You are on page 1of 17

17/5/2014 When is proof by contradiction necessary?

| Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 1/17
Gowers's Weblog
Mathematics related discussions
When is proof by contradiction necessary?
Its been a while since I have written a post in the somewhat philosophical category, which is
where I put questions like How can one statement be stronger than an another, equivalent,
statement? This post is about a question that Ive intended for a long time to sort out in my mind
but have found much harder than I expected. It seems to be possible to classify theorems into three
types: ones where it would be ridiculous to use contradiction, ones where there are equally sensible
proofs using contradiction or not using contradiction, and ones where contradiction seems forced.
But what is it that puts a theorem into one of these three categories?
This is a question that arises when I am teaching somebody who comes up with a proof like this.
Suppose that the sequence is not convergent. Then a few lines of calculation which
implies that . Contradiction. They are sometimes quite surprised when you point out that
the first and last lines of this proof can be crossed out. Slightly less laughable is a proof that is more
like this. We know that . Suppose that . Since the derivative of
has absolute value at most 1 everywhere, it follows that , which is a contradiction.
Therefore, . There, it is clearly better to work directly from the premise that
via the lemma that to the conclusion that
. However, the usual proof of the lemma does use contradiction: one assumes
that the conclusion is false and applies the mean value theorem.
The result of all this is that I dont have a good tip of the form, If your theorem is like this then try a
proof by contradiction, and otherwise dont. For the remainder of this post Ill discuss another
couple of examples that show some of the complications that arise.
Example 1. The irrationality of the square root of 2.
This is of course the classic proof by contradiction, and one can even give a kind of quasi-proof that
it must use contradiction. The reason is that the word irrational means not equal to for any
pair of integers . If that is the definition, then let us suppose that the last two lines of the proof
went: therefore has property ; therefore is irrational. We could then ask, Why does
having property imply that a number is irrational? It might be obvious that property implies
irrationality, but to prove it it would still be necessary to say, Well, take any rational number
therefore does not have property . (Why would that be necessary? For precisely the
same reason! Perhaps this is an induction on proof length or something like that.)
With those thoughts in mind, consider the following argument. We begin by calculating the
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 2/17
With those thoughts in mind, consider the following argument. We begin by calculating the
continued-fraction expansion of . We find that . The
denominator of the fraction is , so we see that the continued-fraction
expansion repeats itself, and, in one reasonably standard notation, is . In particular, it is
infinite. Therefore, is irrational.
At first sight, this looks like a direct argument rather than a proof by contradiction: we used the
hypothesis that to deduce that has a property that obviously implies irrationality.
However, as I mentioned in the general remarks, one could question this by asking, Why is it that
a number with an infinite continued-fraction expansion has to be irrational? The answer? Its
obvious that a rational number has a terminating continued fraction, because as you work it out the
denominators keep decreasing oops, sorry, that was a proof by contradiction.
So perhaps the answer is indeed that if you are trying to prove a negative statement, then you have
to use a proof by contradiction. But what is a negative statement? How about the following
theorem?
Theorem. If and are integers, then .
Aha, you say, the not equals makes that negative. But we can deal with that by a quick
reformulation.
Theorem. If and are integers, then .
Whats negative about that? If you think that its somehow bound up in the notion of strictly
greater than, then how about this?
Theorem. If and are integers, then there exists a real number such that .
That looks pretty positive to me, since it is asserting the existence of something.
It becomes less positive if you imagine how you would establish the existence of such an . The
obvious thought would be, Well, the only thing that could possibly go wrong is if , so all
we have to prove is that . And thats negative again. So does that mean that a statement is
negative if the only sensible way of proving it is to reduce it to a statement that includes the word
not? Even if something like that is correct, it seems quite hard to formalize.
Heres another example of that last type of question. Is being infinite a negative property? One
might say yes, because it means not being finite. But when we were talking about continued
fractions, all we cared about was sequences, and we can define a sequence to be infinite if its terms
can be put in bijection with the natural numbers. (And we could define a set to be infinite if there is
an injection to a proper subset. But is a proper subset a negative concept because it means not
including all elements?)
Example 2. Continuous functions on closed bounded intervals.
Until recently I knew that the following was the case. If you want to prove something using the
compactness of , then you can either prove it directly using the Heine-Borel theorem or you can
prove it by contradiction by reformulating everything in terms of sequences and applying the
Bolzano-Weierstrass theorem. For example, to prove that a continuous function on is
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 3/17
bounded, you either find a neighbourhood of each point on which is bounded (by the definition of
continuity) and cover by finitely many such neighbourhoods (by the Heine-Borel theorem), or
you assume that is unbounded, construct a sequence such that for every , apply
the Bolzano-Weierstrass theorem, and reach a contradiction.
