MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

As the title says, please share colloquial statements that encode (in a non-rigorous way, of course) some nontrivial mathematical fact (or heuristic). Instead of giving examples here I added them as answers so they may be voted up and down with the rest.

This is a community-wiki question: One colloquial statement and its mathematical meaning per answer please!

share|cite|improve this question
Recent answers suggest this question is getting a bit long in the tooth. – S. Carnahan Jul 6 '12 at 5:38

51 Answers 51

Another one: not so much a catchphrase, but a nifty interpretation of a theorem:

"Suppose a human is walking a dog on the leash and they encounter a lamp post. Then, if the leash is kept short enough, the human and the dog wind around the post the same number of times."

I learned this interpretation of Rouche's Theorem from the textbook in complex analysis by Saff and Snider. They include pictures, too.

share|cite|improve this answer
By “the leash is kept short enough”, do you mean it can't be flung over the lamp? – Zsbán Ambrus Jul 5 '12 at 21:31

Being compact is the next best thing to being finite.

As seen for example by the fact that the (uniform) limit of a sequence of continuous real-valued functions on a compact space is continuous. Or the even more down to earth example: The extreme value theorem.

share|cite|improve this answer
I've heard a minor variation: "Compact is the new finite." – S. Carnahan Aug 6 '12 at 0:47

Truth is undefinable,

which is a statement of Tarski's theorem. More precisely,

Truth in a context where one can do arithmetic is undefinable in that context.

share|cite|improve this answer

A straight line is the shortest distance between two points.

share|cite|improve this answer
How can a line be a distance? – Rasmus Bentmann May 10 '11 at 8:11
@Rasmus: I was initially uncertain whether to take your question seriously. But I find some answers to the original question that seem to construe the question in an altogether different way from what appears to me to have been intended. I thought "colloquial catchy statements" meant things that ordinary non-mathematicians would say, giving words the meanings they normally have in the usages of non-mathematicians. My answer here is verbatim the way it's normally heard. – Michael Hardy May 10 '11 at 14:58
This is entirely subjective, but to me this does not feel like the way it's normally heard. Maybe "the shortest way"? – Rasmus Bentmann May 10 '11 at 20:29
Seems to me you're trying to make it fit into the conventions used by mathematicians, not shared by others. – Michael Hardy May 10 '11 at 21:41
That's possible. The google result shocked me. ;) – Rasmus Bentmann May 11 '11 at 12:14

I'm surprised that nobody has mentioned the famous,

"Four colors suffice."

share|cite|improve this answer

The consequences of this little harmless looking statement are deep enough that I worry that the majority will think that it is provably wrong.

Simple statement:

  • "You can pick a real number at random between 0 and 1, so that any number is as likely as any other."

more colloquially, (Freiling):

  • "You can throw a dart at the unit square."

more computer-scienc-y:

  • "The process of coin flipping to determine the binary digits of a real number converges to a unique well defined real number answer in the limit of infinitely many throws".

The interpretation of this statement is not that the probability distribution of the result is a well defined function, nor that statements about whether this number is in this or that Borel set can be assigned a probability--- both these assertions are true and boring. The assertion above is that a real number "x" produced in this way actually exists as an element of the mathematical universe, and every question you can ask of it, including "does x belong to this arbitrary subset S of [0,1]" gets a well defined yes or no answer in the limit. If you believe this assertion is self-evidently true, as I do, beware the implications!

  • The continuum hypothesis is false. (Sierpinsky,Freiling)

For contradiction, well order [0,1] with order type aleph-1, then choose two numbers x,y at random in [0,1]. What is the probability that $y\le x$ in the well-ordering? Since the set $\{ z|z\le y\} $ is countable for any y, the answer is 0. The same thing works whenever sets of cardinality less than the continuum always have zero Lebesgue measure.

  • Every subset of [0,1] has well-defined Lebesgue measure. (Solovay, more or less)

Make a countable list of independent numbers $x_i$, and ask for each one whether the number is in the set S or not. The fraction of random picks which land in S will define the Lebesgue measure of S. In more detail, if you write down a "1" every time $x_i$ is in S, and write down a zero when $x_i$ is not in S, then the number of ones divided by the number of throws converges to a unique real number, which defines the Lebesgue measure of S.

