Finite Differences

AIM Math Teachers’ Circle Summer Immersion Workshop Ted Alper

July 26, 2012 tmalper@stanford.edu

1 Warming up (no formulas on this side!)

Problem 1 For each the following sequences, try to analyze

• What is “the” next number (or two) in “the” sequence?

• What is the a pattern that characterizes your sequence? (What types of descriptions count
as a pattern?)

• Better still, Find as many patterns as you can describing the sequence.

• For each pattern, can you find other sequences that meet the same pattern? Can you char-
acterize (in some way) the family of sequences?

(A) 3, 7, 11, 15, 19, . . .

(B) 13, 6, −1, −8, −15, −23, . . .

(C) −1, 0, 1, 4, 9, 16, . . .

(D) 0, 4, 11, 21, 34, . . .

(E) 1, 1, 3, 13, 37, 81, . . .

(F) −13, 5, 9, 5, −1, −3, 5, 29, 75, 149, 257, . . .

(G) 1, 2, 4, 8, 16, 32, . . .

(H) 1, 1, 2, 3, 5, 8, 13, 21, 34, . . . . . .

1
2 Finite Differences
Given any sequence of numbers: a1 , a2 , a3 , a4 , . . .
The sequence of differences is given by a2 − a1 , a3 − a2 , a4 − a3 , . . .
It is convenient to write them in the following format

a1 a2 a3 a4 a5 a6 a7 . . .
a2 − a1 a3 − a2 a4 − a3 a5 − a4 a6 − a5 a7 − a6 ...

Example:
−8 −1 0 1 8 27 64 . . .
7 1 1 7 19 37 ...
Of course, you can take the difference of a sequence of differences, and take the difference of
that sequence, and so on.

−8 −1 0 1 8 27 64 . . .
7 1 1 7 19 37 ...
−6 0 6 12 18 ...
6 6 6 6 ...
0 0 0 ...

Problem 2 Go back and do this with our example functions. What happens? We should try to
come up with some hypotheses, and maybe gather some evidence.
Useful Notation: If we represent our sequence as an , we can represent the sequence of differ-
ences using the difference operator, ∆:

∆an = an+1 − an .

And, we can call the difference of the difference of a sequence ∆(∆an ), which can also be written
(with some caution) as ∆2 an

3 Working backward
If I know ∆an , can I reconstruct an ? (At least, the terms of the sequence).

? ? ? ? ? ? ?...
6 2 0 0 2 6 ...

What if I know ∆2 an is the sequence 3n − 2?

What if I know ∆3 an is the constant sequence −12, −12, −12, −12, −12, . . .?

2
4 Working diagonally
What if I know:

1 ? ? ? ? ? ?...
−1 ? ? ? ? ...
2 ? ? ? ...
−3 ? ? ...
0 0 ...
Do I know the entire (top row) sequence? What additional assumption might allow me to
complete the sequence?
Can I find a formula for the sequence?
Or how about:
a ? ? ? ? ? ?...
b ? ? ? ? ? ...
c c c c c c...

5 The general problem and an approach to a solution

if I know values on one diagonal d0 , d1 , d2 , . . . dn (and also that the rows below dn is entirely 0)

d0 ? ? ? ? ? ?...
d1 ? ? ? ? ...
d2 ? ? ? ...
d3 ? ? ...
0 0 ...

Can I determine the sequence on the top row? Can I express it in a formula in terms of
d0 , d1 , . . . , dn ?

3
5.1 Repertoire method
A very useful idea that we should verify for ourselves with examples (and maybe even
prove):
If I write the sequence an as the sum of two sequences bn and cn , then the sequence of differences
of an is the sum of the two sequence of differences for bn and cn . In fact, if an = j · bn + k · cn , then

∆an = j · ∆bn + k · ∆cn

(we could say: the difference operator is linear)

Next, can we solve the general problem some special cases? In each of these cases, assume the
row below the last row given is entirely 0.

(case 0)
1 ? ? ? ? ? ?...

(case 1)
0 ? ? ? ? ? ?...
1 ? ? ? ? ...

(case 2)
0 ? ? ? ? ? ?...
0 ? ? ? ? ? ...
1 ? ? ? ? ...

(case 3)
0 ? ? ? ? ? ?...
0 ? ? ? ? ? ...
0 ? ? ? ? ...
1 ? ? ? ...

(case 4)
0 ? ? ? ? ? ?...
0 ? ? ? ? ? ...
0 ? ? ? ? ...
0 ? ? ? ...
1 ? ? ...

4
5.2 Pascal’s Triangle
You probably already know

1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
1 5 10 10 5 1
1 6 15 20 15 6 1
Can we see any connections to finite differences?