I had also tacitly assumed that there is an algorithm for converting proofs of one kind into proofs of
the other, though I had never actually tried to work out the details.
But recently, a colleague and I had a conversation that led to the following proof of the theorem,
which disturbs my cosy view of how things work. The idea is to try to imitate the above proof by
contradiction as much as possible, but without actually bothering to prove the result by
contradiction. Specifically, one builds a sequence that is the most likely sequence to cause trouble,
and then proves that it does not cause trouble. Here is how the argument goes.
Let be the supremum of . By the definition of a supremum, we can
find a sequence such that . Pick a convergent subsequence , by Bolzano-
Weierstrass. Then is a subsequence of , so . But if is the limit of the
sequence , then . So . So is an upper bound for . (Note that this
proof also shows that the bound is attained.)
There seems to be no contradiction involved. But again, if we dig a little deeper it starts to look as
though what is really going on is that the contradiction is hidden in the obvious steps of the proof.
For instance, how do we know that we can find a sequence such that ? Well need to
split into two cases (unless we want to define the topology we are implicitly putting on the extended
real line). The case where is not really worth investigating, since it instantly gives us that is
bounded (though it was by unnecessarily doing this step that we obtained the added information
that attains its upper bound). And if we then look at the case , how is what we are doing
any different from assuming that is unbounded? I find myself rather confused.
Final remarks.
One thing that seems to be coming out of these examples is that the notion of a proof by
contradiction is relative to the definitions you use and the small results that you take for granted. For
instance, I could define a number to be irrational if its continued-fraction expansion is non-
terminating. I wouldnt actually advocate doing that, but if one did, then the direct proof I gave of
the irrationality of would be just that direct. And if I do not allow myself to assume that
then the proof that whenever
appears to stop being direct and require a contradiction.
In which case, perhaps the advice that I give to students proof by contradiction is a very useful
tool, but try not to use it unless you really need it is, though not completely precise, about the
best one can do.
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 4/17
You May Like
1.
This entry was posted on March 28, 2010 at 3:59 pm and is filed under Somewhat philosophical.
You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or
trackback from your own site.
53 Responses to When is proof by contradiction necessary?
Terence Tao Says:
March 28, 2010 at 4:21 pm | Reply
Nice post, as always
One can use the continued fraction expansion of to give a constructive algorithm to solve the
equation with an which is guaranteed to be well-defined (basically, x will be another continued
fraction, in which all denominators can be shown to be non-zero). Once one can solve and , one
can of course solve , though it is amusing that this proof of a purely rational fact proceeds using
the reals (via continued fractions). Though it may be possible to terminate the continued fraction
after a finite time and make this a purely rational proof as well (though this may require an
infinite descent, leading to another way for proof by contradiction to sneak in).
At some point one has to use the fact that the sum of two positive numbers is positive (and thus
has a reciprocal), but if such a statement falls under the category of proof by contradiction
then really theres no escaping that category. There may also be subtleties in the construction of
the real numbers, and the demonstration that every continued fraction converges to a real
number.
For me, the purest examples of proofs by contradiction are those of the non self-defeating
object type, where the existence of an overpowered object is crucially needed to demonstrate the
non-existence of said object. See my post at
http://terrytao.wordpress.com/2009/11/05/the-no-self-defeating-object-argument/
Even in those cases, though, one can often extract a large component of the argument which
does not involve proof by contradiction (Euclids theorem on the infinitude of primes being a
good example of this).
About these ads (http://en.wordpress.com/about-
these-ads/)
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 5/17
Terence Tao Says:
March 28, 2010 at 5:10 pm
Actually, I just realised that the proof I had in mind that the algorithm to solve terminates
relies on infinite descent, which of course uses contradiction. But one can take
contrapositives and prove the solubility of, say, by strong induction and a division into cases
instead. Indeed, if p is odd, or if p is even and q is odd, then is either odd, or twice an odd
number, in which case solubility is clear; and if are even then one can appeal to the strong
induction hypothesis.
The division into cases would be problematic if one were to adopt an intuitionistic stance on
logic, though. (I believe that one can still prove irrationality of sqrt(2) intuitionistically, but
now proof by contradiction is definitely required.)
Ben Lund Says:
April 29, 2010 at 3:41 am
Following Gowers logic above, you still need a contradiction to prove the irrationality of
sqrt(2).
For any rational number a, is soluble. This is not the case for , so is irrational.
porton Says:
March 28, 2010 at 4:36 pm | Reply
Field medalist Gowers, it is really simple indeed.
A direct proof A1 => A2 => => An is equivalent to proof by contradiction not An => not
A(n-1) => => not A1.
What is more natural depends on what statements are more natural Ai or not Ai.
