CS103 Handout 28

Winter 2018 February 9, 2018

Guide to Inductive Proofs
Induction gives a new way to prove results about natural numbers and discrete structures like
games, puzzles, and graphs. All of the standard rules of proofwriting still apply to inductive proofs.
However, there are a few new concerns and caveats that apply to inductive proofs.
This handout details some of the style concerns that often arise in inductive proofs.
The Inductive Proof Template
There are many diferent ways to structure an inductive proof. When you're just getting started
with inductive proofs, we recommend structuring your inductive proofs along the following lines.
As we progress through the quarter and you start getting more comfortable with writing inductive
proofs, we'll start to rely on this structure less and less.
1. Define somee preicatee P(n) ehte you'll prove by cniuaecon. When writing an inductive proof,
you'll be proving that some predicate is true for 0 and that if that predicate holds for k, it
also holds for k + 1. To make explicit what predicate that is, begin your proof by spelling
out what predicate you'll be proving something about by induction. We've typically denoted
this predicate P(n). If you're having trouble with this, don't worry! There's a section later on
in this handout describing what this P(n) thing is all about.
2. Setee ehe ehe proof cs by cniuaecon. After you've written out your predicate P(n) in step (1),
say that you're going to prove, by induction, that it's true for all the numbers you care
about. If you're going to prove P(n) is true for all natural numbers, say that. If you're going
to prove that P(n) is true for all even natural numbers greater than fve, make that clear.
This gives the reader a heads-up about how the induction will proceed.
3. Setee tni prove your btse atse. All inductive proofs require some kind of base case, so it's
probably best to start of by proving it. You've defned what P(n) is in step (1), and now you
need to prove P(0). We recommend explicitly writing out what P(0) actually states before
you try to prove it. This will communicate to your reader what you're going to prove and
helps make it clearer what you need to show. For example, you might write “We will prove
P(0), which states that ...” or “For our base case, we'll prove P(0), namely, that …”
4. Setee tni prove ehe cniuaecve seep. The inductive step in a proof by induction is to show that
for any choice of k, if P(k) is true, then P(k+1) is true. Typically, you'd prove this by as-
suming P(k) and then proving P(k+1). We recommend specifcally writing out both what
the assumption P(k) means and what you're going to prove when you show P(k+1). As with
step (3), this makes it clearer to the reader what they're going to see next and helps you
confrm that you know what you're assuming and what you're proving.
5. Conaluie ehe proof. This usually isn't too much – you can just say something like “complet-
ing the proof by induction” and add the little boxy ■ symbol. You're done!
 2/7

A Guide to P(n)
An important step in starting an inductive proof is choosing some predicate P(n) to prove via mathemat-
ical induction. This step can be one of the more confusing parts of a proof by induction, and in this sec -
tion we'll explore exactly what P(n) is, what it means, and how to choose it.
Formally speaking, induction works in the following way. Given some predicate P(n), an inductive
proof
• proves P(0) is true as a base case;
• proves that if P(k) is true, then P(k+1) must be true as well; and
• concludes that P(n) is true for any natural number n.
At a nuts-and-bolts level, induction is a tool for proving that some predicate P(n) holds for any natural
number n. When writing an inductive proof, you'll want to choose P(n) so that you can prove your over-
all result by showing that P(n) is true for every natural number n.
As an example, suppose that you want to prove this result from Problem Set Two:
For any natural number n, any 2n × 2n grid missing a square can be tiled with right triominoes.
This is a universal statement – for tny natural number n, some predicate holds for that choice of n. To
prove this using mathematical induction, we'd need to pick some predicate P(n) so that if P(n) is true for
every natural number n, the original statement we want to prove is true. One possible choice of P(n)
could be this one:
P(n) is the statement “any 2n × 2n grid missing a square can be tiled with right triominoes.”
Let's look at this statement. By itself, it just says “we're defning some new statement called P(n), where
P(n) is shorthand for “any 2 n × 2n grid missing a square can be tiled with right triominoes.” If you plug
in any value of n, you get back some new statement that might be true and might be false. For example,
P(0) is the statement “any 1 × 1 grid missing a square can be tiled with right triominoes,” which might
be true and might be false. (It turns out that it's true, but that's a separate proof.) The statement P(5) is
that any 32 × 32 grid missing a square can be tiled with right triominoes, and the statement P(10) is that
any 1024 × 1024 grid missing a square can be tiled with right triominoes.
Let's suppose that we do a proof by induction and show that P(n) is true for every possible choice of nat-
ural number n. What would that mean? Well, it would mean that
For any natural number n, the statement P(n) is true.
Expanding out the defnition of P(n) tells us that
For any natural number n, any 2n × 2n grid missing a square can be tiled with right triominoes.
This is the statement that we wanted to prove. Therefore, if we can prove that P(n) is true for any choice
of natural number, we'll have a proof of our overall statement.
As a quick note: the notation P(n) might seem confusing – it looks like it's some function of the number
n. However, P(n) isn't a mathematical quantity. Instead, it's a predicate – a logical assertion saying
something about the number n. Consequently, be careful not to treat P(n) as a number.
 3/7