6 Convenient notation
It is helpful (but not universal) to use the notation for falling powers, that is:

xm = x(x − 1) · · · (x − m + 1)

(Rising powers are similarly defined, xm = x(x + 1) · · · (x + m − 1), but we won’t use them here.)
You may also know the expression
 
n n! n(n − 1) · · · (n − m + 1)
= =
m m!(n − m)! m!

in connection with binomial coefficients and Pascal’s triangle, but we can also consider them as
polynomials in their own right:

xm
 
x x(x − 1) · · · (x − m + 1)
= =
m m! m!
x k (xm )? ∆k ( x )?
What is ∆(xm )? What



is ∆( m )? What is ∆ m
x


The polynomial m is 0 for x = 0, 1, . . . , m − 1 and 1 for x = m, (Let’s verify this!) So we can
see how its succession of finite differences will look. This gives a way to resurrect any polynomial
from the first (well, 0th) diagonal difference sequence, solving the general problem above.
This approach also gives a nice proof of the recurrence relation:
   
n n
p(x + n) = p(x + n − 1) − p(x + n − 2) + . . . + (−1)n−1 p(x)
1 2

for any polynomial of degree less than n.

5
7 Problems that naturally lead to finite differences

Problem 3 Any problem where the sequence of solutions satisfies an+1 = an + P (n) where P (n)
is a polynomial.

• an+1 = an + k

• an+1 = an + n

We might need a starting point a0 or a1 .

Problem 4 In particular, many summations
n
X
Sn = ak = a1 + a2 + a3 + · · · + an
k=1

can be evaluated with this approach, since Sn+1 − Sn = . . ..

Problem 5 Can we evaluate:
n
X n
X n
X
1. k, k2 , k3
k=1 k=1 k=1
n
X
2. k · (k + 3)
k=1
n
X n
X
3. k3 , k4
k=1 k=1

n X
X k
4. j 2 (this last one came up in a problem Josh told me yesterday)
k=1 j=1

Problem 6 (Common) Into how many pieces can a pizza be divided by n straight vertical cuts?
(Assume the pizza is essentially 2-dimensional – also convex. And no moving the pieces around
between the cuts.)
Problem 7 Into how many pieces can a cake be cut with n straight cuts (not necessarily vertical!
The point is that the cake has thickness, so now the shape is 3-dimensional and the cuts are not
lines, but planes!)
Problem 8 (More repertoire method √
than finite differences) The polynomial equation x2 −x−1 = 0
1+ 5
has the two solutions φ = 2 = 1.61803399 . . . and Φ = −0.61803399 . . .. The recurrence
relation an+1 = an + an−1 has many solutions, the most famous being the fibonacci sequence
1, 1, 2, 3, 5, 8, 13, 21, 34, . . .. Show that the geometric sequences φ1 , φ2 , φ3 , . . . and Φ1 , Φ2 , Φ3 , . . .
satisfy the same recurrence relation. Verify that, if you can find a and b fo rwhich 1 = aφ1 + bΦ1
and 1 = aφ2 + bΦ2 , then the nth Fibonacci number must be aφn + bΦn .

6
8 (more advanced) Contest Problems

Problem 9 (AIME 1992) For any sequence of real numbers A = (a1 , a2 , a3 , . . .), define ∆A to be
the sequence (a2 − a1, a3 − a2 , a4 − a3 , . . .), whose nth term is an+1 − an . Suppose that all of the
terms of the sequence ∆(∆A) are 1 and that a19 = a92 = 0. Find a1 .
Problem 10 (From the 1995 Polya Team Mathematics Competition) it will be convenient for us
to list the sequences in this round with initial index 0: that is, each sequence listed here should be
considered to be of the form: a0 , a1 , a2 , a3 , . . .

(1) The sequence 1, 1, 7, 13, 55, 133, . . . is an example of a sequence that satisfies the recurrence
relation
an = an−1 + 6an−2 for all n ≥ 2.

(a) Find all geometric sequences a0 , a1 , a2 , . . . that

(i) satisfy the same recurrence relation an = an−1 + 6an−2 for all n ≥ 2.
(ii) have the first term a0 equal to 1.
(b) For the sequence 1, 1, 7, 13, 55, 133, . . . listed above, find a closed form expression for
the 101st term a100 (that is, an P expression involving only simple sums, products, and
exponentials, without the use of notation or indices).
(c) Prove that there is only one sequence of real numbers satisfying this recurrence relation
with both an infinite number of positive terms and an infinite number of negative terms

(2) The sequence 0, 1, 4, 9, 16, 25, . . . , n2 , . . . is an example of a sequence that satisfies the recurrence
relation
an = 3an−1 − 3an−2 + an−3 for all n ≥ 3.

(a) Find all geometric sequences a0 , a1 , a2 , a3 , a4 , . . . that

(i) satisfy the same recurrence relation an = 3an−1 − 3an−2 + an−3 for all n ≥ 3.
(ii) have the first term a0 equal to 1.
(b) For the general sequence a0 , a1 , a2 , a3 , . . . satisfying the recurrence relation, find a closed
form expression for a100 in terms of a0 , a1 , and a2 .
(c) Prove that there are no sequences of real numbers satisfying the recurrence relation with
both an infinite number of positive terms and an infinite number of negative terms

(3) Prove that the sequence given by a0 = 2 and, for n ≥ 1,

√
an = The integer closest to (5 + 2 7)n

satisfies a recurrence relation of the form an = x · an−1 + y · an−2 for n ≥ 2. (For partial
credit, find the values for x and y.)