No philosophy.
gowers Says:
March 28, 2010 at 4:39 pm
Er thanks.
kazek Says:
March 29, 2010 at 9:52 am
Axiom1: X={A,B,C}
Axiom2: [ (s =A) or (s= B) or (s = C) ] => p(s)
A1= s=A
A2= p(s)
A1=>A2 (axiom 2)
not A1 = not (s = A) or equivalent not A1 = s=/= A
not A2 = not p(s) ( if s=B then s=/=A and p(s) is still true).
So as things are different at least from that You wrote.
kazek Says:
March 29, 2010 at 10:49 am
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 6/17
Last sentence should be: So things are different at least from that You wrote. I suppose
You are saying that proof by contradiction and direct are equivalent, and that is true, but a
way You describe equivalence is probably not correct.
daniel. Says:
April 16, 2010 at 10:18 am
You are thinking of contrapositive and not contradiction.
sincerely, non-Fields medalist nobody.
Qiaochu Yuan Says:
March 28, 2010 at 5:07 pm | Reply
As far as exercises in classes go, Ive found it most natural to use proof by contradiction if the
statement I want to prove is universally quantified, since its negation is an existential statement. I
can then use the object the existential statement gives me to power up my proof in a way that
a direct proof (checking that some condition really does hold everywhere) cannot do, at least
as easily. I think Ive heard similar sentiments expressed somewhere; maybe MO.
As far as the larger question at hand, if proof by contradiction means using the law of the
excluded middle then the problem is, as you say, that many definitions commonly in use in
mathematics are equivalent only if the law of the excluded middle is assumed. Without it one
must make some extra choices, and then one can ask what is provable in intuitionistic logic and
what is not. (Or something like that.)
gowers Says:
March 28, 2010 at 6:28 pm
What Im looking for is a way of recognising in advance which statements are best tackled
by contradiction. I dont think your suggestion captures it there are just too many
counterexamples. For example, suppose I need to prove that there is no smallest positive real
number. Then I am proving that for every positive real number there is a smaller one. If I
followed your advice, the proof I would write out would go like this.
1. Suppose the result is false, and let be the smallest positive real number.
2. The number is positive and smaller than .
3. But this contradicts the assumption that was the smallest positive real number.
It seems to me more natural to say this.
1. Let be a positive real number.
2. Then is positive and smaller than .
The point in this example is that the assumption that is the smallest positive real number is of
no help in finding a smaller positive real number. I think a sensitivity to something like that
should be part of the general principle I am looking for.
Another example is this: prove that the product of any four consecutive integers is divisible
by 24. Contradiction doesnt come into it. One can either argue that is the binomial
coefficient , or one can argue that the set must contain at least one multiple of 4, one other
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 7/17
even number, and one multiple of 3. (Of course, as above it may be that justifying these
further facts eventually requires contradiction to prove some very low-level assertion.)
Qiaochu Yuan Says:
March 29, 2010 at 6:57 pm
Hmm. Maybe I should give an example and leave the extraction of the general principle to
others. Heres a typical example of when I found it useful to negate statements in topology.
Proposition: Suppose a topological space has the property that every open cover has a finite
subcover. Then closed subsets of with the finite intersection property have non-empty
intersection.
Proof. Suppose that has the first property but not the second. Then there exists a collection of
closed subsets of with the finite intersection property with empty intersection. The collection
is then an open cover of , so it admits a finite subcover, but this contradicts the finite
intersection property.
Maybe its equally easy to phrase this proof without the use of contradiction and I am
missing something. But the hypotheses dont seem to match up as nicely if you dont use
contradiction.
Qiaochu Yuan Says:
March 29, 2010 at 7:02 pm
Ah, I am missing something: the statement I wrote above can be proven directly by using the
complement of the intersection to build an open cover. Let me see if I can find an example
with more quantifiers
Tom Ellis Says:
March 28, 2010 at 5:31 pm | Reply
Its utterly obvious because if it has a terminating continued-fraction expansion then you can
just work out what rational number
Am I missing something, or did you mean the converse here?
gowers Says:
March 28, 2010 at 5:55 pm
Whoops no, that was my mistake. Ill change it right away.
Joshua Green Says:
March 28, 2010 at 5:32 pm | Reply
In your first example Suppose that the sequence (a_n) is not convergent. Then a few lines of
calculation which implies that a_n\rightarrow a. Contradiction. it isnt necessarily the case
that the first and last lines can be removed, as that first line might be necessary for one of the
implications along the way. On the other hand, it is often the case that the last line can be
dropped, replaced instead by a contradiction with the line before. Terence Tao points out my
favorite example of this, Euclids proof of the infinitude of primes, where arguably the final
contradiction is with the Fundamental Theorem of Arithmetic but many people use that theorem
to instead derive a contradiction with the opening statement.
gowers Says:
March 28, 2010 at 5:59 pm
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 8/17
Id be inclined to say that the assumption that fails to converge is rather unlikely to be helpful
in proving that . The main point is that I really have seen proofs written by students in which
it really is the case that the initial Suppose not and final contradiction can be crossed out
with no loss to the argument.
