Divided Differences

Kada Williams

June 15, 2017

This article, aimed at beginners, investigates a theory of algebraic expres-

sions, ultimately explaining the celebrated Combinatorial Nullstellensatz.
While this topic is very loosely related to Olympiads, I hope it will be a lot
of fun. 1

Contents

1 Foreword 2

2 Difference sequences 2
2.1 Let’s Take Differences! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Formal Series and Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 More on Finite Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Some problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3 Divided Differences 8
3.1 Introduction and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Application: Roots of Multivariate Polynomials . . . . . . . . . . . . . . . 10

4 More Problems 12
4.1 Difference Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Divided Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1
For the fancy Latex, I used Evan Chen’s evan.sty, from his website http://web.evanchen.cc.

1
Kada Williams (June 15, 2017) Divided Differences

§1 Foreword
My not-so-secret goal with writing this article is to give you an experience

ˆ that math which seems very advanced is at times actually very easy, and

ˆ how us humans can build up complex and deep ideas if we keep looking at what
we know from a distance.

In other words, we can think about complicated things if we can understand them in
simple terms:

1. we can rephrase what we want to prove more compactly,

2. we can visualize objects (that’s what our brains evolved for!),

3. we can use notation/language to think about complex processes in simple terms

(e.g. ”if I do X, then Y happens except if V ; if I also can do U , then V doesn’t
happen, so let’s do X with U ” is how 20-page proofs are created).

Therefore, even if Divided Differences had absolutely nothing to do with anything you
want to get done, it may be worth reading this article to learn about how ideas can build
onto each other.

§2 Difference sequences
§2.1 Let’s Take Differences!
Let’s start with a sequence a0 , a1 , a2 , . . . , and compose the differences of the neighboring
terms, generating the sequence bn = an+1 − an . For example,

0 1 4 9 16 25 36 ...
1 3 5 7 9 11 ...

Starting from a sequence (an )n≥0 , we can generate sequences (bn ), (cn ), . . . with bn =
an+1 − an , cn = bn+1 − bn , . . . , arranging them like this:

a0 a1 a2 a3 a4 ...
b0 b1 b2 b3 ...
c0 c1 c2 ...
... ... ...

Example 2.1
Calculate the first few terms of the difference sequences when an = n, an = n2 ,
an = n3 , and an = n4 .
Here’s the array for an = n3 :

0 1 8 27 64 125 ...
1 7 19 37 61 ...
6 12 18 24 ...
6 6 6 ...
0 0 0 ...

2
Kada Williams (June 15, 2017) Divided Differences

Problem 2.2. Let k be a positive integer. What do you think will happen when an = nk ?

Here’s another question to think about:

Example 2.3
Here’s the array for an = n3 + 5n:

0 6 18 42 84 150 ...
6 12 24 42 66 ...
6 12 18 24 ...
6 6 6 ...
0 0 0 ...

Do you notice anything?

Try to come up with conjectures (observations) and try to imagine possibly vague
reasons for why your conjectures should be true (intuition). We will later find memorable
ways to make sense of that intuition, and concoct proofs out of them.

3
Kada Williams (June 15, 2017) Divided Differences

Observations:
ˆ Eventually, we will end up with a row full of 0’s, and before that, there is a constant
sequence.

ˆ Let an = nk . This constant is actually k!.

ˆ If the first row is Row 0, the second is Row 1, and so on, then Row k will be
k!, k!, . . . and Row k + 1 will be 0, 0, . . . .

ˆ We can get the array for an = n3 + 5n by just adding [the array for an = n3 ] and
[5× the array for an = n].
Intuitively, the experience is that taking differences first creates some chaos, but then
taking differences a few more times ”eats up” the confusion. This makes sense, because
in general, when we take differences, often some terms vanish, for example

(n + 1)3 − n3 = (n3 + 3n2 + 3n + 1) − n3 = 3n2 + 3n + 1.

Observe how when we got bn = 3n2 + 3n + 1 from an = n3 , the formula for the terms in
bn ended up having so-called degree 2, because the n3 terms cancelled.

§2.2 Formal Series and Polynomials

What did we mean by degree before? We said that n3 has degree 3, but 3n2 + 3n + 1 has
degree 2.

