Brilliant Math Circle

Chebyshev polynomials via Trigonometric

definition

April 22, 2015

Ground rules (For Circles):

I’m trying to keep the chat orderly, and that would mean minimizing inter-
ruptions from everyone else. Here are some ground rules to obey:
1. If you have a comment / question, please send me a Direct message in-
stead.

2. If I ask a question, there are the following scenarios

A) DMA means ”DM me your answer”. Replies in channel will likely be deleted.
B) FFA means ”Free for all”. Anyone can answer in the channel.
C) @name means that person (alone) should answer in the channel.
D) Yes/No questions - Simply reply Yes/No.

3. When asking a question, I will sometimes say ”(pause)” to indicate that

we will pause for some time to answer the question. This is equivalent to ”Daniel
Spends Time Thinking”.

4. I will not be able to respond to all DM’s directly or indirectly. If you

have a burning question/issue, keep on DMing me.

5. Unfortunately, I have not gotten Latex set up as yet. I will mainly be

sending screenshots where possible (esp for prepared notes), but otherwise we
will just use Latex syntax to communicate.

Note: Failure to obey these may result in your messages being deleted, or
deportation to the holding-cell.

Questions that I did not cover during the Circle are denoted with a (**).
You should solve them too.

1
1 Introduction
The Chebyshev polynomials (of the first kind) are defined as

Tn (x) = cos(n arccos x), Tn (cos θ) = cos nθ.

Since we know that

cos 0θ = 1, cos θ = cos θ, cos 2θ = 2 cos2 θ − 1, cos 3θ = 4 cos3 θ − 3 cos θ,

hence we can conclude that

T0 (x) = 1, T1 (x) = x, T2 (x) = 2x2 − 1, T3 (x) = 4x3 − 3x.

Y/N: Do you understand how to obtain these formulas? We essentially re-

place cos θ with x.

FFA: Find T4 (x).

FFA: Find T5 (x).

FFA: Find T6 (x).

(Pause) FFA: How do we obtain a more general formula?

In answering the previous question, most people tried to expand

cos(n + 1)θ = cos nθ cos θ − sin nθ sin θ.

We can easily convert the first 2 terms into the Tn form. However, the issue
with this approach is that sin nθ sin θ is not easy to deal with, and will (cur-
rently) require much further expansion.

Instead, we use the fact (from trigonometric sum and product formulas) that
cos(n + 1)θ + cos(n − 1)θ = 2 cos θ cos nθ.
DMA: What is the recurrence relation that the above equation states?

Ans: Tn+1 (x) = 2xTn (x) − Tn−1 (x).

(**) Prove the following statements:

1. Tn (−x) = (−1)n Tn (x)
2. T2n (0) = (−1)n
3. Tn (1) = 1
4. T2n+1 (0) = 0
5. Tn (−1) = (−1)n

2
2 Conjectures about the coefficients
The following is a table of initial values of Tn (x).

T0 (x) = 1
T1 (x) = x
T2 (x) = 2x2 − 1
T3 (x) = 4x3 − 3x
T4 (x) = 8x4 − 8x2 + 1
T5 (x) = 16x5 − 20x3 + 5x
T6 (x) = 32x6 − 48x4 + 18x2 − 1
T7 (x) = 64x7 − 112x5 + 56x3 − 7x
T8 (x) = 128x8 − 256x6 + 160x4 − 32x2 + 1
T9 (x) = 256x9 − 576x7 + 432x5 − 120x3 + 9x.
T10 (x) = 512x10 − 1280x8 + 1120x6 − 400x4 + 50x2 − 1.

(pause) FFA: Based on the above table, what conjectures can we make?

Here are some observations that people made.

1. Coefficients are integers.
2. Constant term is (−1)k for n = 2k and 0 for n = 2k + 1.
3. Leading term is 2n−1 .
4. Linear term of T2n+1 (x) is (−1)n (2n + 1).
5. The non-zero terms have exponents with the same parity as n
6. Sum of coefficients is 1
7. non-zero Coefficients alternate in sign

(pause) FFA: Can we prove these conjectures?