Emmanuel Kowalski Says:
March 28, 2010 at 7:37 pm | Reply
In the proof of irrationality of sqrt(2), I find it interesting to see how it enters in a proof of the a
priori stronger, and direct-looking, statement that gives a positive lower-bound for $|\sqrtt{2}-
p/q|$ for any integers $p$ and $q$, as a function of $q$. This is a special case of Liouvilles
inequality, and the idea is to compare an upper bound for $|f(p/q)|=f(\sqrt{2}-f(p/q)$, where
$f(x)=x^2-2$, which comes from the mean-value theorem and involves $|\sqrt{2}-p/q|$, with a
lower bound coming from the fact that $q^2f(p/q)$ is an integer, which is non-zero because
$\sqrt{2}$ is not rational.
Klas Markstrm Says:
March 28, 2010 at 8:37 pm | Reply
This is certainly biased due to my own way of thinking about things but I have found that
proofs by contradiction are often another way of stating a construction algorithm for an object
with given properties . E.g. the proof by contradiction for the infinitude of primes can be turned
into an algorithm which constructs an infinite list of primes. The proof of the four colour theorem
is another proof by contradiction, which can be turned around and made into a polynomial time
algorithm for constructing a four-colouring of a planar graph.
Mark Bennet Says:
March 28, 2010 at 10:40 pm
The Four Colour Theorem is an interesting example it is necessary to prove that the
algorithm for four-colouring works for all planar graphs, and this seems to me to be done by
contradiction? [Any planar graph must contain one of a finite list of special subgraphs, find
such a subgraph and you can reduce your problem to a smaller case. And to prove it
contains one of the subgraphs - assume it doesn't and establish a contradiction]
Fermats method of descent can be structured to use contradiction in a similar way to many
approaches to 4CT by assuming a minimal counterexample and then showing that there
must be a smaller one contradiction. There are alternate proofs of results proved by this
method but now Ive been tuned to see contradiction where I didnt expect it, I cant
confidently cite one where the alternative is definitely positive.
Sobre las demostraciones por contradiccin Says:
March 28, 2010 at 10:49 pm | Reply
[...] publicado en su blog una reflexin sobre las demostraciones por contradiccin. Se los
recomiendo: When is proof by contradiction necessary? Clase: Matemticas | Etiqueta: [...]
Arnaud Spiwack Says:
March 29, 2010 at 12:21 am | Reply
Im not quite sure, but when I read the sentence: How can one statement be stronger than an
another, equivalent, statement?. It seem to me that you want to know more about constructive
mathematics.
Also it seems that by proof by contradiction you mean two things that are logically rather
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 9/17
Also it seems that by proof by contradiction you mean two things that are logically rather
different. First that too prove that a, it suffices to prove a contradiction while assuming a
(dubbed negation introduction). The second is that a a (or (qp)(pq),
equivalently) (dubbed double-negation elimination). Now, the latter is not a valid principle in
constructive mathematics (it is equivalent to the principle of excluded middle). This results in
having a way to prove negated formula, but no real way to use them to prove stuff (apart from
contradiction, basically). Consequently definition formulated as negative are quite weak,
compared to a classically equivalent definition that would be all positive.
Now that means that constructive mathematicians avoid negative definitions as much as
possible. I would recommend reading Errett Bishops book : Foundations of constructive analysis
which is, unfortunately out of print, and rather hard to find second-hand. He develops a
rather significant amount of analysis without ever resorting to a proof by contradiction of the
second kind, and little, if any, of the first kind.
gowers Says:
March 29, 2010 at 8:27 am
Undoubtedly whether or not a proof is constructive is closely related to whether or not it
relies on contradiction in an essential way. But my question is a less formal one: I want to
know how an ordinary mathematician who doesnt mind non-constructive proofs can look at
a statement and recognise that a proof by contradiction is more or less forced. It sometimes
happens that one proves a result by contradiction and then realizes that it is quite easy to
prove all the contrapositives of the steps. Is there some way of recognising in advance that
this is going to be the case?
Even at the informal level, it may well be that thinking about whether one can prove
something constructively will be relevant, as Terry showed in his comments above.
I can see though that its quite likely that my informal question cannot be made formal and
answered except in more or less the way you indicate.
Arnaud Spiwack Says:
March 29, 2010 at 11:12 am
Well, I doubt it can be made formal, because different people might put the limit of what is
reasonably forced to be proved by contradiction. My guess was something like: maybe
reading some constructive-intensive mathematics can help one make up his mind. Especially
regarding whether this or that definition should contain a negation.
