108
Chapter 5: Mathematical Induction
So far in this course, we have seen some techniques for dealing with stochastic
processes: first-step analysis for hitting probabilities (Chapter 2), and first-step
analysis for expected reaching times (Chapter 3). We now look at another tool
that is often useful for exploring properties of stochastic processes: proof by
mathematical induction.
5.1 Proving things in mathematics
There are many different ways of constructing a formal proof in mathematics.
Some examples are:
• Proof by counterexample: a proposition is proved to be not generally true
because a particular example is found for which it is not true.
• Proof by contradiction: this can be used either to prove a proposition is
true or to prove that it is false. To prove that the proposition is true (say),
we start by assuming that it is false. We then explore the consequences of
this assumption until we reach a contradiction, e.g. 0 = 1. Therefore something
must have gone wrong, and the only thing we weren’t sure about was our initial
assumption that the proposition is false — so our initial assumption must be
wrong and the proposition is proved true.
A famous proof of this sort is the proof that there are infinitely many prime
numbers. We start by assuming that there are finitely many primes, so they
can be listed as p1, p2, . . . , pn, where pn is the largest prime number. But then
the number p1 × p2 × . . . × pn + 1 must also be prime, because it is not divisible
by any of the smaller primes. Furthermore this number is definitely bigger than
pn . So we have contradicted the idea that there was a ‘biggest’ prime called pn ,
and therefore there are infinitely many primes.
• Proof by mathematical induction: in mathematical induction, we start
with a formula that we suspect is true. For example, I might suspect from
109
observation that nk=1 k = n(n + 1)/2. I might have tested this formula for
P
many different values of n, but of course I can never test it for all values of n.
Therefore I need to prove that the formula is always true.
The idea of mathematical induction is to say: suppose the formula is true for
all n up to the value n = 10 (say). Can I prove that, if it is true for n = 10,
then it will also be true for n = 11? And if it is true for n = 11, then it will
also be true for n = 12? And so on.
In practice, we usually start lower than n = 10. We usually take the very easiest
case, n = 1, and prove that the formula is true for n = 1: LHS = 1k=1 k =
P
1 = 1 × 2/2 = RHS. Then we prove that, if the formula is ever true for n = x,
then it will always be true for n = x + 1. Because it is true for n = 1, it must
be true for n = 2; and because it is true for n = 2, it must be true for n = 3;
and so on, for all possible n. Thus the formula is proved.
Mathematical induction is therefore a bit like a first-step analysis for prov-
ing things: prove that wherever we are now, the next step will al-
ways be OK. Then if we were OK at the very beginning, we will be
OK for ever.
The method of mathematical induction for proving results is very important in
the study of Stochastic Processes. This is because a stochastic process builds
up one step at a time, and mathematical induction works on the same principle.
Example: We have already seen examples of inductive-type reasoning in this
course. For example, in Chapter 2 for the Gambler’s Ruin problem, using
the method of repeated substitution to solve for px = P(Ruin | start with $x),
we discovered that:
• p2 = 2p1 − 1
• p3 = 3p1 − 2
• p4 = 4p1 − 3
We deduced that px = xp1 − (x − 1) in general.
To prove this properly, we should have used the method of mathematical
induction.
110
5.2 Mathematical Induction by example
This example explains the style and steps needed for a proof by induction.
n
X n(n + 1)
Question: Prove by induction that k= for any integer n. (⋆)
2
k=1
Approach: follow the steps below.
(i) First verify that the formula is true for a base case: usually the smallest appro-
priate value of n (e.g. n = 0 or n = 1). Here, the smallest possible value of n is
n = 1, because we can’t have 0k=1.
P
First verify (⋆) is true when n = 1.
1
X
LHS = k = 1.
k=1
1×2
RHS = = 1 = LHS.
2
So (⋆) is proved for n = 1.
(ii) Next suppose that formula (⋆) is true for all values of n up to and including
some value x. (We have already established that this is the case for x = 1).
Using the hypothesis that (⋆) is true for all values of n up to and including x,
prove that it is therefore true for the value n = x + 1.
Now suppose that (⋆) is true for n = 1, 2, . . . , x for some x.
x
X x(x + 1)
Thus we can assume that k= . (a)
2
k=1
((a) for ‘allowed’ info)
We need to show that if (⋆) holds for n = x, then it must also hold
for n = x + 1.
111
Require to prove that
x+1
X (x + 1)(x + 2)
k= (⋆⋆)
2
k=1
(Obtained by putting n = x + 1 in (⋆)).
x+1
X x
X
LHS of (⋆⋆) = k = k + (x + 1) by expanding the sum
k=1 k=1
x(x + 1)
= + (x + 1) using allowed info (a)
2
x
= (x + 1) +1 rearranging
2
(x + 1)(x + 2)
=
2
= RHS of (⋆⋆).
This shows that:
n
X n(n + 1)
k = when n = x + 1.
2
k=1
So, assuming (⋆) is true for n = x, it is also true for n = x + 1.