Let us prove these conjectures

1. This follows immediately by induction and using the recurrence relation.

Note that this should be slightly surprising to you , since it is not immedi-
ately obvious why cos(100θ) in terms of cos θ shouldn’t involve a fraction
in its coefficient. We now get this conclusion for ”free”.
2. The easy way to prove this is to remember that the constant term of a
polynomial is equivalent to evaluating it at 0. Plugging in x = 0 in the
recurrence, we get

Tn+1 (0) = 2 × 0 × Tn (0) − Tn−1 (0) = −Tn−1 (0).

The result follows immediately.

3
 3. This follows from induction, where the extra 2 comes from the 2xTn term.
4. The easy way to prove this is to remember that the linear term of a poly-
nomial is equivalent to evaluating the first derivative at 0. Differentiating
the recurrence relation, we get

0
Tn+1 (x) = 2xTn0 (x) + 2Tn (x) − Tn−1
0
(x).

Then, evaluating it at 0 gives us

0 0
Tn+1 (0) = 2Tn (0) − Tn−1 (0).

Alternative solution: Use conjecture 2 to track what happens to the linear

term.

5. This follows from induction on the recurrence relation.

6. The sum of the coefficients is equal to evaluating at 1. From the trigono-
metric defninition, we have Tn (1) = Tn (cos 0) = cos 0 × n = 1.
7. (**) This follows from induction on the recurrence relation.

Some other conjectures that were made. Look at these values and fill in the
blanks.

8. The coefficient of the x2 term has the form ...

9. The sum of absolute value of the coefficients is ...
10. The roots of Tn are ...
11. Fill in your own!

(**) Find more conjectures.

(**) Prove these additional conjectures.

4
3 Complexity of calculating Tn (x)
Let’s say that we want to calculate T1000 (x). How much work would that take?
One way that we could approach it, is to calculate what each of the Tn
values are for n < 1000, and slowly work our way up the list. Since Ti has
(about) 2i terms, we will need to make about 2i + 2i = i calculations to get Ti ,
meaning that we will need about n(n+1)2 calculations to get Tn (x). We say that
the complexity of this calculation is O(n2 ).

Is there a faster way?

Yes! Recall how computers would calculate large exponents like 21000 . They
do not calculate it as

2 × 2 × 2 × 2 × ... × 2

which would take 1000 calculations. The complexity of this process is O(n).
Instead, we would first write 1000 in base 2, to get 1000 = 11111010002 .
This means that

21000 = 2512 × 2256 × 2128 × 264 × 232 × 28

Then, we calculate these power of 2 via square them to get 21 , 22 , 24 , 28 , 216 , . . .

and then multiplying what we need. As such, we have performed much fewer
than 1000 calculations. The complexity of this process is O(ln n), which is much
better than O(n).

Let’s focus back on the Chebyshev polynomial. We would thus want to find
a way that would allow us to double up the indices, to avoid having to calculate
T999 , T998 , . . . T501 . How can we do this?

DMA: Prove that for m > n,

2Tm (x)Tn (x) = Tm+n (x) + Tm−n (x)

Hence, conclude that

T2n (x) = 2Tn2 (x) − 1

T2n+1 (x) = 2Tn+1 (x)Tn (x) − x

The above formulas thus give us a way to calculate T1000 from T500 , which is
obtained from T250 , which is obtained form T125 , etc. You can already see how
this saves the amount of work that needs to be done.
(**) Using the above approach, what is the complexity of calculating Tn ?
Can you do better?

5
4 Polynomial Interpolation
1. FFA In terms of cos A, what are the roots of T3 (x) = 0?
Ans: cos 30◦ , cos 150◦ , cos 270◦ .
2. FFA: In terms of Tn , find a polynomial whose roots are cos 120◦ , cos 240◦ , cos 360◦ .
Hint: From the above, the answer is not T3 (x) = 0.
Note: For those who evaluated the roots to find the polynomial, think
about how we can use T3 (x) directly.
3. DMA In terms of cos A, what are the roots of 2T4 (x) − 1 = 0?
Ans: cos 15◦ , cos 105◦ , cos 195◦ , cos 285◦ .
4. (**) In terms of Tn , find a polynomial whose roots are cos 10◦ , cos 100◦ , cos 190◦ , cos 280◦ .