Andrej Bauer Says:
March 29, 2010 at 4:05 pm
I think what Arnaud is saying is that you are calling two things proof by contradiction
when only one of them deserves this name. If you are trying to prove a negation of a
statement then what you call proof by contradiction is unavoidable in a certain sense
(there are meta-theorems in logic about this). If you are proving a statement which itself is
not a negation by assuming its negation and getting to a contradiction, then very likely you
can find a direct proof. For example, your Bolzano-Weierstrass example has a direct proof.
Its to long to be posted here, but I wrote a blog post which explains all this more carefully
and contains the proof at http://math.andrej.com/2010/03/29/proof-of-negation-and-proof-
by-contradiction/.
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 10/17
gowers Says:
March 29, 2010 at 5:21 pm
Ive just read your blog post and found it very interesting. And I perfectly fit your paradigm
of somebody who was unaware of the difference between proof by negation and proof by
contradiction. Ill try to heighten my awareness from now on. (In fact, your analysis of why I
wrote what I wrote was spot on all the way through: Im so used to cancelling out double
negations that I didnt really appreciate the significance of what Arnaud was saying.)
Two small further remarks. The constructive proof you gave of the boundedness of
continuous functions on [0,1] was essentially the same as one that my colleague came up
with, and the argument in my blog post was a different argument that I produced in
response.
Finally, with reference to your comment about countable choice, one could obtain the
sequence by imitating the proof of the intermediate value theorem: for each one wants to find
such that , so one lets be the set of all such numbers and takes to be the infimum of ; one then
checks, using the continuity of , that .
Andrej Bauer Says:
March 29, 2010 at 5:39 pm
Your suggestion for eliminating countable choice works very well classically.
Intutitionistically there is trouble (how do you know these infima exist?) so people who
usually worry about choice (namely constructive mathematicians) dont think of such tricks.
Anyhow, I think my blog was not explicit enough, so Ill reiterate my answer to your
question: proofs of negation cannot be eliminated, except via previously proved lemmas
which eventually reduce to a proof of negation. Proofs by contradiction can almost always be
eliminated (in analysis), except for some well-known theorems that are equivalent to (weak
forms of) excluded middle.
Kenny Easwaran Says:
April 10, 2010 at 11:15 pm
When I started reading the post I was going to make a point similar to Andrej Bauer, that
when a statement begins with a negation, then negation-introduction is going to be
necessary.
But in the original post, I think Tim Gowers raises some issues that make this
characterization a bit problematic whether or not the formal version of an informal
statement begins with a negation depends on how you formalize it. Statements with strict
inequalities look like positive statements, but are often equivalent to negated equations. I
suppose constructivists/intuitionists (Im not always clear on the distinction) avoid this worry
by allowing that and are not equivalent statements. So a classical mathematician can't just
use this particular distinction.
I think instead the phenomenon of interest has to be about the informal proofs (which
mathematicians actually use), and not the formalized counterparts (which logicians study).
There's something displeasing about using proofs by contradiction in certain contexts, but
often it can only be eliminated by using some lemma that hides it, or by mutilating the proof
in some worse way. I'm not certain if there's any way to characterize the set of statements
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 11/17
whose proofs will be like this. (I suppose in some sense, this problem seems like it might be
strictly harder than characterizing the set of statements that are provable at all, which is of
course impossible given the formalized notions of "characterizing" and "provable".)
Zed Norwood Says:
October 12, 2011 at 4:14 pm
Andrej: Would you mind indicating what meta-theorems youre referring to in your first
reply to this comment?
Sorry to resurrect an old conversation.
andrejbauer Says:
October 12, 2011 at 4:52 pm
@Zed: what I have in mind is the theorem which states that in the intuitionistic propositional
calculus proofs can be converted to normal forms (eliminations followed by introductions).
You can understand a normal form proof as optimal and direct in the sense that it has no
unecessary steps and does not use any lemmas. Because not A is the same thing as A
implies false, the normal form proof of a negation is of the form assume A then prove
false.
porton Says:
March 29, 2010 at 12:24 am | Reply
You ask: How can one statement be stronger than an another, equivalent, statement?
Consider an algorithm (with or without the requirement that it terminates on every input) which
proves a class of mathematical theorems.
The we can define the statement A to be stronger than B if (A=>B) can be proved by this
algorithm.
Sam Says:
March 29, 2010 at 12:05 pm | Reply
I think the reason why it is so difficult for a mathematician to recognise a priori that a proof by
contradiction is forced, is that it is even difficult a posteriori, after writing down the proof, to
recognise whether or not a proof really, essentially is a proof by contradiction.
So I think a question that needs to be answered prior to the one you are posing here would be:
what are the conditions a proof has to satisfy in order for it to be called a proof essentially by
contradiction?