In this forum, somewhere or other, someone had the idea that this process will not converge for sets S which are not measurable, alternating between long strings of "0"s and long strings of "1"s in such a way that it will not have an average frequency of 1's. This is impossible, because the picks are independent. That means that any permutation of the 0's and 1's is as likely as any other. If you have a long string of N zeros and ones, the only permutation invariant of these bits is the number of ones. Any segregation of zeros or ones that has oscillating mean has less than epsilon probability whenever the mean number of ones after M throws, deviates by more than a few times $\sqrt{\ln \epsilon}/\sqrt{N}$ from the mean established by the first N throws.

It is astonishing to me that someone here simultaneously holds in their head the two ideas: "there exists a non measurable subset of [0,1]" and "you can choose a real number at random between [0,1]". The negation of the first statement is the precise statement of the second.

(Solovay defined this stuff precisely, but did not accept the resulting model as true. Others take the axiom of determinacy, thereby establishing that all subsets of R are measurable and that choice fails for the continuum, but determinacy is a stronger statement than "you can pick in [0,1]".)

  • The axiom of choice fails, already for sets of size the continuum.

Since the axiom of choice easily gives a non-measurable set.

  • The continuum has no well order.

This is because you could then do choice on the reals. So the first bullet on this list should really be rephrased as "the continuum hypothesis is just a stupid question".

  • Sorry, you can NOT cut up a grape and rearrange the pieces so that it is bigger than the sun.

Simply because if you put a grape next to the sun, and pick a random point in a big box that surrounds both, the probability that the random point lands in the grape is less than the probability that it lands in the sun. The Lebesgue measure of the pieces is well defined, and invariant under translations and rotations, so it never amounts to more than the measure of the grape.

  • The reals which are in "L", the Godel constructable universe, have measure zero.

When the axiom of Choice holds for all elements of the powerset of Z (i.e. R), then the pea can be split up and rearranged to make the sun. The axiom of choice holds in L, so that the Godel constructible L-points in the pea can be cut up and rotated and translated to fit over the L-points of the sun. This means that these points make a measure zero set, both in the pea and in the sun, when considered as a sub-collection of the real numbers which admit random picks.

To understand the Godel constructible universe, and choice, I will pretend that the phrase "Godel constructible" simply means "computable." This is a bald-faced lie. the Godel constructible universe contains many non-computable numbers, but they all resemble computable numbers, in that they are defined by a process which takes an ordinal number of steps and at each step uses only text sentences of ZF acting on previously defined objects. If you replace ordinal by "omega" and "text sentences acting on previously defined elements" by "arithmetical operations defined on previously defined memory", you get computable as opposed to Godel-constructible. To well order the Godel universe, you just order the objects constructed at each ordinal step by alphabetical order and ordinal birthday. To well order the computable reals, you just order their shortest program alphabetically (like the well-ordering of the Godel-universe, this ordering is explicitly definable, but not computable).

  • That stupid hat trick doesn't work in the random-pick real numbers

There is a recently popularized puzzle: A demon puts a hat, either red or green, on the head of a countably infinite number of people. Each person sees everyone else's hat, and is told to simultaneously guess the color on their heads. If infinitely many get this wrong, everybody loses. If only finitely many people get the answer wrong, everybody wins.

When the demon picks the hat color randomly on each person's head, they lose. Each person has 50% chance of getting their hat right. End of story. Nothing more to say. Really. This is why set theory has nothing to do with weather prediction.

  • The stupid hat trick does work over the computable reals, but is intuitive.

If the demon is forced to place hats according to a fixed definite computer program, there are only countably many different programs, the demon must pick a program, and stick with it. Then it is reasonable that each person can figure out the program used from the infinite answers at their disposal, up to a finite number of errors.

Supposing the people are provided just with some halting oracles and a prearranged agreement regarding computer programs. They do not need a choice function on the continuum. The people see the other hats, and they test the computer programs one by one, in lexicographic order, until they find the shortest program consistent with what they see. They then go through all the programs again, until they find the shortest program on integers which will give be only different from what they see in finitely many places (this requires a stronger oracle, but it still doesn't require a choice function). Then they answer with the value of this program at their own position.

(more precisely, to see everyone else's hat means that the demon provides a program which will give the value of everyone elses hat. You use the halting oracle to test whether each program successively will answer correctly on everyone else's hat, until you find the shortest program that does so.)

