helped crack some of Paul Erdős's unsolved problems, as reported by Quanta Magazine. But speed in searching is not the same as guaranteed correctness. An AI might produce a proof that looks convincing but contains a subtle logical gap, similar to an autocomplete text that sounds fluent but states a wrong fact. That is why mathematicians still verify each AI-suggested step, much like a teacher checking a student's full working, not just the final answer.

What to remember

A proof is only accepted once every step withstands scrutiny, whether it was written by a human or suggested by an AI system. AI is proving useful for testing many possible paths quickly, especially for problems that are easy to state but hard to search through, like several of Erdős's puzzles. The creative judgment of choosing what to try, and the discipline of verifying it, still rests with human mathematicians for now.

Key words

Proof

A logical, step-by-step argument that shows a mathematical claim is true in all possible cases.

Axiom

A basic statement accepted as true without proof, used as a starting point for building other proofs.

Referee

An expert mathematician who carefully reviews a new proof for errors before it is accepted as correct.

Key facts

  • 1A mathematical proof must hold for every possible case, not just the examples that have been tested so far.
  • 2Referees, who are expert mathematicians, formally review new proofs before they are accepted into the wider body of mathematical knowledge.
  • 3Paul Erdős posed over 1,000 open problems in number theory and combinatorics before his death in 1996.
  • 4AI systems have recently helped resolve some long-standing Erdős problems by rapidly testing large numbers of possible proof strategies, according to Quanta Magazine.
  • 5A famous proof can remain valid forever once verified, unlike scientific theories that can be revised with new evidence.

Why it matters

Understanding what counts as a real proof explains why mathematicians celebrate AI progress on Erdős's puzzles cautiously, since every AI-suggested proof still needs careful human verification before it counts as solved.

Sources

  • Quanta Magazine
  • MacTutor History of Mathematics Archive, University of St Andrews

Related stories