The concept degree doesn’t have to do with the actual value of our expressions, rather
with the expressions themselves. This is just like how in grammar, there’s a difference
between the meaning of words (what we think of when reading a word) and the words
themselves (the set of characters we write down).

This section serves to solidify the reader’s understanding of polynomial expressions

while clarifying this distinction.

Consider a so-called polynomial expression f (n), such as f (n) = n3 or f (n) = 3n2 +

3n + 1. The substitution value of the polynomial is what value it takes at a certain
value at n, such as f (5) or f (42). When we rather talk about the symbols themselves,
i.e. the expression ”3n2 + 3n + 1”, we talk about polynomials formally.
Consider some real numbers a0 , a1 , . . . , and look at the formal sum

f (x) = a0 + a1 x + a2 x2 + · · · + ak xk + . . . .

If for some k, ak+1 = ak+2 = · · · = 0, then we call this expression a polynomial. In

general, this sum could be infinite as well – then, we call it a formal series. This way,
each formal series f (x) corresponds to a sequence (a0 , a1 , . . . ) and vice versa.
We can now define addition, subtraction and multiplication on formal sums2 .
Let f (x) = a0 + a1 x + . . . and g(x) = b0 + b1 x + . . . . Then f (x) ± g(x) is the formal
series corresponding to the sequence

(a0 ± b0 , a1 ± b1 , . . . , ak ± bk , . . . ).

2
In abstract algebra, a set R(+, ·) with ’units’ like 0, 1, addition, additive inverses (x 7→ −x), and
multiplication which does a · (b + c) = a · b + a · c is called a commutative ring.

4
Kada Williams (June 15, 2017) Divided Differences

Example 2.4
Let f (x) = x2 + x + 1 and g(x) = x + 3. Then

f (x) − 3g(x) = x2 − 2x − 8.

What is f (x) · g(x)? No, it isn’t (a0 b0 ) + (a1 b1 )x + . . . , for example. We want to
define multiplication in a way that if we substitute any value for x and evaluate the sums
(pretend that we can do infinite addition), the multiplication still remains true3 . This
way, f (x) · g(x) becomes the formal series corresponding to the sequence
 
a0 b0 , (a0 b1 + a1 b0 ), (a0 b2 + a1 b1 + a2 b0 ), . . . , (a0 bk + a1 bk−1 + · · · + ak b0 ), . . . .

Example 2.5
We have

(4 − x + x2 + 0x3 + 0x4 + . . . )(3 + 2x + x2 + 0x3 + 0x4 + . . . ) =

= 12 + 8x+4x2
−3x−2x2 − x3
+3x2 + 2x3 + x4
= 12 + 5x+5x2 + x3 + x4 .

Recall that polynomials are the formal series

P (x) = a0 + a1 x + · · · + ak xk + . . .

where the coefficients a0 , a1 , . . . are eventually all 0. In particular, if ad 6= 0 but

ad+1 = ad+2 = · · · = 0, we say that d is the degree of P , and that ad is the leading
coefficient. On and in the exceptional case a0 = a1 = · · · = 0, we say P is the zero
polynomial, and define its degree as ”−∞”.
We can perform + and · on polynomials, too. The sum and product of polynomials
remains a polynomial.

Example 2.6
What does addition and multiplication do to a polynomial’s degree?
Also, note that we could have defined polynomials without talking about formal
series had I felt like it.

We can also talk about powers of polynomials: f (x)2 = f (x)·f (x), f (x)3 = f (x)2 ·f (x),
and so on. The polynomial f (x+1) comes from f (x) = · · ·+ak xk +. . . by fully expanding
all the terms ak (x + 1)k using the Binomial Theorem:

3
Higher math would say that the evaluation map from the ring of formal power series to the reals defined
by f (x) 7→ f (r) should be a (ring) homomorphism. (Well, OK, that’s just wrong – e.g. 1 + 1 + 1 + . . .
does not exist –, but Calculus does prove that there is a subring of formal series where f (r) exists.)
Think about why it could be worth talking about math like this.

