You are on page 1of 9

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan 1 Arithmetico-Geometric Sequences

An arithmetico-geometric sequence is one which has the form a0 , a1 r, a2 r2 , a3 r3 , . . . , where the sequence {ai } is arithmetic. Sequences like this can be summed using a similar method to the one for geometric series. Problem. What is 5 2n 1 3 1 ? + 2 + 3 + + 2 2 2 2n

Solution. Let S be the sum. Then we have 2S = 1 + Subtracting the original sum, S =1+ 2 2 2 2 2n 1 + + 3 + + n1 . 2 22 2 2 2n 3 5 2n 1 + 2 + + n1 . 2 2 2

Using the usual formula for a geometric series, this is S 2n 1 1 (1/2)n1 1/2 2n 2n + 3 = 3 . 2n = 1+

Telescoping

We illustrate telescoping with an example. Problem. What is 1 ? n(n + 1) n=1 Solution. Call the sum S . To avoid problems with convergence, we use telescoping to compute partial sums rather than the full sum. The pth partial sum Sp is given by

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan

Sp

= =

1 n ( n + 1) n=1
p

n=1 p

1 1 n n+1 1 n n=2
p+1

= =

1 n n=1 1

1 . p+1

By denition, the sum S of the original series is the limit of Sp as p goes to innity. Clearly this is 1. More generally, if you need to determine 0 f (n), and you can nd a function g such that f (n) = g (n) g (n + 1), you can telescope the sum to get g (0) g (p + 1). This technique is especially useful with rational functions like the one in the example, because you can decompose a rational function into a sum or dierence of other rational functions. Sometimes it also helps with trigonometric sums. Problem. What is
90 p

n sin(2n )?
n=1

Solution. Let S be our sum. Since theres no obvious way to break the summand into a dierence of f (n) and f (n + 1), we try multiplying it by sin(1 ), which will allow us to use the product-to-sum formula to create a dierence in the summand. We have
90

S sin(1 ) S sin(1 ) 2S sin(1 )

=
n=1 90

n sin(1 ) sin(2n ) 1 n[cos((2n 1) ) cos((2n + 1) )] 2 n=1


90

=
n=1

n[cos((2n 1) ) cos((2n + 1) )]

Now the only thing stopping us from telescoping is the factor of n in the summand. Rewriting the summation may make it more clear how to proceed:

2S sin(1 ) = (cos 1 cos 3 ) + 2(cos 3 cos 5 ) + + 89(cos 177 cos 179 ) + 90(cos 179 cos 181 )

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan
Using cos(x) = cos(180 x), we notice that we can group the n = 1 term with the n = 89 term, the n = 2 term with the n = 88 term, and so on, which gives us

2S sin(1 ) 2S sin(1 ) 2S sin(1 ) S

= = = =

90[(cos 1 cos 3 ) + + (cos 87 cos 89 )] + 45(cos 89 cos 91 ) + 90(cos 179 cos 181 ) 90(cos 1 cos 89 ) + 90 cos 89 90 cos 1 45 cot 1

Linear Recurrences

A sequence is given by a recurrence if each term is dened in terms of the previous ones. A particularly common type of recurrence is a linear recurrence. A linear recurrence is a sequence x1 , x2 , x3 , in which every term after the k th is given by xn = a1 xn1 + a2 xn2 + + ak xnk . The rst k terms must be specied explicitly to dene the full sequence. Often we want to solve a linear recurrence, which means to nd an explicit formula for the terms. The key to doing this is to guess that the answer is yn = n , where is an unknown constant. Our recurrence then becomes n = a1 n1 + a2 n2 + + ak nk , which simplies to k a1 k1 a2 k2 ak = 0. This polynomial is known as the characteristic polynomial of the recurrence. If is any root of this polynomial, then the sequence yn = n satises the recurrence. In fact, if the roots are distinct values 1 , 2 , . . . , k , the general solution to the recurrence is
n n xn = c1 n 1 + c2 2 + + ck k .