S Says:
March 29, 2010 at 2:00 pm | Reply
The approach I take is usually Which direction has more information?, or rather, Which
approach gives me more to work with?.
Proof of negation and proof by contradiction Mathematics and Computation Says:
March 29, 2010 at 4:00 pm | Reply
[...] have been meaning to write for a while. It was finally prompted by Timothy Gowerss blog
post When is proof by contradiction necessary? in which everything seems to be called proof
by [...]
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 12/17
DC Says:
March 30, 2010 at 3:47 am | Reply
Ignoring formal mathematics, this is a thorny pedagogical issue. We develop intuition for how
proofs of statements of various kinds will go, and want to transmit it to students but
much of what we experience as intuition is really hindsight.
The heart of any tip of the form If your theorem is like [blank] then try a proof by
contradiction, and otherwise dont is not the blank, but the word theorem. We ask students
to find proofs of statements that are known to be true and known to have proofs using what
we have taught them. This is often more significant than anything specific to the phrasing of a
statement being proved. Surely, sometimes we deliberately phrase a statement so that it conveys
information about a proof (e.g. prove that the polynomial x^2 2 has no rational roots may
suggest a classroom theorem), but in general, this information is not explicit making the
relevant connections is part of the students job. Make enough of these connections, and you
develop intuition or is that hindsight?
What does your intuition say about a proof of the Euler-Mascheroni constant is irrational? Just
to speak of a proof, we need to know more than we know.
This seems bound up with the meta-mathematical issue of knowing something is true versus
having a proof. These are certainly different concepts in the classroom: I can know Liouvilles
theorem is true, but not remember any proof during an exam. Outside the classroom it is harder
to say what the difference is. It is the basis of a very unfunny math joke.
Q: Whats the best way to start a proof of X?
A: Know that X is true.
Jonathan Vos Post Says:
March 31, 2010 at 6:51 pm | Reply
I remembered Feynman telling me the last quote here, and googling it I saw this nice page:
http://infoproc.blogspot.com/2006/12/keynes-and-planck.html
Found in the comments on Economists View:
Professor Planck, of Berlin, the famous originator of the Quantum Theory, once remarked to
me that in early life he had thought of studying economics, but had found it too difficult!
Professor Planck could easily master the whole corpus of mathematical economics in a few days.
He did not mean that! But the amalgam of logic and intuition and the wide knowledge of facts,
most of which are not precise, which is required for economic interpretation in its highest form is,
quite truly, overwhelmingly difficult for those whose gift mainly consists in the power to imagine
and pursue to their furthest points the implications and prior conditions of comparatively simple
facts which are known with a high degree of precision. (Keynes, Essays in Biography 1951
158n)
Its true: powerful mathematical minds are not necessarily comfortable with the messiness of the
real world. This observation might be applied as well to string (or more formal or mathematical)
theorists vs theorists who are more data- or intuition-driven (often called phenomenologists, in a
terrible use of terminology). Of course, some people (like Feynman, although he wasnt very
mathematical by todays standards) are good at everything
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 13/17
We know a lot more than we can prove Feynman.
Posted by Steve Hsu
John Says:
April 1, 2010 at 4:37 am | Reply
Your final comments resonate with some of my own informal observations it depends on what
kind of objects and statements you are working from.
For example, like I think you were saying, it seems that you can `bundle some of the non-
constructive or inherently `negative aspects into definitions and give direct proofs based on these
definitions, while different definitions might require a `negative proof, given the definitions you
are using.
In practice, I sometimes like to think about contradiction as about consistency and elimination of
alternatives, usually arising when you think, `this MUST be true, because it cant not be vs `this
IS true because I know this. Whenever you feel like saying `MUST you might think about
contradiction.
John Says:
April 1, 2010 at 4:43 am
Or `MUST not be true because if it was
gowers Says:
April 1, 2010 at 11:31 am
I like that must test. Ill try it out on a few examples and see whether it seems to work.
fan Says:
April 1, 2010 at 10:26 pm | Reply
In graph theory the easiest way to get your induction right is a proof by contradiction. Rather
than trying to figure out in what ways small graphs can extend to bigger ones (messy and error-
prone), you grab a smallest counterexample.
Anonymous Says:
April 2, 2010 at 1:29 am | Reply
As a non-mathematician can I ask why proof by contradiction is so disliked? I get that it would
seem to be inherently non-constructive and that might annoy some people. I suppose if you
arrive at a contradiction you know that something is wrong but is it necessarily the thing you
were trying to disprove, or is the the problem with the method of proof/disproof? Perhaps
equally you might be suspicious of a direct constructive proof as it could be an accident of
flawed logic or other false assumptions?
Getting to know your math tools. Math Society the club Says:
April 2, 2010 at 3:28 pm | Reply
[...] Getting to know your math tools. Timothy Gowers is a Fields Medal winner who writes a
blog about math. He has written a post asking When is proof by contradiction necessary? [...]
