Administrative info
HW2 out! Due Tuesday, 4:59pm
PA's not meant to be long, hard
Review
Direct proof, proof by contrapositive, proof by contradiction
Logical chains of deduction
When proving a claim ∀ x . P(x) => Q(x), what do we get to
assume? Do we assume that the whole statement is true? We can't do
that, since then we would be assuming what we are trying to prove!
What we actually do is pick an arbitrary x and assume that P(x)
holds, since if it doesn't, the implication is vacuously true. Then
we write out a chain of deduction:
P(x)
P1(x)
P2(x)
...
Q(x)
Implicitly, what we are saying is that P1(x) follows from P(x),
P2(x) follows from P1(x), and so on, until we get to Q(x). In terms
of implications, what we mean is
P(x)
=> P1(x)
=> P2(x)
=> ...
=> Q(x)
So we have a chain of implications that leads from P(x) to Q(x).
Since we made no assumptions about x, we can plug in any x into our
reasoning and conclude that P(x) => Q(x).
In a proof by contradiction, we start by assuming that what we are
trying to prove is false. Then we derive a contradiction. In
particular, we provide a chain
¬P
=> P1
=> P2
=> ...
=> R
for some proposition R and another chain
¬P
=> Q1
=> Q2
=> ...
=> ¬R
Putting these two together, we get
¬P => (R ∧ ¬R)
What is the value of the RHS? It must always be false. What is
the value of the LHS? Since the RHS is false, the only way the
implication holds is if the LHS is false. Thus, ¬P is false,
so P must be true. This is why proof by contradiction works.
Warning! Proof by contradiction is powerful, but it's dangerous. We
start with an assumption (that is actually false), and we derive many
deductions from it (which are also all false), until eventually we
reach something we can demonstrate contradicts our assumption or is
simply false. The problem is that it's easy to make a simple mistake
of reasoning; if your chain of reasoning doesn't follow, you can
easily reach a falsehood. Contradiction might come from mistake in the
middle rather than a false assumption at the start. This can be hard
to notice, since you're deliberately deriving all sorts of false
statements, and you can't use your intuition to recognize the first
claim that's false. Best to prove as much as possible outside the
proof by contradiction, using lemmas that actually are true.
Another example
Theorem: There are infinitely many primes.
How can we prove this? We don't know of any formula for generating
primes. So in desperation, we resort to proof by contradiction. It's
dangerous, but when you're desperate...
Proof: Assume (for a contradiction) that there are finitely many
primes p1,...,pk. Let a = p1 * p2 * ... * pk + 1. Notice that none
of the p's divide a, since a % p_i = 1. Thus, a has no prime
factor, since none the p's are a factor of a. Thus, a must be
prime, which is a contradiction since it's not one of the p's.
We used the below lemma in our proof. We will see how to prove it
later.
Lemma: Every natural number n>1 is either prime or has a prime
factor.
Q: In our proof, we concluded that a must be prime. Is it true that
the product of the first k primes must be prime?
2 * 3 * 5 * 7 * 11 * 13 + 1 = 30031 = 59 * 509
Proof by cases
Sometimes we're not sure which of a set of possible cases is true,
but we know at least one of them is. If we can prove our claim holds
in any of the cases, then that suffices as a proof of the claim.
Theorem: There exist irrational x and y such that x^y is rational.
Proof: Consider x = y = sqrt(2). Then either sqrt(2)^sqrt(2) is
rational or irrational, though we don't know which is the case.
Case 1: sqrt(2)^sqrt(2) is rational. Then we are done.
Case 2: sqrt(2)^sqrt(2) is irrational.
Consider new values x = sqrt(2)^sqrt(2), y = sqrt(2). Then x^y =
(sqrt(2)^sqrt(2))^sqrt(2) = sqrt(2)^(sqrt(2) sqrt(2)) =
sqrt(2)^2 = 2. Thus, we have shown irrational x, y such that x^y
is rational.
Since one of the above cases must be true, and since we have shown
that in either case there are irrational x,y such that x^y is
rational, we can conclude that such x,y always exist.
=====
Suppose we are given a difficult statement to prove of the form
∀n∈N . P(n). How can we go about solving it?
(1) We could try writing down separate proofs for P(0), P(1), ...,
but we'd never finish.
(2) We could write down one proof for arbitrary n, like we did in
direct proofs. But this might be too hard. How do we prove that
an arbitrary n is a prime or product of primes?
(3) We could try proof by contradiction, but not only is it
dangerous, it may not help.
Recursion
Sometimes, we are faced with a statement that is really hard to
prove using any of the techniques we've seen so far. We need a new
proof technique.
Suppose instead we are writing a program to compute something that
is difficult to do so directly. What do we do? We turn to recursion!
Recursion, at its very core, is to solve a big problem by breaking
it into one or more smaller problems. We solve those smaller
problems, possibly further breaking them down in the process. Then
we put those smaller problems back together in order to solve the
bigger problem.