(iii) Refer back to the base case: if it is true for n = 1, then it is true for n = 1+1 = 2
by (ii). If it is true for n = 2, it is true for n = 2 + 1 = 3 by (ii). We could go
on forever. This proves that the formula (⋆) is true for all n.
We proved (⋆) true for n = 1, thus (⋆) is true for all integers n ≥ 1.
112
General procedure for proof by induction
The procedure above is quite standard. The inductive proof can be summarized
like this:
Question: prove that f (n) = g(n) for all integers n ≥ 1. (⋆)
Base case: n = 1. Prove that f (1) = g(1) using
LHS = f (1)
.
= ..
= g(1) = RHS.
General case: suppose (⋆) is true for n = x:
so f (x) = g(x). (a) (allowed info)
Prove that (⋆) is therefore true for n = x + 1:
RTP f (x + 1) = g(x + 1). (⋆⋆)
LHS(⋆⋆) = f (x + 1)
some expression breaking down f (x + 1)
=
into f (x) and an extra term in x + 1
= substitute f (x) = g(x) in the line above by allowed (a)
= {do some working}
= g(x + 1)
= RHS(⋆⋆).
Conclude: (⋆) is proved for n = 1, so it is proved for n = 2, n = 3,
n = 4, . . .
(⋆) is therefore proved for all integers n ≥ 1.
113
5.3 Some harder examples of mathematical induction
Induction problems in stochastic processes are often trickier than usual. Here
are some possibilities:
• Backwards induction: start with base case n = N and go backwards,
instead of starting at base case n = 1 and going forwards.
• Two-step induction, where the proof for n = x + 1 relies not only on the
formula being true for n = x, but also on it being true for n = x − 1.
The first example below is hard probably because it is too easy. The second
example is an example of a two-step induction.
Example 1: Suppose that p0 = 1 and px = αpx+1 for all x = 1, 2, . . .. Prove by
mathematical induction that pn = 1/αn for n = 0, 1, 2, . . ..
Wish to prove
1
pn = for n = 0, 1, 2, . . . (⋆)
αn
Information given:
1
px+1 = px (G1 )
α
p0 = 1 (G2)
Base case: n = 0.
LHS = p0 = 1 by information given (G2 ).
1 1
RHS = 0 = = 1 = LHS.
α 1
Therefore (⋆) is true for the base case n = 0.
General case: suppose that (⋆) is true for n = x, so we can assume
1
px = x . (a)
α
Wish to prove that (⋆) is also true for n = x + 1: i.e.
1
RTP px+1 = x+1 . (⋆⋆)
α
114
1
LHS of (⋆⋆) = px+1 = × px by given (G1)
α
1 1
= × x by allowed (a)
α α
1
=
αx+1
= RHS of (⋆⋆).
So if formula (⋆) is true for n = x, it is true for n = x + 1. We have
shown it is true for n = 0, so it is true for all n = 0, 1, 2, . . ..
Example 2: Gambler’s Ruin. In the Gambler’s Ruin problem in Section 2.7,
we have the following situation:
• px = P(Ruin | start with $x);
• We know from first-step analysis that px+1 = 2px − px−1 (G1)
• We know from common sense that p0 = 1 (G2 )
• By direct substitution into (G1 ), we obtain:
p2 = 2p1 − 1
p3 = 3p1 − 2
• We develop a suspicion that for all x = 1, 2, 3, . . . ,
px = xp1 − (x − 1) (⋆)
• We wish to prove (⋆) by mathematical induction.
For this example, our given information, in (G1), expresses px+1 in
terms of both px and px−1 , so we need two base cases. Use x = 1 and
x = 2.
115
Wish to prove px = xp1 − (x − 1) (⋆).
Base case x = 1:
LHS = p1 .
RHS = 1 × p1 − 0 = p1 = LHS.
∴ formula (⋆) is true for base case x = 1.
Base case x = 2:
LHS = p2 = 2p1 − 1 by information given (G1 )
RHS = 2 × p1 − 1 = LHS.
∴ formula (⋆) is true for base case x = 2.
General case: suppose that (⋆) is true for all x up to x = k.
So we are allowed:
(x = k) pk = kp1 − (k − 1) (a1 )
(x = k − 1) pk−1 = (k − 1)p1 − (k − 2) (a2 )
Wish to prove that (⋆) is also true for x = k + 1, i.e.
RTP pk+1 = (k + 1)p1 − k. (⋆⋆)
LHS of (⋆⋆) = pk+1
= 2pk − pk−1 by given information (G1 )
n o n o
= 2 kp1 − (k − 1) − (k − 1)p1 − (k − 2)
by allowed (a1) and (a2 )
n o n o
= p1 2k − (k − 1) − 2(k − 1) − (k − 2)
= (k + 1)p1 − k
= RHS of (⋆⋆)
So if formula (⋆) is true for x = k − 1 and x = k, it is true for
x = k + 1. We have shown it is true for x = 1 and x = 2, so it is true
for all x = 1, 2, 3, . . ..
