Expanding to find patterns and make hypotheses

Last class we learned what a recurrence relation is, how it defines a sequence, how we can immeditely translate a recurrence relation to a recursive function in a program, what a closed form is, and how we can use induction to *prove* that some proposed closed form really is correct for the function defined by a given recurrence relation. What we didn't cover is how to derive good guesses (it sounds better to say "hypotheses") for closed forms to a given recurrence relation. Now we will address that!

Consider the function $a(\cdot)$ defined by the following recurrence relation: $$ \begin{array}{l} a(n) = a(n-1) + n + 1\\ a(0) = 2 \end{array} $$ What we went through in class (and I summarize here) is expanding the recurrence relation, looking for a pattern. $$ \begin{array}{rcl} a(n) &=& \underbrace{{\color{red}{a(n-1)}}}_{a(n-2) + (n-1) + 1} + n + 1\\ &=& {\color{red}{a(n-2) + (n-1) + 1}} + n + 1\\ &=& a(n-2) + n + (n-1) + 1 + 1\\ &=& \underbrace{{\color{red}{a(n-2)}}}_{a(n-3) + (n-2) + 1} + n + (n-1) + 1 + 1\\ &=& {\color{red}{a(n-3) + (n-2) + 1}} + n + (n-1) + 1 + 1\\ &=& a(n-3) + n + (n-1) + (n-2) + 1 + 1 + 1\\ &=& a(n-{\color{red}{3}}) + \underbrace{n + (n-1) + (n-2)}_{{\color{red}{\text{3 terms}}}} + \underbrace{1 + 1 + 1}_{{\color{red}{\text{3 terms}}}}\\ &&\vdots\\ &=& a(n-{\color{red}{k}}) + \underbrace{n + (n-1) + \cdots + (n-(k-1))}_{{\color{red}{\text{k terms}}}} + \underbrace{1 + 1 + \cdots + 1}_{{\color{red}{\text{k terms}}}}\\ &&\vdots\\ &=& \underbrace{a(n-{\color{red}{n}})}_{a(0) = 2} + \underbrace{n + (n-1) + \cdots + 2 + 1}_{{\color{red}{\text{n terms}}}} + \underbrace{1 + 1 + \cdots + 1}_{{\color{red}{\text{n terms}}}}\\ &=& 2\ +\ 1+2+\cdots+n\ + \ n \end{array} $$ We found a pattern and we imagined continuing the expanding process until the end - i.e. we get to $f(0)$. And now we have a hypothesis:

hypothesis: $a(n) = 1+2+\cdots+n\ + \ n + 2$

We can calculate a few terms in the series based on the recurrence, and check the formula and it seems to work. But, of course, we need to prove that it, in fact, always works.

For function $a(\cdot)$ defined by the recurrence relation above, $a(n) = 1+2+\cdots+n\ + \ n + 2$.
we will prove this by induction on $n$.
base case $n=0$: According to the recurrence relation, $a(0) = 2$. At zero, the closed form expression is $1+2+\cdots+n\ + \ n + 2 = 0 + 0 + 2 = 2. ✓
inductive case $n > 0$: $$ \begin{array}{rcl} a(n) &=& a(n-1) + n + 1 \text{, by the inductive hypothesis } a(n-1) = 1+2+\cdots + (n-1) + (n-1) + 2\\ &=& 1 + 2 + \cdots + (n-1) + (n-1) + 2 + n + 1\\ &=& 1 + 2 + \cdots + n + n + 2 \end{array} $$ So $a(n) = 1+2+\cdots+n\ + \ n + 2$ as required. ✓

ACTIVITY

Your job in class was to expand the following two recurrence relations, find the pattern and make a hypothesis. Then, of course, give an inductive proof that your hypotheses were correct.
  1. $f(n) = f(n-1) + 3,\ f(0) = 5$
  2. $g(n) = 2\cdot g(n-1),\ g(0) = 3$

Gauss's proof

In the first example, we showed that $a(n) = 1+2+\cdots+n\ + \ n + 2$. But this isn't really a closed form, because of the "$1+2+\cdots+n$". So let us see how to get a closed form for this sum. There is a beautiful derivation and proof that is said to have been discovered by the famous mathematician Carl Fredrich Gauss as a child. Let us call the sum $S_n$. In other words $S_n = 1 + 2 + \cdots + n$. $$ \begin{array}{rccccccccc} S_n = &1 &+& 2 &+ &\cdots &+ &(n-1) &+& n\\ S_n = &n &+& (n-1) &+& \cdots &+& 2 &+& 1\\ \hline 2\cdot S_n = & n(n+1)\\ %S_n =& \frac{n(n+1)}{2} \end{array} $$ ... and so $S_n =\frac{n(n+1)}{2}$.

It's beautiful, isn't it?

One more beautiful proof

Here's another difficult sum to consider: $$ P_n = 1 + r + r^2 + \cdots + r^n $$ Here's another beautiful derivation/proof of a closed form. Consider $(r-1)\cdot P_n$. $$ (r-1)\cdot P_n = \begin{array}{} && r &+& r^2 &+& \cdots &+& r^n &+& r^{n+1}\\ -1 &-& r &-& r^2 &-& \cdots &-& r^n &&\\ \hline -1 & & & & & & & & & +& r^{n+1} \end{array} $$ ... so we get $(r-1)P_n = r^{n+1}-1$. Dividing both sides by $r-1$ gives $$ P_n = \frac{r^{n+1}-1}{r-1}. $$ Once again, what a beautiful proof of what is, I think, a quite unexpected formula!

Christopher W Brown