5
Kada Williams (June 15, 2017) Divided Differences

Theorem 2.7
Consider the Pascal triangle. Note how its n-th row can be encoded by the polynomial
       
n n 1 n 2 n n
+ x + x + ··· + x .
0 1 2 n

Since the Pascal triangle is an addition table, i.e. n+1 n n

k+1 = k + k+1 , when we
formally multiply this polynomial by (1 + x), we get the polynomial encoding the
(n + 1)-th row. This means that the polynomial encoding row 1 is 1 + x, the one
encoding row 2 is (1 + x)2 , and so on:
       
n n n 1 n 2 n n
(1 + x) = + x + x + ··· + x .
0 1 2 n

We call this idea the Binomial Theorem.

P
One more thing: from now on, I’ll use (sum) Pnotation. So for example, a polynomial
or formal power series could be called f (x) = ak xk . Additionally, if I wanted to
n
X
specify that I only wish to add terms with k = 0, 1, . . . , n, then I’d write f (x) = ak xk .
k=0
Similarly, I can do things like
n
X
(−1)j x2j = 1 − x2 + x4 − x6 + · · · + (−1)n x2n .
j=0

Now let’s return to our observations about finite differences and prove them!

§2.3 More on Finite Differences

Remember how we defined bn = an+1 − an from (an )? Now let f (x) = xk + αk−1 xk−1 +
· · · + α0 be a polynomial, and consider an = f (n). Define

f1 (x) = f (x + 1) − f (x), ..., fm+1 (x) = fm (x + 1) − fm (x).

We already know that the degree of these polynomials decreases, i.e. deg f > deg f1 >
deg f2 > . . . 4 . We can actually say more.

Theorem 2.8
Using the Binomial Theorem and considering the array generated by an = nk and
by an = αk−1 nk−1 + · · · + α0 , prove that deg f1 = deg f − 1. Next, prove that
deg fm = deg f − m (2 ≤ m ≤ k).

In fact, if the leading term of f (x) is xk , then the leading term of f1 (x) is kxk−1 , the
leading term of f2 (x) is k(k − 1)xk−2 , and so on. By the time we get to fk+1 (x), every
term will have vanished, so fk+1 (x) is the zero polynomial. Similarly, fk (x) is a constant,
and it is equal to its leading term

k(k − 1)(k − 2) · . . . · 1 = k! .

4
Recall how we made the zero polynomial have degree −∞; just pretend that −∞ > −∞.

6
Kada Williams (June 15, 2017) Divided Differences

§2.4 Some problems

Problem 2.9. Let f (x) be a polynomial, and suppose that fk+1 (x) ≡ 0, i.e. fk+1 is
formally equal to the zero polynomial. What can we say about f (x)?

Problem 2.10. Invoke the Pascal triangle to find that actually,

  `  
` X n
f` (x) = f (x + `) − f (x + ` − 1) ± · · · + (−1)` f (x) = (−1)n−j f (x + j).
1 j
j=0

Problem 2.11. Let an = 2n . What are the difference sequences? What about an = λn ?

Problem 2.12. Define the Fibonacci sequence by F0 = F1 = 1, Fn+1 = Fn + Fn−1

(n ≥ 1). For m ≤ n, simplify
m  
X m
(−1)m−j Fn+j .
j
j=0

Problem 2.13. Let p be an odd prime number. Prove that Fn+p − Fn − Fn−p is a
multiple of p.

Problem 2.14. Let f (x) be any polynomial with integer coefficients, and let p be a
prime number. Prove that every coefficient of fm (x) is divisible by p if m ≥ p.

Problem 2.15. The difference operator ∆h creates new functions from functions: from
f (x), it creates ∆h f (x) = f (x + h) − f (x). Prove the following:

(a) If ∆nh is ∆h applied n times,

n  
n−j n
X
∆nh f (x) = (−1) f (x + jh).
j
j=0

(b) Difference operators commute, i.e. ∆a (∆b f (x)) = ∆b (∆a f (x)), and we have the
product rule

∆h (f (x)g(x)) = f (x + h) · ∆h g(x) + g(x) · ∆h f (x).

(c) If f (x) is a polynomial of degree d > 0, then ∆h f (x) is a polynomial of degree

d − 1.

Problem 2.16. Research project: suppose that (an ) is periodic with period m. Then
we can encode it as P (x) = a0 + a1 x + · · · + am−1 xm−1 . Look at the difference sequences
of (an ) modulo 2, or maybe modulo some odd prime p. What’s the pattern? Prove stuff.