So to solve the original recurrence, given x1 , x2 , . . . , xk , we just need to solve for the proper ci s. There is an extension to the formula for the general solution which applies when the characteristic polynomial has repeated roots. If a root has multiplicity m, then its contribution to the general solution is (d0 + d1 n + d2 n2 + + dm nm )n

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan
rather than merely cn . Problem. Solve the recurrence xn+1 = 7xn 12xn1 if x0 = 2 and x1 = 7. Solution. The characteristic polynomial here is 2 7 + 12 = 0, which factors to ( 3)( 4) = 0. Thus the general solution is xn = c1 3n + c2 4n . Clearly xn = 3n + 4n matches the initial conditions x0 = 2 and x1 = 7, so that is the answer. Problem. How many ways are there to tile a 2 n board using only rectangles of size 2 2 and 2 1? Solution. Assume the horizontal dimension is n, and let xn be the number of tilings. We can split possible tilings into three cases according to how we tile the left end of the board: Use a 2 2 rectangle. Place a 2 1 rectangle vertically. Place two 2 1 rectangles horizontally. These cases contribute xn2 , xn1 , and xn2 congurations, respectively, to the overall total. Therefore, we have the recurrence xn = xn1 + 2xn2 . The characteristic equation is 2 2 = 0, which has solutions 2 and 1, so xn must be of the form c1 (2n ) + c2 (1)n . We can use some small cases (for example x1 = 1 and x2 = 3) to solve for c1 and c2 , and we end up with xn = 2 n 1 2 + (1)n . 3 3

Problem. Solve the recurrence a0 = 1, a1 = 6, an+2 = 6an+1 9an . Solution. This time the characteristic equation is 2 6 + 9 = ( 3)2 = 0, which has a double root of 3. So the general solution is (d0 + d1 n)(3n ). From the initial conditions, we can solve for d0 and d1 , and we nd that d0 = d1 = 1. So an = 3n (n + 1).

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan 4 Inhomogeneous Recurrences

The recurrences of the previous section were homogeneous, meaning that each term in the denition of xn was a multiple of xni for some integer i. An inhomogeneous linear recurrence is one which has the form xn = a1 xn1 + a2 xn2 + + ak xnk + f (n) (again for n > k ). Suppose that the sequences {yn } and {zn } both satisfy this recurrence. Then by plugging each sequence into the recurrence and subtracting, we can prove that their dierence, wn = yn zn , satises the recurrence xn = a1 xn1 + a2 xn2 + + ak xnk , the homogeneous version of the original recurrence. This means that if we can nd one solution {pn } of the original recurrence, a so-called particular solution, every other solution will have the form
n n xn = pn + c1 n 1 + c2 2 + + ck k ,

where the cs are arbitrary constants and the s are the roots of the characteristic equation (just as in the homogeneous case). The remaining piece of the puzzle is how to nd a particular solution. There is a general rule of thumb which applies when f (n) is a polynomial, and that is that the particular solution is also a polynomial. (To be more precise, one of the many possible particular solutions is a polynomial.) You can nd the polynomial by guessing its degree (try low numbers rst) and then substituting an arbitrary polynomial of that degree into the recurrence, which should let you solve for the coecients. (Note: the particular solution you nd will not in general satisfy the initial conditions; the initial conditions let you solve for the coecients of the n terms.) Problem. What is the (2n)th term in the sequence 1, 2, 4, 5, 10, 11, 22, 23, 46, 47, . . . ? This sequence is simple enough that you can probably guess the answer without recurrences. Take a moment to try. Now, suppose you didnt see it. Can we nd the answer systematically? Solution. Let an be the (2n)th term of the original sequence. Then a1 = 2, and we have the recurrence an+1 = 2an + 1. The corresponding homogeneous recurrence is an+1 = 2an ,

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan
which has the solutions an = c(2n ). Now we must nd a particular solution to the recurrence. It will be a polynomial, so we try the simplest option, a constant. If an = k , we have k = 2k + 1, so k = 1. Thus a particular solution is 1, and the general solution is an = c(2n ) 1. Since a1 = 2, c = 3/2, and our nal answer is an = 3 n (2 ) 1 = 3(2n1 ) 1. 2

Problem. Solve the recurrence a0 = 0, an+1 = 2an + n. Solution. As in the last problem, the solution to the corresponding homogeneous recurrence is would be c(2n ). To nd a particular solution, lets again try setting an = k , where k is a constant. Substituting into the recurrence gives k = 2k + n, which clearly cannot be satised for all n. We should therefore try a higher-degree polynomial. Suppose an = bn + d. Then our recurrence becomes b(n + 1) + d = 2(bn + d) + n, or bn + n + d b = 0. Since this equation must hold for all n, we have b = d = 1, and our particular solution is n 1. The general solution is c(2n ) n 1, and from a0 = 0 we quickly nd that c = 1. So an = 2n n 1.