This version also has application to weather prediction: by studying the weather long enough, you can guess that it is obeying the Navier Stokes equations. Then you can simulate these equations to predict the weather. Come to think of it, this is exactly what we humans did.

  • The stupid hat trick is also intuitive in L, so long as you always think inside a countable model of ZF(C).

The demon again is constrained to definable reals below omega one, which is now secretly a countable ordinal (but ZF doesn't know it). So there is very little difference between the conceptual method to guess the definable real, except that now it isn't so easy to interpret things in terms of oracles.

  • There is no problem with "$R_L$, the L version of R, coexisting inside $R_R$, the actual version of R, in your mental model of the universe.

The axiom of choice is true in L, which includes a particular model for the real numbers (and all powersets). This model is fine for interpreting all the counterintuitive statements of ZFC, since they are just plain true in L. When you read a choicy theorem, you just imagine little "L" subscripts on the theorem, and then it is true (this is called relativizing to L in logic). But you always keep in mind that L is measure zero. Then that's it. There are no more intuitive paradoxes.

Your mental axiom system for no-intuitive-paradox mathematics than can be this

ZF (but you interpret powerset as "L-powerset")

V=L (and therefore Choice for all sets in your universe, and the continuum hypothesis for the constructible reals, and for the constructible powersets)

For each set S in the universe L, (which is well-ordered by V=L), there is a non-well-orderable proper class of subsets of S, the true power-class. Every subclass of a powerclass has a real valued Lebesgue measure, and every subset (meaning well-orderable sub-SET, not a non-well-orderable subCLASS) of a powerclass has measure zero. All powerclasses are the same size, since they are not powerclasses of previous powerclasses, just powerclasses of dinky little sets. The measure of the proper-class completion of the dinky little measure-zero L-Borel sets is the same as the measure assigned to these Borel sets in L.

This system does nothing but shuffle the intuition around. There is no new real mathematics here (no new arithmetic theorems). But with this in your head, you banish all the choice paradoxes to the dustbin of history. No more puzzles, no more paradoxes, no more nothing. This has been a public service announcement.

share|cite|improve this answer

Kronecker was wrong: God did not make the integers. He only made the empty set. Then He made mathematicians so they could make the integers from the empty set.

share|cite|improve this answer
I don't understand the distinction between the empty set and other sets here. If mathematicians construct sets from other sets, then they could construct the empty set from other sets as well. – Zsbán Ambrus May 9 '11 at 9:09
Assume there was no empty set. Consider the set of all empty sets, ... [seen years ago in Martin Gardner; not sure of the original source] – Noam D. Elkies Jul 6 '12 at 4:37
It is a bit like: Assume there were no proofs by contradiction.... – Lennart Meier Sep 17 '14 at 15:13

This sentence is false.

The famous liar paradox. As the wiki article explains:

If "this sentence is false" is true, then the sentence is false, which would in turn mean that it is actually true, but this would mean that it is false, and so on ad infinitum.

Similarly, if "this sentence is false" is false, then the sentence is true, which would in turn mean that it is actually false, but this would mean that it is true, and so on ad infinitum.


All Cretans are liars.

or as pointed out in the comment below the Barber paradox:

The barber shaves only those men in town who do not shave themselves.Who shaves the barber?

share|cite|improve this answer
Don't forget the (male) barber who shaves only those who do not shave themselves: This encodes Russell's antinomy. – Margaret Friedland Jul 4 '12 at 20:48

Surprised this one hasn't appeared yet:

The flap of a butterfly's wings in Brazil can set a tornado in Texas.

share|cite|improve this answer

There is no free lunch. Refers to risk/reward in financial investment and the fact that an efficient market moves to the point where you can only make money by taking risk.