Paulo Oliva Says:
April 14, 2010 at 10:32 am | Reply
Two points. First, in [R. L. Goodstein, Proof by reductio ad absurdum. The Mathematical
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 14/17
Two points. First, in [R. L. Goodstein, Proof by reductio ad absurdum. The Mathematical
Gazette, Vol 32, No. 300, pp 198-204, 1948] the same issue of this blog is also discussed, and a
proof that square root of 2 is irrational is given without using proof by contraction. The article is
short and easy to read. Regarding the question How can an ordinary mathematician look at a
statement and recognise that a proof by contradiction is more or less forced?, I would say, given
my background in proof theory, that if the theorem states that an object x has property P(x),
where P(x) is computer-checkable then proof-by-contradiction can normally be avoided. For
instance, existence of infinitely many prime (object x in this case is a number, and P(x) says that
x is prime and bigger than n), or square root of 2 being irrational (object x is an epsilon, and P(x)
says that square root of 2 is away from p/q by that epsilon). In these two cases n and p/q are
arbitrary parameters, which can be thought of as inputs. If, on the other hand, the theorem
states that an object x with property P(x) exists but P(x) cannot be checked computationally,
then you most certainly will only be able to prove the theorem via a proof by contradiction. For
instance, in the case of the Bolzano-Weierstrass theorem, the object x is a real number, and the
property P(x) saying that x is an accumulation point would require someone to check infinitely
many possibilities. Another example is Brouwers fixed point theorem on [0,1] for instance, in
this case x is a real in [0,1] and P(x) says that fx = x, which is an equality between real numbers
that cannot be checked effectively by computer (think of how you would check two real
numbers are the same, you might check digit by digit and never know). This is of course an
informal answer to an informal question.
Anonymous Says:
April 19, 2010 at 4:29 am | Reply
I agree with Andrej and Paulo generally, the question you are asking is a natural question in
foundations of mathematics and logic, and there is a difference between Reductio ad
absurdum and introduction rule for negation. I just want to add three minor points.
About Andrejs comment, in Intuitionistic logic, one is mainly concerned with true statement,
therefore if you want to show something is false the only thing we can do is to show that by
assuming it we can derive contradiction, i.e. it does not study how to refute a statement directly.
There are statements that we can refute directly without going through this, example $0 \neq
1$. Similarly, to show that $\forall x, \varphi(x)$ is false we can give a specific $x$ such that
$\varphi(x)$ is false. We dont need to derive a contradiction form it. Intuitionistic logic is not
symmetric with respect to true and false statements.
About Paulos comment, his trick work in many cases because computable mathematics is a
model of constructive mathematics. But it is not faithful with respect to constructive reasoning,
i.e. there are statements that hold in computable mathematics which can not be proven
constructively. As a result, a statement can be both true in classical mathematics and
computable mathematics, and still you may need to use proof by contradiction to derive it.
Finally, I want to emphasize the seemingly obvious but not trivial point that to prove a
statement using proof by contradiction can be much easier than proving it without it.
If you are interested, I would suggest taking a look at Beesons book:
Foundations of Constructive Mathematics: Metamathematical Studies, Springer,
Berlin/Heidelberg/New York, 1985.
It is a nice book, and it is not an ideological one so it should make sense to classical
mathematicians.
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 15/17
Andrej Bauer Says:
April 19, 2010 at 5:51 am
Dear Anonymous, your examples of direct refutations ($0 \neq 1$ and giving a
counterexample to a universal statement) work intuitionistically just as well as clasically, and
are frequently used in intuitionistic mathematics, just as in classical mathematics. If you boil
them down to their formal proofs, you will discover however, that they (a) either rely on an
axiom which has the form of a negation, such as $0 \neq 1$ in the theory of fields, or $n+1
\neq 0$ in Peano arithmetic, or (b) your direct method of proving a negation relies on a
lemma of the form $A \implies \lnot B$ (for example $(\exists x, \lnot \phi) \implies \lnot
\forall x. \phi)$) whose proof then contains a proof of negation.
Your view of truth in intuitionistic logic is somewhat strange. What is in intuitonistic logic
one is concerned mainly with true statements supposed to mean? I thought all
mathematicians, no matter what party they belong to, are mostly concerned with true
statements. The trouble is, we dont know which one are true
Anyhow I think this question is not about intuitionistic logic. Its just that having practice
with intuitinstic logic helps one distinguish various logical forms which classical
mathematicians autoamatically view as the same.