Finite dierences

Given a sequence that is generated by a polynomial, its worth knowing how to extract that polynomial. We will illustrate with an example. Problem. Find the (simplest) polynomial sequence an such that a0 = 4, a1 = 10, a2 = 26, a3 = 58, a4 = 112, a5 = 194. Solution. We write the given entries in a row and then compute the dierences of successive terms, writing those values in the next row. We repeat the process until we reach a row of zeros. The resulting table contains the nite dierences of the sequence: 4 6 10 6 0 10 16 16 6 0 26 32 22 6 ... 58 112 54 82 28 . . . ... 194 ... ...

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan
We read o the rst entry in each row to form a new sequence {bn } = 4, 6, 10, 6. It turns out that the formula for the sequence is given by b0 which in our case is 4 n n n + b1 + b2 + , 0 1 2 n n n n +6 + 10 +6 . 0 1 2 3

Expanding out the combinations will lead to the formula an = n3 + 2n2 + 3n + 4. Note that we made an assumption that the fth row consisted entirely of 0s. If this were not true, the polynomial would have more terms. Since the problem asked for the polynomial of lowest degree that ts the given numbers, we stop at the fth row. We will not prove that the method given above works, but it is not too hard to do on your own. Notice that every entry in the top row can be expressed as a linear combination of the bi s, where the coecients of the linear combination vary with n but are xed for all sequences. If you try to work out these coecients, you will quickly see Pascals triangle emerging, which is why the combinations arise.

Other problems

Many problems involve sequences and series that dont fall into any of the usual categories (arithmetic, geometric, arithmeto-geometric, telescoping, recursive), and thus require some creativity. Here is an example. Problem. In a nite sequence of real numbers the sum of any seven successive terms is negative, and the sum of any eleven successive terms is positive. Determine the maximum number of terms in the sequence. (1977 IMO) Solution. The maximum is 16. Suppose the sequence a1 , a2 , . . . , a17 worked. Then we could form the following 7 11 table: a1 a2 . . . a7 a2 a3 . . . a8 ... ... .. . ... a11 a12 . . . a17

The columns would have to sum to negative numbers, whereas the rows would have to sum to positive numbers, a contradiction. One can nd a sequence of length 16 which works by trial and error, or by assuming that the groups of 11 sum to (say) 1 and the groups of 7 sum to 1, and then solving the resulting system of linear equations. An example of a valid sequence is 5, 5, 13, 5, 5, 5, 13, 5, 5, 13, 5, 5, 5, 13, 5, 5.

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan 7 Exercises
1. Find a closed-form expression for 1(1!) + 2(2!) + 3(3!) + + n(n!). 2. Let a, b be integers. Dene a sequence a0 , a1 , . . . of integers by a0 = a, a1 = b, a2 = 2b a + 2, an+3 = 3an+2 3an+1 + an . Determine all pairs (a, b) for which an is a perfect square for all n 1998. (Vietnam 1998) 3. Solve this recurrence: a0 = 0, an+1 = 1 +
i=0 n1

ai

for n 0.

4. How many strings of 1s and 0s having length n do not contain two consecutive 0s? 5. What is 6. What is 7. Evaluate
n=3 1 n=1 (2n1)(2n+1) ? 1 n=1 n(n+1)(n+2) ?

cot1

n2 2

8. The sequence {an } is dened by a1 = a2 = a3 = 1 and an+1 = Find a formula for an . 9. If xn+1 =
xn 1+nxn ,

1 + an an1 . an2

nd an expression for xn in terms of x0 .

10. Let c be a positive integer. The sequence {fn } is dened as follows: f1 = 1, f2 = c, fn+1 = 2fn fn1 + 2 (n 2).

Show that for each positive integer k there exists a positive integer r such that fk fk+1 = fr . (1984 IMO shortlist) 11. A sequence {an } is dened by means of the recursion a1 = 1, an+1 = 1 + 4an + 1 + 24an . 16

Find an explicit formula for an . (1981 IMO shortlist)

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

Art of Problem Solving

WOOT 2008-09 Sequences & Series


Sean Markan
12. Let n1 < n2 < be a sequence of natural numbers such that for i < j , the decimal representation of ni does not occur as the leftmost digits of the decimal representation of nj . (For example, 137 and 13729 cannot both occur in the sequence.) Prove that

i=1

1 1 1 1 + + + . ni 2 9

(Iran 1998) 13. Prove that the sequence {an } dened by a1 = 1 and an = an1 + a
n/2

n = 2, 3, 4, . . .

contains innitely many integers divisible by 7. (Poland 1998)

Worldwide Online Olympiad Training


www.artofproblemsolving.com Sponsored by D. E. Shaw group and Jane Street Capital

You might also like