Directionality in Induction
In the inductive step of a proof, you need to prove this statement:
If P(k) is true, then P(k+1) is true.
Typically, in an inductive proof, you'd start of by assuming that P(k) was true, then would proceed to
show that P(k+1) must also be true.
In practice, it can be easy to inadvertently get this backwards. Here's an cnaorreae proof that the sum of
the frst n powers of two is 2n – 1. (Note that the result that it proves is true, but the proof itself has a
logical error that we'll discuss in a second).

(Incorrect!) Proof: Let P(n) be the statement “the sum of the frst n powers of two is 2n – 1.” We will
prove by induction that P(n) holds for all n ∈ ℕ, from which the theorem follows.
For the base case, we prove P(0), that the sum of the frst zero powers of two is 2 0 – 1. Since the sum of
zero numbers is 0 and 20 – 1 = 0, this result is true.
For the inductive step, assume that P(k) is true for some arbitrary k ∈ ℕ, meaning that
20 + 21 + … + 2k-1 = 2k – 1. (1)
We will prove that P(k+1) is true, meaning that the sum of the frst k+1 powers of two is 2 k+1
– 1. To see
this, note that
20 + 21 + … + 2k-1 + 2k = 2k+1 – 1
20 + 21 + … + 2k-1 + 2k = 2(2k) – 1
20 + 21 + … + 2k-1 + 2k = 2k + 2k – 1
20 + 21 + … + 2k-1 = 2k – 1
We've arrived at statement (1), which we know is true. Therefore, P(k+1) is true, completing the induc-
tion. ■

This proof is, unfortunately, incorrect, but it might not immediately be clear why. The setup of the
proof is fne, as is the proof of the base case. However, things break down when we get to the inductive
step. Take a look at this part:

We will prove that P(k+1) is true, meaning that the sum of the frst k+1 powers of two is
2k+1 – 1. To see this, note that

20 + 21 + … + 2k-1 + 2k = 2k+1 – 1

Here, the proof says that it's going to prove that the sum of the frst k+1 powers of two is 2k+1 – 1, so the
proof should try to establish why this result is true. However, the very frst thing it does is write out an
equality asserting that the sum of the frst k+1 powers of two is 2k+1 – 1. This should give you some
pause: if the proof says it's going to prove something, why is it stating it as if it were a mathematical
fact?
If you keep reading through the proof, you'll see that the proof works by manipulating this equality and
ultimately arriving at the fact that 20 + 21 + … + 2k-1 = 2k – 1, the inductive hypothesis. The proof then
claims that this statement is by assumption true, so the proof is complete.
Take a step back from this proof and think about what it actually ends up doing. It begins by tssumecng
that the sum of the frst k+1 powers of two is 2k+1 – 1, then proceeds from there to prove that under this
 4/7