5. (pause) DMA In terms of cos A, what are the roots of 4T52 (x) − 3 = 0?
6. (**) In terms of Tn , find a polynomial whose roots are cos(θ + 2n 5 π) for
i = 1 to 5.
Solution. Since cos (5 × (θ + 2n


5 π) = cos 5θ, it follows that the polyno-
mial is

T5 (x) = cos 5θ.

Generalization:
α+2k
The solutions to Tn (x) = cos α are n .
2π
The polynomial whose n roots are cos(θ + n ) would be Tn (x) = cos(nθ).

For the following questions, we will apply Vieta’s formula to solve the
problem.
7. (pause) Explain how to evaluate

cos 10◦ × cos 50◦ × cos 70◦ .

Hint: Evaluate

cos 10◦ × cos 130◦ × cos 250◦ .

Hint: Find a polynomial whose roots are cos 10◦ , cos 130◦ , cos 250◦ .

6
 8. (**) (pause) Evaluate, in terms of x,

4
X 2mπ
cos( + x)
m=0
5

Hint: Find a polynomial whose roots are cos( 2nπ

5 + x).

9. (**) (pause) Evaluate

1 1 1 1
+ + + .
cos 25◦ cos 115◦ cos 205◦ cos 295◦

10. (pause) In terms of Tn , find a polynomial whose roots are exactly

cos2 1◦ , cos2 3◦ , cos2 5◦ , . . . , cos2 89◦ .

Hence, evaluate

1 1 1 1
◦
× ◦
× ◦
× ... × .
2
cos 1 2
cos 3 2
cos 5 cos 89◦
2

Hint: The polynomial is (x − cos2 1◦ )(x − cos2 3◦ )(x − cos2 5◦ ) × . . . × (x −

cos2 89◦ ).

Hint: (y 2 − cos2 1◦ ) = (y − cos 1◦ )(y + cos 1◦ ).

Brief solution: First, we know that

T90 (x) = 289 (x − cos 1◦ )(x − cos 5◦ ) . . . (x − cos 357◦ )

Since cos 181◦ = − cos 1◦ , we get that (x − cos 1◦ )(x − cos 181◦ ) = (x2 − cos2 1◦ .
Since cos 357◦ = cos 3◦ and cos 177◦ = − cos 3◦ , we get that (x − cos 177◦ )(x −
cos 357◦ ) = (x2 − cos2 3◦ ).
Thus, this gives us that

T90 (x) = 2179 (x2 − cos2 1◦ )(x2 − cos2 3◦ ) . . . (x2 − cos2 89◦ ).
√
Hence, the polynomial that we are looking for is T90 ( x).
By Vietas, the product of the roots is equal to 2189 , and hence their inverse
is 289 .

7
5 Additional facts
This set of notes had a rather narrow focus, namely interpreting Chebyshev
polynomials through their Trigonometric Cosine definition, and then seeing how
we can apply them in the context of Vieta’s formula and Remainder Factor The-
orem.

In particular, we did not cover the following:

- Chebyshev polynomials of the second kind
- Orthogonality of these polynomials
- Generating function interpretation

Prove the following facts about Tn (x).

1. Prove that
√ √
(x − x2 − 1)n + (x + x2 − 1)n
Tn (x) = .
2

2. Prove that

(−2)n n! p dn n−1
Tn (x) = (1 − x2 ) n (1 − x2 ) 2
(2n)! dx

3. Turan’s inequalities:

Tn2 (x) > Tn−1 (x)Tn+1 (x), ∀ − 1 < x < 1

4. Generating function for Tn (x)

∞
1 − tx X
= Tn (x)tn
1 − 2tx + t2 n=1