7
Kada Williams (June 15, 2017) Divided Differences

§3 Divided Differences
§3.1 Introduction and Motivation
Consider the following identities:
(x + 1)2 − x2 = 2x + 1
(x + a)2 − x2 = 2ax + a2
(x + a)3 − x3 = 3ax2 + 3a2 x + a3
which are essentially ugly, and also consider the identities
y 2 − x2 = (y − x)(y + x)
y 3 − x3 = (y − x)(y 2 + yx + x2 )
 
1 1 1
− = (y − x) −
y x xy
which are fairly nice.
These tell us, for example, that for polynomials P , the expression P (x + a) − P (x) is
messier to work with than P (y)−P
y−x
(x)
, in particular because
y k − xk
= xk−1 y + xk−2 y 2 + · · · + y k−1
y−x
is a polynomial expression – not in 1 variable, but in 2 variables, instead. 5

ak xk , we can
P
Just as how a polynomial in one variable is an expression of the form
talk about multivariate polynomials such as y 2 + x and x1 + x2 − 2x53 . A polynomial
in variables x1 , . . . , xn is defined as an expression which is a finite sum of so-called
monomials, i.e. a finite sum of expressions of the form Cd1 ,d2 ,...,dn xd11 xd22 . . . xdnn . The
degree of such a monomial is d0 +d1 +· · ·+dn , and the degree of a multivariate polynomial
is the maximal degree of its monomials.6

Example 3.1
We can extend this to 3 variables:
y 3 − x3
= y 2 + yx + x2 ,
y−x
y 3 −x3 z 3 −x3
y−x − z−x (y 2 + yx + x2 ) − (z 2 + zx + x2 )
=
y−z y−z
2
y −z 2 x(y − z) x2 (1 − 1)
= + + = x + y + z.
y−z y−z y−z

§3.2 Definition
f (x)−f (y)
Start from a function f (x), and create the function g(x, y) = x−y 6 y). Now
(x =
repeat this step by constructing
g(x, y) − g(x, z)
h(x, y, z) = (y 6= z),
y−z
5
Strictly speaking, we didn’t define division by polynomials, so here it is: if p(x) = q(x)r(x) and
q(x) 6≡ 0, then we define r(x) = p(x)
q(x)
.
6
In a fashion akin to our skirmish with formal series, one can define the sum and product of multivariate
polynomials, but I omit this, because it is intuitively obvious.

8
Kada Williams (June 15, 2017) Divided Differences

which is just doing the same thing, as if x weren’t even there.

Example 3.2
Calculate that
f (x) f (y) f (z)
h(x, y, z) = + + .
(x − y)(x − z) (y − x)(y − z) (z − x)(z − y)

Starting from a function f (x), we can extend this process to n+1 variables x0 , x1 , . . . , xn
by defining

f (x0 ) − f (x1 )
f1 (x0 , x1 ) := ,
x0 − x1
fm−1 (x0 , . . . , xm−2 , xm−1 ) − fm−1 (x0 , . . . , xm−2 , xm )
fm (x0 , x1 , . . . , xm ) := .
xm−1 − xm

In this case, we call

f [x0 , x1 , . . . , xn ] := fn (x0 , . . . , xn )
the n-th order divided difference of f .

Problem 3.3. Prove that

n
X f (xj )
f [x0 , x1 , . . . , xn ] = .
(xj − x0 ) . . . (xj − xj−1 )(xj − xj+1 ) . . . (xj − xn )
j=0

In particular, note that f [x0 , x1 , . . . , xn ] is symmetric: it doesn’t depend on the order of

x0 , x1 , . . . , xn .

Problem 3.4. Prove that if f (x) is a polynomial of degree N ≥ n, then f [x0 , x1 , . . . , xn ]

is a multivariate polynomial of degree N − n. 7

Problem 3.5. Let x, y, z be integers. Prove that

x4 y4 z4
+ +
(x − y)(x − z) (y − x)(y − z) (z − x)(z − y)

is also an integer.