assumption, the sum of the frst k powers of two is 2k – 1. In other words, the proof has gone backwards
– it's assumed P(k+1) is true, and used that to prove that P(k) is true. Consequently, even though the
mathematical reasoning that's written is correct, the proof is establishing the wrong result and is there-
fore incorrect.
When working with an inductive proof, make sure that you don't accidentally end up assuming what
you're trying to prove.
Choosing and Proving Base Cases
Inductive proofs need base cases, and choosing the right base case can be a bit tricky. For example,
think back to our initial inductive proof: that the sum of the frst n powers of two is 2n – 1. In that proof,
we chose as our base case n=0. This might seem weird, since that means that we're reasoning about a
sum of zero numbers. Similarly, in the counterfeit coin example, we chose our base case to be the setup
where we have no weighings on the scale and need to fnd the counterfeit coin. Wouldn't it have made
more sense to talk about cases where we have one weighing, not zero?
There are four main reasons we chose to use these base cases. First, we chose our base cases to make
sure that the theorems we proved were true for every natural number n, not just some of them. For ex-
ample, in the theorem about sums of powers of two, our goal was to prove the theorem true for every
natural number, including zero. Similarly, in the counterfeit coin problem, we wanted to show that with
any number of weighings n, including zero, you can fnd the counterfeit out of 3 n coins with n weigh-
ings. Had we chosen our base case for these proofs to be one instead of zero, then our proofs would
have omitted a case and would not satisfactorily proven the overall result.
Second, we chose these extremely simple base cases because they're as simplifed as they can possibly
be. At some level, each inductive proof works by somehow using the assumption that the theorem is true
in a simpler case (say, that it's true for k) to build up a proof that the theorem is true for a more complex
case (for example, that it's true for k+1). The reason the inductive proof ultimately works is that we
know that it's true for the absolute simplest case, the base case. Philosophically, it's the job of the induc-
tive step to try to simplify, and it's the job of the base case to show that when no more simplifcation is
possible, the result must be true. Therefore, we strongly recommend making your base case something
that is so simple that there's no possible way to simplify any further. After all, you can't make fewer
than 0 weighings, or sum up fewer than 0 numbers.
Third, choosing the absolute simplest possible base case forces you to think about the extreme cases of
the result. There are some times where picking 0 as a base case doesn't actually work because the result
isn't actually true in that case. For example, it is noe true that a square can be subdivided into 0 squares.
By trying to search for the simplest possible base case in an inductive proof, you can sometimes dis-
cover boundary cases that might not have been evident from an intuitive understanding of the theorem.
Conversely, you often discover cases where the result actually ioes work that's not intuitively evident!
Finally, in some cases, choosing the absolutely simplest possible base case can actually make your proof
shorter and simpler. For example, suppose that we changed the base case in the proof of the counterfeit
coin problem to be the case where there are three coins and one weighing. In that case, the base case
would have to explain how to split the coins into three piles, how to weigh those piles against one an-
other, and how to determine which coin from the group was counterfeit. This would lengthen the proof
without adding much, since that exact same line of reasoning appears in the proof of the inductive step.
In fact, that leads to a good general piece of advice: if you fnd yourself using similar reasoning in your
base case and your inductive step, chances are you can make your base case even simpler!
Induction or Complete Induction?
One of the trickier skills when frst learning induction is learning when to use standard induction and
when to use complete induction. This typically depends on the route that your proof takes.
 5/7

At some point in your proof, you'll need to use the assumption that the theorem is true for some smaller
number to establish that the theorem is true for a larger number. The diference between induction and
complete induction is how much smaller “some smaller number” is. Suppose that as part of your induc-
tive step, you're trying to prove that the result is true for k+1. If you can always prove this purely by as-
suming that the result is true for k, then you can just use normal induction. On the other hand, if you
need to assume that the result is true for some number between 0 and k, inclusive, but you're not sure in
advance whcah of those numbers it's going to be, you should use complete induction to make sure you
have those cases covered.
It might be easier to see when this second case might arise by looking at our lecture example of com-
plete induction, where we proved that any tree with n nodes has n-1 edges. In the inductive step, we be-
gan with a tree of k+1 nodes, then deleted an edge from the tree to split it into two smaller trees, one
with r nodes and one with (k+1)-r nodes. Without knowing more about the specifc tree or cut we're
working with, we can't say for certain how big those smaller trees are. Maybe they each have exactly
half of the nodes, or maybe one has one node and the other has k. There's no general way to pin down
how many nodes will be in each tree, but in each case it's guaranteed that there will be between 1 and k.
By using complete induction, we can ensure that no matter how many nodes are in those trees, we can
reason about how many edges are in them.
When deciding to select between induction and complete induction in your proof, try using this general
heuristic:
• If you can determine in advance the size of the smaller problem, use standard induction.
• If you can't determine the size in advance, use complete induction.
An interesting detail here is that complete induction in a sense “supersedes” regular induction. In both
regular and complete induction, you assume that P(k) is true, but only in complete induction do you get
the rest of the assumptions along with it. This gives rise to a reasonable question: why would you ever
use standard induction?
Technically speaking, you never neei to use standard induction. You can always use complete induction
whenever you'd use standard induction. From a stylistic perspective, though, it's not considered good
form to use complete induction unless you actually need those extra assumptions. The type of induction
you choose to use should mirror the structure of your argument. If you always prove that the result is
true for k+1 based purely on the assumption that the result is true for k, there's no reason to clutter the
proof by assuming that the result is also true for 0, 1, 2, 3, etc. Think of it this way: there's all sorts of
true statements that you might want to mention in a proof, but you tend to only include ones that are rel-
evant. If P(0), P(1), …, and P(k) are all relevant assumptions, include them. If you just need P(k), leave
the other ones out.
A Word of Caution
As a closing remark to this discussion, I thought I'd point out one particular issue that comes up in in-
duction that frequently trips people up.
As an example, let's look at the circle game problem problem Problem Set Five. In this problem, you're
asked to prove, by induction, that any circle with n points labeled +1 and n points labeled -1 has a win-
ning point. There are many ways that you can set this problem up inductively. One technique would be
to choose P(n) as the statement “any circle with n points labeled +1 and n points labeled -1 has a win-
ning point .” If you're going about proving this result by induction, at some point you'll need to prove the
inductive step, which says the following:
If P(k) is true, then P(k+1) is true.
Plugging in P(n) tells us that the statement we wish to prove is the following:
 6/7