Here is a real-life example. Suppose I had a heavy box of books that
I wanted to lift up onto a high shelf. Unfortunately, due to too
much time spent typing and not enough time lifting, the box is too
heavy. What do I do? Well, I can take out one book and try again. If
it's still too heavy, I can repeat the process until it's light
enough for mee to lift, but eventually I have the box on the shelf,
minus the book I took out. Then I put that book back into the box to
finish my task.
What elements does the above recursive procedure have? It has a base
case, i.e. the box is light enough for me to lift. It also has a
recursive step, i.e. take a book out, try again with the lighter
box, put the book back in.
Again, the key is to break a hard problem into easier ones. It turns
out, we can follow the same procedure in proofs.
Induction
Now back to that hard to prove statement ∀n∈N . P(n).
Using recursion as an inspiration, how can we solve it?
We need two pieces as in recursion, a base case and a "recursive
step" that breaks a big problem into a smaller problem. The base
case is easy: let's just use the smallest n∈N, i.e. prove P(0).
Then for the "recursive step," we are faced with proving P(k+1). We
break it into a smaller problem of P(k) and then show that given a
proof of P(k), we can turn it into a proof of P(k+1). Logically, we
show P(k) => P(k+1). But we have to be careful about what we are
showing; we can't just show it for a particular k. No, we need to
show that our reasoning holds no matter what k is. So in reality
what we need to show is ∀k∈N . P(k) => P(k+1).
This proof process is called "induction," and the "recursive step"
is actually called the "inductive step." To summarize, we need to
prove two facts:
(1) P(0) [Base case]
(2) ∀k∈N . P(k) => P(k+1) [Inductive step]
Is this any easier? The base case might be easy, but we still have a
universally quantified statement to prove. However, note that the
predicate in that statement is in the form of an implication, so we
know how to prove it: by a direct proof! In particular, we pick an
arbitrary k, assume that P(k) holds, and then show that P(k+1)
follows. The assumption that P(k) holds is called the "inductive
hypothesis."
Let's look at some examples of how to actually carry out this
procedure.
Theorem: ∀n∈N . n^3-n is divisible by 3.
We already saw a direct proof for this, but we can use induction as
well.
Proof:
Base case: P(0)
0^3 - 0 = 0 which is divisible by 3.
Inductive hypothesis: Assume P(k), i.e. k^3-k is divisible by 3.
Inductive step: We must show that P(k+1) follows, i.e.
(k+1)^3-(k+1) is divisible by 3.
[What we need to do here is break (k+1)^3-(k+1) down in terms of
k^3-k (like the procedure we follow in a recursive program) so
we can apply our inductive hypothesis (like calling the
function recursively in a program).]
Then
(k+1)^3 - (k+1)
= k^3 + 3k^3 + 3k + 1 - k - 1
= k^3 + 3k^3 + 2k
= k^3 + 3k^3 + 3k - k
= (k^3 - k) + 3(k^3 + k).
The first term is divisible by 3 by the inductive hypothesis and
the second term is obviously divisible by 3. Thus their sum is
also divisible by 3. QED.
Theorem: ∀n∈N . n >= 1 => 1 + 2 + ... + n = n(n+1)/2.
[Note that we could extend the claim to n = 0 as well, in which
case the LHS would be empty and assumed to be 0 since the
additive identity is 0.]
Proof:
Base case:
[What do we use? We can't use P(0), since if we tried applying
the inductive step to prove P(1), we'd be assuming a false
statement. Instead, we use P(1) as our base case. In general, we
use the smallest element for which the claim we are trying to
prove holds.]
P(1): 1 = 1(1+1)/2 = 2/2 = 1.
Inductive hypothesis: Assume P(k), i.e. 1+...+k = k(k+1)/2.
Inductive step: We need to prove P(k+1), i.e.
1+...+(k+1) = (k+1)(k+2)/2.
[Be careful not to assume that this is true! That's what happens
if we start from this equation and go from there. What we get to
assume is P(k), not P(k+1), and we need to show P(k) => P(k+1),
not the converse!]
Then 1 + ... + (k+1)
= (1 + ... + k) + (k+1)
= k(k+1)/2 + (k+1), by inductive hypothesis
= k(k+1)/2 + 2(k+1)/2
= (k(k+1) + 2(k+1))/2
= (k+1)(k+2)/2.
Theorem: ∀n∈N . n > 1 => (n! <= n^n)
Proof:
Base case: P(1)
1! = 1 <= 1^1 = 1
Inductive hypothesis: Assume P(k), i.e. k! <= k^k.
Inductive step: Prove P(k+1).
(k+1)! = k! (k+1)
<= k^k (k+1), by inductive hypothesis
<= (k+1)^k (k+1)
= (k+1)^(k+1).