Problem 3.6. Prove that in fact, if f (x) = xN , then for n < N , 8

X
f [x0 , x1 , . . . , xn ] = xd00 xd11 . . . xdnn .
d0 +d1 +···+dm =N −n
d0 ,d1 ,...,dm ≥0

Problem 3.7. Prove that if f (x) is a polynomial of degree N , then f [x0 , x1 , . . . , xN ]

is a constant equal to the leading coefficient of f , and that f [x0 , x1 , . . . , xn ] is the zero
polynomial if n > N .

7
Unrelated but interesting: Lexicographical order.
8
This big ugly sum means that we add up the monomial xd00 xd11 . . . xdnn for any possible (n + 1)-tuple of
nonnegative integers d0 , . . . , dn whose sum is N − n.

9
Kada Williams (June 15, 2017) Divided Differences

§3.3 Application: Roots of Multivariate Polynomials

Let K = R, and call K[x] the set of polynomials with coefficients in K. 9

Theorem 3.8
If f (x) = xd + ad−1 xd−1 + · · · + a0 ∈ K[x], then f has at most d distinct roots, i.e.
there are at most d distinct numbers x ∈ K for which f (x) = 0.

Proof. Note that

d
X f (xj )
f [x0 , x1 , . . . , xd ] = ≡1
(xj − x0 ) . . . (xj − xj−1 )(xj − xj+1 ) . . . (xj − xd )
j=0

due to Problem 3.7. But if x0 , x1 , . . . , xd ∈ K are pairwise different, and f (xj ) = 0 for
each j = 0, 1, . . . , d, then f [x0 , x1 , . . . , xd ] = 0. Since 0 6= 1, this shows that f cannot
have more than d distinct roots.

Now bear with me while we generalize this:

Theorem 3.9
Let f (x, y) be a 2-variable polynomial with coefficients from K. Let its degree be d,
in which case it must have some monomial Cxk y ` with C 6= 0 and k + ` = d.
Then there cannot be sets S, T ⊂ K with |S| = k + 1 and |T | = ` + 1 such that
for any x ∈ S and y ∈ T , f (x, y) = 0.

Proof. Recall how when defining divided differences, we took some function g(Y, X0 , X1 , . . . )
and created the new function g(Y,X0 ,X1 ,...Y)−g(Z,X
−Z
0 ,X1 ,... )
. This involved caring about one
variable and leaving the rest alone.
Since f (x, y) has two variables, we’re going to do the same process as before twice.
First, we leave y alone, and repeat taking differences, resulting in the function
k
X 1
g(x0 , x1 , . . . , xk , y) = ·f (xi , y).
(xi − x0 ) . . . (xi − xk )
i=0 | {z }
=:Ai

Second, we leave x0 , x1 , . . . , xk alone, and repeat taking differences, which yields

`
X 1
h(x0 , x1 , . . . , xk , y0 , y1 , . . . , y` ) = g(x0 , . . . , xk , yj )
(y − y0 ) . . . (yj − y` )
j=0 | j {z }
=:Bj
` k
!
X X
= Bj (y0 , . . . , y` ) Ai (x0 , . . . , xk )f (xi , yj ) .
j=0 i=0

Note that f (x, y) is the sum of monomials of the form Axκ y λ . Just like how with finite
differences, the array belonging to (an + a0n ) was the sum of the array belonging to (an )
9
Everything here works for any field K. For example, taking K = Fp in Theorem 3.8 gives Lagrange’s
theorem.

10
Kada Williams (June 15, 2017) Divided Differences

and the array belonging to (a0n ), here too, the function h(x0 , . . . , y` ) generated by f (x, y)
is just the sum of the h’s we get when applying the same process to each monomial
Axκ y λ in the expansion of f (x, y).
Now look at what happens if we replace f (x, y) by xκ y λ in the definition of h. Then h
factors!
` k
!  `  X k 
X X X
κ λ λ κ
Bj Ai x i y j = Bj yj · Ai x i .
j=0 i=0 j=0 i=0

Here, the first factor is the `-th order divided difference of y λ , while the second is the
k-th order divided difference of xκ . Hence, if [κ < k or λ < `], then this product is ≡ 0,
and if [κ = k and λ = `], then this product is 1 · 1 = 1.
Returning to f and h, we thus have h(x0 , . . . , y` ) ≡ C 6= 0. However, if we could have
some S = {x0 , . . . , xk } and T = {y0 , . . . , y` } with f (xi , yj ) = 0 (xi ∈ S, yj ∈ T ), then
just as in Theorem 3.8, we’d get h(x0 , . . . , y` ) = 0, a contradiction.