If any circle with k points labeled +1 and k points labeled -1 has a winning point,
then any circle with k+1 points labeled +1 and k+1 points labeled -1 has a winning point.
One of the most common mistakes we see people make is to try to prove this statement using reasoning
along the following lines:

For the inductive step, assume that for some natural number k that any circle with k points labeled +1
and k points labeled -1 has a winning point. We will prove that any circle with k+1 points labeled +1 and
k+1 points labeled -1 has a winning point. To do so, begin by taking any circle C with k points labeled
+1 and k points labeled -1. Add in a +1 point and a -1 point anywhere in C to form a new circle C’ with
k+1 points labeled +1 and k+1 points labeled -1. […] ⚠

The above setup may look correct, but it's dangerously incorrect.
In a proof by induction, the idea is to use the fact that a result is true for some number k to prove that
it's true for some number k+1. Therefore, it can be tempting to prove a result like the one above by say-
ing “let's start with a circle with k points of each type, then add in some more points to get k+1 points of
each type.” After all, we know something about a circle with k points marked +1 and k points labeled -1
and want to show something about circle with k+1 points of each type, so it seems only reasonable that
we'd start with a circle with k points of each type and then add in one point of each type.
However, think back to the logical structure of what we're trying to prove. We're trying to show that
If
any circle with k points labeled +1 and k points labeled -1 has a winning point,
then
any circle with k+1 points labeled +1 and k+1 points labeled -1 has a winning point.
Ignoring induction for a moment, how would you go about proving this statement? Well, it's an implica-
tion, so we should start of by assuming the antecedent is true:

Assume that any circle with k points labeled +1 and k points labeled -1 has a winning point.

Next, we're going to try to prove the consequent. The consequent here is the statement “any circle with
k+1 points labeled +1 and k+1 points labeled -1 has a winning point.” How do you prove a statement like
this? Well, it's a universal statement, so we should start of by choosing an arbitrary circle with k+1
points labeled +1 and k+1 points labeled -1, then try to show that it has a winning point. In other words,
the proof really should look like this:

For the inductive step, assume that for some natural number k that any circle with k points labeled +1
and k points labeled -1 has a winning point. We will prove that any circle with k+1 points labeled +1 and
k+1 points labeled -1 has a winning point. To do so, consider an arbitrary circle C with k+1 points la-
beled +1 and k+1 points labeled -1. [...]

This can seem backwards the frst time you see it – why are we starting of with a circle with k+1 points
of each type when we only know something about a circle with k points of each type? However, if you
carefully unpack the statement we're trying to prove, you'll see that all we're doing here is proceeding
the way that we normally would if we were trying to prove a universal statement.
 7/7

At this point in the proof, we have a circle with k+1 players of each type – which we know basically
nothing about. If we could reduce it down to a circle with k points of each type, we could start to make
some claims about that smaller circle which might help us in the larger case. In other words, you may
want to think about fnding some way to extract from the larger circle a smaller circle with k points of
each type, then use the inductive hypothesis to conclude something about that smaller circle. If you're
strategic with how you reduce the size, you can use your knowledge about how you found that smaller
circle to arrive at the solution.
Notice how we got here. Rather than beginning with a circle with k points of each type and adding
points into it, we started with a circle with k+1 points of each type and removed points from it from it.
The reason we did so is that the overall structure of the inductive step is itself a giant implication, and if
we follow the normal rules for proving implications we'd end up concluding that this is the right struc -
ture to take.
To summarize – before you write out a proof by induction, write out explicitly what the inductive step
will be. It's going to be a big implication, meaning that you'll assume the antecedent and prove the con -
sequent. Make sure that you take the same steps when assuming the antecedent and trying to prove the
consequent that you would when proving any general implication. That might mean that you need to
start with an object of size k+1 and then massage it down to an object of size k, or it might mean that
you need to start with an object of size k and grow it up into an object of size k+1. You'll know which
case it is based on the structure of the implication you're trying to prove. When it doubt, write it out!