share|cite|improve this answer

You Can’t Unscramble Scrambled Eggs

share|cite|improve this answer
As a statement asserting the existence of one-way functions, P /= NP, or the second law of thermodynamics. – Halfdan Faber Nov 1 '09 at 4:32
I understand the others, but how does this relate to P vs. NP? – Michael Lugo Nov 3 '09 at 15:56
The existence of one-way functions implies P/=NP, in that for any such function p, its inverse function, hp, would, by definition, be hard to compute for any input, but any output would be easy to verify using p. – Halfdan Faber Nov 4 '09 at 4:29
Ah, but theorem: Consider a compact pan with some unscrambled eggs in a closed (no "external" i.e. time-varying physics) classical Newtonian (energy is kinetic, which is positive-def quadratic in velocity, plus potential, which depends only on position) universe. The eggs may be in the process of scrambling. Then at some time in the future (indeed, after some precisely integer number of years, where how long you have to wait can be given an explicit absolute bound in terms of epsilon), the eggs will be within epsilon of unscrambled. – Theo Johnson-Freyd Dec 24 '09 at 20:43
Non-mathematically, this reminds me of the cryptic crossword clue: gegs (9,4). – Andrew Lobb Feb 22 '10 at 21:40

It's better to be lucky than good.

P != NP. A nondeterministic polynomial is one which is always "lucky".

share|cite|improve this answer
Given that P!=NP is unproven, perhaps I hope it's better to be lucky than good. – Ilya Nikokoshev Nov 1 '09 at 20:48
Or "better lucky than smart" – vonjd Feb 23 '10 at 14:24

There is a way to cut up a pea and rearrange the pieces to get the sun.

As a corollary of the Banach-Tarski paradox, we have that if A and B are bounded subsets of R^n (n > 2) with nonempty interior, there exists a partition of A into k pieces {A1, ..., Ak} and isometries of R^n {f1, ... , fk} so {f1A1, ... , fkAk} partitions B.

share|cite|improve this answer
This is not an immediate consequence, and the pea-to-sun statement is a highly misleading translation of the statement of the theorem. Both the pea and sun are roughly spherical, but the theorem shows that you can geometrically decompose a sphere into two spheres of the same radius, not into a sphere of a different radius. If you disagree, please tell me how many pieces you will use to go from a ball of radius 1 into a ball of radius 2. I would use 9 pieces to decompose a sphere into two of the same radius, although that's not minimal. – Douglas Zare Feb 19 '10 at 21:15
Maybe it shoud be formulated as: Give me one pea and I'll feed the world. :-) – Yiftach Barnea Apr 19 '11 at 8:41
@Douglas: There is a generalization of the Banach-Tarski-theorem that applies to almost arbitrary subsets of IR^n. (I think the precise condition is that they have non-empty interior) – Johannes Hahn Apr 19 '11 at 10:14
@Johannes: I think that, in addition to having non-empty interior, the sets need to be bounded. – Andreas Blass Jul 5 '12 at 1:01
Visualize world peas? – Noam D. Elkies Jul 6 '12 at 4:47
There's no such thing as a free lunch.

This refers to the No Free Lunch theorem. The theorem states that it's impossible to develop a search optimization algorithm that works well for all possible problems. Rather, for every class of problems which a given algorithm performs well at, there is a complementary class for which it does not. Thus, you may think you're getting a free lunch, but you're really just paying for it somewhere else.

share|cite|improve this answer
Duplicate of… – Rasmus Bentmann May 10 '11 at 8:17
Not an exact duplicate. Here its about optimization and not about finance. – Dirk Jul 5 '12 at 6:49

I like the phrase ``counting in two different ways'' -- which I learned in a math camp. If you're not sure what it means, I suggest the exercise of trying to prove that $2^n = \sum_{i=0}^n {n\choose i}$ by looking for a proof which could be aptly described by this phrase.

share|cite|improve this answer

Even quite irregularly shaped objects, such as tables and chairs, become approximately spherical if you wrap them in enough newspaper.

(I think this is by J. H. Conway but I heard it through Bill Thurston, who we recently lost. RIP.)

share|cite|improve this answer
What is the serious mathematics that this encodes? – Michael Lugo Aug 25 '12 at 15:02

Poisson arrivals see time averages.

share|cite|improve this answer
Non-random is a special case of random.

When studying probability distributions, and start with the dirac delta function.

share|cite|improve this answer

Nothing contains everything.


There is no universe.

This is how Halmos (pp. 6-7) summarizes the answer of axiomatic set theory to Russell's paradox.

share|cite|improve this answer

GAGA the acronym for Serre's famous Geometrie algebrique geometrie analytique.

share|cite|improve this answer

Numbers are mutually friendly if they share their abundancy

share|cite|improve this answer

protected by François G. Dorais Sep 21 '13 at 23:09

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

Not the answer you're looking for? Browse other questions tagged or ask your own question.