Anonymous Says:
April 20, 2010 at 4:00 am
Hi Andrej,
In classical logic we have symmetry between truth and falsity, so the point I made does not
effect it. Stating something is true is the same as stating its (classical) negation is false. This
symmetry is broken in intuitionistic logic, if we need constructions for proving a statement is
true, we can also have constructions to show that a statement is false. Of course if we dont
treat falsity similar to truth and formalize them as intuitionistic negation we will end up with
what you said. But similar to the situation with true statements, we can have direct
observations showing that a statement is false. To show that 2^2=5 is false, we can just
compute 2^2 and get 4 and compare the normal forms of them to conclude that this
statement is false, where as in intuitionism since falsity is replaced with intuitionistic
negation, what we end up is that assuming 2^2=5 we derive 0=1 (and either define $\bot$ to
be just $0=1$ or have an axiom that states $0=1 implies \bot$) and conclude with $\lnot
2^2=5$. Even a computer does not need to find a proof of contradiction from $2^2=5$ to
claim it is a false statement. My point is there is a more natural way to establish that $2^2=5$
is false. This is a toy example but this holds in general, in place of coming up with a
construction for $\lnot \varphi$ we can show directly $\varphi$ is false by giving a
construction showing its falsity. A white raven is enough to show the universal statement
every raven is black is false, there is no need to assume it and derive a contradiction. But
for this to make sense one has to distinguish between falsity and intuitionistic negation,
similar to the situation in linear logic.
As far as I remember, the negation and falsity in intuitionism was considered problematic
even by some pioneers (Gilevenko?).
Andrej Bauer Says:
April 20, 2010 at 6:51 am | Reply
I will reiterate, but then I hope we can close discussion because intuitionistic mathematics is not
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 16/17
I will reiterate, but then I hope we can close discussion because intuitionistic mathematics is not
relevant to this post. You give examples of how a classical mathematician proves negations and
that equations such as $2^2 = 5$ do not hold. But all of your examples and methods of proof are
badly chosen because precisely the same methods work for the same reasons intuitionistically. I
write papers in constructive mathematics, and I assure you we do not derive false every time we
want to prove a negation. In fact, good math is written in such a way that it is largely irrelevant
whether it is classical or intuitionistic.
If on the other hand we speak about formal proofs (without cut and in normal form as far as
that is possible for classical math, so we cannot hide things inside lemmas), then proofs of
negated statements will generally end with an introduction rule for negation, both in classical
and intuitionistic mathematics. You speak of a symmetry in classical mathematics between truth
and falsity, but even that symmetry has to be proved somehow, does it not? The rules of
inference for negation (in natural deduction style) are the same classically and intuitionistically.
The symmetry you speak of is proved for classical logic from the law of excluded middle. But the
rules of inference come first, and they do not include the symmetry. (You can build the
symmetry into the logic if you use classical sequent calculus, but thats not how mathematicians
write proofs.)
Bivek Says:
April 30, 2010 at 6:22 am | Reply
Hi there,
Why we use this proving technique?
General Application?
Thank you.
Some Mathematical Gifts Gdels Lost Letter and P=NP Says:
December 20, 2010 at 2:40 pm | Reply
[...] that does not use proof by contradiction. See also Tim Gowers interesting discussion on
this and [...]
Mike Says:
January 26, 2012 at 4:55 pm | Reply
I was just wondering, if you dont get a contradiction when using a proof by contradiction, what
does it mean?
Im asking this question because in the books Ive read so far, usually a counterexample was
used to show that a proposition is false. But can we also use the proof by contradiction and not
get a contradiction to show that the proposition is in fact false?
Thanks.
Anonymous Says:
February 29, 2012 at 4:27 am | Reply
According to his biography packet (available on Wikipedia) the late, great statistician David
Blackwell was able to provide a positive approach to the proof of the irrationality of the square
root of 2. Unfortunately, no details were provided.
Quora Says:
17/5/2014 When is proof by contradiction necessary? | Gowers' s Weblog
http://gowers.wordpress.com/2010/03/28/when-is-proof-by-contradiction-necessary/ 17/17
September 5, 2012 at 3:26 am | Reply
What is the most beautiful theorem proof, and why?
Stephen: To answer What specific advantages are there to viewing this as a constructive proof,
one answer is that it will no longer be a false statement, historically speaking (i.e., it will really be
closer to Euclids proof rather than a lat
How proof by contradiction differs of direct proof | Victor Porton's Math Blog Says:
February 26, 2014 at 4:03 pm | Reply
[] famous mathematician Timoty Gowers asked this question: What is the difference between
direct proofs and proofs by []
Friday, February 28 | Discrete Mathematics Says:
February 28, 2014 at 5:09 am | Reply
[] Check out the post on Gowers blog Is contradiction necessary? []
The Kubrick Theme. Blog at WordPress.com.
Entries (RSS) and Comments (RSS).
Follow
Follow Gowers's Weblog
Powered by WordPress.com

You might also like