This method repeats itself in a slightly different form in the key lemma for problem 6
of IMO 2007:

Theorem 3.10
Suppose P (x, y, z) is a polynomial which satisfies P (x, y, z) = 0 if x, y, z ∈ {0, 1, . . . , n}
and (x, y, z) 6= (0, 0, 0). If P (0, 0, 0) 6= 0, then deg P ≥ 3n.

Proof. We do the same thing, except now {x0 , . . . , xn } = {0, 1, . . . , n}, so mere differences
will suffice, actually. To any polynomial Q(x, y, z), assign the real number
 !
n   X n   X n  
X n  n n
hQ := (−1)n−x (−1)n−y (−1)n−z Q(x, y, z)  .
x y z
x=0 y=0 z=0

Again, if Q(x, y, z) = Axα y β z γ , then hQ factors into the finite differences of xα , y β , z γ .

Hence, if one of α, β, γ is less than n, then hQ = 0.
When we split some Q into a sum of polynomials, hQ will be the sum of the corre-
sponding h’s. This means that deg Q ≤ 3n − 1 implies hQ = 0, because then Q only
contains monomials Axα y β z γ where one of the exponents is < n.
Since hP = (−1)n n0 (−1)n n0 (−1)n n0 P (0, 0, 0) 6= 0, this proves that deg P ≤ 3n − 1

is impossible, so deg P ≥ 3n.

The unusual thing here was that we didn’t need to evaluate hQ for xn y n z n . That made
the proof a bit simpler. If we wanted to make the proof look more like Theorem 3.9, we
should have instead taken

P ∗ (x, y, z) = P (x, y, z) − k [(x − 1) . . . (x − n)(y − 1) . . . (y − n)(z − 1) . . . (z − n)]

where k 6= 0 is chosen such that P ∗ (x, y, z) = 0 for any x, y, z ∈ {0, 1, . . . , n}. Then a
version of Theorem 3.9 would have proven that P ∗ (x, y, z) cannot have degree 3n and a
monomial Cxn y n z n with C 6= 0 at the same time, which leads to a contradiction.
The general version of Theorem 3.9 is called the Combinatorial Nullstellensatz:

11
Kada Williams (June 15, 2017) Divided Differences

Theorem 3.11
Let f (x1 , x2 , . . . , xn ) be a polynomial with coefficients from K. Let Cxd11 xd22 . . . xdnn
be a monomial appearing in f which has maximal degree (C 6= 0). Then we cannot
have sets S1 , S2 , . . . , Sn with |Si | = di + 1 for which

f (x1 , x2 , . . . , xn ) = 0 whenever x1 ∈ S1 , x2 ∈ S2 , . . . , xn ∈ Sn .

Proof. The same as Theorem 3.9. Repeatedly take divided differences with respect to each
variable of Q(x1 , x2 , . . . , xn ), getting a polynomial hQ in (d1 + 1) + (d2 + 1) + · · · + (dn + 1)
variables. On one hand, the divided difference of a monomial factors, from which we can
compute that hf ≡ C 6= 0. On the other hand, if such sets S1 , S2 , . . . , Sn did exist, then
plugging in the elements of the sets into hf would make hf = 0, contradiction.

§4 More Problems
§4.1 Difference Sequences
Problem 4.1. Define an = f (n), where
 
x x(x − 1) . . . (x − k + 1)
f (x) = := .
k k!

(a) Find the difference sequences of an .

x x



(b) Prove that every polynomial can be expressed in the form ck k +ck−1 k−1 +· · ·+c0
for some reals ck , ck−1 , . . . , c0 . 10

(c) Prove that f (x) takes on integer values whenever x is an integer if and only if
ck , ck−1 , . . . , c0 are all integers.

(d) Give another proof that when starting from f (x) = xk , we’ll have fk (x) = k!.

Problem 4.2. Let ∆h f (x) = f (x + h) − f (x) be the difference operator. Let s2 (n)
denote the sum of digits of n when written in binary11 . After writing down a few special
cases, prove that
k −1
2X
∆2k−1 ∆2k−2 . . . ∆1 f (x) = (−1)s2 (j) f (x + (2k − j)).
j=0

Problem 4.3. Fix a positive integer k. Prove that for some N , we can partition the
numbers 1k , 2k , . . . , N k into two groups of equal sums.

Problem 4.4. Let a ≥ 3, and let P be a polynomial whose degree is < n. Prove that
we cannot have
δk := ak − P (k) ∈ (−1; 1)
for each k = 0, 1, . . . , n.

10 x x



Higher math would say that 1, 1
, 2
, . . . forms a basis of the vector space R[x] over R.
11
See the Thue-Morse sequence.

12
Kada Williams (June 15, 2017) Divided Differences

Problem 4.5. If you know complex numbers, let ε = cis 2π`

m for some ` ∈ {0, 1, . . . , m−1},
and define the operator Dh by

Dh f (x) = f (x) + εf (x + h) + ε2 f (x + 2h) + · · · + εm−1 f (x + (m − 1)h).

Using Dh , prove that for some N , we can partition the numbers 1k , 2k , . . . , N k into m
groups of equal sums. (If you feel like it, try to also prove this without complex numbers.)

Problem 4.6 (Iberoamerican 2011). Prove that, for any given positive integers m ≤ `,
there is a positive integer n and pairwise distinct positive integers x1 , · · · , xn , y1 , · · · , yn
such that the equality
Xn Xn
xki = yik
i=1 i=1

holds for every k = 1, 2, . . . , m − 1, m + 1, . . . , `, but does not hold for k = m.

§4.2 Divided Differences

Problem 4.7. Check that the finite difference nj=0 (−1)n−j n
P

j f (x + j) of f is n! times
the divided difference f [x, x + 1, x + 2, . . . , x + n].

Problem 4.8. Let f be a polynomial of degree d, and let n > d. Using the Lagrange
Interpolation Formula
n
X (x − x0 ) . . . (x − xk−1 )(x − xk+1 ) . . . (x − xn )
f (x) = f (xk ) ,
(xk − x0 )(xk − x1 ) . . . (xk − xk−1 )(xk − xk+1 ) . . . (x − xn )
k=0

give another proof that f [x0 , x1 , . . . , xn ] is always equal to zero12 .

Problem 4.9 (Based on Komal A.652.). (a) Prove that if f (x) = x1 , then

(−1)n
f [x0 , x1 , . . . , xn ] = .
x0 x1 . . . xn

(b) Prove that if positive integers b0 < b1 < · · · < bn form an arithmetic progression,
then
b0 b1 . . . bn
lcm(b0 , b1 , . . . , bn ) ≥ .
n!
1
(c) Compute the divided differences when f (x) = x2
.

Problem 4.10. Do there exist n periodic functions13 fi : R → R such that for all x ∈ R,

xn = f1 (x) + f2 (x) + · · · + fn (x) ?

Problem 4.11. Let S be a set of odd integers. Prove that P∞ there exists a sequence of
reals (xn )∞
n=1 such that for every odd positive integer k, x
n=1 n
k converges if and only

if k ∈ S.
12
This is not necessarily the same thing as f [x0 , x1 , . . . , xn ] as an n-variate polynomial being identically
zero, it only means that f always evaluates to = 0. But the Combinatorial Nullstellensatz proves that
the two are in fact equivalent.
13
By the way, each linear function is the sum of 2 periodic functions. I’ve heard that this generalizes; if
you’re interested, investigate how many periodic functions we need so that their sum could be xn .

13
Kada Williams (June 15, 2017) Divided Differences

Problem 4.12 (Miklos Schweitzer 2014). Let n ≥ 1 be a fixed integer. Find the distance

inf max |f (x) − p(x)|,

p,f 0≤x≤1

where p runs over polynomials

P∞ with real coefficients of degree less than n and f runs
k
over functions f (x) = k=n ck x defined on the closed interval [0, 1], where ck ≥ 0 and
f (1) = 1.

If you’d like even more problems, take a look at web.evanchen.cc/handouts/BMC_

Combo_Null/BMC_Combo_Null.pdf, or at the example problems in Problems From The
Book by Andreescu and Dospinescu, Chapter 11 (Lagrange Interpolation Formula).
Thanks for reading!

14